CTAD-Net: Cloud detection in cloud-snow coexistence scenarios using a cascaded encoder based on ResCNN and vision transformer

Cloud detection is a crucial preprocessing step in remote sensing image analysis. Despite numerous proposed methods, identifying clouds in mixed cloud/snow scenes remains challenging due to the high spectral similarity between snow/ice and clouds, which significantly interferes with detection performance. To address this, we propose a novel network architecture that integrates a Vision Transformer (ViT) with convolutional networks in order to leverage both global context and local features to enhance spatial and semantic feature extraction for cloud detection. We further improve the encoder’s multi-scale feature representation by incorporating Atrous Spatial Pyramid Pooling (ASPP). To mitigate the loss of low-level semantic information during upsampling, we design a Multi-scale Attention Aggregation Module (MAAM) for the decoder, which effectively fuses multi-branch features for superior image reconstruction. Experimental results on a high-resolution remote sensing dataset demonstrate that our approach outperforms state-of-the-art methods in detecting clouds within mixed cloud/snow regions, achieving a mIoU of 90.81 % and an F1-Score of 91.53 %.