LigCDnet:Remote Sensing Image Cloud Detection Based on Lightweight Framework

Cloud contamination is inevitable in remote sensing images, resulting in a large number of images that cannot be applied in various fields. Therefore, cloud detection is one of the important tasks in remote sensing image preprocessing, aimed at removing images obstructed by clouds. Most existing methods are mostly based on CNN and feature a complex network structure, requiring a significant amount of computational resources, making it challenging to deploy them in practical applications. To tackle this problem, we propose a lightweight cloud detection framework (LigCDnet) with a lightweight feature extraction module (LFEM), a channel attention module (CAM), and a lightweight feature pyramid module (LFPM). The LFEM serves as the backbone of the network to capture rich spatial and contextual information; the CAM adaptively adjusts the channel weights of the feature maps; and the LFPM extracts cloud features at multiple scales. The effectiveness of our approach is evaluated on two public datasets, GF-1 and LandSat8. Extensive experiments have demonstrated that the proposed LigCDnet achieves state-of-the-art detection accuracy while significantly reducing computational burden and having a smaller model size.