SwinGNet: A Hybrid Swin Transform- GoogleNet Framework for Real-Time Grape Leaf Disease Classification

In recent years, automation has been a significant trend in different sectors of the economy, and the agricultural sector is considered the least technological development. Grape framing is an important cultivation, but grape farmers face crop loss due to the need for more analysis of diseases. This research introduces a hybrid approach for classifying grape leaf disease called SwinGNet. Microsoft’s Swin Transform captures the hierarchical features of various sizes and extracts features from the image, and Google’s GoogLeNet to classify grape leaf diseases from this feature. The proposed SwinGNet model is developed and deployed in Raspberry Pi 5 Graphics Processing Unit (GPU) computer. Fusing these concepts on a GPU platform with a deep learning framework improved the performance of the pretrained models. Based on the real-time field tests, SwinGNet performs with a higher validation accuracy (99.2%), sensitivity (98.67%), precision (99.03%), F1 score (98.40, AUC (0.98) and outperforming the pre-trained models.