基于Landsat的多分类器集成遥感影像分类

Multiple classifiers integrated classification based on Landsat imagery

针对传统的单分类器分类精度低,难以满足遥感影像分类精度要求高的问题,该文提出了一种多分类器集成分类方法。该方法有效地将支持向量机算法、C4.5决策树算法和人工神经网络算法进行了组合,实现了多种分类器集成的优势互补,在提高单个类别分类精度的基础上实现了整体精度的提高。该文基于Landsat遥感影像,利用多分类器集成分类技术,获取广州市自1987年以来的土地利用/地表覆盖数据,以平均3年为一个时段,共制作11期数据。实验结果表明,产品分类的平均精度达到88.12%,Kappa系数平均值达到0.868,高于3种基分类器的分类精度,对各种地物的分类精度也明显提高。

Aiming at the problem that the traditional classification method using single classifier can not meet the request of remote sensing image classification with high accuracy,a classification method which integrated multiple classifiers was proposed in this paper,by which the advantages of support vector machine（SVM）,C4.5 decision tree and artificial neural network（ANN）were complemented each other,and the overall accuracy was improved by improved classification accuracy at per-class level.Based on Landsat imageries,land use/cover data at a three year interval with a total of 11 time slots since 1987 were obtained.The results showed that the average classification accuracy reached 88.12%,and the Kappa coefficient reached 0.868,which were higher than the accuracies of all the component classifiers＇ results.