利用SVM与灰度共生矩阵从QuickBird影像中提取枇杷信息

以福建省莆田市东圳水库库区为例,采用QuickBird卫星影像,利用主成分分析方法对灰度共生矩阵方法提取的地物纹理特征进行筛选,选择最佳的影像纹理特征,组成新的波段组合,并应用支持向量机方法(Support Vector Machine,SVM)进行枇杷树的提取分类,最后与只依靠光谱信息来分类的SVM法分类结果进行比较,其分类总精度由原来的71.33%提高到了86.67%,Kappa系数也由原来的0.6410提高到了0.8293,分类精度明显提高,表明光谱信息加入纹理特征信息能辅助并提升高分辨率遥感枇杷树信息提取的精度。

Study on Extracting of Loquat Information Using SVM and Gray-level Co-occurrence Matrix from QuickBird Image

We take Dongzhen Reservoir district of Putian as an example and present a methodology of exracting loquat information using support vector machine\|SVM and gray\|level co\|occurrence matrix from QuickBird image.Firstly,this paper calculating the textural measures using grey level co\|occurrence matrix and determining the optimum parameters for textural information by principal component analysis.Then the support vector machine was applied to classify the remote sensing imagery of the study area.Comparing with the result which depends only on spectrum information.The total classification accuracy for the former method has rised to 86.67% from 71.33%.Kappa coefficient change from 0.6410 to 0.8293.The increase of classification accuracy of exracting loquat information indicates that it is an effective method to fuse spectral and textural information on high\|resolution remote sensing classification.