Simple and Efficient: A Semisupervised Learning Framework for Remote Sensing Image Semantic Segmentation

Semantic segmentation based on deep learning has achieved impressive results in recent years, but these results are supported by a large amount of labeled data, which requires intensive annotation at the pixel level, particularly for high-resolution remote sensing (RS) images. In this work, we propose a simple yet efficient semisupervised learning framework based on linear sampling (LS) self-training, named LSST, to improve the performance of RS image semantic segmentation. Specifically, the classical pseudolabeling-based self-training paradigm is enhanced by injecting strong data augmentations (SDAs) applicable to RS images, based on which a powerful baseline is constructed. Nevertheless, the problem of insufficient data training to generate pseudolabels with a high level of noise persists, and the noisy pseudolabels will continue to accumulate and impede model improvement during the retraining phase. Previous works commonly employ a predefined threshold to remove noise, but it will lead to overfitting the model to easily identified classes. To address it, a method using LS is presented for assigning thresholds to different classes in an adaptive manner, which provides noiseless regions for retraining. Experiments prove that the proposed pixelwise selection is more available for segmentation than image-level selection in RS images. Finally, LSST achieves state of the art on several datasets and different evaluation metrics.