基于C4.5算法的遥感影像分类

随着城市化进程的加快,湿地对整个生态系统的可持续发展具有重要的意义。以洪泽湖湿地为研究区,集合TM影像的光谱信息和纹理信息构建空间数据库,获取训练样本,并从训练样本集中获取分类规则;然后利用C4.5算法构建决策树,并基于知识规则推理得到遥感影像分类结果;最后将分类结果与传统的最大似然法进行比较分析。实验表明,基于C4.5算法得到的分类结果的分类总精度为91.9701%,其分类总精度结果明显高于传统的最大似然法的80.0885%;同样,前者的分类结果的Kappa系数为0.900 3,也远远高于最大似然法的0.746 5。

Classification of Remote Sensing Image Based on C4.5 Algorithm

With the accelerated process of urbanization, wetlands have been crucial to the sustainable development of the entire ecosystem. In this paper, Hongze Lake wetlands were selected as the study area. A collection of spectral information and texture information on TM imagery was used to build a spatial database. Training samples were sampled and classification rules were got from training samples. Then the C4.5 algorithm was used to build decision trees. Classification results of the remote sensing imagery based on knowledge rules. Finally, the C4.5 algorithm was compared with the traditional maximum likelihood method. The results show that the overall accuracy of the results of the C4.5 algorithm, which is 91.970 1%, is significantly higher than that of maximum likelihood method, which is 80.088 5%. The former Kappa coefficient, which is 0.900 3, is also much higher than that of the maximum likelihood method, which is 0.746 5.