Unveiling agricultural insights: leveraging deep learning for enhanced diagnostic accuracy in Maize disease detection with explainable artificial intelligence

Abstract

Maize is a vital crop that feeds over a billion people worldwide and supports numerous industries. However, maize production is threatened by devastating plant diseases such as Maize Lethal Necrosis (MLN) and Maize Streak Virus (MSV), which can lead to significant yield losses and economic impacts, particularly in Sub-Saharan Africa. Therefore, to prevent significant losses of this essential crop, farmers need to be equipped with advanced tool that enables accurate and timely disease detection. In this regards, we have implemented a comparative performance analysis of five Transfer Learning (TF) (EfficientNetV2B2, ResNet50, InceptionV3, VGG16, and Xception) and five Vision Transformer (ViT) (SWIN, DaViT, MobileViT, MaxViT, and Involutional Neural Network (INN)) models for maize crop disease detection. We subsequently developed a fusion model that integrates MobileViT and DaViT. Afterward, the performance of the models was evaluated using multiple metrics such as precision, recall, and f1-score. The proposed fusion model perform best across all the metrics with an accuracy of 96.67%, recall of 95.84%, precision of 96.34%, and a f1-score of 96.54%. For transparent decision-making, three explainable artificial intelligence (XAI) techniques such as saliency map, gradient weighted class activation mapping (Grad-CAM), and local interpretable model agnostic explanations (LIME) have been implemented. Finally, we deployed the proposed fusion model on a Raspberry Pi to facilitate real-time detection of maize diseases.