论文部分内容阅读
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。
Real-time processing can relieve the huge pressure on massive hyperspectral data in storage and downlink transmission, and attract researchers’ attention in the field of hyperspectral anomaly detection. Hyperspectral imaging sensors have become mainstream by acquiring data by sweeping. Therefore, a novel hyperspectral real-time anomaly detection algorithm based on progressive processing frame is proposed. The local causal window model was introduced into the Reed-Xiaoli anomaly detection algorithm, and the abnormal target was detected by sliding the local causal window, which ensured the causality of real-time processing. Aiming at the problem of over-complexity in matrix inversion process, based on the recursive idea of Kalman filter, Woodbury’s inverse lemma is used to update the state information of the current data from the previous data state information, avoiding the problem of large matrix Inverse operation, reducing the computational complexity of the algorithm. Experiments using simulated and real hyperspectral data show that the computational efficiency of the proposed real-time algorithm is significantly improved compared to the original algorithm, while maintaining the accuracy of detection.