A DEEP LEARNING-BASED MODEL FOR MAIZE DISEASE CLASSIFICATION

Abstract

Food security has remained an urgent need for a suitable life all over the world. This study focuses on maize, one of the widely cultivated staple crops, and proposes a solution to address low yields. The aim is the application of a deep learning-based model for maize disease classification. The methodology adopted is the Rapid Application Development (RAD) approach. The data used for this work were collected from three maize farms in Nsukka, Enugu State, Nigeria. The sample size of the collected data consisted of 12,032 images of maize leaves with diseases across eight classes (Yellow curl, Septoria, Gray leaf spot, Healthy, Mold, Bacterial soft rot, Mosaic virus, Late blight). The data were augmented using the PlantVillage dataset as the secondary source. Both data sets formed the new data model for maize disease detection. This was used to fine-tune DenseNet121, a pre-trained algorithm, and generate a maize disease classification model. The Python programming language was used for the implementation. Metrics such as accuracy, precision, recall, and F1 score were applied to evaluate the model's performance. The model achieved an accuracy of 95%, with precision, recall, and F1-scores consistently above 93%. Finally, the findings from the study confirmed that the DenseNet121-based applications offer a pathway to the mitigation of losses in maize yield, thereby strengthening food security and empowering smallholder farmers with precision agriculture tools. In addition to its scientific contribution, this study addresses a sustainable development goal of zero hunger. This is achieved by providing a user-friendly mobile-based precision agriculture application that supports real-time disease detection, reduces crop losses, and empowers farmers to adapt more effectively to the challenges of climate change and food insecurity.