MaizeXNet: A CNN-Based Explainable AI Framework for Maize Pest and Disease Classification

Global food security depends heavily on agriculture, yet crop yield is still severely hampered by pest and disease infestations, especially in the cultivation of maize (Zea mays L.). Accurate and timely identification of such conditions is vital for sustainable crop management. This study introduces MaizeXNet, a lightweight and efficient Convolutional Neural Network (CNN) architecture integrated with Explainable Artificial Intelligence (XAI) for automated classification and interpretation of maize pest and disease classes. The framework employs EfficientNet B0 as a pretrained backbone with a Depthwise Separable Convolutional Block to enhance feature extraction while reducing computational complexity. The dataset used in this study includes diverse maize pest and disease classes, covering both pest infestations and pathogen-induced infections. A balanced dataset was achieved through preprocessing and augmentation to improve generalization and mitigate overfitting. Comparative experiments with EfficientNet B0, ResNet50, and MobileNetV2 demonstrated that MaizeXNet achieved superior performance, attaining 99.58 percent validation accuracy and 99.38 percent test accuracy, along with higher precision-recall statistics, F1 measure, and AUC-ROC values than baseline models. Furthermore, GradCAM++ and LIME were applied to interpret MaizeXNet's predictions and visualize key regions influencing classification. The results confirm that MaizeXNet provides a reliable and interpretable framework for intelligent maize pest and disease detection in modern agriculture.