Transfer Learning for Potato Leaf Disease Detection

Deep learning techniques have demonstrated significant potential in the agriculture sector to increase productivity, sustainability, and efficacy for farming practices. Potato is one of the world's primary staple foods, ranking as the fourth most consumed globally. Detecting potato leaf diseases in their early stages poses a challenge due to the diversity among crop species, variations in symptoms of crop diseases, and the influence of environmental factors. In this study, we implemented five transfer learning models including VGG16, Xception, DenseNet201, EfficientNetB0, and MobileNetV2 for a 3-class potato leaf classification and detection using a publicly available potato leaf disease dataset. Image preprocessing, data augmentation, and hyperparameter tuning are employed to improve the efficacy of the proposed model. The experimental evaluation shows that VGG16 gives the highest accuracy of 94.67%, precision of 95.00%, recall of 94.67%, and F1 Score of 94.66%. Our proposed novel model produced better results in comparison to similar studies and can be used in the agriculture industry for better decision-making for early detection and prediction of plant leaf diseases.