基于无人机可见光影像的人工林单木树冠提取研究

Study on individual tree crown extraction in artificial forest area based on the image of UAV

以人工林区的无人机可见光影像为基础构建冠层高度模型(canopy height model,CHM),以此作为提取单木树冠的基础数据,避免数字正射影像(digital orthophoto map,DOM)中存在的树冠阴影、光照不均、树冠与草地颜色相似等问题对单木树冠提取的影响,采用标记控制的分水岭算法从CHM中提取单木树冠.首先,采用Otsu算法对CHM中的树冠与非树冠进行分离,利用形态学图像细化算法去除树冠之间的粘连,得到粗略的单木树冠;其次,结合距离变换和H-Maxima算法获取单木位置标记,结合距离变换和分水岭算法获取每个粗略单木树冠的影响区域边界,作为背景标记;为增强CHM中的树冠边缘像素,采用Sobel算法生成梯度图像,利用极小值强加算法将单木位置标记设定为分水岭算法分割的起始点,将背景标记用于对分割过程进行约束;最后,对标记过的梯度图像进行分水岭分割,获得单木树冠.采用人工目视解译的方式获取研究区内正确的单木树冠轮廓作为精度验证数据.通过精度验证,提取的单木位置总体精度为92.79%,提取的单木树冠F-score值为84.62%,提取精度较高.3组对比试验,分别是不进行去除树冠粘连处理、用局部极大值算法代替H-Maxima算法、不使用背景标记,得到3个单木树冠提取结果,通过精度验证,F-score值分别为72.33%、10.16%、77.89%,分别比提出方法降低了12.29%、74.46%、6.73%.通过方法对比和通用性检验,试验结果表明以CHM为研究数据,利用标记控制分水岭算法可有效获取人工林区单木树冠.

The canopy height model(CHM)was built based on the visible light image of UAV in the artificial forest area,which was used as the basic data for extracting individual tree crowns,so as to avoid the impacts on the extraction of the individual tree crown,such as the impacts of crown shadow,uneven light,the similar color of tree crown and grass in the digital orthophoto map(DOM)of UAV.Subsequently,individual tree crowns were extracted from CHM,using marker controlled watershed algorithm.First of all,the Otsu algorithm was used to separate the crown and non-crown in CHM,and a morphological image thinning algorithm was used to tackle the problem of adhesion between some crowns and obtain a rough individual tree crown.Then,the distance transformation and H-Maxima algorithm were combined to obtain the individual tree position marker,and the distance transformation and watershed algorithm were combined to obtain as the background marker the rough influence area boundary of each individual tree crown.Furthermore,to enhance the crown edge pixels in CHM,the Sobel algorithm was employed to obtain its gradient image,the minimum value forcing algorithm was used to set the individual tree position marker as the starting point of watershed algorithm segmentation,and the background marker was used to constrain the segmentation process.Finally,the watershed algorithm was applied to segment the marked gradient image and obtain the individual tree crown.And the correct individual tree crown in the study area was obtained by manual visual interpretation,which was used as validation data.Through the accuracy verification,the overall accuracy of the extracted individual tree position was found to be 92.79%,and the extracted individual tree crown F-score value was 84.62%,with relatively high extraction accuracy.Three groups of comparative experiments were carried out so as to obtain three individual tree crown extraction results:by means of no removal of crown adhesion,the local maximum algorithm replacing the H-Maxima algorithm used in this research,and no background marker.Through accuracy verification,F-score values were 72.33%,10.16%,and 77.89%respectively,which were 12.29%,74.46%,and 6.73%lower than those obtained by using the method in this research.The comparison of methods and generality test were also carried out,and the experimental results show that the individual tree crown in the artificial forest area can be effectively obtained by using the markercontrolled watershed algorithm with CHM as the research data.