Grape Leaf Disease Detection Using Deep Learning: A Hybrid Approach with Efficientnet and Mobilenet

Abstract

Viticulture faces challenges from grape leaf diseases that threaten crop yield and quality. This paper aims to introduce a hybrid EfficientNet - MobileNet model and determine how the hybrid model can improve the accuracy and efficiency in identifying leaf diseases from plant images. The datasets used are the grape leaf datasets from the PlantVillage^, dataset, Leaf Health Classification and Disease Detection, and from reliable agricultural websites and research publications which comprise black rot, esca, and leaf blight classes. A total of four thousand one hundred fifty - four (4154) images were used. Fifty images were also used for disease classification. The hybrid model was evaluated using accuracy, recall, precision, and f1 – score. A confusion matrix was also generated. The results revealed a 98.4838% accuracy rating. Forty-six out of fifty images were correctly classified using the hybrid model. The performance of the hybrid model as well as its prediction percentage implies that the hybrid model is reliable for grape leaf disease detection. This further implied that the hybrid model has the potential to enhance image classification techniques in detecting grape leaf diseases. Their use may result in more precise and effective grapevine health monitoring, which could change viticulture's approaches to disease control. This could promote sustainable farming methods and maximize crop productivity by enabling earlier identification, better yield management, and less dependence on chemical treatments. Increasing the dataset, fine-tuning with more feature extraction, and modifying augmentation techniques may be used to enhance the model's performance.