HYBRID DEEP LEARNING MODEL FOR AUTOMATED PLANT DISEASE CLASSIFICATION

Abstract

Plant disease detection is essential to ensure the quality of agricultural production. However, manual identification of disease symptoms is often tedious and susceptible to human error. Deep learning techniques have been widely integrated into agricultural systems, providing innovative solutions thanks to their robustness. Most of these studies are essentially based on Convolutional Neural Networks (CNNs) architectures. This paper presents a lightweight and accurate model for the classification of 29 plant leaf classes using the PlantVillage dataset. We propose a hybrid learning model called Agriculture Network (AgriNet). AgriNet consists of only six layers, and integrates the strengths of three CNN models : Residual Neural Networks-50 (ResNet-50), Visual Geometry Group-16 (VGG-16) and AlexNet. ResNet-50 is a very deep and powerful network, whereas VGG-16, despite having a large number of parameters, achieves good accuracy using (3×3) filters. In contrast, AlexNet presents a relatively simple architecture compared to the other two deep models. An extensive set of experiments was conducted to evaluate the performance of the proposed model for plant disease detection. The analysis, based on assessment measures, including accuracy, precision, recall, and F1-score, demonstrates that our model achieves highly reliable results, with an accuracy of 97.13%. Furthermore, AgriNet maintains low complexity due to its lightweight hybrid architecture and reduced parameters count.