Selective Search Collaborative Representation for Hyperspectral Anomaly Detection

Abstract

As an important tool in hyperspectral anomaly detection, collaborative representation detection (CRD) has attracted significant attention in recent years. However, the lack of global feature utilization, the contamination of the background dictionary, and the dependence on the sizes of the dual-window lead to instability of anomaly detection performance of CRD, making it difficult to apply in practice. To address these issues, a selective search collaborative representation detector is proposed. The selective search is based on global information and spectral similarity to realize the flexible fusion of adjacent homogeneous pixels. According to the homogeneous segmentation, the pixels with low background probability can be removed from the local background dictionary in CRD to achieve the purification of the local background and the improvement of detection performance, even under inappropriate dual-window sizes. Three real hyperspectral images are introduced to verify the feasibility and effectiveness of the proposed method. The detection performance is depicted by intuitive detection images, receiver operating characteristic curves, and area under curve values, as well as by running time. Comparison with CRD proves that the proposed method can effectively improve the anomaly detection accuracy of CRD and reduce the dependence of anomaly detection performance on the sizes of the dual-window.