Label Propagation and Contrastive Regularization for Semisupervised Semantic Segmentation of Remote Sensing Images.

Remarkable progress based on deep neural networks has been achieved in the semantic segmentation of remote sensing images (RSIs). However, pixel-level labeling is expensive for RSIs. Semisupervised semantic segmentation becomes an alternative approach to reduce the cost of annotation, and it is crucial to utilize efficiently a large number of unlabeled data. Nevertheless, inevitably, there is an unbalanced class distribution between labeled and unlabeled data in a remote sensing scene. Existing semisupervised methods train unlabeled images in isolation from labeled images and only learn reliable pixel pseudo-labels, leading to underutilization of unlabeled images. This article proposes a novel semisupervised semantic segmentation approach based on label propagation and contrastive regularization for RSIs. Specifically, the unlabeled images are augmented by randomly copy-pasting the class regions from labeled images. A prototype feature constraint module is used to enforce the constraint on the pixel features of unlabeled images relying on the prototype features from labeled images, achieving feature alignment on the entire dataset. Furthermore, we present the region contrastive learning (RCL) module that guides the model to learn feature consistency under different perturbations and compact feature representations over class regions on unlabeled images. Extensive experimental results on multiple remote sensing datasets demonstrate that our proposed approach achieves superior performance compared with state-of-the-art semisupervised semantic segmentation methods.