Abstract:

Hyperspectral anomaly detection refers to identifying ground objects that deviate from normal background distributions and have low probability and small scales from scenes involving mixed multi-class ground objects, spectral feature overlaps, and noise interference. This technology has received extensive attention in recent years. Although collaborative representation-based anomaly detection algorithms demonstrate excellent performance in hyperspectral image anomaly detection, their time costs are too high to enable widespread application. To address this issue, this paper proposes a hyperspectral image anomaly detection algorithm based on window reconstruction and collaborative representation, which consists of two stages. Window reconstruction is performed on hyperspectral background pixels in scenes involving mixed multi-class ground objects, spectral feature overlaps, and noise interference. The K-means clustering algorithm is adopted to separate the anomaly pixels with similar features, and a reconstructed background dictionary matrix is obtained. Combining with thecollaborative representation idea in reconstructing background pixels—where background pixels can be approximately represented by their spatial neighboring pixels, while abnormal pixels cannot—anomaly detection is conducted on the background dictionary matrix obtained through clustering, which enables more accurate and efficient detection of anomalies. This method is compared with other algorithmsthrough comparative experiments on three real datasets and one synthetic dataset. The results show that it has good results in terms of both time and anomaly detection effect.

Key words: hyperspectral image, anomaly detection, window reconstruction, collaborative representation, experimental verification