A computer vision approach for estimating fruit growth rate in orchards

Fruit development information throughout the growing season is essential for effective orchard management. Repeated measurements of fruit size enable the tracking of its absolute growth rate (AGR), a key physiological parameter for yield prediction and assessment of stressing conditions. Caliper based AGR measurements can be labour-intensive while existing sensor-based solutions are limited by cost and low representativeness at orchard level. In this study, a novel computer vision system (CVS) that employs a depth camera and artificial intelligence algorithm to estimate fruit AGR directly in the field is presented. Images of these fruits were collected at distances of 1.0 and 1.5 m from the tree row, capturing multiple time points during the season. The CVS computed the AGR for each detected fruit in the images. Preliminary results from the ongoing analysis indicate a fruit detection and sizing rate always over 92 and 97%, respectively. Current results are not in line with expected performance for field application, but further improvements in the system algorithm are currently ongoing. It is worth to note that results were obtained from the analysis of only 30% of the whole data set, suggesting that even better results could be reached during further analysis stage. The presented approach wants to harness the large sample size analyzable (up to the entire fruit population), to mitigate errors associated with individual measurements and obtain robust fruit size and AGR estimations, at the orchard level.