基于AdaBoost的组合分类器在遥感影像分类中的应用*

运用组合分类器的经典算法AdaBoost将多个弱分类器-神经网络分类器组合输出,并引入混合判别多分类器综合规则,有效提高疑难类别的分类精度,进而提高分类的总精度.最后以天津地区ASTER影像为例,介绍了基于AdaBoost的组合分类算法,并在此基础上实现了天津地区的土地利用分类.分类结果表明,组合分类器能有效提高单个分类器的分类精度,分类总精度由81.13%提高到93.32%.实验表明基于AdaBoost的组合分类是遥感图像分类的一种新的有效方法.

Application of combining classifier based on AdaBoost to remote sensing classification

The classical classifier combination method based on AdaBoost was used to combine several weak classifiers. Moreover, the mixed combining rule was introduced into the classification. Based on these methods, the classification accuracy for some class which were very difficult to classify was significantly improved, The total accuracy for all the classes was also im- proved. In the end of this paper, taking the ASTER data in Tianjin area as an example, the AdaBoost combining algorithm was developed. The land cover mapping in this area was produced. The results of this case show that the combination classifier can effectively improve the aecuracy of single classifier. The total accuracy is improved from 81.13% to 93.32%. The experimental result also indicates that the combination method based on AdaBoost is a newly effective approach for remote sensing image classification.