Deep Learning-Based Classification of Banana Leaf Diseases Using VGG19: A Robust Approach for Agricultural Diagnostics

Abstract

This study presents a deep learning-based approach using the VGG19 model to classify banana leaf conditions into four categories: Cordana, Healthy, Pestalotiopsis, and Sigatoka. To train the model on the images, a dataset of 973 images was collected from Kaggle and then preprocessed through resizing, normalization and data augmentation to make the model as versatile as possible to simulate real-world scenarios. The VGG19 model using transfer learning gave an overall validation accuracy of 88%. A measure of precision, recall, F1 scores, and confusion matrix presented a nearly equal performance across all classes with insignificant errors. Measures including dropout, batch normalization and early stopping were utilized to enhance the model’s ability to generalize and avoid over complication of the training dataset. The results confirm that the proposed model can be useful in feature extraction and classification hence can be used to automate the banana leaf disease detection process. Further works can include the enlarging of the database, the optimization of the parameters and the implementation of the model to real time diagnostic tools for the general public. This work shows that deep learning has the ability to enhance agricultural diagnostics, thus supporting sustainable agriculture and minimizing crop damage.

Topics & Concepts