结合LiDAR单木分割和高光谱特征提取的城市森林树种分类

Tree Species Classification in Urban Forests based on LiDAR Point Cloud Segmentation and Hyperspectral Metrics Extraction

基于多源遥感数据的城市森林树种分类对城市森林资源调查、森林健康状况评价及科学化管理具有重要意义。以江苏省常熟市虞山国家森林公园内的典型城市森林树种为研究对象,利用同期获取的机载激光雷达(LiDAR)和高光谱数据,针对5个典型城市森林树种进行了树种分类的研究。首先,基于点云距离判断单木分割方法进行单木位置和冠幅提取,并借助实测数据和目视解译结果进行精度验证;然后,在冠幅内提取4组高光谱特征变量,并借助随机森林模型对特征变量进行重要性分析;最后,筛选出重要性高的特征变量进行2个级别的树种分类并借助混淆矩阵进行验证评价。结果表明:基于点云距离判断分割方法的单木位置提取精度较高(探测率为85.7%,准确率为96%,总体精度为90.9%);利用全部特征变量(n=36)对5个树种进行分类,分类的总体精度达到了84%,Kappa系数为0.80;利用优选特征变量(n=9)进行分类,总体精度达83%,Kappa系数为0.79;利用全部特征变量(n=36)对两种森林类型进行分类,分类的总体精度达91.3%,Kappa系数为0.82,其中阔叶树种分类精度为95.6%,针叶树种分类精度为85%;利用优选特征变量(n=9)进行分类,分类的总体精度达90.7%,Kappa系数为0.80,其中阔叶树种分类精度为93.33%,针叶树种分类精度为86.67%。

Urban forest tree species classification using multi-source remote sensing data plays a key role in urban forest resources investigation,forest health assessment and scientific management.This study selected typical tree species in Changshu Yushan forest as research objects.The five tree species were classified using combined airborne hyperspectral and LiDAR data which acquired simultaneously.First,the positions and crowns of individual trees were extracted from LiDAR data based on Point Cloud Segmentation method (PCS) and validated using field and visual interpretation data;second,the four sets of hyperspectral metrics were extracted from hyperspectral data and the importance of metrics were assessed using Random Forest algorithm;finally,the tree species were classified in two levels using Random Forest algorithm and accuracies were evaluated by confusion matrix.The results indicated that the PCS approach had high accuracy (Detection Rate =85.7%,Precisio n=96% the Overall Accuracy=90.9%) in the extraction of individual tree positions;the overall accuracy of five tree species classification using all metrics (n=36) was 84%,Kappa coefficient was 0.80;the overall accuracy of five tree species classification using the optimal metrics (n=9) was 83%,Kappa coefficient was 0.79;the overall accuracy of two forest types classification using all metrics (n=36) was 91.3%,Kappa coefficient was 0.82,the overall accuracies of conifer and broadleaved tree species were 85% and 95.6% respectively;the overall accuracy of two forest types classification using the optimal metrics (n=9) was 90.7%,Kappa coefficient was 0.80,the overall accuracies of conifer and broadleaved tree species were 86.67% and 93.33% respectively.