<p>Deep learning has made significant advancements in the semantic segmentation of UAV images. However, challenges persist, including difficulties in capturing long-range information and inadequate edge segmentation performance. To address these issues, this study proposes an improved remote sensing image segmentation model, termed Semantic Detail Fusion and Contextual Enhancement Network for remote sensing image segmentation using an advanced DeepLabV3+ model (SDFCENet). Specifically, a Multi-scale Context Feature Fusion (MCFF) module is constructed, utilizing a structure that combines multi-scale pooling, convolution, and feature fusion. This structure aims to effectively capture contextual information in complex scenes, enhancing the model’s ability to process details and comprehend global structures. Additionally, a Global Context Enhancement Self-Attention Module (GCESAM) is introduced by integrating self-attention mechanisms with global average pooling. Attention maps are generated to enhance feature representation, and residual connections are incorporated to balance local details and global features, significantly improving the model’s capacity to capture long-range information. Furthermore, by fusing semantic information with spatial details and incorporating local and global attention mechanisms, a Semantic Detail Fusion Module (SDFM) is proposed to effectively capture and fuse multi-level features. This module aims to improve edge segmentation accuracy and enhance the model’s robustness. Experiments on the Aerial Semantic Segmentation Drone and UAVid datasets demonstrate that the improved model achieves 67.2% and 59.1% Mean Intersection over Union (MIoU) values on these datasets, respectively, representing a substantial performance improvement over the conventional DeepLabV3+ model.</p>