Classification of Sugarcane Leaf Diseases Using Vision Transformers and CNN Models

Abstract

A globally prominent economic crop, sugarcane is an indispensable raw material for over 80% of sugar production worldwide. In Thailand, the sugarcane and sugar industry holds a top position in export markets. The loss of sugarcane crops due to diseases is a devastating problem that can never be overstated. Not only does it affect the economy, but it is also the primary source of income for many farmers in the provinces. To prevent a wide spread of any disease, farmers have long been heavily relying on visual inspections and their expertise to detect any signs of disease as early as possible. To assist farmers, this work applied computer vision and machine learning technology to help classifying sugarcane diseases from its leaves. Deployed on mobile devices, our application allows farmers to easily send to our chat-bot a photo of their suspected sugarcane leaves, and get a real-time response specifying the name of the disease or none if it is deemed healthy. Trained and fine-tuned on public data sets, our classifier, which is a vision transformer model, outperformed previous works. Tested on a newly collected local data set, ours achieved a high accuracy 79.64%.