Design and experiment of a binocular vision-based canopy volume extraction system for precision pesticide application by UAVs

Abstract

When unmanned aerial vehicles (UAVs) are used for orchard chemicals application, accurate measurement of the canopy volume can provide decision support for determining pesticide dosages, flight parameters, and droplet sizes. Using binocular camera ranging, this study presents a novel canopy segmentation algorithm that preprocesses light detection ranging data to extract sub-grid canopy volumes. A binocular vision-based canopy volume extraction system for UAV chemical application was developed. The system utilizes multi-degree-of-freedom adaptive balance technology to ensure that the binocular camera can still vertically detect the canopy even when the flight attitude changes. Performance experiments were conducted using artificial fruit trees with different leaf densities and regular cardboard box as measurement targets. The canopy volume measurements indicate that the new model accurately detects target contours. When flying at 2 m/s, the maximum errors between system-measured and actual volumes were 6.58 and 9.37 % for the rectangular and triangular, respectively. Increasing speeds and attitudes lead to increased errors and measurement variations. However, the position of the system relative to the target does not cause significant differences in results. The maximum measurement errors between system-measured and actual LiDAR values were 6.44 and 9.17 % for high- and low-density canopies, respectively. These results demonstrate that the proposed system has high measurement accuracy and provides a reliable precision UAV pesticide-spraying control system for plant protection based on real-time canopy detection.