Detection of Potato Leaf Diseases Using the YOLOv8 Method

Abstract

The cultivation of potatoes faces significant challenges due to leaf diseases such as late blight and early blight, which adversely affect productivity. This study aims to evaluate the YOLOv8 model in detecting potato leaf diseases, particularly under varying object distances and different light intensity conditions. The research dataset consists of primary images collected from potato farms in Dieng and secondary data sourced from Kaggle. The model was tested at various distances (10 cm, 25 cm, 50 cm, and 75 cm) as well as under bright and dim lighting conditions. The test results showed that the model's performance tends to decline as the distance between the object and the camera increases, with the best results observed at a distance of 10 cm. For variations in light intensity, the model performed better under bright conditions compared to dim lighting, although the difference was not significant. The YOLOv8 model achieved an mAP50 score of 0.994, precision of 0.994, and recall of 0.997 during training with 50 epochs. These findings demonstrate the potential of YOLOv8 for automatic detection of potato leaf diseases with strong performance at close distances and across varying lighting conditions.