A Review of Image Semantic Segmentation for Remote Sensing Image

With the rapid development of remote sensing technology, semantic segmentation of high-resolution aerial images has become a research hotspot in fields such as land planning and disaster monitoring. Its core objective is to achieve fine-grained identification and classification of pixel-level geographic elements such as roads, buildings, water bodies, and vegetation. Current mainstream methods include two categories: Generative Adversarial Networks (GANs) and Convolutional Neural Networks (CNNs). GAN-based models generate visually compelling segmentation results through style transfer but often lack deep semantic understanding of geographic features. In contrast, CNN-based encoder-decoder architectures significantly improve segmentation accuracy by leveraging techniques such as atrous convolution and multi-scale feature fusion. In recent years, the introduction of Transformer architecture and hybrid neural networks has further enhanced the balance between global context modeling and local detail extraction, improving the robustness of segmentation in complex terrains. However, challenges remain, including high computational costs for processing high-resolution images, limited model generalization, and class imbalance issues. This review systematically summarizes the strengths, weaknesses, and cutting-edge advancements of typical models, providing theoretical support for improving the performance of intelligent interpretation of remote sensing imagery.