遥感图像场景分类综述
2021-07-19钱园园刘进锋
钱园园 刘进锋
摘要:随着科技的进步,遥感图像场景的应用需求逐渐增大,广泛应用于城市监管、资源的勘探以及自然灾害检测等领域中。作为一种备受关注的基础图像处理手段,近年来众多学者提出各种方法对遥感图像的场景进行分类。根据遥感场景分类时有无标签参与,本文从监督分类、无监督分类以及半监督分类这三个方面对近年来的研究方法进行介绍。然后结合遥感图像的特征,分析这三种方法的优缺点,对比它们之间的差异及其在数据集上的性能表现。最后,对遥感图像场景分类方法面临的问题和挑战进行总结和展望。
关键词:遥感图像场景分类;监督分类;无监督分类;半监督分类
中图分类号:TP391 文献标识码:A
文章编号:1009-3044(2021)15-0187-00
Abstract: With the progress of science and technology, the application demand of remote sensing image scene increases gradually, which is widely used in urban supervision, resource exploration, natural disaster detection and other fields.As a basic image processing method, many scholars have proposed various methods to classify the scene of remote sensing image in recent years.This paper introduces the research methods in recent years from the three aspects of supervised classification, unsupervised classification and semi-supervised classification.Then, combined with the features of remote sensing images, the advantages and disadvantages of these three methods are analyzed, and the differences between them and their performance performance in the data set are compared.Finally, the problems and challenges of remote sensing image scene classification are summarized and prospected.
Key words: remote sensing image scene classification; Unsupervised classification; Supervise classification; Semi-supervised classification
遥感图像场景分类,就是通过某种算法对输入的遥感场景图像进行分类,并且判断某幅图像属于哪种类别。早期主要是依靠人工提取特征,这些方法一般情况下需要依靠有大量专业领域知识和实践经验的专家利用颜色、形状、纹理或者光谱信息来设计各种图像特征,进而进行分类。但是,随着技术尤其遥感技术的发展,遥感图像不仅从数量上急剧增加,而且从质量上来说遥感图像的复杂度以及图像的分辨率也都越来越高,这些因素限制了早期人工方法的继续使用。因此,基于机器学习以及深度学习的自动学习方法应运而生,并在图像处理领域大放异彩。
在过去的几十年里,各方面学者对遥感图像提出了许多自分类方法。……
