Channel Attention Cloud Detection Network for Ground-Based Cloud Detection

Cloud detection is indispensable in ground-based cloud observation, and it can implement automatic cloud cover estimation. Cloud detection is quite challenging because of blurred cloud boundaries and variable shapes. In this paper, we propose a new network Channel Attention Cloud Detection Network (CACDN) for ground-based cloud detection. The proposed CACDN is an encoder-decoder architecture, and we design the cloud channel attention (CCA) module to filter information for accurate cloud detection. We conduct the experiments on TLCDD, and the experimental results show that our method achieves better results than other methods, thus proving the effectiveness of the proposed CACDN.