天绘一号是我国第一代传输型立体测绘卫星,主要用于科学研究、国土资源普查、地图测绘等领域的科学实验任务。以天绘影像为实验数据,利用面向对象的影像分割技术,通过选择合适的尺度参数对影像进行分割。结合SVM对得到的影像对象层进行分类实验。具体分析了SVM分类器核函数的选择以及参数的设置对分类精度的影响。最终分类实验结果的总体精度为90.857 1%,Kappa系数为0.858 1。将分类结果与传统基于像元的马氏距离分类法和最大似然值分类法的分类结果进行比较,总体精度分别提高了约29.29%、5.91%,Kappa系数分别提高了约0.35、0.06。实验结果表明,面向对象的SVM分类法不仅对影像分类的精度有大幅度的提高,同时,也很好地解决了传统基于像素分类法出现的"椒盐"现象,是一种很有优势的影像分类法。

In this paper,taking the images of Mapping Satellite-1 as the experimental data,using the object-oriented image segmentation technique,the images were segmented by selecting the appropriate scale parameters to get the image object layer.And then,the SVM was used to classify the image object layer.The choice of SVM classifier kernel function and the influence of parameter setting on classification accuracy were analyzed in detail.The overall accuracy of the classification was 90.857 1%and the Kappa coefficient was 0.858 1.At last,the classification results were compared with the traditional classification based on the results of Mahalanobis distance classification and the maximum likelihood classification method.The overall accuracy increased by about 29.29% and 5.91%respectively,and the Kappa coefficients increased by about 0.35 and 0.06 respectively.The experimental results show that the object-oriented SVM classification method not only improves the accuracy of image classification,but also solves the"salt and pepper"phenomenon in the traditional pixel-based classification.