Research On Hyperspectral Target Detection Method Based On Blind Signal Processing

With the characteristics of high resolution and map unity, Hyperspectral remote sensing images provide detailed and rich spectral information. Therefore the use of hyperspectral imaging for target detection has unique advantages. Hyperspectral target detection is to distinguish target from other surface features to determine the existence of the problem within the target pixel in each hyperspectral imaging. The nature of the target detection is a classification problem. The target and the background is seen as two types of surface features within images, which is different from other classifications in that its target proportion is very small. In target detection, the selection of the background area not only directly affects the size of the background dimension, but also directly affects the performance of target detection. So how to choose a background area and get background dimension is a key issue for hyperspectral target detection.This paper discusses hyperspectral target detection methods based on background subspace and blind signal processing in detail.The main research work can be summarized as follows:First, this paper analyzes and explains the imaging principle and composition of hyperspectral remote sensing image, summarizes its existing technologies, describes briefly the current applications and development of hyperspectral remote sensing, does detailedly theoretical analysis and experimental research for the classical target detection algorithms, Such paper achieves a series of classic target detection algorithms such as RX, GLRT, MF, ACE, etc., and these algorithms are verified experimentally by using the real data and simulated data.Second, the paper puts forward a kind of hyperspectral target detection method based on blind signal processing, introduces blind signal processing method, which is usually used in signal processing, to the target detection of hyperspectral remote sensing image, uses the target signal AR model as a priori knowledge to bring the core idea of mean square cross prediction error(MSCPE) into algorithms, and combines with blind source extraction(BSE) method to enable detection results more accurate. Experiments show that, compared hyperspectral target detection algorithms based on blind signal processing with classical methods, the former can improve the detection performance.Finally, this paper studies the hyperspectral target detection methods deeply, describes the acquisition methods of the global and regional background dimension in detail starting with the background dimension, and thus presents a regional background subspace estimation method based on algorithm for multiple signals extraction(LBSEAMUSE). While suppressing background energy, such method can improve the efficiency of target detection, and reduce complex calculation steps. Through experiments, the defects of the regional background subspace estimation method(LBSE) are presented, and the fact that the detection effect of LBSE-AMUSE method is better and its running speed is faster, is also proved.