基于无人机遥感的灌区土地利用与覆被分类方法  

Classification Method of Land Cover and Irrigated Farm Land Use Based on UAV Remote Sensing in Irrigation

作　　者：韩文霆[1,2] 郭聪聪[1] 张立元[1] 杨江涛[3] 雷雨[1] 王紫军[1] 

机构地区：[1]西北农林科技大学机械与电子工程学院,陕西杨凌712100 [2]西北农林科技大学水土保持研究所,陕西杨凌712100 [3]西北农林科技大学水利与建筑工程学院,陕西杨凌712100

出　　处：《农业机械学报》2016年第11期270-277,共8页Transactions of the Chinese Society for Agricultural Machinery

基　　金：科技部国际合作项目(2014DFG72150);杨凌示范区工业项目(2015GY-03)

摘　　要：为研究无人机可见光遥感技术在灌区土地利用和覆被分类中的有效性,以河套灌区五原县塔尔湖镇为试验区域,用TEZ固定翼无人机搭载索尼A5100型相机进行航拍试验。应用Agisoft Photo Scan软件对无人机遥感系统获取的可见光高分辨率原始单张影像数据进行拼接处理。除目视提取的特殊用地与水域及水利设施用地外,通过试误法确定分割尺度300、形状权重0.4、紧致度权重0.5为无人机遥感影像数据的最佳分割参数。通过对剩余各地物在光谱、形状、纹理特征参量中表现的特异性,分别建立决策树、支持向量机、K-最近邻分类规则集提取土地利用类型试验。结果表明,支持向量机能较准确地提取各地物的特征,总体精度为82.20%,Kappa系数为0.765 9;决策树分类方法的总体精度为74.00%,Kappa系数为0.667 5;K-最近邻分类方法的总体精度为71.40%,Kappa系数为0.610 7。采用支持向量机结合决策树分类法创建的决策树模型,可以将总体精度提高到84.20%,Kappa系数达到0.790 0。因此无人机可见光遥感技术可以用于提取灌区土地利用类型,但存在农、毛渠错分为交通运输用地的情况,渠系的提取还需进一步研究。In order to verify the availability of UAV （unmanned aerial vehicle） optical remote sensing technology in land use type and classification, Wuyuan county Tal Lake town of Hetao Irrigation Area was chosen as research area and visible images were obtained by using TEZ fixed wing UAV equipment with SONY AS100. After obtaining the visible high resolution images by using the UAV remote sensing system, they were mosaicked in the Agisoft PhotoScan software. In addition to visually extracting ground object, we also adopted object oriented which segmentation scale was 300, shape factor was 0.4, smoothness was 0.5 to divide images. On the basis of visual, according to the specificity of ground object in spectrum, shape and texture feature, we respectively established decision tree, support vector machine, K-nearest neighbor classification to extract land use type. Results indicated that SVM can accurately extract characteristics of ground object, the overall accuracy was 82.20% , Kappa coefficient was 0. 765 9; overall accuracy and Kappa coefficient of decision tree were 74.00% and 0. 667 5, respectively; overall accuracy and Kappa coefficient of K-nearest neighbor classification were 71.40% and 0. 610 7, respectively. In this paper, based on the support vector machine classification method combined with the decision tree model, the overall accuracy was grown up to 84.20% , Kappa coefficient reached 0. 790 0. But there existed the wrong situation of small trench being divided into traffic and transport. The visible UAV remote sensing technology can be used to extract the irrigated land use types, but the extraction ditches need further study.

关 键 词：无人机遥感 可见光波段 灌区土地利用 土地覆被分类 支持向量机 

分 类 号：TP722.4[自动化与计算机技术—检测技术与自动化装置] V279.2[自动化与计算机技术—控制科学与工程]