喀斯特地区遥感影像解译新算法——支持向量机算法

现行的遥感影像解译方法有监督分类和非监督分类。在监督分类中有平行算法,最小距离算法、最大似然算法等,而支持向量机是监督分类中的一种新的算法。本研究选择贵阳市花溪区小碧乡局部地区为研究对象,采用SPOT数据,分别运用最大似然算法和支持向量机算法对研究区遥感影像进行解译。通过建立混淆矩阵,来计算分类精度和Kappa系数。结果表明:支持向量机具有分类精度高,分类图斑完整等优点;但在时间的消耗上,支持向量机算法要比最大似然算法长。对于这两种算法而言,都存在地物光谱特征明显相异的地物易于区别,光谱相似的地物容易造成错分的现象,然而支持向量机分类精度要比最大似然分类精度高一些。支持向量机对样本数量具有敏感性,样本数量过多将导致运算时间过长。因此在实际运用中应根据实际情况,选择适合的算法。 

Application of a new remote sensing image interpretation method in karst area - support vector machine algorithm

The existing methods of remote sensing image interpretation are unsupervised classification and supervised classification. The supervised classification includes parallel algorithm, the minimum distance algorithm and maximum likelihood algorithm. Support Vector Machine is a new supervised classification algorithm. In this study, some parts in the Huaxi District, Xiaobi Township in Guiyang is selected as the research object. Remote sensing images are interpreted by means of the maximum likelihood algorithm and Support Vector Machine algorithm respectively with SPOT data. Through establishing confusion matrix, calculating classification accuracy and Kappa coefficient, it is found that the classification accuracy of support vector machine is high and classification polygon is integrity. But to the time of consumption, the support vector machine is longer than the maximum likelihood algorithm. According to the two algorithms, there are both ground objects easy to be distinguished for their spectral features being quite different from other objects and some ground objects with similar spectrum easy to lead to misclassification. However, in terms of the classification accuracy, SVM classification is higher than the maximum likelihood. SVM is sensitive to the number of samples, so too much sample size will cause too long operation. Selection of the two algorithms in practice still needs to consult the practical situation of the study area and contrast their merit and demerit.