Grape yield spatial variability assessment using YOLOv4 object detection algorithm

Over the last few years, several versions of the machine learning algorithm, YOLO, have been developed, improving its performance. In this study, the last official version of YOLO (v4) was evaluated, to assess grape yield spatial variability. YOLO models were used to classify 24 georeferenced RGB images on an 8 ha vineyard. The models were used to detect the number of bunches, based on different resolution images (320-1,280 pixels) and different confidence thresholds (0.25-0.35). The detected number of bunches was then compared with the actual ones and with the relative final weight harvested from the vines used as a target for the collected images by correlation. According to the results, the best linear regression model for vines yield was obtained with 416 pixels images, which showed an R-2 of 0.59, indicating YOLO as a suitable tool for detecting yield spatial variability.