基于多种分类器的云南蒙自市遥感分类比较研究

A REMOTE SENSING COMPARATIVE RESEARCH BASE ON VARIOUS CLASSIFIERS IN MENGZI CITY YUNNAN PROVINCE

以蒙自市为研究区域,使用ISODATA法、最大似然法、支持向量机法、人工神经网络法、面向对象法共5种分类方法对蒙自市资源3号遥感图像进行分类。利用天绘一号卫星拍摄的高分辨率影像随机选取1 000个点选出真实感兴趣区作为参照,对以上5种分类结果进行精度评价,根据精度检验结果选择了精度最高的面向对象的分类方法作为本次遥感图像分类的分类器。在进一步研究面向对象的分类时,用较大的间隔确定分割尺度的最佳取值所在区间后缩小数据间隔最终确定分割尺度的最佳值,基于精度最高的尺度分割参数的分类结果确定蒙自市2016年的土地利用分类进行统计分析。

We selected the Mengzi City as research area, employed total 5 classification methods to classify ZY -3 remote sensing image, the methods including ISODATA method, maximum likelihood method, support vector machine method, artificial neural network method and object oriented method. Random sampling 1000 in- terest areas of high - resolution TH - 1 satellite image was taken as reference for accuracy evaluation of the above 5 classification results. According to the accuracy results, the object -oriented classifier with the highest accuracy was chosen as the research classification method. In the further object -oriented classification research, we deter- mined the best segmentation scale with a larger interval firstly, and then reduced the data interval to determine the optimal segmental scale. Finally, we determined the classification result and statistics analysis of Mengzi City in year 2016 with the most accurate scale segmentation parameters.