Inspired by the recent rapid progress of l1-norm minimization techniques and the great success of sparse dictionary
learning in image modeling, this paper proposes a sparse multi-way models clustering fusion technique to improve the
classification performance in hyperspectral imagery. Multi-way models consider hyperspectral imagery data as a whole
entity to treat jointly spatial and spectral modes. The whole clustering fusion method is composed three steps. Firstly, the
complete hyperspectral data is grouped into several independent sub-band data sources. Then, sparse multi-way model is
used to feature extraction in every band set, and divide the scene into a series of homomorphic regions. At last, we
propose a fusion method to combine the information provided by each band set, it can acquire approximate supervised
classification performance (such as K-nearest Neighbor classifier).The experimental results on the HYDICE imagery
demonstrate the efficiency and superiority of the proposed clustering method to the classical K-means clustering method.
The classification of hyperspectral image data has drawn much attention in recent years.
Consequently, it contains not only spectral information of objects, but also spatial arrangement of
objects. The most established Hyperspectral classifiers are based on the observed spectral signal, and
ignore the spatial relations among observations. Information captured in neighboring locations may
provide useful supplementary knowledge for analysis. To combine the spectral and spatial information
in the classification process, in this paper, a Multidimensional Local Spatial Autocorrelation (MLSA) is
proposed for hyperspectral image data. Based on this measure, a collaborative classification method is
proposed, which integrates the spectral and spatial autocorrelation during the decision-making process.
The trials of our experiment are conducted on two scenes, one from HYDICE 210-band imagery
collected over an area that contains a diverse range of terrain features and the other is toy car
hyperspectral image captured at Instrumentation and Sensing Laboratory (ISL) at Beltsville
Agricultural Research Center. Quantitative measures of local consistency (smoothness) and global
labeling, along with class maps, demonstrate the benefits of applying this method for unsupervised and
In the past few years, imaging spectroscopy has been used widely. However, it only acquires intensity information in a narrow electromagnetic band, ignoring the polarimetric information of the electromagnetic wave, resulting in inaccurate material classification. Imaging spectropolarimetric technology as a new sensing method can acquire the polarimetric information at a narrow electromagnetic band sequence, but there are few results showing how to combine the redundant and complementary features provided by spectropolarimetric imagery. In this paper, an unsupervised spectropolarimetric imagery classification method is proposed to jointly utilize the spatial, spectral and polarimetric information to make material classification more accurate. First, a spectropolarimetric projection scheme is proposed to divide the spectropolarimetric data set into two parts: a polarimetric spectrum data set and a polarimetric data cube. Then, a kernel fuzzy c-means clustering method is used to cluster the polarimetric spectrum data set and polarimetric data cubes. At last, kernel fuzzy c-means clustering results are combined by evidence reasoning to get better clustering performance. Through experimentation and simulation, the effects of classifying different materials with similar surface colour can be enhanced greatly.
This paper presents an algorithm to calculate polarized images, based on spatially adaptive wavelet analysis, in which image fusion theory is used. According to the principle and method of polarization imaging, the shortcomings of traditional methods in preserving detail information, removing the noise, and dealing with the misalignment of components in polarimetry are analyzed. Polarized-image calculation is a special case of image fusion, in which the combination rule is fixed. At the same time, wavelet-based image fusion method has a special advantage in acquiring rich detail information. To remove the effects of noise, we propose a spatially adaptive wavelet transform method. Then this method is extended to translation-invariant wavelets, which yield better results than the orthogonal wavelet transform when there is misalignment among components in polarimetry. Experiment and simulation results show that spatially adaptive wavelet-based polarization imaging yields significantly superior image quality to the traditional method.
The detection of low signature objects in cluttered backgrounds is a crucial problem in remote sensing. In the past few years, imaging spectral and polarimetric sensors have been evaluated for this application. As the reflection or emission spectral signatures depend on the elemental composition of objects residing within the scene. The polarization state of radiation is sensitive to surface features such as relative smoothness or roughness. But each character (spectral, polarimetric or spatial character) giving an incomplete representation of an object of interest, it expected that the combination of complementary and redundant characters would be contributed to reduce the false alarm rate, improve the confidence in the target identification and the quality of the scene description as a whole. Imaging spectropolarimetry provides effective mean to acquire spatial, spectral and polarimetric information of scene. This paper presents a study of spectropolarimetric image data set recorded from imaging spectropolarimeter located on top of building. The low probability detection algorithm was separately applied to polarimetric data sets of each band (Stokes images, degree of polarization image and angle of polarization image ) to obtain a series of two dimensional map of objects and false detection. As there are some conflictions among these maps, D-S reasoning is used to combine these maps to improve the detection rate and low false rate. Through experiment and simulation, we conclude that this fusion algorithm can be well applied to enhance the detection performance.
Polarization imparted by surface reflections contains unique and discriminatory signatures which may augment spectral target-detection techniques. With the development of multi-band polarization imaging technology, it is becoming more and more important on how to integrate polarimetric, spatial and spectral information to improve target discrimination. In this study, investigations were performed on combining multi-band polarimetric images through false color mapping and wavelet integrated image fusion method. The objective of this effort was to extend the investigation of the use of polarized light to target detection and material classification. As there is great variation in polarization in and between each of the bandpasses, and this variation is comparable to the magnitude of the variation intensity. At the same time, the contrast in polarization is greater than for intensity, and that polarization contrast increases as intensity contrast decreases. It is also pointed out that chromaticity can be used to make targets stand out more clearly against background, and material can be divided into conductor and nonconductor through polarization information. So, through false color mapping, the difference part of polarimetric information between each of the bandpasses and common part of polarimetric information in each of the bandpasses are combined, in the resulting image the conductor and nonconductor are assigned different color. Then panchromatic polarimetric images are fused with resulting image through wavelet decomposition, the final fused image have more detail information and more easy identification. This study demonstrated, using digital image data collected by imaging spectropolarimeter, multi-band imaging polarimetry is likely to provide an advantage in target detection and material classification.
The detection of low signature or camouflaged objects in cluttered backgrounds is a crucial problem in tactical reconnaissance. In the past few years, imaging spectral and polarimetric sensors have been evaluated for this application. As the reflection or emission spectral signatures depend on the elemental composition of objects residing within the scene. And the radiation polarization is sensitive to surface features such as relative smoothness or roughness. But each character giving an incomplete representation of an object of interest, it expected that the combination of complementary and redundant characters would be contributed to reduce the false alarm rate, improve the confidence in the target identification and the quality of the scene description as a whole. Anomaly detection and fuzzy integral are used in this paper to combine the spectral and polarimetric information captured by imaging spectrometer and polarimeter. Through experiment and simulation, the effects of detection of low signature or camouflaged targets can be enhanced greatly.
With the development of sensory technology, new image sensors have been introduced that provide a greater range of information to users. But as the power limitation of radiation, there will always be some trade-off between spatial and spectral resolution in the image captured by specific sensors. Images with high spatial resolution can locate objects with high accuracy, whereas images with high spectral resolution can be used to identify the materials. Many applications in remote sensing require fusing low-resolution imaging spectral images with panchromatic images to identify materials at high resolution in clutter. A pixel-based false color mapping and wavelet transform integrated fusion algorithm is presented in this paper, the resulting images have a higher information content than each of the original images and retain sensor-specific image information. The simulation results show that this algorithm can enhance the visibility of certain details and preserve the difference of different materials.