几种遥感影像分类方法比较与分析
Comparison and Analysis of Several Remote Sensing Image Classification Methods

遥感技术是一种重要的自然资源调查监测手段，国内外研究学者提出多种监督分类算法，具有不同的分类精度和特点，各种监督分类算法被集成到不同的软件平台。为获取各分类算法及其在不同平台的分类精度，本文以某县为研究区域，选取哨兵Sentinel-2多光谱影像为实验影像，利用多种监督分类方法进行分类，并将2021年国土变更调查数据作为标准数据，采用全样本定量评价方法检验不同分类方法的精度及特点。经实验证明，基于ArcGIS的最大似然法分类精度最高，基于ENVI的支持向量机法对各地类分类效果更为均衡，基于简译的面向对象分类方法在10米分辨率影像的分类精度与传统分类方法差距不明显，本文实验结果对研究及工作人员选取监督分类方法具有重要的参考价值。
Remote sensing technology is an important means of natural resource investigation and monitoring. Domestic and foreign researchers have proposed various supervised classification algorithms with different classification accuracy and characteristics. Various supervised classification algorithms are integrated into different software platforms. In order to obtain the classification accuracy of various classification algorithms on different platforms, this article takes a certain county as the research area, selects Sentinel-2 multispectral images as experimental images, uses multiple supervised classification methods for classification, and uses the 2021 land change survey data as standard data. The accuracy and characteristics of different classification methods are tested using a full sample quantitative evaluation method. Experimental results have shown that the maximum likelihood method based on ArcGIS has the highest classification accuracy, while the support vector machine method based on ENVI has a more balanced classification effect on different regions. The object-oriented classification method based on simplified translation has no significant difference in classification accuracy compared to traditional classification methods in 10-meter resolution images. The experimental results of this article have important reference value for research and staff to choose supervised classification methods.