遥感影像场景分类研究进展

Research Progress on High-Resolution Remote Sensing Image Scene Classification

遥感影像场景分类是遥感影像解译最基础的任务之一,在土地利用调查与分析、地质灾害监测、地理空间情报获取等方面具有重要的应用价值。文章对遥感影像场景分类方法研究进展进行了系统性的总结和分析:首先,梳理了常用的遥感影像场景分类的数据集,分析了遥感影像的特性给场景分类任务带来的挑战和困难;其次,对现有典型的遥感影像场景分类方法——基于手工设计特征的场景分类方法和基于深度学习的场景分类方法,进行了总结归纳,针对遥感影像场景分类任务分析了现有方法的优化改进方案;然后,对比了主流遥感影像场景分类方法的性能;最后,对遥感影像场景分类技术仍未解决的问题以及下一步遥感影像场景分类应用研究方向进行了总结和展望,探讨了遥感影像场景分类在高精度细粒度分类任务、高精度轻量化模型设计、少样本学习技术、遥感影像场景解译大模型等方面的研究前景,以期推动遥感影像场景分类任务实现更加深入的研究和广泛的应用。

Remote sensing image scene classification is a fundamental task in remote sensing image interpretation,which has important application value in land-use investigation,geologic disaster monitoring,geospatial intelligence acquisition and so on.This paper systematically summarizes and analyzes representative methods for remote sensing image scene classification.Firstly,the commonly used datasets for remote sensing image scene classification are sorted out,and the challenges and difficulties brought by the characteristics of remote sensing images to scene recognition are elaborated.Secondly,some typical remote sensing image scene classification methods are summarized,which includes hand-crafted features and deep learning.And the optimization and improvement for remote sensing image scene classification task are analyzed.Then,the performance of some mainstream methods is compared.Finally,the unsolved problems and the next research directions of remote sensing image scene classification are summarized and explored.And the research prospects of high-precision fine-grained scene classification task,high-precision lightweight algorithms,few-shot learning,and large model for remote sensing image scene interpretation are explored with the aim of promoting the remote sensing image scene classification and recognition tasks to achieve more in-depth research and a wide range of applications.