ViT vs CNN: A Comparative Study of Wheat Disease Classification for Custom Data

Wheat plays an important role in Pakistan's agricultural sector, ensuring food security and livelihoods for many. However, the persistent danger of wheat diseases needs immediate attention. This research focuses on advanced wheat disease classification, using Artificial Intelligence (AI) for timely identification. To validate this approach, we used a self-compiled dataset containing a wide range of wheat disease occurrences in Pakistan. This strategy focuses on the Vision Transformer (ViT) concept, specifically the ViT-B32, which processes images as 32x32 patches and captures complex features with efficiency. We tailor ViT-B32 to detect plant health anomalies by fine-tuning it on the wheat disease dataset. This research contains a comparison of ViT-B32 with the VGG-19 Convolutional Neural Network (CNN) model. The outcomes are instructive since VGG-19 only manages an accuracy of 98.4% whereas ViT-B32 surpasses it by about 99.2%. The difference highlights ViT's capacity in accurately recognizing wheat diseases, boosting its potential for precision agriculture. Through the integration of AI and harnessing ViT-B32's capabilities, this research advances disease classification, fostering refined crop management practices and bolstering agricultural sustainability.