HAPNet: A Hybrid Adaptive Perception Network for Remote Sensing Semantic Segmentation

Semantic segmentation of remote sensing images is widely applied in land use monitoring, urban planning, and other fields. However, mainstream models such as Deeplabv3+ struggle to accurately identify small objects and to simultaneously capture both channel and spatial information in complex scenes. To address these limitations, this paper proposes a lightweight network that dynamically adjusts the receptive field and effectively captures both spatial and channel-wise features. MobileNetV2 is adopted as the backbone to reduce the number of parameters and computational complexity. Experiments conducted on the Potsdam datasets demonstrate that the proposed method significantly improves segmentation accuracy while maintaining a lightweight structure, achieving superior performance in terms of mIoU and inference speed compared to existing mainstream methods.