Leveraging EfficientNet for Enhanced Grape Leaf Disease Detection: A Novel Approach to Precision Viticulture
Pratham Kaushik, Pooja Sharma
Abstract
The grape industry is infected by different diseases caused by fungal, bacterial, and viral pathogens. These diseases bring about fungi, bacteria, and viruses in grape yields, thus diminishing yields and fruit quality while affecting plant health. Thus, detection of these diseases in their early stages and accurately is considered significant to ensure grapevine health for the management of high-quality grapes. The paper presents a proposal for a new EfficientNet-based approach for the classification of grape diseases. This was one of the state-of-the-art models in convolutional neural networks, with extraordinary effectiveness and efficiency. In this study, grape diseases have been classified into four different categories: black rot, ESCA, leaf blight, and healthy. This dataset contains 1,806 images, each belonging to one of the aforementioned categories. The EfficientNet model does quite well in this classification task, producing an overall accuracy of 99%. Now, at a fine-grained level of analysis, one could expound on the fact that results turn out to achieve very high precision, recall, and f1-scores across all categories; actually, Healthy and Leaf Blight both show perfect scores. Black Rot and ESCA metrics are very impressive, with more than 95% accuracy and recall rate. This clearly emphasizes the model's robustness and effectiveness for differentiating among grape diseases and can be a valuable tool in vineyard management and prevention. This paper shows how state-of-the-art machine learning techniques can exploit current agricultural practices to really improve grape production outcomes.