VHSR图像基于分割对象分类器性能评价

THE PERFORMANCE OF OBJECT-BASED CLASSIFIERS IN THE CLASSIFICATION OF VHSR IMAGE

对比研究了平行六面体、最近邻分类法、最大似然法、神经网络等经典分类算法以及近年来新发展的支持向量机分类算法在基于分割对象的高分辨率遥感图像分类中的性能,详细分析了不同内积核函数对于支持向量机分类的影响。对两个试验区进行试验的结果表明,支持向量机分类算法分类精度得到明显改善,同时分类结果受参数、样本选择等影响较小,稳定性好。

The parallelepiped classifier （PC）, minimum distance classifier （MDC）, Maximum Likelihood Classifier （MLC）, Neural network （NN） and, especially, the newly developed Support Vector Machines （SVM） were assessed in the object - based image analysis of VHSR data. The impacts of kernel configuration on the performance of the SVM and of the selection of training data of the four classifiers were also evaluated. The result reveals that SVM can improve the accuracy significantly, and is by far more stable than other algorithms in the classification of VHSR data based on OBIA.