基于拉普拉斯矩阵图的高光谱图像异常目标检测方法  

Laplacian Matrix Graph-based Anomaly Target Detection Method for Hyperspectral Graphs

作　　者：张峰干 何芳 肖磊 李婷 ZHANG Fenggan;HE Fang;XIAO Lei;LI Ting(Rocket Force University of Engineering,Xi’an 710025,China;General Office of the Beijing Municipal Committee,Beijing 100007,China)

机构地区：[1]火箭军工程大学,陕西西安710025 [2]北京市委办公厅,北京100007

出　　处：《火箭军工程大学学报》2025年第1期1-12,共12页Journal of Rocket Force University of Engineering

摘　　要：为提高高光谱图像异常目标检测的精度,提出一种基于拉普拉斯矩阵图的异常目标检测方法(Laplacian Matrix Graph for Anomaly Detection,LGD)。通过构造全连接图和高斯核函数构造的近邻矩阵,将高光谱图像中异常目标的位置和光谱信息进行联合处理,实现了高光谱数据不同波段之间的信息融合;利用图的傅里叶变换和拉普拉斯矩阵性能,将图信号的总变差作为判断异常目标的评价函数,实现了异常目标像素点的准确检测,避免了常规检测算法中的矩阵求逆问题,降低了算法的复杂度。在异常检测常用的AVIRIS-I、AVIRIS-II、ABU-urban-2、ABU-urban-4和EI Segundo 5种高光谱数据集上,进行了算法性能验证。实验结果表明:该算法在5种数据集上异常目标检测的AUC值与ROC曲线均优于其他算法,在检测精度上具有明显优势。To further improve the accuracy of anomaly target detection for hyperspectral graghs,an anomaly target detection method based on Laplacian matrix graph was proposed.First,by constructing a full-connection graph and a nearest neighbor matrix obtained by the Gaussian kernel function,the method connected the information fusion between different bands of hyperspectral data,and jointly processed the position and spectral information of the anomaly target in hyperspectral graphs.Second,by utilizing the Fourier transform and Laplacian matrix of the graph,the total variation in the graph signal was used as an evaluation function to determine the anomaly target.Accordingly,the accurate detection of anomaly target pixels was realized simultaneously,the matrix inversion calculation in the conventional detection algorithms was avoided and complexity of the algorithm was reduced.Finally,the proposed algorithm was validated on five hyperspectral data sets commonly used for anomaly detection:AVIRIS-Ⅰ,AVIRIS-Ⅱ,ABU-urban-2,ABU-urban-4,and EI Segundo.The experimental results showed that the AUC values and the ROC curves of the proposed algorithm on the five data sets for anomaly target detection outperform other algorithms,indicating that this method has advantage in detection accuracy.

关 键 词：高光谱图像 异常目标检测 图信号 拉普拉斯矩阵 图傅里叶变换 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]