Klasifikasi Penyakit Daun Mangga Menggunakan YOLOv11 Berbasis Deep Learning dan Computer Vision

Abstract

Indonesia’s mango agriculture sector continues to face significant challenges due to leaf diseases that reduce crop productivity. Conventional disease identification methods remain inefficient because they rely on subjective visual observation. This study aims to develop a mango leaf disease classification model using the YOLOv11 deep learning algorithm. YOLOv11 is chosen for its capability in real-time object classification with an optimal balance between accuracy and processing speed. The research will utilize the Mango Leaf Disease dataset from Kaggle, consisting of eight classes (seven disease types and one healthy class). The planned methodology includes preprocessing, image augmentation, data splitting using K-Fold Cross Validation, and hyperparameter tuning on optimizer, learning rate, epoch, and batch size. Model performance will be evaluated using the Confusion Matrix. This research is expected to produce an accurate and efficient classification model that enables objective and rapid early detection of mango leaf diseases. The research utilizes a dataset from Kaggle consisting of 4,000 images across eight classes—comprising seven disease types and one healthy leaf class. The methodology involves preprocessing (resizing to 640x640 pixels and normalization), image augmentation, and data splitting using 10-Fold Cross Validation. Performance was optimized through hyperparameter tuning of the Adam optimizer, a learning rate of 0.001, a batch size of 16, and various epoch settings. The experimental results demonstrate that the YOLOv11s model achieves exceptional and stable performance. Evaluation using a Confusion Matrix shows that the model reached a 100% accuracy, precision, recall, and F1-score on the dataset used in this study. The model recorded an average training loss of 0.0979 and a validation loss of 0.0027. These findings confirm that YOLOv11s is not only highly accurate but also computationally efficient, making it a viable candidate for real-time detection systems on mobile or edge computing devices to support early disease detection in mango orchards. As the main contribution, this study provides a comprehensive evaluation of YOLOv11s for mango leaf disease classification using a 10-Fold Cross Validation scheme, stability analysis based on validation loss, and an assessment of its potential for real-time deployment on mobile and edge computing devices.