Sugarcane And Mango Leaf Disease Detection Using Data Augmentation And Yolov8

Abstract

The identification of crop diseases is essential for sustainable agricultural practices and preservation of farm income. This research emphasizes the detection of diseases in sugarcane and mango leaves using contemporary object detection methods and augmentation strategies. These two crops are known to suffer from diseases that can cause economic losses if left unmitigated, so early detection and diagnosis is needed for effective disease control. For this study, the YOLOv8 model was selected as it has high detection accuracy, runs in real-time, and can detect multiple diseases from the same image. Through the application of extensive data augmentation, such as flipping, rotation, scaling, and addition of noise, we modeled the YOLOv8 process to make it as generalizable, robust, and reliable as possible. Applying augmentation increases diversity in the dataset, but also helps combat overfitting when developing the model, which ultimately enhances the model's predictive ability on previously unseen data. We selected the augmented dataset and trained it on the YOLOv8 model and analyzed performance metrics including mean Average Precision (map), Precision, Recall, and F1-measure. The results prove that YOLOv8, in conjunction with augmentation can achieve very high levels of detection accuracy, and highlighted the increased effectiveness of being able to note diseased areas better than previous models. This work provides an example of the evolving nature of artificial intelligence and its impact in a rapidly changing agricultural world, and supports a vision of a day when drone or imaging technologies could be practically adopted for use in agriculture.