Hyperspectral Anomaly Detection With OCF- And Otsu-Based Isolation Forest

Hyperspectral anomaly detection (HAD) aims to identify anomalous pixels that differ significantly from their surrounding environment by analyzing the spectral characteristics of each pixel in a hyperspectral image (HSI). This technique is widely used in remote sensing, environmental monitoring, military reconnaissance, and other fields. Existing HAD methods face high computational costs, feature redundancy, and poor segmentation, which results in high-dimensional data processing and detection efficiency. To address these issues, this article proposes an isolation forest (iForest) method based on OCF and Otsu (OOIF), which is based on the assumption that anomalous pixels are more easily isolated than normal pixels. Specifically, when constructing the isolation tree (iTree) structure, OOIF first uses the optimal clustering framework (OCF) to select band combinations that are information-rich and low-correlated. Secondly, this method introduces Otsu’s method to optimize the selection of hyperplane intercepts for iTree partitioning. By calculating the slope that maximizes data separation, the accuracy of each split is further enhanced. Experimental results show that the algorithm performs better than the comparative algorithms on multiple datasets.