Apple crop-load estimation with over-the-row machine vision system

A crop load estimation is important for efficient apple orchard management. This information is important for planning and assigning appropriate labor pool and equipment for harvesting and transportation of fruits from orchard to packing houses. Current machine vision-based techniques for crop load estimation have achieved only limited success. Two primary factors affecting the accuracy of such systems are the occlusion of fruit, and variable outdoor lighting condition. In order to minimize the effect of these factors, an over-the-row platform with a tunnel structure was developed to take images of apple trees from two opposite sides. Average visibility of apples increased from 70% to 97% when imaged from both sides of a row of apple trees in a modern commercial orchard. The tunnel structure minimized illumination of apples with direct sunlight; hence reducing the variability in lighting condition. The platform, equipped with artificial lighting, was capable of nighttime operations also. Images captured during day and night time were processed for identifying apples. Location of apples in three-dimensional space was used to eliminate repeated counting of apples that were visible from both sides of the tree. Rootmean-squared error on identifying apples and repeated apple counting were estimated to be 12% and 9.5% respectively. Overthe-row machine vision system showed a promise for accurate and reliable apple crop-load estimation that may substitute for traditional way of crop load estimation using visual inspection. Accurate identification and 3D localization of apples will also provide a foundation for the development of robotic harvesting system.