LeafDNet: Transforming Leaf Disease Diagnosis Through Deep Transfer Learning.

Abstract

The health and productivity of plants, particularly those in agricultural and horticultural industries, are significantly affected by timely and accurate disease detection. Traditional manual inspection methods are labor-intensive, subjective, and often inaccurate, failing to meet the precision required by modern agricultural practices. This research introduces an innovative deep transfer learning method utilizing an advanced version of the Xception architecture, specifically designed for identifying plant diseases in roses, mangoes, and tomatoes. The proposed model introduces additional convolutional layers following the base Xception architecture, combined with multiple trainable dense layers, incorporating advanced regularization and dropout techniques to optimize feature extraction and classification. This architectural enhancement enables the model to capture complex, subtle patterns within plant leaf images, contributing to more robust disease identification. A comprehensive dataset comprising 5491 images across four distinct disease categories was employed for the training, validation, and testing of the model. The experimental results showcased outstanding performance, achieving 98% accuracy, 99% precision, 98% recall, and a 98% F1-score. The model outperformed traditional techniques as well as other deep learning-based methods. These results emphasize the potential of this advanced deep learning framework as a scalable, efficient, and highly accurate solution for early plant disease detection, providing substantial benefits for plant health management and supporting sustainable agricultural practices.