This PDF file contains the front matter associated with SPIE Proceedings Volume 11735, including the Title Page, Copyright Information, and Table of Contents.

Flashover is a dangerous phenomenon caused by near-simultaneous ignition of exposed materials. It is one of the major causes of firefighter fatalities. Research has been done using CMOS vision cameras combined with thermal sensors to perform remote detection and dynamic classification of fire and smoke patterns. Tests and experiments have been done to detect fire and smoke remotely. The inexpensive visible and infrared sensors used in the tests corroborate and closely follow the detailed trends recorded by the more expensive (and less mobile) radiometers and thermocouples. Deep neural networks (DNN) have been used to detect, classify and track fire and smoke areas. Real-time segmentation is utilized to measure the fire and smoke boundaries. The segmentations are used to dynamically monitor fluctuations in temperature, fire size and smoke progression in the monitored areas. A fire and smoke progression curve has been drawn to predict the flashover point. In the paper, data analysis and preliminary results will be shown.

In this paper, we propose architectures for the implementation 16 Boolean optical gates from two inputs using externally pumped phase- conjugate Michelson interferometer. Depending on the gate to be implemented, some require single stage interferometer and others require two stages interferometer. The proposed optical gates can be used in several applications in optical networks including, but not limited to, all-optical packet routers switching, and all-optical error detection. The optical logic gates can also be used in recognition of noiseless rotation and scale invariant objects such as finger prints for home land security applications.

The interference theory is developed for of the phase conjugate Michelson interferometer in which its ordinary mirrors are replaced by a single externally pumped phase conjugate mirror. According to the theory, it was found that for an interferometer with two equal arms, the path length difference depends solely on the initial alignment of the two input beams, and the vertical alignment readout. Small vertical misalignments in the readout beam by mrad causes a huge change in the phase difference in the phase between the two interferometer arms beam. The phase difference is proportional to the interferometer arm lengths. The overlap between the phase conjugate beams is not affected by the interferometer beam alignment. The interferometer is proposed for nondestructive testing and the design all optical logic and associated fuzzy logic for ultrafast optical pattern recognition.

To address safety issues caused by the increase availability and application of UAVs, automated detection systems are required. While such systems typically rely on a variety of sensor modalities, EO sensors are often a key modality. This is due to their comparatively low cost and direct intepretability by human operators. Besides automated detection of UAVs in EO imagery, classification of a UAV’s type is an important task to assess the degree of potential safety risk. While the applicability of deep learning based UAV detection in EO imagery has already been demonstrated, this work is the first to examine the potential of deep learning based UAV type classification in EO imagery. We evaluate multiple deep learning based detectors trained to classify UAV types as well as a cascade of an initial detector followed by a separate classifier. Our evaluation is carried out on publicly available data, supplemented by a few self-recorded sequences, and focuses on aspects like class balancing, UAV size, image scale and classification backbone architectures. We further analyse a class confusion matrix to better understand occurring classification errors.

Over the past decade, deep learning approaches have been applied to the detection of malicious software, otherwise known as malware. Despite their improved performance compared to conventional detection methods such as static and dynamic analysis, however, deep learning-based malware detection systems have been shown to be vulnerable to adversarial attacks. Few image-based malware detection systems have been proposed, especially those that evaluate their performance against adversarial attacks. Furthermore, little research has been done beyond the classification of malware targeted at Windows (PE) or Android systems, leaving entire realms such as Mac (Mach-O), Linux (ELF), and embedded software unexplored and unprotected. These realms, specifically embedded software, are used in critical technology such as avionic systems and special care must be taken to ensure their safety. In this paper, we present an image-based malware detection system on PE, ELF, Mach- O, and embedded C code files. The system’s architecture incorporates layers of encoders that are taken from independently-trained autoencoders and multi-layer perceptron that returns the output of the network. We evaluate the performance of the system against adversarial attacks, or the misclassification of a malware file as a benign, by adding gradient based perturbations to unused sections of the malware often referred to as the slack bits. The network achieves an accuracy of 96.51% on non-adversarial PE and ELF files, 95.45% on transfer learned non-adversarial Mach-O files, and 99.2% on transfer learned non-adversarial synthetic plane files. For the classification of adversarial examples, the network achieved a 81% success rate of misclassification on adversarial PE and ELF files and a 99% success rate of misclassification on adversarial synthetic plane files.

Aerial object detection is one of the most important applications in computer vision. We propose a deep learning strategy for detection and classification of objects on the pipeline right of ways by analyzing aerial images captured by flying aircrafts or drones. Due to the limitation of sufficient aerial datasets for accurately training the deep learning systems, it is necessary to create an efficient methodology for object data augmentation of the training dataset to achieve robust performance in various environmental conditions. Another limitation is the computing hardware that could be installed on the aircraft, especially when it is a drone. Hence a balance between the effectiveness and efficiency of object detector needs to be considered. We propose an efficient weighted IOU NMS (intersection over union non-maxima suppression) method to speed up the post-processing time that satisfies the onboard processing requirement. Weighted IOU NMS utilizes confidence scores of all proposed bounding boxes to regenerate a mean box in parallel. It processes the bounding box score at the same instant without removing the bounding box or decreasing the bounding box score. We perform both quantitative and qualitative evaluations of our network architecture on multiple aerial datasets. The experimental results show that our proposed framework achieves better accuracy than the state-of-the-art methods for aerial object detection in various environmental conditions.

This paper present class activation mapping in conjunction with transfer learning to investigate and explain the predictions of a deep learning based satellite image classification. Deep learning based classifiers offer no way of gauging what a network has learned or which part of an input to the network was responsible for the prediction of the network. When these models fail and give incorrect predictions, they often fail spectacularly without any warning or explanation. The presented class activation mapping (CAM) technique can address this issue and provide the visual explanations of the predictions of convolutional neural networks. In addition, the proposed method can also reveal what specific part of an input image can confuse the network resulting an incorrect prediction. The proposed approach employing transfer learning using ImageNet pretrained models is implemented on the xView dataset. The experimental results are very promising and provide an insight into the satellite based image classification.

As one of the classic fields of computer vision, image classification has been booming with the improvement of chip performance and algorithm efficiency. With the rapid progress of deep learning in recent decades, remote sensing land cover and land use image classification has ushered in a golden period of development. This paper presents a new deep learning classifier to classify remote sensing land cover and land use images. The approach first uses multi-layer convolutional neural networks to extract the image features, attached through a fully-connected neural network to generate the sample loss. Then, a hard sample memory pool is created to collect the samples with large losses during the training. A batch of hard samples is randomly extracted from the memory pool to participate in the training of the convolutional fully connected model so that the model becomes more robust. Our method is validated by testing the classic remote sensing land cover and land use dataset. Compared with the previous popular classification algorithm, our algorithm can classify images more accurately with a shorter training iteration.

Stone accumulation in kidney is a typical disease/ sickness in most countries all around the world. Its frequency rate is continually expanding. It has been observed that, the classification of renal stones prompts an imperative decrease of the re-occurrence. The classification of stones based on particular texture, surface highlights and lab examinations are a standout amongst the most utilized strategies. In this paper we use dataset of explicitly intended for top captured pictures of 454 expelled kidney stones which extracted through urinary or surgical procedure for classification purpose. In this paper different techniques have been learned and applied the specialist’s defined framework to arrange them into defined classes then perform classification process. In this paper we use feature fusion technique to collect as much as possible features. We select VGG16, InceptionV3 and ALEX features for fusion using serial feature fusion method. We choose C-SVM and F-KNN classifier to get improved accuracy of same dataset and predict better correctness’s with the possibilities of expansion of the dataset measure. In initial testing classification accuracy recorded at 83.43%, FNR 19.25%, Precision Rate 88.48% and Sensitivity of 86.51% on CSVM, later on the best testing classification accuracy recorded at 99.5%, FNR 0.1%, Precision Rate 99.90% and Sensitivity of 99.96% on F-KNN.

We explore different 3D imaging techniques suitable for topographical measurement of partially reflective metal surfaces produced in an additive manufacturing environment. Three candidate techniques are investigated: digital holography, ptychographic coherent diffractive imaging, and structured light. The pros and cons of each method in terms of applicability in the AM environment are discussed. Selected simulation results and experimental findings are reported.

In this article, we explore the role and usefulness of parts-based spatial concept learning about complex scenes. Specifically, we consider the process of teaching a spatially attributed graph how to utilize parts-detectors and relative positions as attributes in order to learn concepts and to produce human oriented explanations. First, we endow the graph with parts detectors and relative positions to determine the possible range of attributes that will limit the types of concepts that are learned. Next, we enable the graph to learn concepts in the context of recognizing structured objects in imagery and the spatial relations between these objects. As the graph is learning concepts, we allow human operators to give feedback on attribute knowledge, creating a system that can augment expert knowledge for any similar task. Effectively, we show how to perform online concept learning of a spatially attributed graph. This route was chosen due to the vast representational capabilities of attributed graphs, and the low-data requirement of online learning. Finally, we explore how well this method lends itself to human augmentation, leveraging human expertise to perform otherwise difficult tasks for a machine. Our experiments shed light on the usefulness of spatially attributed graphs utilizing online concept learning, and shows the way forward for more explainable image reasoning machines.

Cities development accelerates with galloping urbanization on the surface on the world [1,2]. They must face significant threats linked to risks of human origin, like terrorism. In this paper, we present our approach for intrusion detection composed of 3 phases. The first one consists in selecting, via a GUI interface, zones supposed to be prohibited zones in an image. The second one, based on a Neural Network method, is applied for the person detection. The third one verifies if the detected person is present in one of the prohibited zones or not. If so, an alarm goes off automatically. Real tests were performed to secure an elementary school in the city of Nice in France. The obtained results showed the efficiency of our method in terms of good detection. Other work is in progress with the aim of deeply analyzing the intrusion to detect the abnormal behavior.

Behavioral analysis in an urban environment is a complex task that requires material and human resources, due to the difficulty of interpreting the situations. This paper presents a method to improve the detection of dangerous behaviors by assisting surveillance stations. Our objective is to alert when one of these behaviors is captured by a surveillance camera. To do this, we analyze the positions and paths of the persons in a global way, through a group of parameters. These parameters are determined by an automatic image analysis algorithm such as DBSCAN computed on an NVIDIA Jetson TX2. This analysis allows to detect, through the evolution and clustering of points in each cloud, phenomena qualified as abnormal, such as dispersion and rapid clustering, as well as poaching. The data used to feed our algorithm come from simulations that allow testing new and different scenarios. The performance of our proposed method is evaluated on videos representing real case situations.

In this paper, we propose a U-net architecture that integrates a residual skip connections and recurrent feedback with EfficientNet as a pretrained encoder. Residual connections help feature propagation in deep neural networks and significantly improve performance against networks with a similar number of parameters while recurrent connections ameliorate gradient learning. We also propose a second model that utilizes densely connected layers aiding deeper neural networks. EfficientNet is a family of powerful pretrained encoders that streamline neural network design. The proposed networks are evaluated against state-of-the-art deep learning based segmentation techniques to demonstrate their superior performance.

Nowadays, due to various challenges such as large-scale variation of population, mutual occlusion, perspective distortion and so on, crowd counting has gradually become a hot issue in computer vision. To address the large- scale variation exists in the images, in this paper, we propose a novel multi-scale network called MSNet which aims to maintain continuous variations and count the number of pedestrians accurately. While most state-of-the- arts multi-scale and multi-column networks aim to integrate the scale information of heads with different size, lots of researches still need to do to achieve continuous variations. In MSNet, specifically, the first ten layers of the visual geometry group network(VGG) are used as the backbone to extract the rough features of images and a multi-scale block is employed to maintain the scale information which contains several receptive kernels to obtain a better performance towards the difficulty of scale-variation. Inspired by the knowledge that using multiple small receptive field kernels to replace a single large receptive field will get a better performance, we utilize two dilated convolutions with the receptive field of 5 to replace the large kernel. Our MSNet has moderate increase in computation, and we evaluate our method on three benchmark datasets including ShanghaiTech (Part A: MAE=59.6, RMSE=96.1; Part B: MAE=7.5, RMSE=12.1), UCF-CC-50(MAE=207.9, RMSE=273.8) and UCF-QNRF(MAE=93, RMSE=158) to show the outperformance of our method.

Current models for determining dementia progression are network diffusion models derived from the heat equation without diffusion sources, and they do not model the disease agents (misfolded β-amyloid and τ -protein) transmission dynamics. In this paper, we derive from a SIRI (Susceptible-Infected-Recovered-Infected) epidemic model a simplified model under the information-centric paradigm over a network of heterogeneous agents and including the long-range dispersal of disease agents. The long-range disease agent dispersal is implemented by including the Mellin and Laplace transforms in the adjacency matrix of the graph network. We analyze the influence of different transforms on the epidemic threshold which shows when a disease dies off. Further we analyze the dynamical properties of this novel model and prove new conditions on the structure of the network and model parameters that distinguish important dynamic regimes such as endemic, epidemic and infection-free. We demonstrate how this model can be used for disease prediction and how control strategies can be developed for disease mitigation.

As the technological advances of the last decade have led to increased performance and availability of video cameras, along with the rise of deep learning-based image recognition, the task of person re-identification has almost exclusively been studied on datasets with ground-based, static camera settings. Yet re-identification applications on aerial-based data captured by Unmanned Aerial Vehicles (UAVs) can be particularly valuable for monitoring public events, border protection, and law enforcement. For a long time no publicly available UAV-based re-identification datasets of sufficient size for modern machine learning techniques existed, which prevented research in this area. Recently, however, two new large-scale UAV-based datasets have been released. We examine re-identification performances of common neural networks on the newly released PRAI-1581 and P-DESTRE aerial-based datasets for UAV-related error sources and data augmentation strategies to increase robustness against them. Our findings of common error sources for these UAV-based datasets include occlusions, camera angles, bad poses, and low resolutions. Furthermore, data augmentation techniques such as rotating images during training prove to be a promising aid for training on the UAV-based data with varying camera angles. By carefully selecting robust networks in addition to choosing adequate training parameters and data augmentation strategies we are able to surpass the original re-identification accuracies published by the authors of the PRAI-1581 and the P-DESTRE dataset respectively.

In several particularly secure applications such as the entrance to a school, it is important to know whether the person entering is an adult or a child. In this article, we propose a human body morphology detector that distinguishes whether the person is an adult or a child. This detector could be included in a smart portal to detect whether the entry person is an adult or a child to apply a different treatment depending on the morphology. A person detector module1 is deployed to detect the presence of a person within a predefined radius. When the location of the person is detected, our system can measure the height of the person and determine if the person is an adult or a child based on its height.

A new coherent superposition based single-channel color image encryption using gamma distribution and biometric phase keys, is proposed. In this method, first the original color image is split into R, G, and B channels and corresponding iris print image is also split into R, G, and B channels. Second, each channel of iris print image is phase encoded to use as biometric phase key. Third, the biometric phase key of each channel is independently multiplied with gamma distribution phase mask. Finally the original color channel can be directly separated into two phase masks: one is a combination of gamma distribution and biometric phase keys and the other is a modulation of the combination by the original color channel. The remarkable benefit of the proposed scheme is introduction of gamma distribution and biometric phase keys. The proposed system can be implemented by using a simple hybrid optoelectronic system. Numerical simulation results validate the feasibility and effectiveness of the proposed method.

Pneumonia, an infectious disease that can influence the lungs, is a severe medical field topic. Therefore, how to correctly classify images of pneumonia is very important. The limitations of traditional machine learning algorithms and the significant improvement of computing performance make deep learning widely used. At present, using a convolutional neural network to classify pneumonia is still the mainstream method. This paper provides a modified capsule network to detect and classify pneumonia by using X-ray pictures. The model consists of two parts: encoder and decoder. Encoder contains convolutional layer, primary capsule layer, and digital capsule layer. The primary capsule layer and digital capsule layer convert a scalar into a vector and then try to cluster vectors of the same category by dynamic routing. The decoder contains a deconvolutional layer. The image is reconstructed by up-sampling the vector generated by the encoder, and the reconstructed image is compared with the original image to make the features extracted by the encoder more representative. The training and testing process takes place on the dataset "Labeled Optical Coherence Tomography (OCT) and Chest XRay Images for Classification." This dataset contains a total of 5856 pictures. We divide the images into the training set and testing set at a ratio of 8:2. The accuracy rate on this dataset is 98.6%. This model has a more straightforward structure and fewer parameters than other popular models, which means that it can be more easily deployed in various conditions in practical applications.