. | Application of YOLOv8L Deep Learning in Robotic Harvesting of Persimmon (Diospyros kaki)

turc. .

anglais. Deep learning has been a branch of science that has been used by many researchers and has gained popularity in recent years. Deep learning techniques perform better than traditional methods by providing high accuracy in analyzing and processing agricultural data. Therefore, the use of deep learning techniques in agriculture is increasing. The persimmon used in this study is a fruit tree belonging to the Ebenaceae family and is cultivated in various regions of Turkey, including the Trabzon region. The persimmon harvest is typically done during the fall season when the fruits reach optimal maturity. It is recommended to harvest the persimmon when they are hard but slightly soft to the touch. In this study, using the deep learning method, the classification was made by considering the color feature of the fruit. The aim here is to develop a method to be used in robotic harvesting systems. YOLOv8L was chosen as the method. The metric values of the model were analyzed and it was observed that the 'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5' and 'metrics/mAP_0.5:0.95' values of the model increased as the number of epochs increased. In the last epoch, precision was measured at about 71%, recall was measured at 79%, mAP_0.5 was measured at about 84% and mAP_0.5:0.95 was measured at about 76%. These values indicated that the model was able to detect and classify objects with high accuracy in the validation set. Measured value Size: 640x640, Batch: 16, Epoch: 102, Algorithm: YOLOv8L. It was concluded that YOLOv8L was the best detection model to be used in robotic persimmon harvesting to separate the persimmon from branch.