Attention-Based Octave Dense Network for Hyperspectral Image Denoising
Ziwen Kan, Suhang Li, Yi Zhang · 2021
Inevitable corruption and degeneration make performance of high-level semantic tasks in Hyperspectral images (HSIs) unsatisfactory. To suppress noise and preserve HSIs spatial-spectral structure, we propose an Attention-Based Octave Dense Network (AODN) to extract spatial-spectral features consistent to the structure prior. The features are fine-tuned by attention module and then the octave network focus on high frequency noise feature learning through. The simulated and real-world experiments demonstrate that the proposed method outperforms existing traditional and learning-based methods in quantitative evaluations, visual effects and classification accuracy.