Weighted Discriminative Collaborative Competitive Representation With Global Dictionary for Hyperspectral Target Detection

Hyperspectral target detection (HTD) is a promising yet challenging endeavor in remote sensing image processing. Representation learning-based detectors have become one of the mainstream methods to address the task. However, these methods often suffer from weak separability between the target and the background, which results in inferior target detection performance. The reason is that their detection model cannot effectively distinguish the subtle differences between the target and the background. To tackle this issue, this article proposes a new weighted discriminative collaborative competitive representation (WDCCR) model for HTD. In WDCCR, the separability between targets and backgrounds is enhanced by integrating discriminative, competitive, and weight constraints. Meanwhile, to obtain pure background pixels for the representation model, we investigate a new category-based pixel selection method with target orthogonal purification (CPSTOP). The proposed WDCCR target detector is evaluated on six hyperspectral datasets. Experimental results demonstrate that WDCCR outperforms other advanced methods, achieving good detection performance in HTD. The code will be available at
https://github.com/liurongwhm
.