Leveraging Transfer Learning for Efficient Classification of Coffee Leaf Diseases

Coffee leaf diseases pose a significant threat to the quality and yield of coffee crops, necessitating early and precise identification for effective disease management. This study introduces a robust approach leveraging transfer learning to classify seven prevalent coffee leaf diseases. By employing a ResNet50v2 model, the research aims to enhance classification accuracy while mitigating bias. The proposed methodology integrates data preparation, preprocessing, data augmentation, partial layer freezing, feature fusion, and fully connected layers to develop a reliable disease classifier. The ResNet50v2 model initially distinguishes healthy from unhealthy leaves, achieving an impressive test accuracy of ${96.99\%}$. In subsequent stages, the model classifies unhealthy leaves into sooty molds, brown spots, and rust leaf diseases with ${94.40\%}$ accuracy, and further identifies red spider mite, leaf miner, phoma, and cercospora diseases with ${92.66\%}$ accuracy. Overall, the model demonstrates a classification accuracy of ${94.20\%}$ across the entire dataset, underscoring its efficacy in detecting and classifying multiple coffee leaf diseases.