Smart Agriculture: Leveraging ResNet50 for Precise Detection and Classification of Banana Leaf Diseases
Ruchika Bhuria
Abstract
Various diseases that greatly lower output and damage growers financial stability provide major difficulties for banana farming. Images of diseased (Sigatoka, Cordana, Pestalotiopsis) and healthy banana leaves comprise the dataset, taken with smartphone cameras and painstakingly annotated by a trained plant pathologist in Bangabandhu Sheikh Mujibur Rahman Agricultural University and surrounding areas in Bangladesh. The research paper developed and tested the ResNet50 model on this dataset to classify the diseases. The model developed a general classification accuracy of 96.4%. With accuracy, recall, and F1-scores of above 94% for all classes, the classification report reveals that the model is robust in detecting the disorders. Especially, the model shown amazing accuracy in identifying healthy leaves with minimal misclassification. Given the Pestalotiopsis class has the best misclassification rate, the confusion matrix emphasizes even more the capacity of the model. This paper emphasizes the opportunities of deep learning in automating disease diagnosis by means of early detection and quick response, therefore supporting the global banana supply chain and enhancing disease management strategies.