Enhancing crop health monitoring: A deep dive into GNN-integrated models for wheat disease detection

Abstract

Plant diseases pose an important threat to agricultural productivity, affecting both the quality and quantity of crops. Early detection and severity assessment of infections in plant crops are critical for effective disease management and minimizing crop loss. This paper proposes a methodology for detecting wheat crop diseases using hybrid deep learning models that combine graph neural networks (GNNs) with convolutional architectures. By leveraging GNN + convolutional neural network (CNN), GNN + ResNet, and GNN + Visual Geometry Group 16 (VGG16) models, we aim to enhance the ability to detect diseases from images of wheat leaves accurately. The proposed models were trained on a comprehensive dataset of wheat crop images, with extensive preprocessing, model training, and hyperparameter tuning to optimize their performance. Our results indicate that the GNN + CNN model achieved the highest accuracy at 93%, followed by GNN + ResNet at 86% and GNN + VGG16 at 82%. These findings suggest that GNN + CNN is particularly effective for disease detection, providing a high degree of accuracy and robustness. This approach shows promise for automated, precise crop disease management, offering a valuable tool for advancing agricultural productivity and disease control.