Leveraging Transfer Learning-based Fine-tuned ResNet50 Model for Maize Leaf Disease Classification

The anticipation of corn leaf disease is crucial in precision agriculture, facilitating prompt actions to protect crop output and quality. This study utilizes a pre-trained ResNet50 deep-learning model as a predictor for maize leaf disease classification. This model has been implemented on an extensive dataset comprising images of both healthy and diseased maize leaves collected from Kaggle. The proposed ResNet50 architecture has been implemented for multiclass classification containing the corn blight, corn common rust, corn grey spot and healthy class. Along with the multiclass classification, the binary class classification results have been also analyzed for corn blight vs corn common rust, corn blight vs corn grey spot, and corn common rust vs corn grey spot. The accuracy results have been shown at different numbers of epochs for multiclass classification as 94.70%, 95.02%, 94.80%, 95.49%, and 96.40% at epochs 10, 15, 20, 25, and 30. Further, the accuracy of corn blight vs corn common rust, corn blight vs corn grey spot, and corn common rust vs corn grey spot have been also identified as 91.10%,98.30%, and 98.70%, respectively. The proposed model has shown remarkable performance for both multi and binary classification. The performance can further be enhanced by performing data augmentation for increasing the number of images in the dataset.