LDBNet: A Lightweight Semantic Segmentation Network with Dual-Branch

With the advancement of UAV (Unmanned Aerial Vehicle) technology, UAV imagery has increasingly become a critical component in the field of semantic segmentation. However, due to the challenges posed by varying scales, complex backgrounds, and the demand for real-time processing, there is an urgent need in the industry for a precise and efficient lightweight segmentation method. To address these challenges, we propose a lightweight dual-branch semantic segmentation model. This dual-branch architecture processes global semantic information and local edge details separately, enabling the model to capture both broad contextual features and fine-grained details. The global branch leverages a Multi-Scale Feature Enhancement module to extract and fuse multi-scale features, thereby enhancing the model's ability to understand complex scenes. Concurrently, the local branch incorporates a Similarity Add module, which refines edge information through attention-based fusion of local and global features. Finally, the Attention Fusion Module dynamically adjusts the weights of the feature maps, achieving an optimal balance between accuracy and computational efficiency. Extensive experiments demonstrate that our model significantly improves segmentation performance on UAV imagery, maintaining high precision while achieving lightweight parameters, making it highly suitable for on-board UAV applications in diverse and complex environments.