Automatic Bunch Detection in White Grape Varieties Using YOLOv3, YOLOv4, and YOLOv5 Deep Learning Algorithms

Abstract

Over the last few years, several Convolutional Neural Networks for object detection have been proposed, characterised by different accuracy and speed. In viticulture, yield estimation and prediction is used for efficient crop management, taking advantage of precision viticulture techniques. Convolutional Neural Networks for object detection represent an alternative methodology for grape yield estimation, which usually relies on manual harvesting of sample plants. In this paper, six versions of the You Only Look Once (YOLO) object detection algorithm (YOLOv3, YOLOv3-tiny, YOLOv4, YOLOv4-tiny, YOLOv5x, and YOLOv5s) were evaluated for real-time bunch detection and counting in grapes. White grape varieties were chosen for this study, as the identification of white berries on a leaf background is trickier than red berries. YOLO models were trained using a heterogeneous dataset populated by images retrieved from open datasets and acquired on the field in several illumination conditions, background, and growth stages. Results have shown that YOLOv5x and YOLOv4 achieved an F1-score of 0.76 and 0.77, respectively, with a detection speed of 31 and 32 FPS. Differently, YOLO5s and YOLOv4-tiny achieved an F1-score of 0.76 and 0.69, respectively, with a detection speed of 61 and 196 FPS. The final YOLOv5x model for bunch number, obtained considering bunch occlusion, was able to estimate the number of bunches per plant with an average error of 13.3% per vine. The best combination of accuracy and speed was achieved by YOLOv4-tiny, which should be considered for real-time grape yield estimation, while YOLOv3 was affected by a False Positive–False Negative compensation, which decreased the RMSE.

Computer Vision and Deep Learning for Precision Viticulture

“…For this purpose, their model reached an inference time of only 12 ms per image on an Nvidia GTX 1050Ti GPU. Similarly, a counting error of 13.3% was reported in the work of Sozzi et al [112] with a Yolov5x model. The performance obtained by Aguiar et al [94] seems to be closer to the natural condition of many vineyards: their SSD-MobileNet model reached 49.85 and 53.34% mAP on grapes before veraison, without defoliation, and at two different phenological stages.…”

End-to-end deep learning for directly estimating grape yield from ground-based imagery

“…While many previous studies have focused on berry counting, some have examined bunch detection using the YOLO model. Sozzi et al [21] demonstrated an AP value of 0.76 with YOLOv5. However, grape bunches visualized in the study appear in higher resolution, with less occlusion than seen in the present work.…”

“…Much of the previous work has been performed in orchard crops, such as apples [11,14], oranges [15], and mangos [16]. Studies in other specialty crops, including tomatoes [9,17] and grapes [18,19,20,13,21] have also been conducted. In vineyards, much of the existing literature in yield estimation from proximal imagery have consisted of methods leveraging feature engineering and computer vision to count individual grape berries [22,2,23], although pixel count has also been related to grapevine yield [1].…”

“…In vineyards, much of the existing literature in yield estimation from proximal imagery have consisted of methods leveraging feature engineering and computer vision to count individual grape berries [22,2,23], although pixel count has also been related to grapevine yield [1]. In very recent years, studies on proximal imaging for fruit detection have moved away from hand-crafted feature and algorithm development towards deep learning methods [12,18,24,13,21]. Deep learning methods used for fruit detection reduce the bias involved in model development via their flexibility in learning both the optimal features and optimal relationships between features to converge to the desired result.…”

A Deep Learning-Based Model to Reduce Costs and Increase Productivity in the Case of Small Datasets: A Case Study in Cotton Cultivation

“…A deep neural network is utilized to build the predictive model. DNN is an artificial neural network (ANN) algorithm with several hidden layers [30,31]. The proposed DNN model in this paper has four hidden layers.…”