Innovative Leaf Area Detection Models for Orchard Tree Thick Canopy Based on LiDAR Point Cloud Data

Abstract

Orchard spraying can effectively control pests and diseases. Over-spraying commonly results in excessive pesticide residues on agricultural products and environmental pollution. To avoid these problems, variable spraying technology uses target canopy detection to evaluate the leaf area in a canopy and adjust the application rate accordingly. In this study, a mobile LiDAR detection platform was set up to automatically measure point cloud data for a thick canopy in an apple orchard. A test platform was built, and manual measurements of the canopy leaf area were taken. Then, polynomial regression, back propagation (BP) neural network regression, and partial least squares regression (PLSR) algorithms were used to study the relationship between the orchard tree canopy point clouds and leaf areas. The BP neural network algorithm (86.1% and 73.6% accuracies for the test and verification data, respectively) and the PLSR algorithm (78.46% and 60.3%, respectively) performed better than the Fourier function of the polynomial regression (59.73% accuracy). The leaf area model obtained using PLSR was intuitive and simple, while the BP neural network algorithm was more accurate and could meet the requirements for high-precision variable spraying.