Performance and Computational Complexity Analysis of Pre-Trained Deep Learning Models for Citrus Fruit Disease Classification

Citrus fruit diseases are a real threat affecting agricultural productivity and can lead to losses that are hard to recover if they are not detected and managed promptly.This research focuses on using transfer learning and deep learning architectures to classify citrus diseases to reduce agricultural losses and increase crop productivity.The study uses a dataset of 1,463 high-resolution images of citrus fruits and leaves that fall under these classes: canker, black-spot, greening, and healthy.The images were processed for analysis using a pre-trained Neural Network (CNN) such as MobileNetV2, InceptionV3, ResNet50, and VGG19.Transfer learning was used to fine-tune these models on the task of disease classification, which depends on features previously learned from larger datasets.Model performance was evaluated through accuracy, precision, recall, F1 score, and computational efficiency metrics.Out of the four models analyzed in this research study, MobileNetV2 attained the highest accuracy at 97.29% while ResNet50 did not perform well with an accuracy value of 58.21%.This emphasizes the importance of the selection of models.This research showcases how transfer learning and deep learning can bring scalability into automated systems for managing citrus diseases providing better accuracy, and quicker diagnosis times.