基于无人机多光谱影像的单木树冠提取方法

Crown extraction method of individual tree based on multi-spectral image of UAV

基于无人机多光谱影像,选取郁闭度较高的阔叶林区作为研究对象,在提取植被特征的基础上,采用双边滤波和面向对象的多尺度分割方法,选取最佳分割参数组合,得到单木树冠.结果表明,与直接对原始真彩色影像采用多尺度分割的结果相比,改进方法的过分割现象明显减少,分割准确率达到76.63%,F测度为80.24%,说明该方法能有效减少背景对分割精度的影响,有效抑制传统多尺度分割方法造成的过分割问题,可对郁闭度较高的阔叶林区单木树冠进行自动提取.

Bilateral filtering and multi-scale segmentation were applied to vegetation characteristics information extracted from UAV multi-spectral images of broadleaf forest with high canopy density.Upon the optimum combinations of segmentation parameters,single crown images were generated.Compared with the results directly segmented from the original true color multi-scale images,oversegmentation was noticeably reduced from the improved method,resulting in a segmentation accuracy of 76.63%and F measure of 80.24%.It can be concluded that the proposed method can effectively lower the impact of background on compromising segmentation accuracy,and effectively reduce over-segmentation arisen from multi-scale segmentation by conventional methods,which can be used for the automatic tree crown extraction of broadleaf forest with high canopy density.