Advanced Grape Leaf Disease Diagnosis Using EfficientNetV2L with Data Augmentation and Grad-CAM Visualization in Precision Agriculture

Grape cultivation is essential to the global agricultural economy, but grape plants are highly vulnerable to diseases such as downy mildew, powdery mildew, black rot, and anthracnose. These diseases cause significant yield losses and impact plant health, necessitating accurate and timely identification for effective management and to maintain vineyard productivity. Conventional grape leaf disease identification relies on visual inspections by trained agronomists, a process that is time-consuming, labor-intensive, and susceptible to human error. Recent advancements have introduced machine learning and deep learning techniques, utilizing models like VGG, ResNet, and EfficientNet, and leveraging annotated datasets to automate disease classification. Existing methodologies often face challenges such as scalability, handling class imbalances, and adapting to environmental variability. Additionally, some approaches may lack sufficient accuracy or efficiency for real-time applications in the field, limiting their practical utility. This study proposes using the EfficientNetV2L deep learning model to address these limitations by leveraging transfer learning and extensive data augmentation. The EfficientNetV2L model is designed to balance model size and accuracy, providing robust performance across various scales of input images and ensuring reliable disease classification. The proposed model employs a pre-trained EfficientNetV2L architecture, enhanced with custom fully connected layers and comprehensive data augmentation techniques such as resizing, rescaling, and random transformations. The training process incorporates techniques like early stopping and checkpointing to optimize performance and prevent overfitting, ensuring the model’s robustness and generalizability.