Subpixel hyperspectral target detection using local spectral and spatial information Academic Article uri icon

abstract

  • Ben-Gurion University of the Negev, Department of Electrical and Computer Engineeringand The Earth and Planetary Image Facility, P.O.B. 653, Beer-Sheva 84105, IsraelE-mail: srotman@ee.bgu.ac.ilAbstract. We present two methods to improve three hyperspectral stochastic algorithms fortarget detection; the algorithms are the constrained energy minimization, the generalized like-lihood ratio test, and the adaptive coherence estimator. The original algorithms rely solely onspectral information and do not use spatial information; this usage is normally justified in sub-pixel target detection, since the target size is smaller than the size of a pixel. However, we foundthat since the background (and the false alarms) may be spatially correlated and the point spreadfunctioncandistributetheenergyofapointtargetbetweenseveralneighboringpixels,theimple-mentation of spatial filtering algorithms considerably improved target detection. Our firstimprovement used the local spatial mean and covariance matrices, which take into accountthe spatial local mean instead of the global mean. While this concept has been found in theliterature, the effect of its implementation in both the estimated mean and the covariance matrixis examined quantitatively here. The second was based on the fact that the effect of a target ofphysical subpixel size will extend to a cluster of pixels. We tested our algorithms by using thedata set and scoring methodology of the Rochester Institute of Technology Target DetectionBlind Test project. The results showed that both spatial methods independently improvedthe basic spectral algorithms mentioned above, and when the two methods were used together,the results were even better.

publication date

  • March 7, 2012