Small Target Detection Based On Hyperspectral Imagery

Hyperspectral imagery can provide hundreds even thousands bands, it has the unique characteristic of acquiring spectral and spatial information simultaneously, it also has higher spectral resolution and richer images and spatial information than multispectral imagery. By comparing and analyzing target signature we can achieve good performance in hyperspectral target detection which is hard to achieve in other imagery model. Hyperspectral image target detection includes data reduction, unmixing and target detection and recognition et al. Based on small target detection, the research of this paper are as follows:We have a work on the hyperspectral data reduction and unmixing, several algorithms have been implemented. We have done the work of data reduction, endmember extraction and abundance maps' inversion and analysis their performance through the experiments.Several classic detection algorithms have been proposed and analysis, with limited dataset, we have evaluated these algorithms' performance and made a conclusion through the linear implant targets detection experiment.Aiming at the over-fitting problem which is often occurred in hyperspectral target detection, we proposed a regularization framework based on the sparse matrix transform(SMT). By combining this framework with detection algorithms which need to calculate the inverse matrix we can improve their performance. Target detection algorithms such as SMT-Re-CEM and SMT-Re-MF are proved to achieving better performance through our experiment.