面向对象的土地荒漠化信息提取研究  

Research on land desertification information extraction by object-oriented classification method

作　　者：韩军良 韩留生[1] 穆豪祥 张至一 郭宇晨 刘晓亚 HAN Junliang;HAN Liusheng;MU Haoxiang;ZHANG Zhiyi;GUO Yuchen;LIU Xiaoya(School of Civil Engineering and Geomatics,Shandong University of Technology,Zibo,Shandong 255000,China)

机构地区：[1]山东理工大学建筑工程与空间信息学院,山东淄博255000

出　　处：《测绘科学》2023年第4期149-160,共12页Science of Surveying and Mapping

基　　金：山东省自然科学基金项目(ZR2020MD018,ZR2020MD015)。

摘　　要：针对基于像元的荒漠化信息提取中存在光谱混淆、椒盐现象等问题,该文选取2021年8月份的Landsat 8遥感影像为数据源,利用均值方差法和最大面积法确定最优分割尺度;在最优尺度分割的基础上,采用面向对象的随机森林分类方法提取了毛乌素生态保护区土地荒漠化信息。研究结果表明:不同最优分割尺度选取方法确定的最优尺度相近但并不相等,通过分析发现利用最小的尺度分割影像的对象完整且均一,各地类也不会产生欠分割和过分割现象;与基于像元的分类方法相比,面向对象方法提取的荒漠化分类结果中的“椒盐现象”明显减弱,并且制图精度和用户精度均有显著提高,总体精度提高了8.06%,Kappa系数提高了0.111 4。本研究为土地荒漠化的防护与治理提供了有效的决策与支撑。Aiming at the problems of spectral confusion as well as salt and pepper phenomenon in extracting land desertification information by pixel-based method,Landsat 8 remote sensing images in August 2021 were collected,and the optimal segmentation scale was determined by applying the mean variance method and the maximum area method.Based on the optimal scale segmentation,object-oriented random forest classification method was used to extract desertification information of Mu Us ecological reserve.The results showed that the optimal scales determined by different optimal segmentation scale selection methods were similar but not exactly the same.It was also found that the image segmentation object by using the smallest scale was complete and uniform,and the phenomenon of under-segmentation as well as over-segmentation did not occur.Compared with the pixel-based classification result,the salt and pepper phenomenon significantly disappeared in object-oriented classification result,and the mapping accuracy as well as user accuracy were both significantly improved.The overall accuracy was improved by 8.06% and the Kappa coefficient was increased by 0.111 4.This study could provide effective decision support for the protection and control of desertification.

关 键 词：荒漠化 最优分割尺度 面向对象 随机森林分类 毛乌素沙地 

分 类 号：P237[天文地球—摄影测量与遥感]