Cholesky分解的逐像元实时高光谱异常探测

发布时间：2018-07-03 18:16

本文选题：高光谱 + 异常探测　；参考：《遥感学报》2017年05期

【摘要】：传统的实时异常探测算法需对高维的背景样本统计矩阵进行求逆运算,数值稳定性差、时间复杂度高。而基于Cholesky分解,将高维矩阵的求逆运算转换为求解下三角线性系统,采用Cholesky分解因子的一阶修正方法快速更新背景统计信息,降低逐像元实时处理的时间复杂度并且保持数值稳定性。由于算法仅涉及下三角矩阵的更新过程,压缩了数据存储空间,适用于机载或星上实时处理。采用3维接收器曲线(3D-ROC)以及计算机实际处理时间对实验结果进行定量化分析,结果表明,该算法在不降低异常探测精度的同时,对当前时刻像元的实时处理时间约缩短为基于QR分解算法的0.4%—0.65%,或减少至基于Woodbury矩阵引理算法的27%—33%,有效提高实时高光谱异常探测器的计算性能,并且保持处理过程中的数值稳定性。
[Abstract]:The traditional real-time anomaly detection algorithm needs to calculate the inverse of the high-dimensional background sample statistical matrix. The numerical stability is poor and the time complexity is high. Based on Cholesky decomposition, the inversion of high dimensional matrix is transformed into solving the lower triangular linear system. The first order correction method of Cholesky factorization factor is used to update the background statistical information quickly. The time complexity of pixel-by-pixel real-time processing is reduced and numerical stability is maintained. Because the algorithm only involves the updating process of the lower triangular matrix, the data storage space is compressed, which is suitable for real-time processing on board or on board. The 3D receiver curve (3D-ROC) and the actual processing time of the computer are used to quantify the experimental results. The results show that the algorithm does not reduce the accuracy of anomaly detection at the same time. The real-time processing time of the current pixel is reduced to 0.4-0.65 based on QR decomposition algorithm, or to 27-33 based on Woodbury matrix Lemma algorithm, which effectively improves the computational performance of the real-time hyperspectral anomaly detector. And maintain the numerical stability in the process of processing.
【作者单位】：中国科学院遥感与数字地球研究所;中国科学院大学;
【基金】：北京市科委项目(编号:238796489321) 国家自然科学基金(编号:41501396)~~
【分类号】：TP751

【相似文献】