Investigation of the suitability of existing Maize Plant Leaf Disease detection and classification approaches: Challenges and Open Issues

Abstract

Maize, a crucial staple crop in Zambia and many other regions, is seriously threatened by several leaf diseases, such as Gray Leaf Spot, Maize Streak Virus, and Northern Corn Leaf Blight. Early detection and accurate classification of these diseases are challenging due to the time-consuming and error prone nature of traditional detection techniques, such as visual inspections by farmers or experts. In recent years, deep learning has shown promise as an automated method for identifying and categorizing plant diseases. This paper describes a deep learning-based framework for identifying leaf diseases in maize plants and identifies the main obstacles and unresolved problems in the field. The lack of large enough and diverse datasets is one of the main obstacles to using deep learning for maize disease diagnosis, particularly in certain regions such as Zambia. The usefulness of existing models is limited in real-world scenarios because they frequently fail to generalize across various environmental circumstances, such as variances in climate, illumination, and soil type. In addition, the class imbalance creates a big gap in datasets with overrepresentations of specific diseases, which skews model predictions. The lack of lightweight, deployable models appropriate for low-resource settings, including rural farms with limited access to high-end computing equipment, is another significant gap. Furthermore, deep learning models are frequently perceived as "black boxes," and because farmers and other agricultural specialists need explicable insights into disease forecasts, they are less likely to be adopted due to the lack of interpretability of the models. This paper addresses the requirement for ongoing model updates to deal with changing disease patterns and investigates the possibilities of domain adaptation and transfer learning approaches in enhancing model generalization across conditions and locations. The report concludes by urging a concentrated effort to incorporate regional farmers and agricultural stakeholders in the development process to guarantee that the solutions are workable, approachable, and contextually appropriate. Even though deep learning has a lot of promise to improve the detection of maize leaf disease, there are still several issues that need to be resolved to produce more scalable and successful solutions. This study identifies these gaps and makes recommendations for how to close them in the future to support food security and sustainable agricultural development in areas like Zambia.