Optimization and Application of Grape Disease Identification Algorithm Based on Deep Learning

Grape, as an important cash crop in the world, is susceptible to many diseases during its growth. Traditional disease identification methods are inefficient and subjective. In this study, deep learning (DL) technology, especially convolutional neural network (CNN), was used to significantly improve the accuracy and efficiency of grape disease identification through model design optimization, transfer learning, data enhancement and other technologies. A data set containing about 10,000 high-quality grape leaf images was constructed, and MobileNetV3 was used as the basic model, and residual connection, depth separable convolution and attention mechanism were introduced to optimize it. The experimental results show that the optimized model has significantly improved the accuracy, recall and F1 score, and effectively solved the fine-grained classification problem in grape disease identification. In addition, this study also discussed the application of data enhancement technology and weighted cross-entropy loss function in dealing with category imbalance, which provided a new solution for intelligent identification of grape diseases and was of great significance for promoting the healthy development of grape industry.