This PDF file contains the front matter associated with SPIE Proceedings Volume 13091, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

The linear canonical wavelet transform (LCWT) is the generalization of the classical wavelet transform (WT) and the linear canonical transform (LCT). It has been proven to be a powerful mathematical tool and is widely used in signal processing, image processing, optics, and other fields. However, some basic results of this transform are not yet mature, such as convolution and correlation theorems. Therefore, this paper discusses the convolution and correlation theorems of the LCWT. Firstly, we review the basic theory of the WT, the LCT, and the LCWT. Secondly, we define the new convolution and correlation operators, and deduce the convolution and correlation theorems of the LCWT. The results show that they are similar in other joint space/spatial-frequency or time/frequency representations. Finally, we give the filter design method of the proposed convolution theorem in the LCWT domain, which provides us with more possibilities to consider performing spatially varying filtering operations in the LCWT domain.

The estimation of nonlinear systems has always been an interesting research topic. Meanwhile, heavy-tailed and skew non-Gaussian noise exists widely in the actual environment. At present, most of the measurement noise of linear systems is estimated by univariate skew-t distribution (UST) or multivariable skew-t distribution (MST). There is little research on non-Gaussian noise in linear systems where both process noise and measurement noise are subject to heavy-tailed and skew, and even less on nonlinear systems. In this paper, the heavy-tailed and skew non-Gaussian noise is modeled using the multivariate skew-t distribution. In addition, a hierarchical Gaussian state space model for stochastic uncertain systems is presented, and the system estimation problem of heavy-tailed and skew non-Gaussian noise is transformed into the estimation problem of hierarchical Gaussian state space model. And a robust Bayesian smoother based on variable Bayesian inference is proposed to approximate the system state and the measured unknown noise parameters. On this basis, the proposed algorithm is simulated through the target tracking scenario, and the simulation results are compared with the existing extended Rauch-Tung-Striebel smoother (ERTSS) to verify the effectiveness of the proposed algorithm. Finally, the advantages and disadvantages of the proposed algorithm and the possible development direction in the future are summarized.

Radar signal sorting is an important issue in radar electronic warfare, and its accuracy directly affects the identification of radar radiation sources and the effectiveness of electronic reconnaissance systems. This article proposes an efficient radar sorting algorithm. Firstly, the number of pulses in small intervals is counted, and the number of pulses in the interval is calculated. Then, the potential pulse interval is searched for, and the corresponding pulse string of the signal with the potential pulse interval is obtained using an improved sequence retrieval algorithm based on vector dot product. The interference between pulse strings with different PRIs (Pulse Repetition Intervals) is reduced by using the method of successive elimination. Simulation experiments show that the algorithm greatly reduces the calculation time while maintaining high accuracy in radar sorting.

This article proposes the unified short-time Wigner-Ville distribution (USWD) and explores its various properties, which is a generalized integral transform suitable for time-varying signals with varying features. Recognizing the complexity involved in this transformation and aiming for practical applications, we focused our research on a specific case, denoted as SWDL. In this study, we derived the Heisenberg’s uncertainty principle for SWDL and presented its discrete form. Furthermore, we investigated the potential applications of SWDL in the analysis of stepped-frequency linear frequency modulation (SF-LFM) signals.

In this paper, we propose a modified augmented Lagrange multiplier method to improve the estimation performance of the direction of arrival (DOA) of sparse arrays. We use the duality of the augmented Lagrangian multipliers to optimize the dual solution based on the residual term generated during the iteration process, and the Artificial Fish Swarm Algorithm (AFSA) is used to adaptively update the coefficient of the residual term, so as to improve the accuracy of the original solution of the matrix completion problem. Simulation results show that, this method has better DOA estimation performance compared to the traditional Augmented Lagrangian Method (ALM) and Singular Value Thresholding (SVT), and can be applied to coherent sources.

Generalized monopulse angle measurement technique is widely used in target angle estimation of digital multi-channel planar array active phased array in interference environment because of its simplicity and small amount of calculation. Due to the non-uniform, asymmetric and anisotropic characteristics of the conformal array, the conventional generalized monopulse angle measurement technique for planar array cannot be directly applied. As to this problem, this work proposes a generalized monopulse angle measurement method for conformal arrays. Firstly, based on the anisotropic element pattern model of conformal array, the subarray echo model of conformal array is established. Secondly, the conventional generalized monopulse angle measurement technique of planar array is modified for the conformal arrays, and the direct solution formulas of target azimuth and elevation angles are derived. Finally, the angle estimation accuracy of the proposed method is analyzed theoretically. The simulation results show that compared with the conventional generalized monopulse angle measurement method, the proposed method greatly improves the target angle estimation accuracy of the conformal array, and is consistent with the theoretical estimation accuracy. The method proposed in this work can effectively solve the problem of generalized monopulse angle measurement of conformal array in interference environment.

Modern battlefield environment is complex and changeable. In addition to having higher target resolutions, airborne radar should also have good radio frequency (RF) stealth performance such as large time-bandwidth product, low interception factor, low power spectrum density (PSD) amplitude, etc. There are the high distance grating lobes after pulse compression in the large frequency interval of Costas signal. After Costas is modulated by linear frequency modulation, we can obtain the inter-pulse frequency coded and intra-pulse linear frequency modulation (Costas-LFM) signal. Compared with Costas signal with large frequency interval, Costas-LFM signal eliminates distance grating lobes, obtain the large bandwidth, increases the complexity of the intra-pulse frequency modulation features and the difficulty of sorting and identification of the passive detection system. Simulations show that Costas-LFM signal is significantly improved in ambiguity function, autocorrelation function (ACF), PSD and interception factors. While realizing radar detection, the signal also has good RF stealth performance, which has a wide application prospect in modern war.

The linear canonical Stockwell transform (LCST) is an extension of the Stockwell transform (ST) and the linear canonical Fourier transform (LCT). It not only characterizes signals in the time-linear canonical frequency plane but also inherits the advantages of the Stockwell transform. This study aims to generalize LCST into a two-dimensional linear canonical Stockwell transform (2D LCST) in response to the widespread interest in 2D ST across various fields.

We begin by examining the fundamental aspects of the two-dimensional linear canonical Stockwell transform, including its definition, basic properties, and Parseval formula. Subsequently, we introduce and investigate a comprehensive reconstruction formula and an energy formula. As we approach the conclusion, we derive the convolution theorem and the cross-correlation theorem associated with the two-dimensional linear canonical Stockwell transform.

The synchroextracting transform is a time-frequency post-processing method that allows signal reconstruction to obtain a more focused time-frequency representation. This paper proposes a modified second-order synchroextracting transform (MSSET) time-frequency analysis method which can handle multi-component signals and strong frequency-modulation signals more effectively. Different from the conventional time-frequency analysis methods, this method pre-detects the target area through joint image processing, and then designs a new local optimal window width (LOWW) selection scheme to achieve the optimal effect in the two dimensions of time scale and frequency scale. Finally, the superiority of the method in instantaneous frequency (IF) estimation and the energy concentration of the time-frequency representation (TFR) is verified by simulation experiments.

Direction of arrival (DOA) estimation on uniform linear array with single snapshot has always been a hot topic in radar signal processing. The traditional subspace class estimation methods, such as multiple signal classification (MUSIC), and sparse recovery algorithms, such as subspace pursuit (SP) algorithm, are suitable for different situations, meanwhile have certain requirements for observation data and signal-to-noise ratio (SNR). Combining the advantages of these two methods, a novel MUSIC subspace pursuit (MSP) algorithm is proposed in this paper. Firstly, an error evaluation function of the residual signal is constructed. The angles measured by both methods are replaced and selected one by one. And then, the framework of the greedy algorithm is iteratively applied until convergence. In this algorithm, the subspace information of the signal is added to the greedy algorithm, and more conditions are used to improve the angle measurement accuracy with single snapshot. Simulation results illustrate that the proposed MSP algorithm outperforms the SP and MUSIC algorithms at different SNR levels and it keeps the faster reconstruction speed of SP algorithm.

Airborne radars usually face non-uniform clutter environments, and it is difficult to obtain enough independent and identical distributed (i.i.d.) training samples, which degrades the clutter suppression performance of space time adaptive processing (STAP). To address the problem, a log-determinant sparse recovery based STAP (LSR-STAP) method is proposed in this paper. Through the sparse recovery theory of low-rank matrices and the prior knowledge of clutter covariance matrices, the corresponding problem is modeled using the log-determinant (LogDet) approximation of the rank function, and the solution of the resulting nonconvex optimization problem is derived in the framework of the symmetric alternating direction method of multipliers (S-ADMM). The simulation results show the superiority of the proposed method over similar algorithms.

When observing high-speed maneuvering targets, the relative motion between the target and the radar will produce linear range walking, range bending, Doppler spread and other phenomena, resulting in the failure to obtain higher SNR gain by prolonging the coherent integration time. A coherent integration algorithm based on Wigner Ville Distribution (WVD) is proposed for target detection with uniform acceleration in any direction. The algorithm first uses the second-order Keystone transform and combines the linear detection algorithm to correct range walking and range bending, then introduces the delay variable and constructs a third-order matrix based on the traditional WVD algorithm to convert the echo signal to the time-frequency domain plane for third-order phase parameter estimation, and compensates the third-order phase coefficient, and then uses the LVD algorithm to estimate and compensate its second-order phase coefficient to achieve the correction of Doppler spread. The algorithm has excellent parameter estimation performance in a strong noise environment and has lower algorithm complexity compared with the traditional full-dimension search algorithm.

This article proposes and designs a new type of joint graph filter, which is based on the design of two commonly used general filters, that is, the vertex sampling method and the sum-of-squares integration method. A joint optimization of the graph filter is proposed to organically combine the two methods, and redesign the graph filter coefficients. This method can reflect the advantages of both design methods, thereby obtaining a more stable and smoother graph filter. The obtained filter can effectively control the noise amplitude and the ripple amplitude of the cutoff band of the graph filter in the initial phase of the response. This article demonstrates through theoretical analysis and numerical experiments that this method outperforms the basic method in designing the desired graph filter in terms of frequency response.

In order to improve the convergence speed, stability and state estimation accuracy of the traditional consensus Kalman filter algorithm, this paper proposes a consensus Kalman filter optimization algorithm based on fractional powers, that is, on the basis of the traditional consensus Kalman filter algorithm, fractional powers are introduced into the local Kalman filter part and the consensus fusion part respectively. The two better fractional power values are selected respectively and added to the traditional consensus Kalman filter algorithm at the same time. Through simulation experiments, it is validated that adjusting the fractional powers can notably expedite the convergence speed. Additionally, introducing fractional powers into the Kalman filtering process can also smooth error curves, enhancing stability and estimation accuracy. In comparison to introducing fractional powers separately in the Kalman filtering part and consensus fusion part, simultaneously introducing appropriate fractional powers in both parts demonstrates superior performance.

Recently, an affine projection tanh (APT) algorithm has been designed based on optimization criteria with hyperbolic tangent function constraints. However, there is a significant steady-state error in the APT algorithm. To address this drawback, this article presents an improved APT (IAPT) algorithm based on the optimization framework with hyperbolic tangent function square constraints. It is shown that the proposed IAPT algorithm displays strong robustness, higher convergence speed, and smaller estimation error compared to affine projection (AP) algorithm, AP sign algorithm (APSA) and APT algorithms under the impulsive noise disturbance in the system identification application scenario.

Constant false alarm detection plays an important role in radar communication imaging technology. The doped noise in the signal is one of the main reasons affecting the detection efficiency of constant false alarm, and the filtering algorithm can remove the noise and improve the detection performance. Most of the existing filtering algorithms have good filtering effect only for the noise in a specific environment. In this paper, a universal adaptive filtering selection algorithm is proposed by combining the adaptive filtering algorithm and the classical mean-class constant false alarm algorithm, which can improve the detection probability under different background noises. Finally, simulation experiments are given to verify that the adaptive filter selection algorithm proposed in this paper can be selected for different environments, and can maintain a better detection probability of mean class constant false alarm than other existing single algorithms.

Due to the complexity of deep features, discriminative correlation filter (DCF)-based trackers only extract single-layer deep features for object description and the characteristic of different layers cannot be fully exploited. Therefore, a framework of integrating multi-layer deep features is proposed for trackers based on DCF. In the training stage, multi-layer deep features are extracted using convolutional neural network. Max pooling and channel compression is applied to deep features to reduce feature dimensions and channels. Then the compressed deep features are cascaded with hand crafted features for correlation filter training. In the tracking stage, only hand-crafted features are extracted in multi-scales and deep features are extracted in single scale. Then tracking responses from single scale deep features are added to responses from multi-scale hand crafted features. Experiment results show that the tracking precision and success rate is 3.7% and 3.9% higher than original LADCF respectively. Meanwhile, the speed is 48.6% faster. It indicates that the proposed method is feasible which can maintain the performance of DCF trackers and improve the speed of the algorithm.

In agriculture, obtaining the remaining fertilizer weight is crucial for achieving accurate fertilizer application when using drones for spreading fertilizer. To address the problem of inaccurate weighing of remaining fertilizer caused by environmental and system factors during the fertilizer application process, a fusion filtering algorithm suitable for dynamic weighing of fertilizer is proposed. First, a second-order model is established between the wheel speed and the fertilizer application rate. The three coefficients in the second-order mapping relationship formula between the fertilizer weight and the wheel speed are used as the four-dimensional state variables of the Kalman filter, with the wheel speed on the drone set as the control variable. Then the weight data measured by the weighing sensor is sent to the Kalman filter for primary filtering. At the same time, the sliding average filtering algorithm is used to smooth out the oscillation phenomenon in the filtering results. Simulation experiments and real data processing results show that the drone dynamic weighing fusion algorithm, based on the four-dimensional Kalman filter and sliding average filtering has good noise reduction and smoothing effects on fertilizer weight data.

Pulse compression technique is used in many radar systems to achieve long-range detection with high range resolution. A drawback associated with the use of pulse compression is the unwanted range sidelobes which may cause spurious targets or may mask small targets nearby. The conventional approach uses window functions to reduce the effect of range sidelobes at a cost of loss in SNR and widening the main lobe width. In this paper, we investigate the P3 polyphase coded waveforms and its matched filter response characteristics. Based on the position characteristic of the main lobe and the sidelobes, a novel approach is proposed to blank out the spurious targets caused by the range sidelobe of the waveform. The approach operates on the same received signal and does not require two processing channels to set the blanking threshold. Simulation is conducted and compared to the conventional approach to show the effectiveness of the proposed approach.

In recent years, with the wide application of remote photoplethysmography (rPPG) in heart rate (HR) estimation based on facial video. How to effectively overcome the Motion Artifact (MA) to estimate the heart rate in the moving scene has become an urgent problem to be solved. In the paper, a heart rate estimation framework combining adaptive filtering and spectral subtraction is proposed to improve the accuracy of heart rate estimation during exercise. The framework will choose different signal processing methods for different motions, and use the extracted motion information to adaptively select adaptive filtering or spectral subtraction to reduce the noise of the signal. In the final heart rate estimation part, the Hidden Markov Model (HMM) is used to track the frequency trace of the signal in the frequency domain, to improve the robustness of heart rate estimation. We tested the proposed framework for heart rate detection in a variety of exercise scenarios on a publicly available dataset containing a variety of fitness exercises. The experimental results show that the proposed framework can effectively improve heart rate estimation performance during vigorous exercise.

Heart rate is a key parameter for evaluating a person’s physiological condition. In recent years, there have been many researches on remote heart rate measurement. However, these methods are mostly conducted in close-range scenarios, making them inapplicable in many scenarios. Remote photoplethysmography provides more possibilities for heart rate measurement in far-field environments. Moreover, the performance of heart rate measurement will be significantly reduced when the subject’s movement and the illumination changing. We propose a rPPG framework for heart rate detection, which selects a larger region of interest using feature point tracking in far-field environments. The combination of fast wavelet transform and second-order blind identification is used to resist illumination interference and most of the motion interference. Singular spectrum analysis is then used to resist residual motion interference. In addition, we collected a database of illumination changes in far-field environments and tested our framework with it. The results show that our method is superior to all previous methods.

Remote photoplethysmography (rPPG) monitors heart rate (HR) without requiring physical contact, which has applications. However, accurate measurement can be challenging due to different illuminations of the surrounding environment, and the impact of different illuminations on deep learning-based methods is more severe than that of traditional methods. In this study, to improve the robustness of the model to extract rPPG signals and estimate HR under different illuminations, we proposed a method called TNDR (temporal normalization and DC removal) to reduce lighting information in facial video and tested it using 3D CNN network (PhysNet). We evaluated our proposed method using four publicly available datasets (PURE, UBFC-rPPG, UBFC-PHYS, LGI-PPGI) and a proposed dataset containing three illuminations. The results show that the proposed method can significantly improve the accuracy of rPPG extraction and HR estimation under different illuminations.

In order to solve the problem of poor concealment and easy forgery of traditional authentication, and ensure good identification effect. We propose contactless authentication using heartbeat information collected by millimeter wave radar. The Frequency Modulated Continuous Wave (FMCW) signal is transmitted to the human heart by millimeter Wave radar, and the IF signal is extracted by mixing the echo signal. The phase of IF signal is extracted, and the phase difference of human heartbeat is obtained by frame difference. A deep learning network model including input layer, Long Short Term Memory (LSTM) layer, fully connected layer, softmax layer and classification output layer was constructed. Wavelet scattering is used to automatically extract features from heartbeat data extracted by millimeter wave radar. The LSTM model is trained and tested with the extracted features, and the recognition results are output. This method has achieved good recognition effect, and the recognition accuracy of three people can reach 83.8%.

In the process of monitoring the degree of vascular access occlusion in the patient with a surgically created arteriovenous fistula, premature beats can affect the calculation accuracy of a critical indicator - the bilateral pulse peak arrival time difference (bilateral difference) variations derived from the photoplethysmographic (PPG) signals of the bilateral thumbs. To address this calculation error issue caused by premature beats, we propose an algorithm based on PPG waveform morphology to remove the influence of premature beats. This algorithm establishes PPG waveform templates for three types of heartbeats and utilizes template matching during the PPG signal peak detection analysis to remove pulse peak points corresponding to premature beats before computing the bilateral difference values. The experimental results show that removal of premature beats can greatly reduce the computational error in the variation of the true bilateral difference, and thus can improve the accuracy of monitoring the degree of vascular occlusion. The proposed algorithm exhibits a sensitivity of 97.6% and a false positive rate of 2.03%. Moreover, it requires minimal prior thresholds and offers a straightforward and efficient computation process, making it easily adaptable to various embedded system platforms.

Heart rate is closely related to physiological and psychological states, and video-based techniques such as Imaging Photoplethysmography (IPPG) have been developed for heart rate detection. Although there have been some methods based on IPPG that are used to address the impact of illumination changes on heart rate detection, these methods perform poorly in environments with intense or complex illumination. This study proposes a framework that uses normalized least mean square adaptive filtering and singular spectrum analysis to combat the effects of illumination changes on heart rate detection. Experimental results on a dataset comprising 13 men and women aged 20 to 28 demonstrate the feasibility of our method under illumination changes.

Despite the remarkable recent developments in the field of remote photoplethysmography (rPPG), extracting a robust pulse signal still remains a challenge. In this study, we analyze the existing rPPG methods and unify them into a projection operation that exploits a common spatio-temporal scheme for pulse extraction. An adaptive projection plane is then proposed for pulse extraction that is orthogonal to the direction of maximum variance of the RGB trajectory. Finally, the optimal projection vector on the fitted plane is determined by utilizing Signal-to-Noise Ratio (SNR) and Kurtosis as feature selection criteria. We evaluate the performance of proposed method on PURE and UBFC2 database. Experimental results show that the proposed method enables accurate heart rate detection, and the introduced plane exhibits significant advantages.

A comprehensive assessment of the performance of maritime radar needs to rely on sea clutter data with different characteristics. However, the limitations of radar field tests lead to the fact that measured sea clutter data with different characteristics are often difficult to obtain. In practice, the problem of limited sea clutter data can be effectively solved by computer simulation. For the sea clutter simulation problem, this paper investigates the sea clutter simulation method with spatial correlation, temporal correlation and variable non-Gaussianity. Firstly, through the statistical analysis of the measured sea clutter data, it is concluded that the temporal correlation of sea clutter in a short time is determined by its speckle component, the spatial correlation is determined by its texture component, and the non-Gaussian property is determined by the probability distribution of the texture component. Based on the aforementioned sea clutter properties and the theory of spherically invariant stochastic processes, a sea clutter simulation method with generalized inverse Gaussian texture and spatial-temporal correlation is proposed. Finally, the proposed simulation method is evaluated experimentally, and the experimental results show that the proposed sea clutter simulation method can effectively simulate sea clutter with generalized inverse Gaussian texture and spatial-temporal correlation.

In this paper, a clutter suppression algorithm for velocity measurement of water flow signals is presented to improve the signal-to-noise ratio. Considering the characteristics of water flow signals and the properties of moving target indication (MTI) algorithm in signal processing, we use Texas Instrument's IWR1843 development card and DAC1000 acquisition card for data acquisition and apply the Doppler frequency shift of the signal as a reference to solve the problems encountered in MTI. First, we perform distance-dimension FFT to extract the distance unit where the target is located. Then, MTI is used to eliminate static clutter. Finally, we define a kind of adaptive weighted sliding window based on the Doppler frequency shift to suppress the abnormal growth of spurious signals after MTI processing and compress the main signal of the velocity target. The measured data and simulation results are provided to confirm the effectiveness of the proposed algorithm.

A synthetic aperture radar three-dimensional imaging system based on frequency diversity array (3D-FDA-SAR) has the characteristics of low cost and flexible transmission signal and is used in 3D imaging of complex environments. However, due to the range-angle dependence of the frequency diversity array (FDA), the information in each dimension will be coupled with each other when imaging the target, and the space-time-frequency sparse characteristic of the echo signal leads to high side lobes in the imaging results. In this paper, the deconvolution algorithm is applied to the imaging of 3D-FDA-SAR, the coupling is removed according to the coupling generation characteristics, and the effect of reducing side lobes is achieved at the same time. Using MATLAB simulation and comparing the simulation results with the simulation using the BP algorithm directly, the results show that the 3D-FDA-SAR after adding the algorithm in this paper has a better effect on multi-target imaging and is more suitable for the real imaging environment.

Ground surveillance radar has the advantages of far-distance detecting, full-weather, and working condition not affected by day and night. However, the existence of natural and man-made interference, complex landforms and other factors seriously affect the target detection performance. As a key technology in radar signal processing system, constant false alarm detection technology directly determines the performance of radar detection, so it is of great value to study the ground radar constant false alarm detection algorithm. To solve the problem of performance degradation of minimum description length constant false alarm algorithm (MDL) in multi-target environment detection, the minimum description length constant false alarm algorithm based on sparse regularization is proposed. The non-convex regularization term is used to regularize the outliers, the indicator function is introduced to improve the maximum likelihood estimation process, and the robustness of outlier vector determination is improved based on the median idea. The algorithm solves the problem that the detection threshold is greatly raised, suppresses the “target masking effect”, and improves the detection performance in multi-target environment. Furthermore, the minimum description length variability index constant false alarm algorithm based on sparse regularization is proposed. Based on the improved mean ratio and the improved variability index statistic, combined with the minimum description length idea, the two-level edge detection and the first-level homogeneity judgment are used to evaluate the clutter environment.

Aiming at the problem of spectrum aliasing of micro-Doppler signature generated by millimeter-wave radar when many people move at the same time, a multi-person micro-Doppler signal decomposition method based on millimeter-wave radar is proposed. By preprocessing the data collected by radar, the position information is obtained by using the time-distance map, and the range of single person's position is divided by using the range bin according to the position information, and the micro-Doppler signature is obtained by short-time Fourier transform of different range bins to extract human posture characteristics. The micro-Doppler signature obtained by decomposing micro-Doppler signals are used to extract different gesture features when multiple people act at the same time. The separated micro-Doppler signatures are recognized in VGG16, ResNet50, DenseNet121 and MobileNetV2 convolutional neural networks, and the recognition accuracy is as high as 99%, 98%, 99.5% and 98.5% respectively. Experiments verify the effectiveness of the multi-person micro-Doppler signal decomposition method based on millimeter-wave radar.

In this paper, a new 2D convolutional neural network (CNN) model is proposed for the classification of people, cars, and UAVs detected by low-resolution ground surveillance radars. Process the signals of people, cars, and UAVs in different domains to obtain the unique characteristics of the target signal, which can be used for target classification and recognition. Using the newly designed model to classify radar targets is divided into three steps. First, the Toeplitz matrix is used to reconstruct the 1D radar signal into 2D signals, and then a multi-channel adaptive attention module is constructed to classify radar targets. In the first step, since 2D radar signals are required for training, the 2D data is reconstructed using the Toeplitz matrix. In the second step, a channel attention module and a coordinate attention module weighted with adaptive coefficients are constructed, and the two are combined to form a multi-channel adaptive attention module. This module extracts features from the input through receptive fields of different sizes on different channels, and performs feature fusion. Since the overall network introduces a residual structure, the possible gradient disappearance/explosion problem in backpropagation is effectively solved. Tested on the actual human, car and UAV data sets, the accuracy rate reached 98.7% on the original time domain test set, 96.9% accuracy rate was achieved on the original frequency domain test set.

With the rapid development of new stealth technology and materials for aircraft, the traditional radar detection technology faces many difficulties in the detection of lower RCS targets. In order to improve the detection accuracy of space targets, this paper adopts the radar-ultraviolet cooperative detection method based on the radar working principle and ultraviolet radiation theory, and proposes a radar-ultraviolet cooperative detection algorithm for lower RCS targets. This paper takes the DF-15 missile as an example, and studies the radar-ultraviolet cooperative detection results of the missile in the detection of lower RCS characteristics, and the results show that when encountering the lower RCS target characteristics, the leakage detection rate is higher when only using radar detection, while the radar-ultraviolet cooperative detection can improve the accuracy of the target recognition, and the leakage detection rate is lower.

Flow velocity measurement is an important part in hydrological monitoring activities. Aiming at the problems of low accuracy and long time of velocity measurement in current flow velocity radar, we design a millimeter wave cross-sectional flow velocity radar system by extracting water flow echo for velocity measurement, and propose an Enhanced Squeeze-and-Excitation Network (ESE-Net) which is used for the intelligent target classification. The structure uses a multi-scale wide residual network to broaden the network width and enhance the representation ability of the network, while avoiding the degradation problem caused by network deepening. The attention mechanism is introduced into the network model to strengthen the role of informative features in the network, and realize the recognition and classification of three types of targets, water flow, floater and land. Meanwhile, the hyper-parameters of this network are optimized using the manta ray foraging optimization algorithm (dFDB-MRFO), and the recognition effect was further improved.

To address the limitations of traditional radar in time-varying and complex environments, a novel closed-loop structure for cognitive radar is introduced in this paper. The proposed system begins by estimating the power spectrum density of clutter using a clutter inversion algorithm. Next it predicts the upcoming clutter information using an autoregression model and a clutter prior information matrix as input. Finally, it employs a water-filling method and a time-synthesis algorithm to design an optimal spectrum and a constant modulus transmit sequence utilizing the prediction clutter priori information. Moreover, experimental results using real-measured data under varying parameters demonstrate that the proposed strategy outperforms traditional radar process.

A new adaptive CFAR (Constant False Alarm Rate) detector based on CA/GO/OS three-dimensional fusion is proposed in this paper. The detector integrates the best detection performance of CA-CFAR, GO-CFAR and OS-CFAR in homogeneous environment, clutter edge environment and multi-target environment respectively, and uses the convex hull learning algorithm to use the decision convex hull for target detection in three-dimensional space. Through detailed simulation analysis in homogeneous environment, multi-target environment, clutter edge environment, multi-target and clutter edge simultaneous environment, it is proved that the detection performance of the CFAR detector in different environments has maintained a very high level and has a good ability to adapt to clutter environment.

Falls are the leading cause of injuries and even fatalities among elderly individuals in home environments, resulting in the development of fall detection technology particularly crucial. In this paper, we propose a robust human fall detection method based on the millimeter-wave radar and 4D point cloud imaging technology. The main objective of this method is to detect various types of fall actions in real-time and provide timely alerts to assist the fallen individuals. In our proposed method, we first perform range-FFT and static clutter suppression on the radar echo data. Subsequently, we conduct range-domain target detection and angle estimation to generate initial point cloud information. Next, we introduce the median absolute deviation (MAD) based outlier removal method to eliminate non-human body outliers from the point cloud. Lastly, we present a suspected fall detection (SFD) method and a secondary fall detection method based on support vector machines (SVM) to maintain high detection accuracy while minimizing false alarms. The experimental results demonstrate that the average detection accuracy of our method for different types of falls is 97.5%, with an average false alarm rate of 0.4%.

Low-velocity small target detection in maritime surveillance radars is always a challenging task. Low signal-to-clutter ratio requires long-time coherent integration to obtain enough gain of target returns. However, long-time coherent integration encounters insufficient secondary data due to the spatial inhomogeneity of sea clutter. In this paper, considering the decorrelation time of speckle component of sea clutter short up to a dozen of milliseconds, the spherical invariant random vector (SIRV) model with block tridiagonal speckle covariance matrix and the inverse Gamma distributed texture is proposed to model sea clutter sequences in several tenths of a second. In this model, a long-time adaptive generalized likelihood ratio test with linear threshold detector (GLRT-LTD) is constructed. Owing to the block tridiagonal structure of speckle covariance matrices, the adaptive detection requires much less reference cells for speckle covariance matrix estimation and much lower computational cost for its inversion. The proposed detector is verified by an X-band CSIR radar database with low-velocity small boats as test targets. The experimental result shows that it obtains competitive detection performance in comparison with the state-of-the-art detectors.

Polarimetric radar is significant in radar signal processing to improve the detection performance. The research and development of target detection methods in multi-polarization channels have positive significance in complicated marine environments. In this paper, we present the problem of polarization diversity detection in compound Gaussian sea clutter with inverse Gamma texture. In this respect, a fusion detector using HH and HV dual-polarization data is proposed to detect targets in the compound-Gaussian clutter with inverse Gamma texture. On the basis of the shape parameter matched MTD detector, the test statistics of different channels are weighted and fused by using the signal-to-clutter ratio information of sea clutter. In this way, a new detector is obtained. Thresholds of the detector are calculated by Monte Carlo experiments. Numerical results of real measured radar data show that the proposed detector outperforms its competitors.

Non-contact sleep apnea detection based on radar is a technology of great significance and becomes a research hotspot because of the outstanding advantages of the radar sensors. In this paper, we propose a low-complexity sleep apnea detection method using millimeter-wave radar. The core idea of the proposed method is utilizing the reference thresholds to obtain strong generalization ability for different human targets and sleep postures. Firstly, we perform pre-processing and target detection to obtain the range bins occupied by the target. Secondly, we detect the big movement based on the body movement index and extract the features based on resting energy and respiratory waveform of multiple scatterers. Finally, the reference thresholds are utilized for sleep apnea detection to avoid the impact of different human targets and sleep postures. Experiments with ten participants and three sleep postures are conducted and results show that the proposed method can detect sleep apnea in different sleep postures including supine, lateral, and prone, with an average accuracy of 99.8%.

The issue of target detection for the single-base co-prime MIMO radar is studied, and propose a target detection algorithm for fusion of 2D-FFT algorithm and Unitary Root-MUSIC algorithm. Compared with the traditional MIMO radar, the coprime MIMO radar is able to enhance the virtual degrees of freedom by increasing the array element spacing, thus improving the angular resolution of target detection. The algorithm utilizes a co-prime array as the transceiver array of the radar, and adopts 2D-FFT target detection algorithm to obtain the target's Range Doppler map (RD map), then performs peak detection on the target's RD map to obtain the target's distance and velocity information. Finally, the Unitary Root- MUSIC algorithm is used to solve the target DOA information. Under the condition of the same number of array elements, the DOA estimation performance of the co-prime MIMO array outperforms that of the uniform MIMO array and the conventional uniform line array.

This work addresses the problem of adaptive target detection in compound-Gaussian clutter for multistatic radar. We exploit a-priori knowledge of the clutter to alleviate the performance degradation due to the non-Gaussian characteristic of the clutter. Here, the inverse Gaussian distribution is adopted to describe the texture of the clutter. In addition, the speckle covariance matrix of the clutter is modeled as a random matrix following the complex inverse Wishart distribution. Using different fusion rules between the a-priori knowledge of the texture and the speckle component within the Bayesian framework, two adaptive detectors are derived based on the Generalized Likelihood Ratio Test (GLRT) criterion. Finally, the detection performance of the proposed detectors is verified using simulated data. The results show the superiority of the proposed adaptive detectors over the existing techniques.

This paper mainly studies an adaptive MTI, through the analysis of the input signal, the spectral center of the clutter spectrum is calculated, and then combined with the zero assignment method, the most suitable filter weight coefficient is adaptively calculated, so as to achieve the purpose of calculating the appropriate MTI filtering method for the input signal with different frequency characteristics. The filtering method studied in this paper is based on zero assignment method, and the notch of the band-reject filter is set in the center of the clutter spectrum to maximize the suppression of clutter. For the estimation of the center of the clutter spectrum, it is estimated using the centroid method. The simulation results show that the proposed method has a good suppression effect on the motion clutter with narrow spectrum.

The effective perception of marine detection scenarios is essential for maritime search pulse radar to detect and track maritime targets. However, the dynamically changing and complex marine environment makes it challenging for maritime search pulse radar to accurately perceive the sea and land regions. To improve the accuracy of sea-land region perception, and reduce the computation and complexity of algorithm of the network, this paper proposes a MobileNet-V3-based maritime search pulse radar sea-land segmentation method. Firstly, a radar sea-land segmentation dataset based on PPI (Plan Position Indicator) is constructed using real measured data from different scenarios and sea conditions. Subsequently, the MobileNetV3 network is trained on this dataset to achieve sea-land segmentation for maritime search pulse radar. Experimental results demonstrate that the segmentation accuracy of the MobileNetV3-L-based radar sea-land segmentation method surpasses its competitors.

Due to the excellent advantages of the radar sensor, it is considered to be one of the most potential technologies for the sleep monitoring. In this paper, we propose a sleep stage estimation method based on state transition using frequency-modulated continuous wave (FMCW) millimeter-wave (MMW) radar. The core of the proposed method is to utilize the physiological characteristics of different state transitions during sleep to achieve state transitions for different human targets. Firstly, we conduct signal preprocessing and target detection to determine the presence of the target. Secondly, we extract features from the respiratory rate and body movement to determine the start of sleep and the end time of sleep. Finally, we employ reference thresholds to determine the state transition of sleep for sleep staging. A total of more than 138 nights of data from 10 participants were tested and compared with the Mi Band 6, Mi Band 7, and Huawei Band 6. The results demonstrate the effectiveness of the proposed method.

This paper considers a multi-input multi-output (MIMO) dual functional radar communication (DFRC) system and focuses on the joint design of transmit beamforming matrix and sparse antenna array. We propose to minimize the Cramér-Rao bound (CRB) of radar sensing while preserving a predetermined level of signal-to-interference-plus-noise ratio (SINR) for the communication users. The hybrid analog-digital (HAD) beamforming technology with fewer radiofrequency (RF) chains is considered to reduce the cost of hardware system as well as to deal with the rank-deficient problem of radar sensing. Due to the complex representation of the CRB matrix, the learning-based method is proposed to simultaneously optimize the HAD beamforming matrix and antenna selection matrix. Numerical simulations are conducted to demonstrate the effectiveness of the proposed method.

In this paper, a joint path planning and beampattern (JPP-BD) strategy is proposed for multi-target tracking (MTT) in airborne colocated multiple-input multiple-output (C-MIMO) radar system. The key mechanism of the proposed strategy is to collaboratively coordinate the waveform correlation matrix (WCM), kinematic velocity and heading angle of the airborne radar, in order to improve MTT performance and low sidelobe performance under the constraints of maneuverability limitations and system resource budgets. The predictive Bayesian Cramér-Rao lower bound (BCRLB) and peak sidelobe level (PSL) are derived and adopted as the metrics to characterize the target tracking accuracy and low sidelobe performance, respectively. As the formulated JPP-BD problem is non-linear and non-convex, we propose a partition-based three-stage approach to solve it effectively. Simulation results show that the proposed JPP-BD strategy achieves the best system performance in comparison with other benchmarks.

Due to the limitations of clock synchronization technology and non-ideal clocks of the sensors, it is not feasible to achieve precise clock synchronization for distributed multiple-input multiple-output (MIMO) radar systems. In this paper, a closed-form solution for locating moving target in distributed MIMO radar systems with clock synchronization errors by using the time delay and Doppler shift measurements is proposed. Specifically, a transmitter is utilized as the reference and the clock synchronization errors are introduced to establish estimation formulations about the target position and velocity, and the formulations are solved by best linear unbiased estimator firstly. Then, the estimates of the target position and velocity are refined based on the relation between the reference parameters and unknowns. The proposed solution is demonstrated to be approximately unbiased and able to reach the Cramer-Rao lower bound under weak noise conditions through the theoretical derivation and numerical simulations. Furthermore, the simulations show that the proposed solution enjoys better target localization accuracy than the state-of-the-art algorithms.

High-Performance Automotive Millimeter-Wave Radar usually uses Doppler Division Multiple Access (DDMA) to build virtual MIMO arrays. However, the DDMA approach leads to blurring of the radar velocity interval and severe antenna interference problems. In this paper, we analyze these two problems in the DDMA mode, and also propose an improved DDMA target demodulation method, which can efficiently identify the interference target and correctly estimate the velocity of the interference target by means of differentiating the Empty-band of two adjacent frames. Simulation experiments show that this method can effectively solve the interference problems arising from Empty-band demodulation, and has certain practical significance.

In this work, we consider target parameter estimation of phase modulated continuous wave (PMCW) multiple-input multiple-output (MIMO) radar systems with few-bit quantization observations. We formulate the parameter problem as a sparse recovery problem and then jointly estimate the targets’ amplitudes, time delays, Doppler shifts, and directions under the generalized sparse Bayesian learning (Gr-SBL) framework. Under this framework, this proposed algorithm decomposes the original nonlinear problem into a sequence of standard linear model (SLM) problems. Therefore, we can apply the standard sparse Bayesian learning (SBL) algorithm to solve the above SLM. Numerical results demonstrate the effectiveness of the proposed Gr-SBL algorithm for the parameter estimation of a PMCW MIMO radar systems with few-bit analog-to-digital converters (ADCs).

The mapping of the radar echo dataset into a graph signal offers a novel perspective for solving the radar target localization problem. However, the published graph-based methods are mostly applicable to the uniform array configuration. In this paper, we propose an enhanced graph-based target localization method that can be applicable to the non-uniform frequency diversity array radar to fill this gap. Following the previous studies, we establish a space-domain graph model for the echo signal acquired from a non-uniform frequency diversity array radar. Subsequently, we employ the graph signal processing method to solve the target localization problem. Numerical simulations demonstrated that the proposed graph-based localization method provides a high resolution and accurate estimation, surpassing conventional methods.

Radar communication integration is one of the development directions of multifunctional combat systems. Design of integrated signal that can realize radar and communication functions is the main method to build an integrated system. In this paper, a constant envelope OFDM-Chirp integrated signal based on phase coding (CE-PC-OFDM-Chirp) is proposed. By analyzing the ambiguity function of the signal, the simulation results show that the proposed integrated signal has low distance sidelobe, narrow velocity main lobe and good error performance.

In modern warfare, using a UAV swarm as an effective countermeasure to deceive and interfere with hostile networked radar is a common tactic. However, when facing a complex combat environment, the digital storage and forwarding equipment carried by the UAV swarm has limited accuracy, resulting in poor forwarding delay. This delay leads to a deviation between the actual generated false target point and the preset false target point, significantly reducing the effectiveness of deception. To address this issue, this paper proposes a solution based on the concept of "homologous detection" and the certain spatial resolution of networked radar. The boundary conditions for effective deception jamming by digital storage and forwarding technology are analyzed, and the influence of forwarding delay error on deception jamming effect in typical networked radar systems is identified. Simulation results demonstrate that the analysis and deductions can provide a reliable assessment of whether the actual jammer performance can effectively deceive and jam the networked radar.

In this paper we propose a novel signal processing architecture for millimeter wave flow velocity radar. FPGA and DSP are combined to make the system more compact while achieving real-time signal processing. We focus on the workflow and software design for DSP, and propose an improved CFAR based on the median absolute difference and quantile for the target characteristics, which improve the detection performance of the radar in dense target condition as well as clutter edge environment. Finally, experiment results show that the signal processing system can work stably and achieve real-time and high-accuracy measurements.

In recent years, the demand for target detection in Synthetic Aperture Radar (SAR) under low signal-to-noise ratio conditions is increasing. To improve the ability to detect weak moving targets, this paper proposes a moving target detection method based on quadratic rational kernel function for sequential SAR images. This paper begins by presenting background information on weak moving target detection, followed by the simulation model for sequential SAR images and the specific process for detecting moving target in those sequential images using kernel trick. Finally, numerical experiments on detecting moving target in sequential SAR Images are carried out, and those results demonstrate that the effectiveness of the proposed method.

A low-complexity rear vehicle detection approach based on a dual-channel millimeter-wave radar is presented in this paper. The main task is to detect whether there are vehicles approaching on both rear sides of the vehicle, and to give an early warning when the relative distance between the two vehicles is less than the early warning range. The difficulty in accurately detecting an approaching target vehicle lies in the fact that the vehicle is driving in a complex traffic scene and there are metal roadblocks in the road environments. In this paper, firstly, in order to reduce the complexity, the cascaded distance and angle of the potential target calculation is adopted to achieve the localization of the potential target. Then, rough recognition of vehicle targets and roadblocks is performed by clustering algorithm and structural features extraction. Since the features of some of the roadblocks are similar to those of the vehicle targets, secondary recognition is required to classify the vehicle targets and roadblocks by the target motion trend and trajectory length. The experimental results show that the average detection rate is 91.13% and the average false detection rate is 0.48% in different traffic scenarios.

With the development of artificial electromagnetic metamaterials, the new target stealth technology represented by metasurface stealth can achieve fast modulation within the pulse, which poses a huge challenge to the traditional radar detection method based on pulse integration. In this paper, the echo model of the metasurface stealth target is constructed, and the short-term coherence of the echo is mined through windowed time-frequency analysis. The time-frequency spectrogram of the target echo is correlated with the time-frequency spectrogram of the transmitted signal to obtain one-dimensional correlation matching coefficients. Target detection is achieved based on coefficient peaks. This paper proves through simulation that the time-frequency correlation matching method can obtain good detection performance under different pulse modulation speeds.

Feature pyramid network (FPN) is a critical component in modern object detection frameworks. The performance gain in most of the existing FPN variants is mainly attributed to the increase in computational burden. An attempt to enhance the FPN is enriching the spatial information by expanding the receptive fields, which is promising to largely improve the detection accuracy. In this paper, we first investigate how the expanding receptive fields affects the accuracy and computational costs of FPN. We explore a baseline model called inception FPN in which each lateral connection contains convolution filters with different kernel sizes. Moreover, we point out that not all objects need such a complicated calculation and propose a new dynamic FPN (DyFPN). The output features of DyFPN will be calculated by using the adaptively selected branch according to a dynamic gating operation. Therefore, the proposed method can provide a more efficient dynamic inference for achieving a better trade-off between accuracy and computational cost. Extensive experiments conducted on MS-COCO benchmark demonstrate that the proposed DyFPN significantly improves performance with the optimal allocation of computation resources. For instance, replacing inception FPN with DyFPN reduces about 40% of its FLOPs while maintaining a similar high performance.

In the domain of wideband radar, detecting targets often involves the dispersion of target echo energy across multiple range resolution cells. A critical aspect of enhancing wideband radar target detection performance lies in effectively utilizing this discrete energy information. This paper introduces the Hough transform and weighted amplitude iteration detector (HT-WAID) for wideband radar, which effectively mitigates the effects of range spread on detection performance. Firstly, the analysis commences for wideband radar targets under both single-rank and multi-rank conditions. Subsequently, a weighted amplitude iteration range spread target detector (WAI-RSTD) is introduced. Furthermore, the WAI-RSTD algorithm is extended to a two-dimensional detection space using Hough transform theory. Simulation results demonstrate that the proposed algorithm outperforms Adaptive generalized likelihood ratio test (GLRT) and Wald-like detectors, showcasing superior target detection and clutter suppression capabilities in complex white Gaussian noise and Weibull clutter backgrounds.

Defect detection in Photovoltaic (PV) cell Electroluminescence (EL) images is a challenge in industry. In this paper, a novel defect detection method YOLOv4 with an improved Convolutional Block Attention Module (YOLO-iCBAM) is proposed for PV cell EL images. We first propose an improved CBAM to enhance the network’s ability to capture multi-scale defects in complex image backgrounds. Then, we modify the conventional YOLOv4 architecture for defect detection. Specifically, we adjust the backbone network to make a fast convergence. Then, we adopt the iCBAM to YOLOv4 to refine the feature map before YOLO Head. Then, we train a K-Means++ model based on PV cell EL images to generate anchors for bounding box regression. Moreover, we conduct experiments in the PVEL-AD dataset to evaluate the proposed YOLO-iCBAM. The experimental results indicated that the proposed YOLO-iCBAM achieves a better F1-Score of 0.716 and mAP of 0.748.

This paper proposes a human presence detection method based on the combination of long- and short-term micro-motion features using millimeter wave (MMW) radar. It can be divided into three parts: potential moving target detection, micromotion parameters estimation, and multi-timescale integrated human presence detection. For the first part, we used the constant false alarm rate (CFAR) detector to detect potential moving targets. For the second part, we estimated the micromotion parameters, such as target status, respiratory rate, body movement index, and target location. For the third part, we extracted the long-term and short-term micro-motion features related to the above obtained parameters by considering different cases of human activities. Then, we perform human and interference recognition based on the above features. Finally, a large number of experiments, considering 11 situations under natural wind or fan blowing curtains, bed curtains, mosquito nets moving, and fan rotation, are conducted. Compared with the traditional CFAR detector, the proposed method can significantly improve the detection of human presence under conditions of interference in sleeping scenarios.

With the rapid development of automated driving technology, it becomes crucial to accurately detect and recognize targets in complex scenes. Camera sensor detection and recognition accuracy is not high enough, has poor stability, and cannot adapt to the target detection of complex scenes. Millimeter wave radar is less affected by the environment and more accurate in speed and distance measurement, but the current mainstream millimeter wave radar sensor point cloud is sparse, so it cannot identify the target feature information. Therefore, the fusion of radar and vision is gradually becoming a mainstream solution for accurate obstacle detection. This paper presents a novel parallel attention fusion module (PAFM) to enhance the performance of the SAF-FCOS target detection network. By incorporating depth-separable convolution and parallel dual attention modules, the RetinaNet network of SAF-FCOS is enhanced to effectively integrate information from upper and lower feature maps of diverse scales. Compared to the direct summation approach, PAFM enables the network to selectively focus on image regions of interest, both in terms of channel and spatial dimensions, without introducing excessive algorithmic complexity. Experimental results on the NuScenes dataset demonstrate that the proposed improved network achieves superior performance compared to the original network.

In order to solve the problem of difficulty in segmenting the foreground of lawn weed images due to the similarity between the foreground and background grayscale, this paper proposes a Retinex enhancement algorithm based on local density fusion. Firstly, image preprocessing is completed based on local variance to realize prominent image foreground and smooth cluttered background; Secondly, using multithreshold segmentation and operational differentiation, pixels are divided into three categories: foreground, background, and pixels to be subdivided. Local density is used to extract spatial information of pixels to be subdivided; Finally, the sigmoid function is used to optimize the reflection component grayscale transformation coefficient, achieving the fusion of local density information and obtaining the final image. The results show that the algorithm proposed in this paper expands the gray difference between weeds and lawn grass while suppressing background noise; Compared with other enhancement algorithms, the peak signal-to-noise ratio, average running time, and segmentation accuracy have all been improved greatly. At the same time, through image analysis, it has been found that the algorithm results in occasional problems of uneven brightness and over segmentation of fine leaf weeds in the image. In response to these two problems, this article finally proposes two corresponding optimization measures, namely global brightness adjustment based on a new adaptive gamma function and Retinex reflection component saturation adjustment based on AHE.

The frequency space characteristic is an important characteristic of steady-state visually evoked potential (SSVEP). The frequency space characteristic of traditional SSVEP is generally obtained by analyzing the power spectrum of EEG of scalp electrodes, but it is vulnerable to noise interference and data quality. This article proposes a new SSVEP frequency space feature analysis method. The proposed method consists of two main parts: 1) A shallow convolutional neural network, called EhythmNet in this paper, is designed for EEG rhythm analysis; 2) Based on multi channels EEG data obtained from a set of stimulus frequencies, single channel EEG of SSVEP dominant electrodes (such as Oz) was used as the training data for EhythmNet, and data from other channels were used as the test set to obtain the recognition rate of each channel, namely single channel recognition rate (SCA). The experimental results indicate that SCAs can accurately reflect the spatial distribution of SSVEP in the scalp electrode, and it is found that the spatial distribution of SCA has good individual stability and differences between individuals. In order to verify this feature, an identity recognition test based on SCA was conducted in the article, and more than 98.5% of the recognition results were achieved.

The radar's high-resolution range profile (HRRP) data structure is complex, and extracting stable and reliable features from it is crucial for HRRP target recognition. In this paper, we propose to use the convolution module to extract local spatial features of HRRP and use the positional encoding to embed the position information to generate new temporal features, and then capture the long-term dependency within the distance unit of HRRP through the multi-head new self-attention mechanism of the Transformer encoder, to construct a reliable feature extraction method for the HRRP target. Finally, a new deep learning model CNN-TEAN (CNN TransEncoder-Attention Network), based on one-dimensional residual convolution, Transformer encoder, and attention mechanism, is formed using the attention mechanism, fully connected layer, and softmax for classification. Using six simulated ship target data types for experimental validation, the CNN-TEAN model proposed in this paper can achieve a higher recognition rate than RNN, LSTM, and SVM models.

Indoor human activity recognition, when deployed on edge devices, holds promising application prospects in security surveillance. However, conventional CNN-based lightweight recognition methods perform poorly in indoor environments due to wall interference and signal attenuation. To address this issue, this paper proposes an effective lightweight model that balances parameters and performance to recognize seven types of human activities occurring behind walls. Specifically, the BiRAMNet consists of four pyramid-like stages and utilizes the Bi-level Routing and Residual (BRR) block, a fundamental block combining the strengths of CNNs and Transformers, as its building block. We introduce Bi-level Routing Attention to focus on the motion characteristics of the Doppler spectrogram, sidelining background information. The proposed model is benchmarked against baseline networks, including ResNet50, ResNet101, MobileNetv2, MobileNetv3 and EMO. The results highlight that our model achieves superior recognition accuracy and a lighter size. Future work includes extending BiRAMNet to more diverse human activity recognition scenarios and improving its performance for real-world applications.

Sparse regularization is an effective tool for synthetic aperture radar (SAR) image despeckling. Designing effective sparse regularization terms plays a very important role in this kind of method. Existing sparse regularization despeckling methods use conventional patch-based sparse representation to design regularization term. This patch-based manner will lose some important spatial information along edges between patches, resulting in staircase effect. In this paper, we propose a new Gradient domain Convolutional Sparse Coding-based (GCSC) method for SAR image despeckling, and derive a feasible algorithm to efficiently solve the corresponding nonconvex optimization problem. In contrast to the well-known sparse regularization despeckling methods that divide a SAR image into patches and process patches individually in the spatial domain or the transform domain, GCSC works on the whole SAR image to learn a convolutional sparsifying regularizer in gradient domain. By taking advantage of the gradient domain convolutional sparse coding, GCSC can capture the correlation between local neighborhoods and exploit the gradient image global correlation to produce better edges and sharp features of SAR image. Experiments conducted on real SAR images demonstrate that the proposed GCSC outperforms those state-of-the-art SAR despeckling methods in terms of subjective and objective evaluation.

Panoptic segmentation is a critical technology in the field of multimedia, applicable to various domains such as autonomous driving and image recognition. However, due to the enormity and complexity of the task, enhancing the efficiency and accuracy of panoptic segmentation remains a challenge. In this paper, we propose a stage-enhanced panoptic segmentation method which improves the feature extraction network of the backbone, incorporates a stage feature fusion network, and designs a module for adaptive stage feature weight allocation. These enhancements optimize the overall network and enrich the stage features. Experimental results on the publicly available COCO-2017 dataset confirm the performance of Stage-Enhanced Panoptic and demonstrate its superiority compared to other methods.

Image forgery has been a serious issue in real life during this boosting big data era. There have been many methods depending on detecting footprints such as edge inconsistency, camera noise, JPEG artifacts, etc., proposed. Unlike the existing methods, we propose a more general method, CRAC-formerNet to capture more stable manipulation traces, not only including RGB domain but also from high-frequency features generated from Steganalysis Rich Model (SRM) and District Cosine transformer in the frequency domain. A shared query is generated from features from both the RGB and frequency domains. The keys and values are generated from each domain respectively. Moreover, features of manipulated and manipulated regions should belong to different distributions, especially features in edge regions. We use a center contrastive loss to learn fine-grained features. A manipulated region proposal module is proposed for improving computing efficiency when calculating the loss. We empirically demonstrate that our method achieves competitive results on four benchmark image manipulation datasets.

High-resolution range profiles (HRRPs) of ships are significant for ship classification and recognition, sparse recovery methods are a major tool to attain ship HRRPs from high-resolution radar echoes. Aiming at the problem that complex HRRPs of ships are not finely modeled in the sparse recovery via iterative minimization (SRIM) method, this paper proposes a new SRIM method to estimate complex HRRPs of ships from radar echoes corrupted by non-Gaussian sea clutter. In the new SRIM method, ship HRRPs are modeled by a bi-parametric lognormal distribution (LND), and high-resolution sea clutter is characterized by the compound-Gaussian distribution with inverse Gaussian texture (CG-IG). The new SRIM method is compared with the classic sparse learning via iterative minimization (SLIM) method, the linear programming-based (LP-based) method, and the recent SRIM method by adopting simulated and measured radar data, the experimental results show that the new SRIM method effectively reduces the central processing unit (CPU) time of ship HRRP estimation while attaining better performance than the existing methods.

Orthogonal time-frequency space (OTFS) is a novel modulation scheme that enables reliable communication in high-mobility environments. In this paper, we propose a Transformer-based channel estimation method for OTFS systems. Initially, the threshold method is utilized to obtain preliminary channel estimation results. To further enhance the channel estimation, we leverage the inherent temporal correlation between channels and a new method of channel response prediction is performed. Based on this, Transformer neural network is applied to our work, which is specialized in processing time series, to refine the preliminary results. Simulation results demonstrate that the performance of the proposed scheme in terms of normalized mean squared error (NMSE) and bit error rate (BER) is better than the threshold method and other deep learning (DL) methods.

Popular machine learning based fingerprint localization methods often struggle to effectively capture non-Euclidean characteristics present in fingerprint data, while geometric deep learning can effectively process such data. In this paper, we propose a geometric fingerprinting based graph neural network indoor localization algorithm (GFGNN), which is models access points (APs) and reference points (RPs) using received signal strength (RSS) fingerprint. This approach maximizes the utilization of the unstructured nature of fingerprint data to enhance indoor localization accuracy and stability in dynamic environments. The algorithm establishes the fingerprint data as a graph feature representation, we first employ a graph convolutional network at the AP level to aggregate RSS values containing spatial relationships. Subsequently, graph isomorphism networks are employed at the RP level to further extract and update the aggregated fingerprint features. Finally, a multi-layer Perceptron is utilized to regressively predict the localization of the target to be located. We evaluate the proposed GFGNN on a self-built dataset, and the localization accuracy remains within 0.43 meters at the 80th percentile of the cumulative distribution function, with stable localization performance even in dynamic scenarios.

In this paper, an iterative algorithm for phase synchronization in a distributed coherent transmission is proposed to achieve phase alignment of multiple transmitter signals at the receiver. In each time slot, each transmitter computes and adjusts its phase offset based on the received feedback and its own phase perturbation. Transmitters get the phase change effect near the best known phase adjustment on the received signal power, and guide the phase adjustment in the direction of increasing the received signal power. The simulation results show that the convergence speed of our algorithm proposed in this paper has been increased compared with the classical 1-bit feedback algorithm, and the convergence accuracy and stability are basically the same as those of the classical 1-bit feedback algorithm.

Millimeter wave (mm-Wave) multiple-input-multiple-output systems have the unique characteristics of high temporal resolution and high directivity, enabling very accurate localization. By utilizing the time delay, angle of departure, and angle of arrival measurements from both line-of-sight and non-line-of-sight paths, provided by the mm-Wave system, we propose a two-stage solution method to jointly estimate the user equipment (UE) position and orientation as well as the scatterer positions. Specifically, based on the geometric relations between the UE and scatterer positions, a coarse estimate is obtained by solving a set of equations, which is then refined by a linear weighted least squares estimator. The computational complexity of the proposed method is pretty low in that both solutions in the two stages are in closed form. Simulation results show that the proposed method is able to reach the Cramér-Rao lower bound performance at low noise levels.

Post-disaster networking is the foundation and the primary work of emergency rescue. In this paper, we adopt the rapid deployment method of artificial bee colony (ABC) based on rapid depth-first search (RDFS) to keep the UAV cluster in the relay state, and use rapid depth-first search in the period of employment bees, observation bees and scout bees, respectively, to quickly find out the connectivity links between the ground control centre and the ground nodes, and by optimizing the depth-first search algorithm, we can stop searching as soon as any link is found, which greatly improves the relaying efficiency and shortens the relaying deployment time, and identifies an optimal relaying deployment strategy to improve the throughput. Simulation experiments show that compared with the deployment method before optimization, the rapid deployment method of artificial bee colony based on rapid depth-first search after optimization improves the throughput of the network by up to 12 times.

The time delay among neighboring microgrids is inevitable. Besides, the perturbations in the microgrid control process may make the time-delay stability more complicated. In order to consider the stability issues, this paper builds the time-delay dynamic model of microgrids with eigenvalue tracking method. To address the effect of communication delays and perturbations on the frequency stability of the microgrid, an effective frequency domain method called eigenvalue tracking is used. This method accurately determines the stable delay margin and the critical eigenvalue on the imaginary axis of the interconnected time-delay microgrids. It demonstrates that the method is easier to use to obtain accurate delay margins than the Lyapunov stability principle. Finally, a three-zone delayed power system and a multi-zone power system with AC-DC interconnection are used as examples to illustrate the feasibility and superiority of the method.

The rapid growth in usage and Instant Messenger Applications (IMAs) has made them a potential target by cyber criminals to conduct malicious activities such as identity theft, illegal trading, etc. For example, WeChat is extending its services to mobile platforms, making it an important source of evidence in cyber investigation cases. Therefore, understanding the types of potential evidence of users activities available on mobile devices is crucial for forensic investigations and research. In this article, we examined WeChat, the most popular application on the Android platform. We created various artifacts (e.g., texts messages, images, audio, and document) that are of forensic interest to perform targeted spoofing tampering with WeChat data. This helps in analyzing the authenticity of WeChat data during forensics, thereby proving the innocence of people or the existence of certain criminal activities.

Flexible electronics, as an emerging technology, has shown potential applications in various fields. In recent years, gallium-based liquid metals have gained significant attention in the field of flexible electronics due to their exceptional metal conductivity and infinite ductility. However, achieving cost-effective and efficient patterning of liquid metal (LM) is crucial for the production of flexible electronic products. This article presents a novel liquid metal circuit printing technology that offers a rapid, affordable, and versatile solution to this challenge. Liquid metal patterned masks can be efficiently produced using laser cutting machines, eliminating the requirement for costly processing steps or equipment. Furthermore, direct mask printing enables the achievement of liquid metal circuit patterning. By employing this technique, a stretchable resistance strain sensor was successfully fabricated, exhibiting a remarkably extensive strain sensing range of 480% and commendable repeatability. These findings suggest that the aforementioned technology holds significant promise for the economical and expeditious realization of wearable health monitoring, intelligent sensing, and human-computer interaction.

A high-sensitivity and stable ethylene glycol sensor was successfully developed by employing a facile hydrothermal method and annealing process to fabricate flower-shaped NiO/ZnO hierarchical nanostructures. The flower-shaped NiO/ZnO was integrated as the sensing material into a Microelectromechanical Systems (MEMS) gas sensor. The MEMS-based hotplate provided favorable heating conditions for the sensitive nanomaterial. Gas-sensing tests demonstrated excellent sensing performance of the sensor based on the flower-shaped NiO/ZnO heterostructure towards ethylene glycol. At the optimal operating temperature of 300 °C, the sensor to 10 ppm of ethylene glycol exhibited a response value of approximately 17.46 with response and recovery times of approximately 18 and 150 s, respectively. Moreover, its selectivity towards ethylene glycol was significantly higher than other common volatile organic compounds. This enhanced sensing performance is primarily attributed to the formation of p-n heterojunctions at the interface, the porous structure of NiO nanosheets, and their effective catalytic activity, resulting in a remarkable enlargement of the surface depletion region and an increase in barrier height. This study provides a simple synthetic process not only applicable to the preparation of other semiconductor oxide composite materials but also offers an effective method for ethylene glycol detection.

Partial discharge detection is an effective method for evaluating the insulation condition of cable accessories. Ultra-high-frequency method for detecting partial discharge has strong anti-interference ability. Therefore, a broadband, circularly polarized Archimedean planar spiral antenna is used as the partial discharge detection sensor. After determining the antenna parameters through theoretical calculations, simulate and design the antenna. Analyze the influence of different inner diameter values on antenna performance. Exponential gradient microstrip balun is added to achieve impedance matching between the antenna and the coaxial line, and a reflection cavity is added to achieve unidirectional radiation and gain improvement of the antenna. Based on parameter analysis, optimize the design of Archimedean planar spiral antenna and complete the physical production. The test results indicate that the designed Archimedean planar spiral antenna has a gain greater than 6dB and a VSWR less than 2. The performance of the antenna meets the requirements.

The development of sensing interfaces with high selectivity and sensitivity is a significant focus on electrochemical biosensor. A hydrogen peroxide biosensor with good properties is developed by constructing a Prussian blue (PB)/Au nanoparticles-chitosan (AuNPs-CS) highly electroactive bilayer composite membrane sensing interface on an Au electrode. The AuNPs-CHi nanocomplexes are assembled with PB via layer-by-layer assembly method to construct a CHi-AuNPs/PB sensing interface. Protection of PB by the network CS membrane and the synergistic effect of PB and AuNPs enhance the electrocatalytic activity of this sensing interface towards H₂O₂. The PB/AuNPs-CS composite membrane sensing interface has a steady-state response time of less than 2 s and high sensitivity to H₂O₂ with a linear response range of 0.01 ~ 7.95 μM, a low limit of detection of 0.269 μM (S/N = 3) and a sensitivity of 511.82 mA·mM- ¹·cm-². The biosensor provides a platform to detecting H₂O₂ with a wide linear range and low detection limit for future rapid characterisation of relevant physiological information.

In this study, the performance of temperature, humidity, and pressure sensors that can be used in smart mattress pads has been investigated. The measurement accuracy of pressure sensors has been compared for different sponge thicknesses and weights. Among 9 different test cases, the RP-S40-ST force sensor yielded the best results. As for temperature sensors, heat transfer measurements were compared for specific temperatures using different sponge thicknesses. The AHT15 and DS18B20 temperature sensors successfully measured 74% of the applied heat on a 2 cm thick viscoelastic sponge, and 52.4% on a thicker 4 cm sponge. Additionally, the humidity sensor's performance in measuring humidity on the mattress pad was evaluated. Based on these assessments, the sensors that will provide the best results for the design of the mattress pad have been proposed. The influence of the sponge, an essential component of the mattress pad, on the functioning of these sensors was also observed. Moreover, apart from their measurement capabilities, the impact of these sensors on the patient's comfort was also taken into consideration.

Chemical indicators in biofluids could provide tremendous amount of information, leading to great application potential of stretchable chemical sensors in the field of advanced clinical therapy and health monitoring. Graphene possesses excellent optical, electrical, and mechanical properties, making it highly promising for many applications in electrochemical sensing and biosensing. However, traditional methods of graphene preparation have notable limitations, such as low production efficiency or complex manufacturing processes. In this study, using laser induced graphene (LIG) produced on polyimide (PI) substrate, we propose a method to transfer LIG onto a flexible substrate using the semi-solidification state of PDMS (Polydimethylsiloxane) and explore effects of temperature and time on transfer printing. As a result, the LIG/PDMS could serve as a flexible electrode and the electrode’s resistance only changes 14% after 70 cycles of 10% stretching. The practicality of the LIG/PDMS flexible electrode is effectively demonstrated through the electrodeposition of Prussian blue and detection of 50μM glucose, which could be used as stretchable glucose sensor in the future integrated chemical health monitoring systems.

Single triaxial acceleration sensor was used to acquire acceleration information generated by physical activities, and classify physical activities with widely varying degrees of intensity into three types: resting (including sitting, lying down and standing), walking (including running, walking up and down stairs), and running and jumping (including running and jumping in place), so as to provide an objective basis for assessing the level of physical activity. Methods: Forty-four participants wearing accelerometers performed eight physical activities, namely meditation, lying down, standing, running, going upstairs, going downstairs, sprinting, and jumping in place. The accelerometer data were collected, cleaned, and preprocessed using a band-pass filter. The signal amplitude region, mean, standard deviation, correlation coefficient of triaxial, frequency domain entropy and other characteristic quantities are then extracted from the filtered signal. Lastly, a radial basis kernel function support vector machine recognition model is used for physical activity recognition and validated using a leave-one-subject-cross-validation method. The results show that The highest recognition rate of 98.50% was achieved using the frequency domain composite feature subset, whereas the time domain feature subset achieved a slightly lower recognition rate of 98.3%. The time taken was calculated to be 223ms and 146ms. The window size was found to have the highest recognition time of 5.12s. In conclusion: A single triaxial acceleration sensor can collect substantial data about human activities for identification purposes. Creating a physical activity type recognition model that uses the radial basis function kernel and support vector machine (SVM) can allow for real-time and accurate identification of physical activity types in natural environments.

In distribution networks, safety and stability are of utmost importance. When a ground fault occurs, accurately determining the fault phase and avoiding misoperation of switches in other phases become crucial. This paper proposes a method using a magnetic sensor to measure the third phase current, which is then applied to ground fault line selection in distribution networks. The research aims to find solutions to eliminate interfering magnetic fields in complex electromagnetic environments, conduct studies on current measurement technologies using various mathematical models based on magnetic sensors, and develop low-cost, easy-to-install key technologies for achieving ground fault line selection in complex electromagnetic environments of distribution networks.

Particle pollution has seriously affected the environment and human health. The monitoring of ambient air particles has become more normalized, and more and more cities are using micro online environmental air quality monitors as a supplement to existing national control stations. Now, high-precision β The Random forest model was established based on β-ray instrument, and the calibration of light scattering sensor was carried out. The correlation coefficient R2 increased from 0.77 before calibration to 0.97, and the relative expanded uncertainty of Random forest prediction results was 0.46%. The results indicate that studying the algorithm model can effectively reduce the measurement error of light scattering sensors, improve the accuracy and availability of micro station data.

To monitor the changes in cutting force during the machining process of machine tools, this article takes ZL205A aluminum alloy as the research object and builds a multi-sensor acquisition system, including cutting force signal, load current signal, and vibration signal acquisition system. Cutting experiments were conducted to obtain cutting force, load current, and vibration amplitude under different process parameters. The time-frequency characteristics of cutting force signals were analyzed. The mapping relationship between cutting process parameters and cutting force, as well as load current and cutting force, was established. This research enables indirect monitoring of dynamic cutting force signals by monitoring load current signals, providing reference and guidance for monitoring dynamic cutting force in actual machining processes and process optimization.

In order to simulate any phase-angle signal over a wide frequency range, this paper designs a DAC-based constant amplitude low-frequency phase shifter. The circuit mainly consists of three parts: a phase shift circuit with feedback, a vector attenuation circuit, and a vector synthesis circuit. The 90° phase shift is achieved by automatically adjusting the feedback-controlled variable resistance of the field-effect transistor through phase angle detection. This ensures high-precision phase shifting for different frequency occasions. The 12-bit high-precision multiplicative DAC chip is used to attain attenuation of the standard vector signal. The high stability of the DAC chip guarantees the accuracy of the target attenuation signal. To minimize errors in vector synthesis, the circuit employs highly matched network matching resistors and operational amplifiers. Experimental results demonstrate that the entire phase-shifting circuit can accomplish 360° phase shifting within the frequency range of 10Hz to 10kHz, with a phase-shifting accuracy better than ±0.1°

Digitally coded metasurface (DMS) is a novel class of electromagnetic materials that possess the ability to manipulate electromagnetic waves at scales significantly smaller than the wavelength. They hold great potential for a wide range of applications, including wireless communication, millimeter-wave imaging, data storage, and sensing. However, the conventional full-wave simulation methods currently employed often require precise geometric modeling, grid planning, and consideration of complex physical models such as material parameters and boundary conditions. These time-consuming and costly approaches face significant challenges in large-scale electromagnetic response studies. Therefore, there is an urgent need to explore more efficient and cost-effective methods to address these issues. In this paper, we propose a deep learning-based approach for predicting the electromagnetic response of individual units in a DMS. We construct a convolutional neural network (CNN) model that can accurately and real-time predict the electromagnetic response based on the input coding pattern. To train and validate the model, we utilize a substantial amount of electromagnetic simulation data and incorporate amplitude constraints into the loss function for model optimization. A way to abandon the dual network structure and achieve simultaneous prediction of dual parameters with smaller computational requirements. Experimental results demonstrate that our model achieves highly accurate predictions of both amplitude and phase responses of DMS, surpassing traditional numerical methods in terms of efficiency and scalability. The proposed deep learning approach offers a promising solution for efficient and low-cost prediction of electromagnetic responses in DMS.

The gas sensor based on Micro-electromechanical system (MEMS) technology has the advantages of high sensitivity, small size, good batch uniformity and low power consumption. It has become an important development direction of the next-generation semiconductor gas sensor. This paper focuses on the design of the micro-hotplate chip for MEMS gas sensors. A type of micro-hotplate chip design with an isothermal hot area (±15 K) accounting for 90% is demonstrated through both the thermal theory analysis and the finite element simulation of the physical field, effectively resolving the issue of broad area uniform heating in the micro-sized chip in MEMS gas sensor designs. The infrared thermal image test results show the temperature of four points from edge to center of heating area are 295.5 °C, 287.5 °C, 289 °C, 294.8 °C respectively, which indicates the heat uniformity of micro-hotplate. Due to the limitation of a low temperature film deposition process, the maximum stress of the micro-hotplate films is about 1500 MPa, and at 375 °C operating temperature, the power consumption per area is only 4.5×10-4 mW/μm².

At present, the effective length of CT ionization chamber commonly used in China is 100mm. When the nominal line is wider than 60mm, the measurement accuracy decreases, and multiple step measurements are required, which is troublesome to operate. In this paper, the energy response of the extended CT ionization chamber with different materials is studied. The MCNP5 program is simulated and verified by the standard radiation field energy response experiment. The experimental results show that the energy response of the collector electrode with 1mm Al core coated with 0.1mm graphite is ±2% in the energy range of CT (100-150) kV, which has good application value.

In the process of hypersonic flight in the dense atmosphere, the loading environment of the vehicle is extremely harsh. The air at the front end is severely compressed and fricts with the windward side, resulting in a sharp rise in the surface temperature of the aircraft. With the increase of the flight Mach number, the phenomenon of aerodynamic heating becomes more and more serious. Therefore, the influence of high temperature must be considered in the ground modal test of the aircraft structure. However, the measurement of vibration response signal in high temperature environment has become a big problem. This paper introduces a non-contact vibration signal test method, which is applied to the modal test of carbon fiber resin matrix composites in high temperature environment. The modal parameters of composite materials in different temperature environments are obtained successfully, and good results are achieved.

In order to accurately measure the dose distribution of high-energy X-ray in water, in this paper, based on the analysis of the requirement of high-energy X-ray dose distribution measurement in water, the manufacturing process of the ionization chamber is studied, the structure design of the ionization chamber is completed, and a thimble ionization chamber with a small sensitive volume is developed in the end. It can be seen from the test experiment that the fingertip ionization chamber has good angular response, directivity, repeatability, time linearity and saturation characteristics. The angular response, directivity and repeatability are 2.3%, 0.4% and 0.09%, respectively. The time linearity test shows that the measured values of the developed thimble ionization chamber remain well linear with the cumulative time over the cumulative time of 200 seconds. Through the saturation characteristic test, it can be determined that the saturation region of the developed thimble ionization chamber is 100V~300V, the plateau length is 200V, and the maximum change rate of absorbed dose rate in the saturation region is 0.70%. Through the comparison and verification experiment of the percentage depth dose measurement between this thimble ionization chamber and PTW's thimble ionization chamber, the results show that the maximum dose point depth and X-ray radiation quality of the two kinds of thimble ionization chamber are in good agreement. So the thimble ionization chamber developed in this paper can be applied to the high energy X-ray dose distribution measurement in water, and its technical level is equivalent to that of foreign products such as PTW Company.

A three-dimensional finite-difference time-domain (FDTD) approach for electromagnetic scattering from a randomly rough surface is developed. In order to save computing time, a graphics processing unit (GPU) based parallel computing framework is proposed to accelerate FDTD method. The parallel FDTD method assigns a thread for each Yee cell in the computing region for update calculation, and thus can achieve higher computational efficiency than the serial CPU algorithm. The parallel program is executed by a NVIDIA’s Quadro P5000 card. The FDTD model for calculation of EM wave scattering from rough surface is developed. VV polarization and HV polarization bistatic scattering coefficients are calculated.

A fast and accurate transient stability assessment method is essential for the safe and stable operation of power systems. Existing deep learning algorithms have achieved good results in transient stability assessment (TSA), but the training time of these models is extended as the number of network layers continues to increase. There-fore, in order to improve the accuracy of power system transient stability and the efficiency of operation, the improved graph attention broad learning system (IGAT-BLS) model is proposed in this paper, which combines broad learning system with graph attention network. This model leverages the deep feature extraction capability of graph attention networks to enhance its feature representation, and utilizes the flat structure of broad learning system (BLS) to reduce the complexity of the network model. Simulation experiment of the New England 39-bus benchmark system show that, compared with other TSA methods based on deep learning, the proposed method can make more accurate transient stability judgments with shorter training time.

Aiming at the problem of dose measurement inaccuracy due to the large volume of ionization chamber in gamma knife small field measurement, a micro-volume semiconductor detector for gamma knife dose measurement was developed. The semiconductor detector was designed with six semiconductor detectors attached to a ceramic cylinder, thus avoiding the measurement directivity defect caused by the placement of a single semiconductor detector. At the same time, it also improves the sensitivity of the entire semiconductor detector. The experimental data show that the developed gamma knife semiconductor detector has the characteristics of good angular response, high sensitivity and good stability. The output dose of gamma knife in each field is measured with the developed semiconductor detector, then the calculated field output factor is consistent with the manufacturer's factor, so the developed semiconductor detector is suitable for measuring the output dose in small field.

In order to solve the complex module dependencies of dialogue systems and improve the system's ability to understand deep knowledge in natural language and produce more coherent texts, this paper introduces an end-to-end dialogue system based on large language models. First, low-rank adaption is used to fine-tune sequence-to-sequence large language models, which reduces system complexity and model fine-tuning cost. Then, the training method of reinforcement learning from human feedback is adopted to make the generated responses more aligned with human expectations. Finally, in-context learning is used to adapt to specific tasks, improving model flexibility and adaptability. Experimental results show that the system performs well in both automatic evaluation and practical use and has strong application value.

In order to solve the problem of fire detection in the heat exchange station, an intelligent fire early warning system based on image analysis is proposed, which realizes the network, digital and intelligent design of video monitoring system. This paper introduces the framework of the platform and the functional structure of each subsystem, gives the overall structure diagram of the system, focuses on the basic algorithm of background modeling, flame segmentation and morphological filtering in the fire identification and analysis module, and puts forward the design scheme and workflow of the alarm and linkage module. Experiments show that the system has low complexity and good real-time performance, which can improve the efficiency of fire emergency command in heat exchange station and meet the actual requirements of heating security.

Tracking unmanned aerial vehicles (UAVs) using infrared cameras is an important technology for the anti-UAV task. Most of the existing trackers adopt the local-search method that crops out an image patch for tracking targets, which is not effective in dealing with challenging situations such as target disappearance and tracking failure. And attaching a global detection module to the local tracker or directly using a global tracker will require too much computation and cannot meet the real time requirement. Therefore, we propose a real-time global-search method for tracking UAVs in infrared videos, dubbed RGT, which is based on a one-stage anchor-free deep framework and introduces an attention module to enhance the ability to discriminate between the target and the background. In addition, according to the multiscale output characteristics in our model, we propose an adaptive multi-scale template updating (AMTU) and a multiscale spatial constraint (MSC) to address the problems of target scale variation and background clutter, respectively. We use the 1st, 2nd and 3rd Anti-UAV datasets for training and testing, and implement comparison experiments with 10 deep trackers. Our algorithm shows excellent performance and can run in real time at 30FPS on an NVIDIA RTX 2080Ti GPU.

In harsh environments, the detected target track is easily interrupted to form fragmented track segments, which creates a great challenge for track management and tracking. The traditional track segment association (TSA) method is based on a hypothetical target motion model and utilizes a large amount of a priori information to complete the association task. When the hypothetical a priori model does not match the actual motion pattern, the inference time is too long, and the performance decreases significantly. In this article, we propose a track segment association algorithm based on Gaussian regression analysis, in which the new and old segments of the track are back predicted to the associated moment for pairwise discrimination respectively. The Hungarian algorithm is introduced to solve the correlation problem of the track, and the track correlation problem is transformed into an allocation problem by establishing the correlation matrix, and the Hungarian algorithm is used to find the optimal solution. The results show that the proposed method can effectively improve the accuracy of the discrimination.

In 2022, lots of tweets consisting of emoji squares and a few words describing moods emerge on Twitter. In fact, it is the result of a word guessing game called Wordle. In order to figure out the relationship among words in Wordle, this paper tries to study the feedback data of its players and extract useful information to further improve the design of Wordle as well as other games of its kind. Specifically, from the perspective of word structure, strict and fuzzy matching algorithms were designed to simulate the data distribution of target word, while some common structure features of words were extracted and weighted by word frequency. On one hand, the data distribution of the target word is simulated with reference to the words with the most similar structures. On the other hand, the difficulty of a word in the Wordle game was quantified by analyzing the number of player’s tries, and then the difficulty coefficient of each word was calculated. Combined with the above numerical algorithms, the difficulty of the target word was evaluated objectively by the difficulty of its similar word sequences. Experiments showed that the proposed framework achieved good performance on the Wordle dataset. Our work was awarded Honorable Mention in 2023 Mathematical Contest in Modeling (MCM).

This article proposes a cooperative optimization algorithm of task allocation and trajectory planning for multi-mission in heterogeneous unmanned aerial vehicle (UAV) cluster. The basis of the proffered algorithm is to establish constraint and threat models to simultaneously minimize range, maximize value gain and survival probability under the constraints of task payload, range and task requirement. With this, derive the objective function for heterogeneous UAV cluster within multi-mission, and adopt it as a metric for assessing the performance of the cooperative optimization in task allocation and trajectory planning. It is revealed that the formulated cooperative optimization problem is a multi-objective, nonlinear and non-convex optimization model due to its multiple decision variables and constraints. By introducing the roulette wheel selection (RWS) principle and the elite strategy (ES), an ant colony optimization (ACO) with ES the capability to solve the complex optimization model. The simulation results indicate that the proposed algorithm is superior and more efficient compared to other approaches.