Subpixel Target Research Articles

Performance Prediction of Matched Filter and Adaptive Cosine Estimator Hyperspectral Target Detectors

Many applications of hyperspectral remote sensing involve the detection of subpixel targets for search and rescue or defense and intelligence operations. The design and potential capabilities of these systems depends on their target detection performance. Therefore, it is important to have tools that reliably predict the performance of target detection systems under different realistic situations. The purpose of this paper is to present a hyperspectral target performance prediction model for the widely used matched filter (MF) and adaptive cosine estimator (ACE) detectors. We use a replacement signal model for resolved and subpixel targets and a finite probability mixture of t-elliptically contoured distributions ( t-ECDs) for the background. A major contribution of this paper is the development of a robust analytical and numerical approach to determine the output distribution of ACE for mixtures of t-ECDs. The proposed technique can be a very useful tool for evaluating target detection performance for highly complex backgrounds.

Compression of Hyperspectral Images Containing a Subpixel Target

Hyperspectral (HS) image sensors measure the reflectance of each pixel at a large number of narrow spectral bands, creating a three-dimensional representation of the captured scene. The HS image (HSI) consumes a great amount of storage space and transmission time. Hence, it would be desirable to reduce the image representation to the extent possible using a compression method appropriate to the usage and processing of the image. Many compression methods have been proposed aiming at different applications and fields. This research focuses on the lossy compression of images that contain subpixel targets. This target type requires minimum compression loss over the spatial dimension in order to preserve the target, and the maximum possible spectral compression that would still enable target detection. For this target type, we propose the PCA-DCT (principle component analysis followed by the discrete cosine transform) compression method. It combines the PCA's ability to extract the background from a small number of components, with the individual spectral compression of each pixel of the residual image, obtained by excluding the background from the HSI, using quantized DCT coefficients. The compression method is kept simple for fast processing and implementation, and considers lossy compression only on the spectral axis. The spectral compression achieves a compression ratio of over 20. The popular Reed-Xiaoli (RX) algorithm and the improved quasi-local RX (RX <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">QLC</sub> ) are used as target detection methods. The detection performance is evaluated using receiver operating characteristics (ROC) curve generation. The proposed compression method achieves maintained and enhanced detection performance, compared to the detection performance of the original image, mainly due to its inherent smoothing and noise reduction effects. Our proposed method is also compared with two other compression methods: PCA-ICA (independent component analysis) and band decimation (BandDec), yielding superior results for high compression rates.

Using mixture-tuned match filtering to measure changes in subpixel vegetation area in Las Vegas, Nevada

In desert cities, accurate measurements of vegetation area within residential lots are necessary to understand drivers of change in water consumption. Most residential lots are smaller than an individual 30-m pixel from Landsat satellite images and have a mixture of vegetation and other land covers. Quantifying vegetation change in this environment requires estimating subpixel vegetation area. Mixture-tuned match filtering (MTMF) has been successfully used for subpixel target detection. There have been few successful applications of MTMF to subpixel abundance estimation because the relationship observed between MTMF estimates and ground measurements of abundance is noisy. We use a ground truth dataset over 10 times larger than that available for any previous MTMF application to estimate the bias between ground data and MTMF results. We find that MTMF underestimates the fractional area of vegetation by 5% to 10% and show that averaging over multiple pixels is necessary to reduce noise in the dataset. We conclude that MTMF is a viable technique for fractional area estimation when a large dataset is available for calibration. When this method is applied to estimating vegetation area in Las Vegas, Nevada, spatial and temporal trends are consistent with expectations from known population growth and policy changes.

On Efficiency of Detection of Subpixel Targets with Hypothesis Dependent Structured Background Power

In this paper, matched detector (MD) and matched subspace detector (MSD) are studied when the structured background power is different under the null and the alternative hypotheses. The distributions of two test statistics are derived under these conditions. It has been analytically shown that these detectors can suffer a drastic degradation in performance for background power deviations under alternative hypothesis. We discuss the differences between the performances of these detectors in the case of the structured and unstructured backgrounds with uncorrelated Gaussian noise. The theoretical results are compared with simulated data and good agreement is reported. We present experimental results of small floating object detection on an agitated sea surface using spectral digital video experiments which validate the theoretical results.

English

English

Adaptive Detection of Subpixel Targets With Hypothesis Dependent Background Power

We design and assess an adaptive scheme to detect a subpixel target in a sequence of images in the presence of an additive correlated Gaussian background. The presence of the subpixel target decreases the background power that hence may be different under the null and alternative hypotheses. We use the generalized likelihood ratio test (GLRT) to adapt the recently proposed modified matched subspace detector (MMSD) to unknown background variances under the null and alternative hypotheses using the secondary and primary data, respectively. We derive a modified adaptive subspace detector (MASD) that is sensitive to both energy in the target subspace and reduced energy in the orthogonal subspace. We contrast it with the MMSD and the well-known adaptive cosine estimator (ACE). Numerical simulations attest to the validity of the theoretical analysis and show that the proposed detector performance outperforms the ACE, especially in the case of dark subpixel targets. The performance-degrading effects of limited secondary data are presented for the proposed detector.

A Kernel-Based Target-Constrained Interference-Minimized Filter for Hyperspectral Sub-Pixel Target Detection

The target-constrained interference-minimized filter (TCIMF) method has been successfully applied to various hyperspectral target detection applications. This paper presents a nonlinear version of TCIMF, called kernel-based TCIMF (KTCIMF), employing the kernel method to resolve the issue of nonlinear endmember mixing in hyperspectral images (HSI). Input data are implicitly mapped into a high-dimensional feature space, where it is assumed that target signals are more separable from background signals. Conventional TCIMF performs well in suppressing undesired signatures whose spectra are similar to that of the targets, thereby enhancing performance, and with less false alarms. KTCIMF not only takes into consideration the nonlinear endmember mixture but also fully exploits the other spectrally similar interference signatures. In this way, it is effective in suppressing both the background and those undesired signatures that may cause false alarms in traditional methods. Experimental results with both simulated and real hyperspectral image data confirm KTCIMF's performance with intimately mixed data. Compared with conventional kernel detectors, KTCIMF shows improved ROC curves and better separability between targets and backgrounds.

Subpixel Target Enhancement in Hyperspectral Images

Hyperspectral images due to their higher spectral resolution are increasingly being used for various remote sensing applications including information extraction at subpixel level. Typically whenever an object gets spectrally resolved but not spatially, mixed pixels in the images result. Numerous man made and/or natural disparatetar gets may thus occur inside such mixed pixels giving rise to subpixel target detection problem. Various spectral unmixing models such as linear mixture modeling (LMM) are in vogue to recover components of a mixed pixel. Spectral unmixing outputs both the endmember spectrum and their corresponding a bundance fractions inside the pixel. It, however, does not provide spatial distribution of these abundance fractions within a pixel. This limits the applicability of hyperspectral data for subpixel target detection. In this paper, a new inverse Euclidean distance based super-resolution mapping method has been presented. In this method, the subpixel target detection is performed by adjusting spatial distribution of abundance fraction within a pixel of an hyperspectral image. Results obtainedat different resolutions indicate that super-resolution mapping may effectively be utilized in enhancing the target detection at sub-pixel level. Defence Science Journal, 2013, 63(1), pp.63 -68 , DOI:http://dx.doi.org/10.14429/dsj. 63.3765

A Decision Fusion Framework for Hyperspectral Subpixel Target Detection

Target detection is one of the most challenging issues of remotely sensed data. Due to high spectral resolution of the hyperspectral images and their limited ground sampling distance, targets of interest occur at subpixel level. In such cases, spatial characteristics of targets are hard to acquire and the only way to overcome such problem is to take advantage of the spectral information. Based on the spectral characteristics of background and the targets to be detected, several methods have been proposed. Some of these methods assume a physics-based approach, while the other may be purely statistical. So, all of these methods are based on some assumptions each of which can be challenged in one way or another. One possible way to take advantage of these differences to improve the final results is the fusion of the detectors' outputs. In this paper, eight subpixel target detectors are employed as the ensemble detectors. It is also worth mentioning that the detectors should be different from each other; otherwise the overall decision will not be better than the individual detectors. So, we suggest using the genetic algorithms to select the most suitable detectors for a given decision fusion rule. Experimental results on a real world hyperspectral data as well as a synthetic dataset show the efficiency of the proposed method to improve the detection performance.

Subpixel hyperspectral target detection using local spectral and spatial information

We present two methods to improve three hyperspectral stochastic algorithms for target detection; the algorithms are the constrained energy minimization, the generalized likelihood ratio test, and the adaptive coherence estimator. The original algorithms rely solely on spectral information and do not use spatial information; this usage is normally justified in subpixel target detection, since the target size is smaller than the size of a pixel. However, we found that since the background (and the false alarms) may be spatially correlated and the point spread function can distribute the energy of a point target between several neighboring pixels, the implementation of spatial filtering algorithms considerably improved target detection. Our first improvement used the local spatial mean and covariance matrices, which take into account the spatial local mean instead of the global mean. While this concept has been found in the literature, the effect of its implementation in both the estimated mean and the covariance matrix is examined quantitatively here. The second was based on the fact that the effect of a target of physical subpixel size will extend to a cluster of pixels. We tested our algorithms by using the data set and scoring methodology of the Rochester Institute of Technology Target Detection Blind Test project. The results showed that both spatial methods independently improved the basic spectral algorithms mentioned above, and when the two methods were used together, the results were even better.

A Method of Target Detection in Remote Sensing Image Captured based for Sensor Network

A refined energy constrained minimization method is developed for target detection in hyperspectral remote sensing images captured by unmanned aerial vehicles (UAVs) during their surveillance missions, which has been tested in the experiment under this paper. The experiment result proves, in the detection process, this method can effectively restrain noises so far as the spectral characteristics of any potential target are known, and find sub-pixel targets out effectively from the hyperspectral remote sensing image in unknown background spectrum.

Prototype development and field-test results of an adaptive multiresolution PANOPTES imaging architecture

The design, development, and field-test results of a visible-band, folded, multiresolution, adaptive computational imaging system based on the Processing Arrays of Nyquist-limited Observations to Produce a Thin Electro-optic Sensor (PANOPTES) concept is presented. The architectural layout that enables this imager to be adaptive is described, and the control system that ensures reliable field-of-view steering for precision and accuracy in subpixel target registration is explained. A digital superresolution algorithm introduced to obtain high-resolution imagery from field tests conducted in both nighttime and daytime imaging conditions is discussed. The digital superresolution capability of this adaptive PANOPTES architecture is demonstrated via results in which resolution enhancement by a factor of 4 over the detector Nyquist limit is achieved.

Automated Image Processing for on-Line Photogrammetry

In order to monitor the deformation of the surface or dynamic adjustment and measurement, the real-time and on-line close range photogrammetry system is needed. Before camera calibration and 3D reconstruction, the targets in the images must be recognized and located automatically with high accuracy and speed firstly. This paper describes the automated image processing for the on-line photogrammetry system. Experimental results show that compared with the other measurement system, the accuracy of subpixel target centroid can reach to about 2% of the pixel size, while the process time of an image with 3008×2000 pixels is about 0.1S.

An improved method to detect remote sensing image targets captured by sensor network

In order to detect targets from the hyper-spectral images captured by unmanned aerial vehicles, the images are moved into a new characteristic space with greater divisibility by making use of the manifold learning theory. On this basis, a furation impulse response (FIR) filter is developed. The output energy can be minimized after images passing through a FIR filter. The target pixel and the background pixel are distinguished according to the restrained conditions. This method can effectively suppress noises and detect sub-pixel targets in the hyper-spectral remote sensing image of unknown background spectrum.

Asymptotically Optimum Quadratic Detection in the Case of Subpixel Targets

This work extends the optimum Neymann-Pearson methodology to detection of a subspace signal in the correlated additive Gaussian noise when the noise power may be different under the null (H0) and alternative (H1) hypotheses. Moreover, it is assumed that the noise covariance structure and power under the null hypothesis are known but under the alternative hypothesis the noise power can be unknown. This situation occurs when the presence of a small point (subpixel) target decreases the noise power. The conventional matched subspace detector (MSD) neglects this phenomenon and causes a consistent loss in the detection performance. We derive the generalized likelihood ratio test (GLRT) for such a detection problem comparing it against the conventional MSD. The designed detector is theoretically justified and numerically evaluated. Both the theoretical and computer simulation results have shown that the proposed detector outperforms the conventional MSD. As to the detection performance, it has been shown that the detectivity of the proposed detector depends on the additional adaptive corrective term in the threshold. This corrective term decreases the value of presumed threshold automatically and, therefore, increases the probability of detection. The influence of this corrective term on the detector performance has been evaluated for an example scenario.

Performance of the Matched Subspace Detector in the Case of Subpixel Targets

This Letter presents the matched subspace detection in the presence of Gaussian background with known covariance structure but different variance for hypothesis H0 and H1. The performance degradation has been evaluated when there are the following mismatches between the actual and designed parameters: background variance in the case of hypothesis H1 and one-lag correlation coefficient of background. It has been shown that the detectability depends strongly on the fill factor of targets in the case of the mode signal matrix with high rank for a prescribed false alarm probability and a given signal-to-background ratio. These results have been also justified via Monte Carlo simulations for an example scenario.

Low-bit rate exploitation-based lossy hyperspectral image compression

Hyperspectral image compression has become increasingly important in data exploitation because of enormous data volumes and high redundancy provided by hundreds of contiguous spectral channels. Since a hyperspectral image can be viewed as a 3-dimensional (3D) image cube, many efforts have been devoted to extending 2D image compression techniques to perform 3D image compression on hyperspectral image cubes. Unfortunately, some major issues generally encountered in hyperspectral data exploitation at low or very low-bit rate compression, for example, subpixels and mixed pixels which do not occur in traditional pure pixel-based image compression are often overlooked in such a 2D-to-3D compression. Accordingly, a direct application of 2D-to-3D compression techniques to hyperspectral image cubes without taking precaution may result in significant loss of crucial spectral information provided by subtle substances such as small objects, anomalies during low bit-rate lossy compression. This paper takes a rather different view by investigating lossy hyperspectral compression from a perspective of exploring spectral information, referred to as exploitation-based lossy compression and further develops spectral/spatial hyperspectral image compression to effectively preserve crucial and vital spectral information of objects which are generally missed by commonly used mean-squared error (MSE) or signal-to-noise ratio (SNR)-based compression techniques when lossy compression is performed at low bit rates. In order to demonstrate advantages of the proposed spectral/spatial compression approach applications of subpixel target detection and mixed pixel analysis are used for experiments for performance evaluation.

Detection Thresholds for Rare, Spectrally Unique Targets within Semiarid Rangelands

Many factors influence classification accuracy, and this study assessed detection thresholds for various sub-pixel targets using QuickBird multispectral imagery. Six iterations of maximum-likelihood classification were used to determine classification accuracy for 100 spectrally unique targets randomly placed over a semiarid rangeland site. Error matrices were calculated using independent validation sites and producer’s, user’s, and overall accuracy, Kappa Index of Agreement, and transformed divergence were analyzed to compare the performance of each classification and determine detection thresholds. Results indicate a strong relationship between target size and classification accuracy (R 2 0.94) as well as an increasingly prominent role played by training site selection as target size decreased. Strong spectral separability and good classification accuracies were achieved for targets 25 percent cover. Sub-pixel targets 25 percent in size were not detectable. This study highlights the effect of target size upon classification accuracy and has direct implications for invasive plant research and rare target detection.

Hybrid Detectors Based on Selective Endmembers

Subpixel target detection is a challenge in hyperspectral image analysis. As the spatial resolution of hyperspectral imagery is usually limited, subpixel targets only occupy part of the pixel area. In such cases, the spatial characteristics of the targets are hard to acquire, and the only information we can use comes from spectral characteristics. Several kinds of method based on spectral characteristics have been proposed in the past. One is the linear unmixing method, which can provide the abundances of different endmembers in the hyperspectral imagery, including the target abundance. Another focuses on providing statistically reliable rules to separate subpixel targets from their backgrounds. Recently, hybrid detectors combining the aforementioned two methods were put forward, which cannot only figure out the quantitative information of the endmembers but also put this quantitative information into an adaptive matched subspace detector or adaptive cosine/coherent estimate detector to separate the target pixels from the background with statistically reliable rules. However, in these methods, all the endmembers are used to construct the statistical rule, while in most cases only some of the endmembers are actually contained in the pixels. This paper proposes hybrid endmembers selective detectors in which different kinds of endmembers are used according to different pixels to ensure that the true composition of endmembers in each pixel is applied in the detection procedure. Three different types of hyperspectral data were used in our experiments, and our proposed hybrid endmember selective detectors showed better performances than the current hybrid detectors in all the experiments.

Image-Derived Prediction of Spectral Image Utility for Target Detection Applications

The utility of an image is an attribute that describes the ability of that image to satisfy performance requirements for a particular application. Building on previous research that defines the assessment of the utility of a spectral image based on the detectability of subpixel targets, this paper examines the prediction of spectral image utility. It first reviews existing methods for predicting spectral image utility and then proposes a new approach in predicting spectral image utility for target detection applications that derive statistical parameters directly from the spectral image. This so-called image-derived approach predicts the likelihood of finding synthetically implanted subpixel targets. Using three airborne hyperspectral images, we benchmark prediction performance by comparing the predicted and assessed utilities, quantifying the accuracy of the prediction, and discussing the time savings associated with prediction. Initial results show utility calculation time savings of up to 40% for a single image, with the potential for greater savings if multiple target types are considered. This method of predicting spectral image utility offers a simple and efficient way of predicting image utility, which can be directly compared with our utility assessments.