DeepCitrus: Leveraging Integrated CNN and VGG16 for Automated Orange Leaf Disease Detection

Abstract

Citrus cultivation is confronted with significant risks posed by a multitude of diseases, which demand novel methodologies to ensure prompt and precise detection. The current research presents a novel approach to disease detection on orange leaves by combining the VGG16 model with Convolutional Neural Networks (CNN). The hybrid model under consideration exploits the respective merits of both architectures by utilizing the VGG16 capabilities for feature extraction and the VGG16 model for classification. We have curated an extensive dataset consisting of images depicting both healthy and diseased citrus leaves, which includes a wide range of pathogens that impact orange trees. By undergoing training on this dataset, the integrated CNN-VGG16 model named "DeepCitrus" acquires the capability to differentiate complex visual patterns that are linked to various manifestations of diseases. The integration of these two robust models significantly improves the precision and dependability of disease identification as a whole. The results of the proposed DeepCitrus have been provided in terms of training accuracy, testing accuracy, training loss and testing loss at different numbers of epochs. The CNN model in the proposed DeepCitrus has been fine-tuned by making the changes in various hyperparameters such as the learning rate set to 0.001, batch size as 16, and number of epochs to 10, 20, 30, 40, and 50. The results depict that at epoch count 50 has shown the highest testing accuracy value of 98.43%, which was 97.31% in the testing phase. The training and testing loss has been also identified as the lowest at epoch 50 as 0.0678, and 0.0432, respectively.

Topics & Concepts