基于逐行处理的高光谱遥感异常快速检测方法

Fast Anomaly Detection Algorithm for Hyperspectral Imagery Based on Line-by-line Processing

针对基于逐像元处理的因果实时异常(Causal Real-time Relationship Reed-X Detector,CR-RRXD)检测算法计算量大,以及基于逐像元方式边检测边成像显示的时间过长而不能满足快速处理要求的缺陷,提出了一种基于逐行处理的CR-R-RXD检测算法.与基于逐像元处理的CR-R-RXD检测算法相比,该方法将高光谱图像整行像元向量作为输入,即处理一行高光谱数据只需计算一次,极大地减少了计算次数.实验结果表明,与R-RXD和基于逐像元处理的CR-R-RXD算法相比,本文算法可在获得与R-RXD算法几乎相同的检测准确度的情况下,实现快速实时处理,其检测准确度相较于基于逐像元处理的CR-R-RXD算法有所提高,且算法检测时间大大缩短,增强了算法的时效性.

The Causal Real-time Relationship Reed-X Detector （CR-R-RXD） detecting algorithm based on the pixel-by-pixel processing for hyperspectral imagery, which has the problems of a large amount of computation, a long display time and a slow running speed. A CR-R-RXD detecting algorithm based on line by line was proposed in this paper. Compared with the CR-R-RXD method based on pixel by pixel processing, the whole row pixel vector of hyperspectral image was used as input in this proposed method. That is, dealing with a row of hyperspectral data needs to be calculated only once, which greatly reduces the calculation times. Experimental results show that, to compare with the R-RXD algorithm and CR-R- RXD method based on pixel by pixel processing, the proposed algorithm can achieve the process of fast real-time processing with almost the same accuracy as the R-RXD algorithm, the detection accuracy is improved to compare with the CR-R-RXD algorithm based on pixel by pixel processing, and the testing time of the algorithm is reduced, which enhances the timeliness of the algorithm.