空谱联合的核光谱角异常检测及GPU实现  

Anomaly detection and GPU implementation for spatial-spectral kernel spectral angle

作　　者：赵春晖[1] 李佳伟 刘务 田明华[1] ZHAO Chunhui LI Jiawei LIU Wu TIAN Minghua(College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001, China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2017年第9期1497-1504,共8页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(61405041;61571145);黑龙江省自然科学基金重点项目(ZD201216);哈尔滨市优秀学科带头人基金项目(RC2013XK009003)

摘　　要：针对高光谱图像空间信息利用不充分影响检测性能的问题,本文提出结合高光谱图像空间信息与光谱信息的异常目标检测算法。该算法无需假设背景模型,通过计算待检测像元与其空间邻域像元的核光谱角累加和,初步得到每个像元的异常程度。利用扩展形态学的腐蚀操作进行异常修正,有效去除噪声干扰,并降低虚警率,从而得到最终的异常检测结果。为提高算法的执行效率,本文进一步提出了基于GPU/CUDA模型下的并行优化处理方法。通过仿真实验证明,该算法在保证较高检测精度的同时,充分利用GPU的并行特性,明显缩减了检测时间。Anomaly target detection has become a hotspot of hyperspectral remote-sensing information processing. To fully utilize the potential information of hyperspectral images and to develop an efficient detection algorithm for hyperspectral imagery anomalies, SS-KSAM, a novel algorithm that integrates spatial and spectral information, was proposed. Without assuming a background model, the proposed algorithm can obtain intermediate results by calculating the summation of the kernel spectral angle between the target pixel and its spatial neighbor pixels. Then, the final results can be obtained through corrosion in extended morphology. Results showed that noise interference is removed and the false alarm rate decreased. A parallel optimization method that is based on the graphics processing unit （GPU）/compute-unified device architecture model was further proposed to improve algorithm efficiency. Simulation results show that the utilization of the parallel characteristics of GPU can shorten detection time and ensure high detection precision.

关 键 词：高光谱图像 异常检测 核光谱角 空谱联合 数学形态学 图形处理单元 统一设备架构 并行处理 

分 类 号：TN911.73[电子电信—通信与信息系统]