XHVCT: Explainable Hybrid Vision Convolutional Transformer for Maize Disease Detection

Maize is a staple crop that suffers significant yield losses worldwide due to various diseases. Current deep learning techniques struggle to accurately capture intricate patterns and provide clear reasoning for their predictions, limiting their usefulness in the agricultural sector. To address these limitations, this study introduces the Explainable Hybrid Vision Convolutional Transformer (XHVCT) architecture for automated maize disease detection. The XHVCT model combines convolutional neural networks (CNNs) with attentionbased transformers, allowing it to extract long-range dependencies in image features and enhance its ability to recognize complex patterns associated with various maize diseases. Additionally, the incorporation of Local Interpretable Model-agnostic Explanations (LIME) facilitates interpretability, providing explanations for the model's predictions that can be validated by domain experts. Experiments were conducted using diseased and healthy samples, comprising a total of 9,149 maize leaf images across four distinct classes. The findings showcase the effectiveness of the proposed XHVCT model, which attains a test accuracy of 97.34% with only 0.33 million parameters. In conclusion, this study underscores the importance of accurate and interpretable disease detection systems in aiding farmers with timely disease management decisions. The XHVCT model offers promising practical applications in agriculture, enhancing crop protection strategies by providing valuable insights into disease presence.