DSmT Based RX Detector for Hyperspectral Imagery

Anomaly detection is very useful for hyperspectral detection with no a priori information of spectral signature. However, most prevailing anomaly detectors are applied directly to the all-bands data without considering the characteristics of hyperspectral imagery in local spectrum range and some computation problem incurred from high dimensional data. In this paper, a Dezert-Smarandache Theory (DSmT) based anomaly detector is presented to handle these problems. The all-bands data are first partitioned into several lower dimensional band-subsets. Then the generalized basic belief assignment of DSmT reasoning are constructed according to target signal to noise ratio (TNR) of different band subsets and probability densities of the detection value from different band subsets. Experimental results from the real hyperspectral imagery of Operative Modular Imaging Spectrometer I (OMIS-I) show that our method outperforms the benchmark RX detector (RXD).