基于激光雷达的树形靶标冠层叶面积探测模型研究

果园靶标冠层叶面积有效探测是施药量在线计算的基本依据。针对树形靶标稠密和稀疏2种冠层类型,搭建叶面积测量三维立体试验平台和激光雷达(Light detection and ranging,LiDAR)探测移动试验平台,构建不同厚度和稠密度树形靶标,采用偏最小二乘回归（Partial least squares regression,PLSR）算法与BP（Back propagation）神经网络算法建立了冠层叶面积探测模型。试验结果表明：PLSR算法获得稠密厚冠层、稀疏厚冠层、稠密薄冠层和稀疏薄冠层叶面积探测模型的决定系数（R2）分别为：0.9626、0.4130、0.8896、0.2699;BP神经网络算法获得模型的R2依次为：0.9727、0.5302、0.8993、0.4290。基于LiDAR的冠层叶面积探测模型对稠密冠层探测精度较高,R2不低于0.8896,对稀疏冠层探测精度较低,不高于0.5302,该探测方法可用于稠密冠层叶面积在线计算,指导果园精准变量喷药。

Detection Model of Tree Canopy Leaf Area Based on LiDAR Technology

The effective detection of leaf area of target canopy in the orchard is the basic for the online calculation of the pesticide application rate. A three-dimensional test platform for leaf area measurement and a light detection and ranging (LiDAR) detection mobile test platform were built. Tree targets of different thickness and density were constructed for the two canopy types of dense and sparse tree targets. Partial least squares regression (PLSR) algorithm and back propagation (BP) neural network algorithm were used for canopy leaf area detection model among the number of LiDAR point clouds data, canopy thickness and canopy leaf area. The experimental results showed that the coefficients of determination (R2) of the equations of dense thick canopy, sparse thick canopy, dense thin canopy and sparse thin canopy obtained by PLSR algorithm were 0.9626, 0.4130, 0.8896 and 0.2699, and the R2 obtained by BP neural network of the canopies were 0.9727, 0.5302, 0.8993 and 0.4290, respectively. Based on the LiDAR canopy leaf area detection model, the detection accuracy of the dense canopy was high, the value of R2 was not less than 0.8896, and the detection accuracy of the sparse canopy was relatively poor, which was not higher than 0.5302. Comparing the PLSR algorithm and the BP neural network algorithm, the latter can significantly improve the accuracy of the model, and the R2 value can be increased by 0.1591. The proposed three-dimensional space tree target canopy leaf area detection method can be used to calculate dense canopy leaf area online to guide orchard accurate variable spraying.