Comparative Study of Advanced Deep Learning Techniques for Citrus Disease Classification

Abstract

The accurate and timely classification of citrus diseases is essential for improving agricultural productivity and reducing crop losses. In this study, an advanced deep learning-based framework is developed to classify citrus diseases with high precision using multiple state-of-the-art convolutional neural network (CNN) architectures. The proposed system incorporates a MATLAB-based Graphical User Interface (GUI) that facilitates the selection, training, and evaluation of six prominent deep learning models: VGG16, InceptionV3, MobileNetV2, GoogLeNet, ResNet50, and DenseNet. Each model is assessed using three widely recognized optimizers—Stochastic Gradient Descent with Momentum (SGDM), ADAM, and RMSprop—under both baseline learning and transfer learning configurations. The system quickly calculates important performance measures, such as accuracy, precision, recall, sensitivity, and specificity, for each combination of model and optimizer. Extensive experiments reveal that transfer learning significantly enhances classification performance across all models. Among the evaluated architectures, ResNet50 consistently outperforms others, achieving superior accuracy, sensitivity, and precision across all optimization techniques. The proposed system also integrates visualization tools that generate comparative performance graphs, offering an intuitive way to analyze results and identify optimal model configurations. This study shows that deep learning models, especially ResNet50 using SGDM optimization, can effectively classify citrus diseases with high accuracy. The findings provide important lessons about precision agriculture and can assist farmers and agricultural specialists in making timely, data-driven decisions for disease management. The developed GUI provides a user-friendly platform for rapid deployment and real-time evaluation of deep learning models in agricultural applications.