面向对象分类方法的镇西厂乡地类信息提取分析

Object Oriented Method for Extracting and Analyzing Land Type Information in Zhenxichang Township

传统影像解译方法在高分影像信息提取逐渐失去优势,而面向对象分类方法的信息提取能利用高分影像的纹理、光谱等空间信息,使其成为影像信息提取研究的热点。基于高分辨率航片对实验区域贵州省镇西厂地区使用eCognition软件进行面向对象分类方法的信息提取方案分析,重点讨论了航片的多尺度分割,分割最优尺度的确定,分类特征描述与应用、并进行分类精度评价等。同时,还使用ENVI软件基于像元的地类提取,对比两者分类结果与精度,面向对象的最邻近方法信息提取总体精度为89%,Kappa系数为87%,而面向像元的最大似然信息提取总体精度为78%,Kappa系数为75%,面向对象的信息提取结果更加接近人工目视解译,对高分影像的信息提取具有更好的优势。

Traditional image interpretation methods are gradually losing their advantages in extracting high-resolution image information,while object-oriented information extraction can utilize spatial information such as texture and spectrum of high-resolution images,making it a hot research topic in image information extraction.Based on high-resolution aerial photographs of the experimental area of Zhenxi Factory in Guizhou Province,eCognition software is used to analyze object-oriented information extraction schemes.The focus is on multi-scale segmentation of aerial photographs,determination of the optimal segmentation scale,description and application of classification features,and finally,evaluation of classification accuracy.At the same time,ENVI software was also used for pixel based land class extraction.The classification results and accuracy of the two methods were compared.The overall accuracy of object-oriented nearest neighbor method information extraction was 89%,with a Kappa coefficient of 87%,while the overall accuracy of pixel oriented maximum likelihood information extraction was 78%,with a Kappa coefficient of 75%.The object-oriented information extraction results were closer to manual visual interpretation,which has better advantages for information extraction of high-resolution images.