An Improved MFFCD-Net Method for Day and Night Cloud Detection in Remote Sensing Images

Cloud detection is a fundamental task in remote sensing image processing, with significant applications in climate monitoring and Earth observation. Despite recent advances, challenges such as thin cloud leakage, boundary blurring, and nighttime cloud-light confusion remain unresolved. This paper proposes an enhanced MFFCD-Net framework with two key innovations: (1) A redesigned Multi-scale Feature Extraction and Fusion (MFEF) module that integrates asymmetric dilated convolutions and attention mechanisms to capture contextual cloud features; (2) A hybrid loss function combining Dice loss and focal loss to address extreme class imbalance. The ablation studies further validate the contribution of each proposed component. Comprehensive experiments on VIIRS DNB datasets demonstrate that our method achieves 90.5% overall accuracy (OA) for daytime images and 81.5% OA for nighttime scenes, better than U-Net and DeepLabV3+.