Waterbody Area Segmentation of Satellite Images Using U-Net with Dense Block Enhanced Encoder

This paper presents an advanced approach for waterbody segmentation in satellite images using a Dense U-Net model with a Dense Block-enhanced encoder. Accurate segmentation of water areas in satellite imagery is crucial for environmental monitoring, flood management, and resource planning. Traditional U-Net models are effective but often lack the capacity to capture fine details required for high-resolution satellite data. To address this, the proposed Dense U-Net replaces standard convolutional layers in the encoder with Dense Blocks, promoting feature reuse and deeper extraction of spatial information. Additionally, multi-scale fusion is incorporated to handle various spatial scales, enabling the model to accurately segment both large and small water regions. The simulation outcome shows that the proposed method segments the waterbody region with high Accuracy (96.7%), Recall (95.4%), and Dice coefficient (96.1%), with robust Specificity (99<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) and F1 scores (96.6%), underscoring its balanced performance in segmenting water regions. Performance metrics indicate the model's effectiveness in isolating water areas with minimal false positives and accurate boundary delineation. This enhanced approach provides a reliable solution for detailed waterbody segmentation in satellite images.