Abstract
The performance of a multi-view machine vision method was documented at an orchard level, relative to packhouse count. High repeatability was achieved in night-time imaging, with an absolute percentage error of 2% or less. Canopy architecture impacted performance, with reasonable estimates achieved on hedge, single leader and conventional systems (3.4, 5.0, and 8.2 average percentage error, respectively) while fruit load of trellised orchards was over-estimated (at 25.2 average percentage error). Yield estimations were made for multiple orchards via: (i) human count of fruit load on ~5% of trees (FARM), (ii) human count of 18 trees randomly selected within three NDVI stratifications (CAL), (iii) multi-view counts (MV-Raw) and (iv) multi-view corrected for occluded fruit using manual counts of CAL trees (MV-CAL). Across the nine orchards for which results for all methods were available, the FARM, CAL, MV-Raw and MV-CAL methods achieved an average percentage error on packhouse counts of 26, 13, 11 and 17%, with SD of 11, 8, 11 and 9%, respectively, in the 2019–2020 season. The absolute percentage error of the MV-Raw estimates was 10% or less in 15 of the 20 orchards assessed. Greater error in load estimation occurred in the 2020–2021 season due to the time-spread of flowering. Use cases for the tree level data on fruit load was explored in context of fruit load density maps to inform early harvesting and to interpret crop damage, and tree frequency distributions based on fruit load per tree.
Highlights
A timely and accurate pre-harvest estimation of crop load is necessary to inform harvest, storage and transport logistics and effective marketing
In the current study we report on the performance of this hardware in conjunction with the multi-view method, and act on the recommendation of a review of yield forecast methods [5] for reporting of estimates at an orchard level
The method used for this estimate varied between farms, with the best practice being a manual count of 25 trees by each of two people per orchard, i.e., a count of 50 trees, or approximately 5% of trees
Summary
A timely and accurate pre-harvest estimation of crop load is necessary to inform harvest, storage and transport logistics and effective marketing. Direct sunlight can create difficult imaging conditions when imaging whole orchards, given the multiple orientation of the camera with respect to the sun Various approaches to this issue have been reported, including use of an ‘over the row’ shade for consistent imaging of fruit on apple trees [1], use of intense strobe lighting and short exposure times [2] and imaging at night with use of artificial lighting [4,7]. Our group has documented the creation of a modified YOLOv3 detection algorithm [4], with implementation of fruit tracking to enable multiview estimation [9] and development of an imaging system mounted to a ground vehicle, e.g., [12] This hardware has been used in conjunction with the dual-view method in estimation of yield mango orchards [12,13]. Use cases for the orchard fruit load density map produced through the machine vision method are presented
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
