摘要
为了有效缓解海量高光谱数据存储与传输压力并快速精确检测异常目标,提出一种以滑动阵列窗像元为局部背景的高光谱图像非因果实时RXD异常检测方法.利用随数据逐像元接收而滑动的阵列窗确定局部背景像元,运用Woodbury引理,通过矩阵与向量的乘法和矩阵的加减实现局部背景协方差矩阵的求逆运算,在逐像元接收数据的同时实现阵列窗口中心像元的异常检测.模拟和真实高光谱图像实验结果表明,与现有实时检测方法相比,所提方法在检测性能或运行效率上有所提升;相比非实时的滑动阵列RXD异常检测,所提方法时间复杂度更低,处理大小为200×200含189波段的图像,其加速比达到近26倍.实验结果验证了该方法能在不降低检测精度的同时满足低运算量和低存储空间的实时性要求.
To reduce the storage and transmission burden of massive hyperspectral data as well as keep detecting the anomaly targets accurately and rapidly,a novel non-causal real-time anomaly detection algorithm based on sliding array window is proposed.During receiving the data pixel-by-pixel,the array window slides and determines the local background pixels;and according to the Woodbury lemma,the computation of the matrix inverse of local background can be replaced by the vector multiplication and matrix additions equivalently,then the anomaly pixel in the center of the sliding array window would be detected when receiving the data pixel-by-pixel.The experiments on synthetic and real-world hyperspectral images demonstrate that,compared with several existing real-time detection methods,the proposed method can improve the performance of detection accuracy or computational efficiency.Compared with the non-real-time sliding array RXD anomaly detector,the proposed algorithm has a lower time complexity,and the speedup ratio is nearly 26 times when processing an image with 200×200 pixels and 189 bands.Experimental results verified that the proposed algorithm could maintain the detection accuracy as well as meet the real-time requirements of low computational complexity and low storage space.
作者
赵辽英
林伟俊
王玉磊
厉小润
ZHAO Liao-ying;LIN Wei-jun;WANG Yu-lei;LI Xiao-run(School of Computer Science Technology, Hangzhou Dianzi University, Hangzhou 310018, China;Information and Technology College, Dalian Maritime University, Dalian, Liaoning 116026, China;College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China;Key Laboratory of Spectral Imaging Technology of Chinese Academy of Sciences, Xi＇an 710119, China)
出处
《光子学报》
EI
CAS
CSCD
北大核心
2018年第7期18-29,共12页
Acta Photonica Sinica
基金
国家自然科学基金(No.61571170)
教育部联合基金(No.6141A02022314)
上海航天科技创新基金(No.SAST2015033)
中央高校基本科研业务费专项资金(Nos.3132017080,3132016331)
中国科学院光谱成像重点实验室开放基金(No.LSIT201707D)资助
关键词
高光谱异常目标检测
实时算法
递归计算
协方差矩阵
滑动阵列
Hyperspectral anomaly target detection
Real-time algorithm
Recursive calculation
Covariance matrix
Sliding array
作者简介
第一作者：赵辽英（1970-），女，教授，博士，主要研究方向为图像处理与模式识别．Email：zhaoly@hdu．edu．cn;通讯作者：王玉磊（1986-），女，讲师，博士，主要研究方向为高光谱图像目标检测．Email：wangyulei@dlmu．edu．cn
