联合Gabor滤波的改进低秩稀疏矩阵异常检测算法  

Improved low-rank sparse matrix anomaly detection algorithm combined with Gabor filter

作　　者：徐智元 王晓飞[1] XU Zhiyuan;WANG Xiaofei(College of Electronic Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080

出　　处：《黑龙江大学自然科学学报》2025年第1期108-115,共8页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61871150);黑龙江省自然科学基金资助项目(PL2024F026)。

摘　　要：高光谱异常检测作为一种不依赖目标先验知识的无监督技术,存在诸如异常与背景难以区分、异常目标的检测精度易受背景污染等难题。为了解决这些问题,提出了一种联合Gabor滤波(Gabor filter,GF)的改进低秩稀疏矩阵异常检测算法。将Gabor滤波器引入低秩和稀疏矩阵分解(Low-rank and sparse matrix decomposition,LRaSMD)中,用来提取高光谱图像数据的空间纹理特征。对LRaSMD进行改进,选取Gabor变换后的初始结果构建子集,并对这个子集进行约束,通过固定子集的n个最小值并减去背景均值向量,最大程度地保留样本的异常值信息。采用马氏距离对构建的背景协方差矩阵进行检测。所提出的算法在5个数据集上的AUC值(Area under the curve)分别为0.9934、0.9967、0.9995、0.9991和0.9964。实验结果证明,所提出的算法极大地减弱了背景对异常像元的污染,并且可以很好地将背景和异常区分开,具有非常好的异常目标检测性能。As an unsupervised method that does not depend on prior knowledge of targets,hyperspectral anomaly identification confronts a number of difficulties,including difficulty differentiating anomalies from the background and the susceptibility of anomaly detection accuracy to background noise.This study suggests an enhanced low-rank sparse matrix anomaly detection technique in conjunction with Gabor filtering to overcome these problems.To extract spatial texture features from hyperspectral image data,the Gabor filter is added to the low-rank and sparse matrix decomposition(LRaSMD).To maximize the retention of anomalous information in the samples,LRaSMD is improved by first choosing the initial results from the Gabor transform to produce a subset,which is limited by fixing the n lowest values and subtracting the background mean vector.Lastly,Utilizing the Mahalanobis distance,the generated background covariance matrix is found.The suggested algorithm's AUC values across five datasets are,in order,0.9934,0.9967,0.9995,0.9991,and 0.9964.The experimental findings reveal that the suggested method performs very well in anomaly detection by effectively differentiating between the background and anomalies and drastically reducing background contamination of anomalous pixels.

关 键 词：高光谱图像 异常检测 低秩稀疏矩阵分解 特征提取 GABOR滤波器 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]