SSPNet- A Strip Pooling and SCSE Enhanced Multi-Scale U-Net for Fine-Grained Water Body Extraction

Accurate extraction of water bodies from remote sensing imagery is crucial for environmental monitoring and water resource management. To address the challenges of blurred boundaries and insufficient multi-scale feature representation, this paper proposes SSPNet, an improved U-Net model integrating Pyramid Pooling (PPM), Strip Pooling, and SCSE attention. PPM enhances global context awareness, Strip Pooling captures elongated water structures, and SCSE adaptively emphasizes boundary features. Trained on the LoveDA dataset with a hybrid BCE–Dice loss, SSPNet achieves an mIoU of 87.34%, mPA of 92.26%, and accuracy of 97.74%, improving by 3.25%, 0.83%, and 0.75% over the baseline U-Net. The results demonstrate superior boundary clarity and small-scale water recognition in complex urban–rural scenes.