基于逐行处理的高光谱实时异常目标检测

发布时间：2018-08-11 10:23

【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 relieve the huge pressure brought by massive hyperspectral data in storage and downlink transmission, and has attracted extensive attention in the field of hyperspectral anomaly detection. The method of data acquisition by push-sweep has become the mainstream of hyperspectral imaging sensor. Therefore, a real-time hyperspectral anomaly target detection algorithm based on line-by-line processing framework is proposed. The local causality window model is introduced into the Reed-Xiaoli anomaly detection algorithm, and the real time processing causality is ensured by moving the local causal window to detect the abnormal target. Aiming at the problem that the complexity of matrix inversion process is too large, based on the Kalman filter recursive thought, using Woodbury inverse Lemma, the state information of the current data is updated iteratively from the data state information of the previous moment. The inverse operation of large matrix is avoided, and the computation amount of the algorithm is reduced. The simulation and real hyperspectral data are used to carry out experiments. The results show that the proposed real-time algorithm is more efficient than the original algorithm on the premise of keeping the detection accuracy unchanged.
【作者单位】：哈尔滨工程大学信息与通信工程学院;
【基金】：国家自然科学基金(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(RC2013XK009003)
【分类号】：TP751

【相似文献】