Source Based Small Targets Detection for Hyperspectral Imagery Using Evidential Reasoning

Detecting unknown small targets in an unknown environment for hyperspectral imagery is a great challenge since the prior knowledge about targets and backgrounds is not available. Low probability detection (LPD) is one of the most commonly used methods dealing with this problem. The disadvantage of LPD is that local discriminabilities of spectral signature aren´t utilized sufficiently. For this reason, a source based detection methods using evidential reasoning is proposed to improve the detection performance. First, hyperspectral imagery data is slit into some data sources corresponding to data of a specific spectral range using correlation analysis; Then, features are extracted from each data source via LPD; Finally, fusion algorithm for detection is implemented by evidential reasoning while basic belief assignment function is constructed involving high-order moments of features. Theoretical analysis and results of experiment verify the effectiveness of the method