Constrained Energy Minimization Algorithm Based on Selected Sample Correlation Matrix for Hyperspectral Imagery

Constrained Energy Minimization (CEM) is a classic target detection algorithm of hyperspectral imagery. CEM uses the background correlation matrix to describe the image ground, whereas the background correlation matrix is very sensitive to target information. In order to solve the problem a CEM algorithm based on selected sample correlation matrix is developed, which compares the similarity between each sample pixel and the target signal using a spectral similarity measure, and then selects the sample vectors most dissimilar with the target signal to estimate the background sample correlation matrix. Two real AVIRIS hyperspectral data sets were tested for target detection. The experimental results demonstrate that the proposed algorithm yields better detection performance and is more suitable for detecting targets occupying a number of pixels.