Abstract: In response to the suboptimal performance of flood water extraction from single-source remote sensing images, a novel multi-source remote sensing image semantic segmentation network called MACFNet is proposed based on MCANet. This model simultaneously utilized optical and SAR images. A factorized depth-wise asymmetric split-shufflenon-bottleneck(FDSS-nbt) module was introduced to enhance the network’s capability to capture dense features. A global feature context adaptation module(GFCAM) was designed to effectively model global context. Experimental results on the WHU-OPT-SAR dataset demonstrate a 15% improvement in water body mIoU. These results validate the effectiveness and superiority of the proposed model, which can be utilized for more accurate extraction of water bodies and other targets.