基于逐行处理的高光谱遥感异常快速检测方法

发布时间：2018-03-18 20:43

  本文选题：高光谱　切入点：异常检测　出处：《光子学报》2017年04期 　论文类型：期刊论文

【摘要】：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.
[Abstract]:The algorithm of causality Real-time Relationship Reed-X detector CR-RRXD based on pixel by pixel processing is computationally large, and the time of image display while detecting edge by pixel is too long to meet the requirement of fast processing. A new CR-R-RXD detection algorithm based on line by line processing is proposed. Compared with the CR-R-RXD detection algorithm based on pixel processing, the whole line pixel vector of hyperspectral image is used as input, that is to say, only one calculation is needed to process a row of hyperspectral data. The experimental results show that compared with R-RXD and CR-R-RXD algorithm based on pixel processing, the proposed algorithm can achieve fast real-time processing with almost the same detection accuracy as R-RXD algorithm. Compared with the CR-R-RXD algorithm based on pixel processing, the detection accuracy of the algorithm is improved, and the detection time of the algorithm is greatly shortened, which enhances the timeliness of the algorithm.
【作者单位】： 中国农业大学信息与电气工程学院;
【基金】：国家自然科学基金(Nos.61201415,61571170)资助~~
【分类号】：TP751
，

本文编号：1631267