Plant disease detection using vision transformers

Abstract

Plant diseases present a major risk to worldwide food security and the sustainability of agriculture, leading to substantial economic losses and hindering rural livelihoods. Conventional methods for disease detection, including visual inspection and laboratory-based techniques, are limited in their scalability, efficiency, and accuracy. This paper addresses the critical problem of accurately detecting and diagnosing plant diseases using advanced machine learning techniques, specifically vision transformers (ViTs), to overcome these limitations. ViTs leverage self-attention mechanisms to capture intricate patterns in plant images, enabling accurate and efficient disease classification. This paper reviews the literature on deep learning techniques in agriculture, emphasizing the growing interest in ViTs for plant disease detection. Additionally, it presents a comprehensive methodology for training and evaluating ViT models for plant disease classification tasks. Experimental results demonstrate the effectiveness of ViTs in accurately identifying various plant diseases across a balanced 55 classes dataset, highlighting their potential to revolutionize precision agriculture and promote sustainable farming practices.