果树冠层LiDAR点云边界自适应提取与体积测量方法

Adaptive Boundary Extraction and Volume Estimation Method for Fruit Tree Canopies Using LiDAR Point Clouds

摘要:

基于果树冠层体积调整施药量是植保无人机变量作业的重要方式。机载激光雷达(LiDAR)作为一种非接触的传感技术,对于果树冠层体积探测具有应用前景,在实际应用中,常用的凸包类算法易忽略冠层边界凹陷导致体积高估,而体素、α-shape 等体积算法的体素大小或参数α选取依赖人工经验,主观性强,易造成体积测量结果偏差较大。为此,研究建立了基于点云密度的α-shape 算法收缩因子(Shrink factor,SF) 自适应,以中位数点间距(Median point spacing,MPS)量化点云密度,构建MPS与SF的关系模型,解决人工设定SF的局限,实现冠层边界的自适应提取,结合分层切片和棱柱体近似法实现对冠层体积的测量。实验结果显示,α-shape自适应算法对切片边界的提取效果与人工标定边界一致性较好,切片面积的均方根误差为0.0090~0.0197m2,体积的均方根误差为0.0001~0.0002m3,冠层体积测量结果的相对偏差为-4.88%~2.80%。α-shape 自适应算法无需人工调整收缩因子即可提取复杂冠层边界特征,且保证了冠层体积的测量精度,可为植保无人机精准变量施药提供可靠参数支撑。

Abstract:

Adjusting pesticide application rate based on fruit tree canopy volume is a key approach for variable-rate operations of plant protection UAVs. UAV-LiDAR as a non-contact sensing technology, exhibits strong application potential in fruit tree canopy volume detection and measurement. In practical applications, common convex hull-based algorithms tend to ignore canopy boundary indentations, leading to volume overestimation. Meanwhile, volume measurement algorithms such as voxel and α-shape rely on manual experience for selecting voxel size or α parameter, which is highly subjective and likely to cause large measurement errors. To address this issue, an adaptive shrink factor mechanism for the α-shape algorithm was established based on point cloud density. Specifically, the median point spacing was used to quantify point cloud density, and a relationship model between MPS and SF was constructed. This model eliminated the need for manual SF setting, enabling the adaptive extraction of canopy boundaries. Combined with the layered slicing and prism approximation methods, the measurement of canopy volume was realized. Experimental results demonstrated that the α-shape adaptive algorithm achieved good consistency between the extracted slicing boundaries and the manually calibrated boundaries. The RMSE of slicing area ranged from 0.0090m2 to 0.0197m2, and the RMSE of volume ranged from 0.0001m3 to 0.0002m3, with a relative deviation of canopy volume measurement between -4.88% and 2.80%. The proposed α-shape adaptive algorithm can extract complex canopy boundary features without manual SF adjustment, while ensuring high measurement accuracy. It can provide reliable parameter support for precise variable-rate pesticide application of plant protection UAVs.