基于最大似然和支持向量机方法的遥感影像地物分类精度评估与比较研究

Accuracy Evaluation and Comparison of Ground Objects Classification in Remote Sensing Images Based on ML and SVM Methods

遥感影像的监督分类算法在环境监测、地质调查等领域均有重要应用。本文利用最大似然(ML)分类器和支持向量机(SVM)分类器对土地利用和地表覆盖问题中地物类型的提取和识别进行研究,系统分析两种不同分类方法对地物分类结果的影响。通过选取Landsat LT5和LE7卫星遥感影像数据及定义训练样本,对比分析利用ML和SVM分类器的分类成果精度,其中Landsat LT5和ML、SVM组合的分类精度分别达94.64%和94.98%,而Landsat LE7和ML、SVM组合的分类精度则分别达97.63%和99.29%。研究表明,对于LT5影像,ML和SVM两种分类器的精度相当,而对于LE7影像,SVM分类器的精度明显高于ML分类器。

Supervised classification algorithm for remote sensing image has been significantly applied in the field of environmental monitoring and geologic survey.A comparison of Maximum Likelihood（ML）and Support Vector Machine（SVM）classifiers was conducted on extracting and recognizing the types of ground objects for land use and surface cover.The impacts of these two methods on the classification results were analyzed systematically.By selecting Landsat LT5 LE7 satellite remote sensing image and defining training samples,the classification accuracies of ML and SVM classifiers were compared.It is found that the classification accuracies of combining Landsat LT5 with ML SVM are 94.64% and 94.98%,while the classification accuracies of combining Landsat LE7 with ML SVM are 97.63% and 99.29%.The experiment results show that,for LT5 image,the accuracies of these two classifiers are almost the same,but for LE7 image,the accuracy of SVM classifier is significantly higher than that of ML classifier.