High-Resolution Impervious Surface Area Segmentation Network Based on Multimodal Data Fusion

With accelerated urbanization, the increase in impervious surface area (ISA) leads to urban heat island effects and heightened flood risks, making accurate high-resolution ISA extraction crucial. Traditional methods face limitations, while deep learning models, particularly U-Net architectures, demonstrate superior performance. However, traditional methods relying on a single data source for ISA identification have limitations. Utilizing multimodal remote sensing data fusion can enhance the accuracy of ISA extraction. Therefore, this study proposes a multimodal data fusion and segmentation network that integrates synchronized multimodal data augmentation and a hybrid sliding window post-processing strategy to improve model generalization and robustness. The approach was validated by achieving the runner-up position in the 2025 IEEE WHISPER Multimodal Semantic Segmentation Challenge, demonstrating its effectiveness.