| Possible methods to help a remote sensing analyst to find a static or moving single pixel target over vast areas of terrain were examined in this work. Specifically, the research deals with the particular problem of how to find these targets using multiple images of the same area that were collected with the same multispectral (6 bands) imaging sensor but with a background that changes between images.;For this, hyperspectral quadratic covariance-based anomalous change detection algorithms were investigated to see if they could be used with multispectral data to find a moving target. In addition, a new method based on change vector analysis was developed to find a static target. In the case of the moving target problem, the performance of the Chronochrome, Covariance Equalization, and the Hyperbolic anomalous change detection algorithms were compared relative to each other and to a straight target detection algorithm. In addition, modifications to the covariance-based algorithms were developed that improved the results. For the static target case, various multispectral images were "layer stacked" together. Then, the Spectral Matched Filter hyperspectral target detection algorithm was applied on these data cubes to explore if this method could help separate a real target from false alarms obtained when simply running a target detection algorithm on a multispectral data cube.;The analysis demonstrated that a significant reduction in the number of false alarms can be obtained with these methods when compared to traditional Spectral Matched Filter (SMF) algorithm to find either static or dynamic single pixel targets of interest. In addition, the analysis shows the limitations and behavior of these methods under some of the issues normally encountered in remote sensing imaging. Overall, it was demonstrated that periodic multispectral imagery collections over a wide area can be very useful to find targets of interest. |
