Cassava Leaf Disease Detection using VGG16 Transfer Learning Based on Digital Images

Cassava leaf diseases reduce crop productivity, and farmers still diagnose them mainly by visual inspection, which is error-prone because the symptoms of different diseases look alike. This study develops an automated system for detecting cassava leaf diseases from digital images using transfer learning with the VGG16 convolutional neural network. A balanced subset of 1,000 images from the Cassava Leaf Disease Classification dataset, covering five classes (cassava bacterial blight, cassava brown streak disease, cassava green mottle, cassava mosaic disease, and healthy leaves), was divided into 70% training, 15% validation, and 15% test data. The images were resized to 224 × 224 pixels, normalized, and augmented. A VGG16 network pre-trained on ImageNet was fine-tuned with its first 14 layers frozen, a dropout rate of 0.5, and the Adam optimizer. The best model reached a validation accuracy of 94.37% and a test accuracy of 93.66%, with macro-averaged precision, recall, and F1-score of 94.38%, 93.88%, and 93.97%, respectively. These results show that VGG16 transfer learning can classify cassava leaf diseases accurately even with a limited dataset. A Flask-based web interface was also designed to present the predictions to users.