Hsuan Ren

Automatic spectral target recognition in hyperspectral imagery

Automatic target recognition (ATR) in hyperspectral imagery is a challenging problem due to recent advances of remote sensing instruments which have significantly improved sensors spectral resolution. As a result, small and subtle targets can be uncovered and extracted from image scenes, which may not be identified by prior knowledge. In particular, when target size is smaller than pixel resolution, target recognition must be carried out at subpixel level. Under such circumstance, traditional spatial-based image processing techniques are generally not applicable and may not perform well if they are applied. The work presented here investigates this issue and develops spectral-based algorithms for automatic spectral target recognition (ASTR) in hyperspectral imagery with no required a priori knowledge, specifically, in reconnaissance and surveillance applications. The proposed ASTR consists of two stage processes, automatic target generation process (ATGP) followed by target classification process (TCP). The ATGP generates a set of targets from image data in an unsupervised manner which will subsequently be classified by the TCP. Depending upon how an initial target is selected in ATGP, two versions of the ASTR can be implemented, referred to as desired target detection and classification algorithm (DTDCA) and automatic target detection and classification algorithm (ATDCA). The former can be used to search for a specific target in unknown scenes while the latter can be used to detect anomalies in blind environments. In order to evaluate their performance, a comparative and quantitative study using real hyperspectral images is conducted for analysis.

New hyperspectral discrimination measure for spectral characterization

The spectral angle mapper (SAM) has been widely used in multispectral and hyperspectral image analysis to measure spectral simi- larity between substance signatures for material identification. It has been shown that the SAM is essentially the Euclidean distance when the spectral angle is small. Most recently, a stochastic measure, called the spectral information divergence (SID), has been suggested to model the spectrum of a hyperspectral image pixel as a probability distribution, so that spectral variations among spectral bands can be captured more effectively in a stochastic manner. This paper develops a new hyper- spectral spectral discrimination measure, which combines the SID and the SAM into a mixed measure. More specifically, letr and r8 denote two hyperspectral image pixel vectors with their corresponding spectra speci- fied bys and s8. Then SAM(s,s8) measures the spectral angle between s and s8. Similarly, SID(s,s8) measures the information divergence be- tween the probability distributions generated by s and s8. The proposed new measure, referred to as the SID-SAM mixed measure, can be imple- mented in two versions, given by SID(s,s8)3tan(SAM(s,s8)) and SID(s,s8)3sin(SAM(s,s8)), where tan and sin are the usual trigonomet- ric functions. The spectral discriminability of such a mixed measure is greatly enhanced by multiplying the spectral abilities of the two mea- sures. In order to demonstrate its utility, a comparative study is con- ducted among the SID-SAM mixed measure, the SID, and the SAM. Our experimental results have shown that the discriminatory ability of the (SID,SAM) mixed measure can be a significant improvement over the SID and SAM.

Real-time processing algorithms for target detection and classification in hyperspectral imagery

The authors present a linearly constrained minimum variance (TCMV) beamforming approach to real time processing algorithms for target detection and classification in hyperspectral imagery. The only required knowledge for these LCMV-based algorithms is targets of interest. The idea is to design a finite impulse response (FIR) filter to pass through these targets using a set of linear constraints while also minimizing the variance resulting from unknown signal sources. Two particular LCMV-based target detectors, the constrained energy minimization (CEM) and the target-constrained interference-minimization filter (TCIMF), are presented. In order to expand the ability of the LCMV-based target detectors to classification, the LCMV approach is further generalized so that the targets can be detected and classified simultaneously. By taking advantage of the LCMV-based filter structure, the LCMV-based target detectors and classifiers can be implemented by a QR-decomposition and be processed line-by-line in real time. The experiments using HYDICE and AVIRIS data are conducted to demonstrate their real time implementation.

A comparative study for orthogonal subspace projection and constrained energy minimization

We conduct a comparative study and investigate the relationship between two well-known techniques in hyperspectral image detection and classification: orthogonal subspace projection (OSP) and constrained energy minimization. It is shown that they are closely related and essentially equivalent provided that the noise is white with large SNR. Based on this relationship, the performance of OSP can be improved via data-whitening and noise-whitening processes.

Target-constrained interference-minimized approach to subpixel target detection for hyperspectral images

Hsuan Ren Chein-I Chang, MEMBER SPIE Remote Sensing Signal and Image Processing Laboratory Department of Computer Science and Electrical Engineering University of Maryland Baltimore County Baltimore, Maryland 21250 Abstract. Due to significantly improved spatial and spectral resolution, hyperspectral sensors can now detect many substances that cannot be resolved by multispectral sensors. However, this comes at the price that many unknown and unidentified signal sources, referred to as interferers, may also be extracted unexpectedly. Such interferers generally produce additional noise effects on target detection and must therefore be taken into account. The problem associated with this interference is challenging because its nature is generally unknown and it cannot be identified from an image scene. This paper presents an approach, called the target-constrained interference-minimized filter (TCIMF), which does not require one to identify interferers, but can minimize the effects caused by interference. It designs a finite-impulse-response filter that specifies targets of interest in such a way that the desired targets and undesired targets will be passed through and rejected by the filter, respectively; the filter output energy resulting from unknown signal sources is also minimized. More precisely, the TCIMF accomplishes three tasks simultaneously: detection of the desired targets, elimination of the undesired targets, and minimization of interfering effects. A recently developed technique, constrained energy minimization (CEM), can be considered as a suboptimal version of the TCIMF. Computer simulations and hyperspectral image experiments are conducted to demonstrate advantages of the TCIMF over the CEM.

Automatic target recognition for hyperspectral imagery using high-order statistics

Due to recent advances in hyperspectral imaging sensors many subtle unknown signal sources that cannot be resolved by multispectral sensors can be now uncovered for target detection, discrimination, and identification. Because the information about such sources is generally not available, automatic target recognition (ATR) presents a great challenge to hyperspectral image analysts. Many approaches developed for ATR are based on second-order statistics in the past years. This paper investigates ATR techniques using high order statistics. For ATR in hyperspectral imagery, most interesting targets usually occur with low probabilities and small population and they generally cannot be described by second-order statistics. Under such circumstances, using high-order statistics to perform target detection have been shown by experiments in this paper to be more effective than using second order statistics. In order to further address a challenging issue in determining the number of signal sources needed to be detected, a recently developed concept of virtual dimensionality (VD) is used to estimate this number. The experiments demonstrate that using high-order statistics-based techniques in conjunction with the VD to perform ATR are indeed very effective

Generalized constrained energy minimization approach to subpixel target detection for multispectral imagery

Subpixel detection in multispectral imagery presents a chal- lenging problem due to relatively low spatial and spectral resolution. We present a generalized constrained energy minimization (GCEM) ap- proach to detecting targets in multispectral imagery at subpixel level. GCEM is a hybrid technique that combines a constrained energy mini- mization (CEM) method developed for hyperspectral image classification with a dimensionality expansion (DE) approach resulting from a gener- alized orthogonal subspace projection (GOSP) developed for multispec- tral image classification. DE enables us to generate additional bands from original multispectral images nonlinearly so that CEM can be used for subpixel detection to extract targets embedded in multispectral im- ages. CEM has been successfully applied to hyperspectral target detec- tion and image classification. Its applicability to multispectral imagery is yet to be investigated. A potential limitation of CEM on multispectral imagery is the effectiveness of interference elimination due to the lack of sufficient dimensionality. DE is introduced to mitigate this problem by expanding the original data dimensionality. Experiments show that the proposed GCEM detects targets more effectively than GOSP and CEM without dimensionality expansion.

Data fusion of hyperspectral and SAR images

A novel technique is proposed for data fusion of earth remote sensing. The method is developed for land cover classification based on fusion of remote sensing images of the same scene collected from multiple sources. It presents a framework for fusion of multisource remote sensing images, which consists of two algorithms, referred to as the greedy modular eigenspace (GME) and the feature scale uniformity transformation (FSUT). The GME method is designed to extract features by a simple and efficient GME feature module, while the FSUT is performed to fuse most correlated features from different data sources. Finally, an optimal positive Boolean function based multiclass classifier is further developed for classification. It utilizes the positive and negative sample learning ability of the minimum classification error criteria to improve classification accuracy. The performance of the proposed method is evaluated by fusing MODIS/ASTER airborne simulator (MASTER) images and the airborne synthetic aperture radar (SAR) images for land cover classification during the PacRim II campaign. Experimental results demonstrate that the proposed fusion approach is an effective method for land cover classification in earth remote sensing, and improves the precision of image classification significantly compared to conventional single source classification.

New hyperspectral discrimination measure for spectral similarity

Spectral angle mapper (SAM) has been widely used as a spectral similarity measure for multispectral and hyperspectral image analysis. It has been shown to be equivalent to Euclidean distance when the spectral angle is relatively small. Most recently, a stochastic measure, called spectral information divergence (SID) has been introduced to model the spectrum of a hyperspectral image pixel as a probability distribution so that spectral variations can be captured more effectively in a stochastic manner. This paper develops a new hyperspectral spectral discriminant measure, which is a mixture of SID and SAM. More specifically, let xi and xj denote two hyperspectral image pixel vectors with their corresponding spectra specified by si and sj. SAM is the spectral angle of xi and xj and is defined by [SAM(si,sj)]. Similarly, SID measures the information divergence between xi and xj and is defined by [SID(si,sj)]. The new measure, referred to as (SID,SAM)-mixed measure has two variations defined by SID(si,sj)xtan(SAM(si,sj)] and SID(si,sj)xsin[SAM(si,sj)] where tan [SAM(si,sj)] and sin[SAM(si,sj)] are the tangent and the sine of the angle between vectors x and y. The advantage of the developed (SID,SAM)-mixed measure combines both strengths of SID and SAM in spectral discriminability. In order to demonstrate its utility, a comparative study is conducted among the new measure, SID and SAM where the discriminatory power of the (SID,SAM)-mixed measure is significantly improved over SID and SAM.

An ROC analysis for subpixel detection

ROC (Receiver Operating Characteristic) analysis has been widely used to evaluate detection performance. It is based on the Neyman-Pearson detection theory, which solves binary hypothesis testing problems. In mixed pixel classification many algorithms that are developed to estimate abundance fractions (of image endmembers) generally produce gray scale images. As a result, they are not directly applied to hypothesis testing problems. Instead of using the standard ROC curve generated by the detection power versus the false alarm probability, a 3-dimensional (3D) ROC curve is developed in this paper for subpixel detection. It is a 3D plot derived from the mean-detection probability versus the mean-false alarm rate with the third dimension specified by abundance fractions produced by subpixel detection algorithms. In order to illustrate the utility of the proposed 3D ROC analysis in subpixel detection, several linear unmixing-based algorithms are used for performance evaluation.