集成面向对象影像分析与KNN算法的多光谱遥感影像分类探讨

Discussion on Multispectral Remote Sensing Image Classification Based on Object-oriented Image Analysis and KNN Algorithm

文章分别使用基于像元和基于对象的KNN分类器算法对1024*1024像元大小的宁夏中卫市地区15m空间分辨率Landsat8融合影像进行分类,比较二者分类效率和准确率,探讨其在影像分类上的不同。研究表明无论是基于对象还是基于像元的KNN分类器算总体分类精度都在90%以上。但基于对象的KNN分类器算法相比基于像元的总体分类精度提高1.9%,Kappa系数提高0.026。且使用相同的训练样本进行训练和分类,基于对象的KNN分类器算法仅耗时0.281秒,而基于像元的KNN分类器算法耗时53分7.275秒。

This paper uses the pixel-based and object-based KNN classifier algorithm to classify the 15m spatial resolution Landsat8 fusion image of Ningxia Zhongwei City with a 1024*1024pixel size,compare the classification efficiency and accuracy,and explore its image classification.The research shows that the overall classification accuracy of both the object-based and pixel-based KNN classifiers is above 90%.However,the object-based KNN classifier algorithm improves the overall classification accuracy by 1.9%and the Kappa coefficient by 0.026.And using the same training samples for training and classification,the object-based KNN classifier algorithm only takes 0.281 seconds,while the pixel-based KNN classifier algorithm takes 53 minutes and 7.275 seconds.