A quasi-Newton-based spatial multiple materials detector for hyperspectral imagery

Hyperspectral remote sensing in space provides information related to surface material characteristics of spacecraft or planets that can be exploited to perform automated detection of targets of interest. At present, the detection in space environment is definitely a hot spot all over the world. So, developing the technique of multiple materials detection makes great sense. In this paper, we propose an algorithm for spatial multiple materials detection in hyperspectral images, which is based on high-order statistics and quasi-Newton method. The proposed detection algorithm, quasi-Newton-based multiple materials detector (QNMMD), exploits spectral information exclusively to make decisions by considering that each pixel contains the interesting materials or not. After single time detection, the pixel containing multiple interesting materials spectra can be exactly detected. The proposed detector has three superiorities. Firstly, due to the quasi-Newton method the proposed algorithm is relatively fast. It needs few times iteration for detecting calculation. Secondly, it performs well when the interesting materials are in low probabilities or small population with the non-Gaussian statistics. Thirdly, with regularization items the algorithm is robust to noise and works well when there are various kinds of interesting materials needing to be detected. Experimental results based on the hyperspectral image of Hubble Space Telescope prove the QNMMD algorithm is effective.