光学学报, 2017, 37 (1): 0128002, 网络出版: 2017-01-13
基于逐行处理的高光谱实时异常目标检测
Hyperspectral Real-Time Anomaly Target Detection Based on Progressive Line Processing
遥感 高光谱异常目标检测 实时算法 逐行处理 局部因果窗 remote sensing hyperspectral anomaly target detection real-time algorithm progressive line processing local causal window
摘要
实时处理可以缓解海量高光谱数据在存储及下行传输方面带来的巨大压力, 在高光谱异常检测领域引起了研究人员的广泛关注。高光谱成像传感器通过推扫获取数据的方式已成为主流, 因此, 提出了一种基于逐行处理框架的高光谱实时异常目标检测算法。将局部因果窗模型引入Reed-Xiaoli异常检测算法中, 通过滑动局部因果窗来检测异常目标, 保证了实时处理的因果性。针对矩阵求逆过程复杂度过大的问题, 在卡尔曼滤波器递归思想的基础上, 利用Woodbury求逆引理, 由前一时刻数据状态信息迭代更新当前数据的状态信息, 避免了大矩阵的求逆运算, 减少了算法的计算量。利用模拟和真实高光谱数据进行实验, 结果表明, 在保持检测精度不变的前提下, 提出的实时算法的运算效率相比于原始算法得到显著提高。
Abstract
Real-time processing can reduce the pressure of data storage and downlink transmission caused by the ever-expending hyperspectral dataset, which has received more and more attention in hyperspectral anomaly detection. Since acquiring data with pushbroom has become main stream for hyperspectral imaging sensors, a real-time anomaly target detection method is proposed based on the framework of progressive line processing. In order to make sure the causality of real-time processing, the local causal window model is introduced into the Reed-Xiaoli anomaly detection algorithm, and the sliding local causal window is used to detect anomaly targets. In terms of the high computational complexity caused by the inversion of matrix, the recursive principle of the Kalman filter and the Woodbury′s lemma are employed to update the status information of current data through iterating data status information at the previous moment, which avoids the inversion of large matrix. The simulated and real hyperspectral data are adopted for the experiment. The results show that under the premise of maintaining the detection accuracy, the proposed real-time algorithm improves the processing efficiency significantly compared with the original algorithm.
赵春晖, 邓伟伟, 姚淅峰. 基于逐行处理的高光谱实时异常目标检测[J]. 光学学报, 2017, 37(1): 0128002. Zhao Chunhui, Deng Weiwei, Yao Xifeng. Hyperspectral Real-Time Anomaly Target Detection Based on Progressive Line Processing[J]. Acta Optica Sinica, 2017, 37(1): 0128002.