基于分水岭算法的无人机不同飞行高度下林木冠幅提取研究

树冠是林木重要的组成部分之一，林木冠幅信息精确提取对森林资源调查和树木生长动态监测有着重要的意义。通过设置不同的无人机飞行高度，以哈尔滨市城市林业示范基地中的樟子松样地为对象，分别利用传统分水岭算法和改进分水岭算法对单木树冠和林隙进行提取，并对树冠冠幅和树冠投影面积进行估算，最后与实测数据进行对比分析。结果表明:1）基于传统分水岭算法平均单木冠幅识别率为51.11%，平均欠分割率为25.18%，平均过分割率为11.11%；树冠冠幅和树冠投影面积平均提取精度分别为69.72%和53.59%，说明传统分水岭算法对单木冠幅提取效果一般。2）改进分水岭算法平均单木冠幅识别率为80.74%，平均欠分割率为8.15%，平均过分割率为6.67%；树冠冠幅和树冠投影面积平均提取精度约分别为79.84%和76.04%，表明改进的分水岭算法对林木单木冠幅提取精度较高。3）50 m飞行高度下样地中林隙面积在0~5 m²和5~10 m²各占57.89%和31.58%；林隙形状指数分布在1.14~1.85，平均值为1.36；研究表明，利用改进分水岭算法在50 m无人机飞行高度获取的林木影像可以有效提取林木树冠和林隙面积信息，研究结果可为森林资源调查提供有效参考。

Extraction of Forest Stand Crown Width at Different Flying Heights of UAV Based on Watershed Algorithm

Canopy is an essential part of the forest,and accurate extraction of canopy width information is of great significance for forest resource investigation and tree growth dynamic monitoring.Taking the sample plots of Pinus sylvestris var.mongolica selected from the Harbin Urban Forestry Demonstration Base as research objects,different unmanned aerial vehicle (UAV) flying heights were set up and the individual tree crown and forest gap were extracted by using traditional watershed algorithm and improved watershed algorithm,respectively.The crown width and the projected area of the crown were estimated and then compared with the measured data.The experimental results showed that 1) the average individual tree recognition rate,the average under segmentation rate and the average over segmentation rate by using the traditional watershed algorithm were 51.11%,25.18%,and 11.11%,respectively; the average extraction accuracy of the crown width and the projected crown area were 69.72% and 53.59%,respectively,which indicated that the traditional watershed algorithm on the extraction of forest individual tree crown width was not satisfied.2) The average individual tree recognition rate of the improved watershed algorithm was 80.74%,the average under segmentation rate was 8.15%,and the average over segmentation rate was 6.67%; the average extraction accuracy of the crown width and the projected crown area were 79.84% and 76.04%,respectively,indicating that the improved watershed algorithm had a higher accuracy in extracting forest individual tree crown width.3) At 50 m flying height,the forest gaps with areas of 0-5 m² and 5-10 m² accounted for 57.89% and 31.58% respectively; the shape index of the forest gaps was between 1.14-1.85,with an average of 1.36.In conclusion,the improved watershed algorithm can effectively extract the forest individual tree crown width and forest gaps from the images obtained by UAV at the flying height of 50 m in this study,and the research results can provide effective reference for forest resources survey.