Search

In the field of hyperspectral target detection, algorithms based on collaborative representation have shown excellent performance. However, the pollution of background dictionaries by anomaly has always been a problem, which affects the detection performance of this algorithm partly. This paper proposes an improved collaborative representation algorithm for hyperspectral anomaly detection, and designs a double-layer collaborative representation structure. Firstly, the first layer collaborative representation algorithm is used to detect most of the anomaly. Its neighborhood pixels are used for background purification to eliminate the pollution of the detected anomaly on the background dictionary. Then, the purified background dictionary is used to predict the background, and in the second layer, the collaborative representation is used for anomaly detection. Simulation experiments show that a simple double layer collaborative representation structure can effectively alleviate the background pollution caused by anomaly. The detection performance of this algorithm is significantly improved compared to the basic collaborative representation algorithm, and it also has relatively good detection performance.