Interpretable Deep Learning for Diagnosis of Maize Streak Disease

Maize Streak Disease (MSD) caused by the maize streak virus remains a significant agricultural challenge during the 4th industrial revolutions, as it poses a substantial threat to global maize crops, resulting in severe yield reductions. Effective and timely identification of the disease is pivotal in curbing its propagation and enacting appropriate control strategies. The emergence of the 5th industrial revolutions has witnessed the integration of advanced technologies into various sectors, with deep learning methodologies standing out for their prowess in diverse domains, including disease detection. Nonetheless, the inherent complexity of deep learning models hinders visibility into their transparency, preventing users from comprehending the underlying features and decision-making processes. This opacity often leads to reduced user trust, even when the models demonstrate high accuracy rates.In this research, we introduce an interpretable machine learning framework through explainable AI (XAI) that harnesses the capabilities of deep learning, specifically Convolutional Neural Networks (CNNs), for the identification of the Maize Streak Disease. This framework not only exploits the potential of deep learning for precise disease classification but also integrates techniques for model interpretability. The proposed methodology capitalizes on CNN architectures to extract discerning features from an extensive dataset of MSD-affected leaf images. The trained CNN model is then coupled with interpretable techniques such as SHAP and LIME.Empirical findings highlight that this easy-to-understand deep learning approach is very effective in correctly diagnosing the Maize Streak Disease. The TensorFlow-designed model achieves a commendable accuracy of 96%.In essence, this study presents a novel interpretable machine learning framework that seamlessly integrates deep learning with interpretability methods to tackle the challenge of Maize Streak Disease. Beyond achieving exceptional classification precision, this research provides a transparent window into the decision-making mechanisms of the model. With a pronounced emphasis on interpretability, this study endeavors to bridge the gap between the potent predictive capacities of deep learning and the imperative for intelligible and dependable diagnostic systems within the agricultural disease domain.The implications of this work extend to the 5th industrial revolutions by bolstering agricultural productivity, mitigating yield losses, and enhancing the efficiency and efficacy of Intelligent Agricultural Support Systems.