Multi-Scale Anomaly Detection in Hyperspectral Images Based on Sparse and Low Rank Representations

Abstract

Anomaly detection is a hot topic in hyperspectral data processing since no prior information about the target is required. Meanwhile, multi-scale information can improve the detection performance. This paper proposes a multi-scale anomaly detection algorithm in hyperspectral images based on the sparse and low rank representation. Some pixels are randomly selected to construct the dictionary, and the pixels belonging to large abnormal targets are selected with high probability. Therefore, the low rank matrix and dictionary constitute the pure background component and large abnormal targets, and the sparse matrix contains noise and smaller abnormal targets. Using recursive sliding array RX detection algorithm, large abnormal targets can be detected in the reconstructed image, and small abnormal targets can be detected in the residual between sparse matrix and the reconstructed image. The final detection result is the combination of the two results. Experimental results d <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> emonstrate that the algorithm achieves very promising performance.