Hyperspectral Image Mixed Noise Removal Using a Subspace Projection Attention and Residual Channel Attention Network

Although the existing deep-learning-based hyperspectral image (HSI) denoising methods have achieved tremendous success, recovering high-quality HSIs in complex scenes that contain mixed noise is still challenging. Besides, these not fully explored local and global spatial–spectral information of HSIs. To address above issues, a novel HSI removal network called subspace projection attention residual channel (SPARCA-Net) proposed. Specifically, we propose an orthogonal (OSPA) module to adaptively learn generate bases signal project input into such space remove noise. By leveraging spatial relations, OSPA able reconstruct structure feature maps more precisely. We further (RCA) emphasize interdependence between exploit correlation them, which could enhance channel-wise adaptive learning. In addition, multiscale joint learning strategies are employed capture features reduce degradation problem, respectively. Synthetic real data experiments demonstrated proposed outperforms many advanced both quantitative qualitative assessments.