Anomaly Target Detection in Hyperspectral Imagery Based on Band Subset Fusion by Fuzzy Integral

An anomaly target detection method based on the high correlation band subsets and fuzzy integral fusion is presented to deal with detecting unknown target in unknown background for hyperspectral imagery. Original hyperspectral data is divided into several continuous band subsets according to the high correlation within the subset. Applying nonparametric kernel density estimation to the RX detector output of each subset to obtain its probability density function (pdf), and a nonparametric fuzzy membership function is constructed; based on the eigenvalues in spectral dimension, a target signal-noise-ratio is defined to measure the degree of importance of detection result from each subset; finally, decision fusion is implemented through Sugeno fuzzy integral method. Experiments on visible/near-infrared OMIS-I hyperspectral imagery justify the effectiveness of the algorithm.