Answers are grounded in this paper and cite their sources.
Hyperspectral anomaly detection based on representation learning has received much attention in recent years. Due to the lack of prior knowledge about anomalies, it is difficult for a collaborative representation (CR) model to obtain a pure dictionary in the ideal case. Some algorithms proposed to remove anomalous pixels from a dictionary, which may result in the removal of contributing background atoms. To address such a problem, this paper introduces a competitive regularization constraint term into the CR model, and divides the dictionary into anomaly and background classes using an outlier searching strategy, while adding competition weights to improve the competitiveness of the background. To better reconstruct the pixels, the Jaccard similarity coefficient is combined with the distance-weighted regularization matrix to adjust the representation coefficients. In addition, to make the most of the information from the hyperspectral data, a significance mechanism is introduced to construct an anomaly saliency weight to achieve the purpose of suppressing the background and highlighting anomalies. Experiments on five real datasets show that the detection performance of the proposed method is better than other advanced algorithms.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2023) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: