Hyperspectral Anomaly Detection: A survey

Abstract

Hyperspectral imagery can obtain hundreds of narrow spectral bands of ground objects. The abundant and detailed spectral information offers a unique diagnostic identification ability for targets of interest. Hyperspectral anomaly detection aims to find targets without prior knowledge, which has attracted attention as a branch of target location. In this article, current hyperspectral anomaly detection methods, anomaly detection performance evaluation techniques, and hyperspectral anomaly detection data sets are widely investigated. Among them, hyperspectral anomaly detection methods can be classified into seven categories: statistic-based, distance-based, reconstruction-based, subspace-based, spatial–spectral-based, deep learning-based, and real-time anomaly detection. The performance of different types of detection methods is also verified with three real hyperspectral data sets. Finally, conclusions about hyperspectral anomaly detection are summarized, and challenges for future research are discussed.