Study On Anomaly Target Detection Technology In Hyperspectral Images

With the rapid development of the electronic imaging technology and space technology, modern remote sensing technology has become an indispensable access to information technology with distinguished features. As a new type of remote sensing data, hyperspectral image is rich with detail information and spectral diagnostics, and has the capability of distinguish with military targets and identification for military targets, so it gets more and more attention from all the country. Anomaly target detection algorithm in hyperspectral image can detect differences with the target of the background in the absence of a priori spectral information, it has become a hot research field of target detection.This paper studies the anomaly target detection technology in hyperspectral images, combination of statistical properties of hyperspectral images, fully exploits the anomaly target of spectral and spatial information and with the help of detection algorithm performance analysis, uses the signal processing and multivariate statistical theory means to detect the target. This paper researches in several aspects as following.Firstly, from the binary hypothesis and likelihood ratio test to start, based on the RX anomaly target detection algorithm model, the improved algorithm is presented based on RX algorithm, and through the AVIRIS simulation data, the analysis of the advantage of correlation matrix is done in hyperspectral target. UTD algorithm is verified to have a good detection result with specific band dependence in the target detection.Secondly, the false alarm probability and detection probability as the performance index formed ROC receiver operating characteristic curve for the performance evaluation is proposed. Analysis of the main parameters affecting the algorithm performance is done, and through the establishment of performance assessment framework, the anomaly target detection algorithm performance of RX is given. Performance of the algorithm for parameter optimization design is proposed.Thirdly, because of limitation for traditional anomaly target detection algorithm using only the spectral information, and according to the additive target signal and unstructured background model, the three-dimensional GMRF algorithm is given based on the spectral correlation and spatial correlation. By estimating the second-order statistics to solve the inverse background data, the algorithm time-consuming is reduced. The theoretical analysis and simulation result shows that the algorithm is effective.Finally, through the experimental test platform DM642, the algorithms in this paper are verified, and combination with the DSP hardware characteristics the DSP program and the source code is optimized. An automatic threshold detection method for anomaly target in hyperspectral image that based on local confidence is proposed, and by experimental test the validity of the method is proved.