Center-Guided Classifier for Semantic Segmentation of Remote Sensing Images

Wei Zhang, Mengting Ma, Yizhen Jiang, Yun Chen, Zhenhua Huang, Wangyu Wu, Kangning Cui, Rongrong Lian, Zhenkai Wu, Xiaowen Ma

Abstract: Compared with natural images, remote sensing images (RSIs) have a unique characteristic, i.e., larger intraclass variance, which makes semantic segmentation for RSIs more challenging. Moreover, existing semantic segmentation models for RSIs usually employ a vanilla softmax classifier, which has three drawbacks: 1) nondirect supervision for the pixel representations during training; 2) inadequate modeling ability of parametric softmax classifiers under large intraclass variance; and 3) an opaque process of classification decision. In this article, we propose a novel classifier (called CenterSeg) customized for RSI semantic segmentation, which solves the above-mentioned problems with multiple prototypes, direct supervision under Grassmann manifold, and interpretability strategy. Specifically, for each class, our CenterSeg obtains local class centers by aggregating corresponding pixel features based on ground-truth masks and generates multiple prototypes through hard attention assignment (HAA) and momentum updating. In addition, we introduce the Grassmann manifold and constrain the joint embedding space of pixel features and prototypes based on two additional regularization terms. Especially during the inference, CenterSeg can further provide interpretability to the model by restricting the prototype to a sample of the training set. Experimental results on three remote sensing segmentation datasets validate the effectiveness of the model. Besides the superior performance, CenterSeg has the advantages of simplicity, lightweight, compatibility, and interpretability. Code is available at https://github.com/xwmaxwma/rssegmentation

External IDs:doi:10.1109/tgrs.2026.3657453