ViL-UNet: Multi-Scale Global–Local Feature Fusion for High-Resolution Remote Sensing Segmentation

Efficiently extracting and integrating multi-scale global and local information remains challenging in remote sensing semantic segmentation. Convolutional neural networks (CNNs) excel at capturing fine-grained details but cannot model long-range dependencies, while Transformer-based and Mamba-based models capture broad context yet often overlook local details. Recent hybrid architectures combine both of them to balance these strengths but still lack the ability to fuse multi-scale global and local features in a unified, efficient manner. In this work, we propose ViL-UNet, a dual-branch encoder–decoder network that addresses this challenge by combining a Vision-LSTM (ViL) global encoder with a CNN-based local encoder. The global encoder captures long-range dependencies across multiple scales, while the local encoder preserves fine spatial details. A novel Multi-Scale Fusing Module then adaptively merges these complementary representations, resizing all global and local features to a common spatial grid, concatenating them and producing a unified feature map that includes broad contextual cues and precise local information. The decoder applies Efficient Global-Local Attention and Squeeze-and-Excitation, and via skip connections brings the fused multi-scale features back to higher resolutions. To demonstrate ViL-UNet’s effectiveness, we compared it against state-of-the-art methods on the ISPRS Vaihingen and LoveDA Urban benchmarks. Across both datasets, ViL-UNet consistently delivers more accurate semantic segmentation by efficiently capturing multi-scale global and local context in large-scale remote-sensing scenes.