ECAE: Edge-Aware Class Activation Enhancement for Semisupervised Remote Sensing Image Semantic Segmentation

Remote sensing image semantic segmentation (RSISS) remains challenging due to the scarcity of labeled data. Semisupervised learning can leverage pseudolabels to enhance the model's ability to learn from unlabeled data. However, accurately generating pseudolabels for RSISS remains a significant challenge that severely affects the model's performance, especially for the edges of different classes. To overcome these issues, we propose a semisupervised semantic segmentation framework for remote sensing images (RSIs) based on edge-aware class activation enhancement (ECAE). First, the baseline network is constructed based on the average teacher model, which separates the training of labeled and unlabeled data using student and teacher networks. Second, considering local continuity and global discreteness of object distribution in RSIs, the class activation mapping enhancement (CAME) network is designed to predict local areas more remarkably. Finally, the edge-aware network (EAN) is proposed to improve the performance of edge segmentation in RSIs. The combination of the CAME with the EAN further heightens the generation of high-confidence pseudolabels. Experiments were performed on two publicly available remote sensing semantic segmentation datasets, Potsdam and ISPRS Vaihingen, which verify the superiorities of the proposed ECAE model.