Rare signals detection in nonwhite noise environment based on multidimensional signal subspace for hyperspectral image

摘要

Due to the resolution limitation of the hyperspectral imaging sensors, some targets only take one or two pixels in the hyperspectral image (HSI). These targets are called small targets. When HSI is impaired and has to be denoised, small targets might be suppressed in the denoising process, which can degrade the detection performance. In this paper, we propose a method to improve the small target detection performance of HSI which is damaged by nonwhite thermal noise. One recently proposed nonwhite noise reduction algorithm prewhitening-multiway-Wiener-Filter (PMWF) and a usually used spatial-domain wavelet packet transform with SureShrinkage (SWPT-SURE) denoising approach are compared with the algorithm proposed in this paper. Finally, a real-world HYDICE HSI is employed to investigate the noise whitening capability and the improvement of small target detection performance.