ABSTRACT With the recent advancements in deep learning and remote sensing technology, several semantic segmentation methods based on convolutional neural networks have been utilized for the semantic segmentation of remote sensing images. However, these models tend to lose valuable shallow detail information, particularly boundary information, due to feature compression during the encoding process. To address this issue, we propose an enhanced semantic segmentation model, EBANet, that focuses on boundary perception. EBANet consists of two parts—boundary path and spatial path, with boundary prediction serving as an independent subtask to enhance the segmentation model. Our model effectively aggregates boundary features and semantic features to achieve boundary enhancement segmentation. To achieve this, we employ a boundary filtering module (BFM) based on the gating mechanism that allows gating the low‐level boundary features through high‐level semantic understanding, thus effectively eliminating noisy data. The boundary‐guided contextual aggregation module (BCAM) enhances the interaction of boundary features with semantic features by establishing non‐local response relationships from boundary features to semantic features. We evaluated EBANet on high‐resolution remote sensing datasets Vaihingen and Potsdam, and the results showed that EBANet outperformed other models in terms of mIoU, F1‐Score, and OA, achieving the highest accuracy levels. On the Vaihingen dataset, EBANet's accuracy levels reached 68.31%, 81.05%, and 84.31%, respectively, while on the Potsdam dataset, it achieved 73.06%, 86.37%, and 89.64%, respectively.

Summary Semantic segmentation of remote sensing images plays a significant role in many applications such as urban planning and ecological protection, but its semantic segmentation suffers from large intra‐category variation and large differences in the scale of objects, so it is prone to misclassification. To cope with this issue, an embedded channel's categorical attention module (ECCA) is proposed to extract contextual information from the perspective of categories, and a channel attention module is embedded in it to achieve multiple contextual information extraction. Combined with the remote sensing atrous spatial pyramid pooling module (RSASPP), which is composed of atrous convolution with different expansion rates, feature fusion of objects at different scales is achieved. The refinement module (RM) is added for boundary refinement to achieve finer segmentation. Experiments are conducted on the WHDLD dataset to prove the effectiveness of the method.