PlantLDNet : Disease Detection in Leaf Images Using Deep Learning Models

Automated disease diagnosis in the agricultural field has benefited from the recent advances in big data, deep learning, and computer vision across a wide range of applications. However, intelligent disease prediction is crucial for safely maintaining a good agricultural production. Moreover, unclear disease prognosis is caused by the degree of leaf disease inhomogeneity, the complexity of the background-which includes the low contrast with the surrounding area and the potential for shadows with a similar intensity. This research provides a custom Convolutional Neural Network (CNN) model called PlantLDNet that has hyperparameters configured optimum to distinguish between leaf images showing disease and those without it, hence avoiding the problems associated with vanishing gradient. A comparative study of the proposed custom CNN model PlantLDNet and earlier research works was conducted using the following metrics: precision, recall, accuracy, F1 score, loss accuracy, receiver operating characteristic (ROC) curves, and Area Under Curve (AUC) measure. The comparison findings show that the proposed PlantLDNet model has the highest classification prediction effectiveness (99.85%) with less computational complexity. It also has an elevated high F1-score of 93% when trained for Early Bright, 98% for Healthy, and 92% for Late Bright.