VitiNet: Precision Grape Disease Identification with Deep Learning
K. Palanivel Rajan, J. Relin Francis Raj, R. Santhana Krishnan, Shobha Shankar, V. Vinoth Kumar, S. Kavitha
Abstract
Grape diseases pose a critical threat to vineyards worldwide, affecting both yield and crop quality. To address this, VitiNet, a deep learning-based algorithm utilizing EfficientNetB7, is designed for precise and scalable grape disease identification. The model focuses on five major grapevine diseases: Powdery Mildew, Downy Mildew, Anthracnose, Black Rot, and Esca. Two custom datasets, GrapeVine Dataset Dataset and VineHealthDataset, were curated to provide a diverse set of images for model training and evaluation. EfficientNetB7, known for its scalability and high performance in image classification, was fine-tuned by unfreezing the last five layers to allow for the extraction of disease-specific features. This fine-tuning enables the model to adapt to the complexities of grape disease identification while maintaining computational efficiency. The model’s implementation includes data augmentation techniques to enhance the diversity of the training set, while Stochastic Gradient Descent (SGD) and Categorical Cross-Entropy optimize the model’s learning process. While accuracy, precision, recall, and F1 score were evaluated on both datasets, the core strength of VitiNet lies in its ability to generalize well across varying environmental conditions and disease manifestations. VitiNet presents a significant advancement in precision viticulture, offering a reliable and automated solution for real-time disease monitoring. By enabling early and accurate disease identification, this model supports proactive disease management, ultimately improving vineyard productivity and sustainability.