空间加权的孤立森林高光谱影像异常目标检测  

Hyperspectral anomaly detection based on isolation forest with spatial weighting

作　　者：薛园园 黄远程[1] 苏远超 XUE Yuanyuan;HUANG Yuancheng;SU Yuanchao(Xi'an University of Science and Technology,Xi'an 710054,China)

机构地区：[1]西安科技大学,西安710054

出　　处：《测绘科学》2021年第7期92-98,共7页Science of Surveying and Mapping

基　　金：国家自然科学基金青年基金项目(42001319)。

摘　　要：针对孤立森林算法在高光谱影像异常目标检测中易产生大量虚警的问题,该文将异常目标在空间分布的稀缺性与目标光谱的差异性两个先验结合,提出了一种空间加权的孤立森林异常目标检测方法。首先利用孤立森林算法计算目标的光谱异常度,得到初步的检测结果;然后分析初步结果中目标区域的连通面积,以连通域面积为变量,基于高斯核计算目标的空间稀缺性,得到目标的空间权重属性;最后将表达空间稀缺性的属性与光谱异常度加权相乘,实现了对异常目标的准确检测。在五组高光谱数据集上的实验结果表明,该方法具有较好的检测性能。In view of the problem that the iForest(isolation forest)method can easily produce a number of false alarms in hyperspectral anomaly detection,this paper combines the two prior knowledge of the spatial distribution scarcity of anomalous instances with the spectral difference of the targets and proposes a hyperspectral anomaly detection based on isolation forest with spatial weighting method.Firstly,the spectral anomaly scores of the targets are calculated by iForest algorithm,and then preliminary detection result is obtained.Secondly,the connected area of the target region in the preliminary result is analyzed.Taking the connected area as a variable,the spatial scarcity of the target is calculated based on the Gaussian kernel,and the spatial weight attribute of the target is obtained.Finally,the expression space scarcity property data is weighted multiplied by the spectral anomaly scores to achieve the accurate detection of abnormal targets.The experimental results obtained with five hyperspectral data sets demonstrate that the proposed algorithm shows an outstanding detection performance compared with the other methods.

关 键 词：高光谱影像 孤立森林 空间加权 异常目标检测 

分 类 号：P237[天文地球—摄影测量与遥感]