The purpose of this paper is to review recent advances in semi-supervised learning based on semisupervised learning in the field of semantic segmentation of remotely sensed images, with a special focus on semi-supervised learning methods based on generative and discriminative models. The definition, task and application background of semantic segmentation of remote sensing images are first introduced, followed by an overview of the limitations of traditional supervised learning methods in this field. Then the application of semi-supervised learning framework in semantic segmentation of remote sensing images is discussed in detail, including methods based on generative model and discriminative model. In the generative model, the generative adversarial network (GAN) method and the variational autoencoder (VAE) method are discussed in detail, and their applications in semantic segmentation of remote sensing images are explored. After that, we focus on self-training, pseudo-labeling and consistency training methods in discriminative modeling, discuss their principles, advantages and limitations, and explore the effectiveness of their applications in semantic segmentation of remote sensing images. Finally, the challenges of current research and future directions are summarized to provide reference and outlook for further research in this field.

Zekun Li, Weidong Sun, Lin Zheng, Yichong Wang. A Review of Semi-Supervised Learning Frameworks for Semantic Segmentation of Remote Sensing Images. Academic Journal of Computing & Information Science (2024), Vol. 7, Issue 6: 76-80. https://doi.org/10.25236/AJCIS.2024.070612.

[1] Zhuang F, Qi Z, Duan K, et al. A Comprehensive Survey on Transfer Learning[J]. Proceedings of the IEEE, 2021, 109(1): 43-76. 

[2] Locatello F, Tschannen M, Bauer S, et al. Disentangling factors of variations using few labels [C]//ICLR. Proceedings of the 8th International Conference on Learning Representations. Addis Ababa: ICLR, 2020. 

[3] Dittadi A, Träuble F, Locatello F, et al. On the transfer of disentangled representations in realistic settings[C]//ICLR. Proceedings of the 9th International Conference on Learning Representations. addis Ababa: ICLR, 2021. 

[4] Tschannen M, Bachem O, Lucic M. Recent Advances in Autoencoder-Based Representation Learning[J]. 2018.DOI:10.48550/arXiv.1812.05069.

[5] Husnain M, Missen M M S, Mumtaz S, et al. Visualization of high-dimensional data by pairwise fusion matrices using t-SNE[J]. Symmetry, 2019, 11(1): 107.