基于DOM及LiDAR的多尺度分割与面向对象林隙分类

摘要

为研究分割尺度对航空正射影像(DOM)与LiDAR数据协同面向对象林隙分割与分类的影响,以东北典型的天然次生林帽儿山实验林场东林施业区为试验区,对DOM与LiDAR数据进行多尺度分割与面向对象林隙分类.分割过程中,采用基于DOM分割、基于LiDAR数据分割、DOM&LiDAR协同分割3种分割方案.每种分割方案采用10种尺度.在每种尺度应用两种数据提取的光谱和高度两种特征,采用支持向量机分类器(SVM)进行林隙分类.研究结果表明:3种分割与分类方案分类精度随尺度的增大整体呈现下降的趋势,与ED3 (Modified)趋势相反.基于LiDAR数据在尺度参数10获得了最优分割结果.在所有尺度上(10 ～ 100),基于LiDAR数据分割与分类精度高于其他两种数据源的分类精度,相比单独使用DOM优势更加明显.基于LiDAR数据分割与分类方案在尺度参数10时获得了最高分类精度(Kappa系数为80％).3种分割与分类方案最优尺度的分类精度显著高于其他尺度分类精度.分割尺度对面向对象林隙分类结果有重要影响.%Aiming to study the effect of segmentation scale on object based segmentation and classification of forest gap through fusion of aerial orthophoto (DOM) and LiDAR data,the typical natural secondary forest in Maoershan Experimental Forest Farm Donglin Industry Zone of northeastern China was selected as the experimental area.The DOM and airborne LiDAR were used for multiscale segmentation and object-oriented forest gap classification.In the process of image segmentation,three segmentation schemes (segmentation of DOM,segmentation of LiDAR data and segmentation of a fusion of DOM and LiDAR data) were adopted.For each segmentation scheme,10 segmentation scales were set,then based on the segmentation results,spectral and height features extracted from DOM and LiDAR data were used for object-oriented forest gap classification with the support vector machine (SVM) classifier.The results showed that the classification accuracies of three segmentation and classification schemes showed a decline trend with the increase of scale,which was opposite with trend of ED3 (Modified).Based on the LiDAR data at scale parameter of 10,the best segmentation result was got.At all scale (10 ～ 100),the classification accuracy based on LiDAR segmentation and classification was higher than that based on two other data segmentation and classification schemes,and had the more obvious advantage than using only DOM.Based on scheme of LiDAR data segmentation and classification at scale parameter of 10,the highest classification accuracy was got with Kappa coefficient of 80％.The classification accuracies of three segmentation and classification schemes at the optimal scale were significantly higher than these at other scales.The segmentation scale had important effect on the object-oriented forest gaps classification.