Dual-domain dehazing algorithm for remote sensing images with multi-concentration non-uniform cloud and fog
Optical remote sensing satellite images are easily affected by densely non-uniform clouds and hazy,leading to se-vere degradation in image quality.Existing dehazing algorithms mostly struggle to effectively handle densely non-uniform hazy,so this paper proposed a self-attention enhancement mechanism called a simple,parameter-free attention module(SimAM)that strengthens the representation and generalization capabilities of self-attention,enhancing perception of non-uniform clouds.To fur-ther improve the network's ability to represent texture details,a novel differential convolution detail enhancement block was de-signed,utilizing differential convolution operators to introduce gradient-level information and enhance the network's ability to re-store texture details.To achieve joint dehazing in both the RGB domain and the adaptive wavelet domain,a deep adaptive wavelet transform based on lifting scheme was introduced to realize adaptive wavelet space,enabling collaborative dehazing in dual do-mains.Experimental results show that the proposed method achieves 0.52 dB PSNR gain compared with the backbone model on popular remote sensing image dehazing datasets.
remote sensing image dehazingself-attention enhancementdifferential convolutiondeep adaptive wavelet transform based on lifting scheme
NETL
NSTL
万方数据
