Leveraging Artificial Intelligence for Early Aflatoxin Detection in Groundnuts to Advance Sustainable Postharvest Food Safety Management

Abstract

Aflatoxin contamination of groundnuts remains a significant food safety and economic issue in Uganda; however, the traditional laboratory methods for its detection. ELISA, a high-performance liquid chromatography, and thin-layer chromatography are expensive, time-consuming, and largely unattainable for smallholder farmers and postharvest handlers. This work investigates whether a deep learning model, trained solely on images, can provide a quicker and more accessible method for screening groundnuts for aflatoxin-related problems during sorting. A convolutional neural network, Inception-ResNet-V2, was created using transfer learning to categorize groundnuts into four postharvest classifications: healthy, moldy, pest-infested, and physiologically disordered, each associated with varying levels of aflatoxin risk. On the reserved test dataset, the model attained an accuracy of 99.29%, precision of 100%, recall of 98.44%, and an F1-score of 99.21%, with the moldy class being categorized nearly flawlessly (AUC = 1.00). The findings indicate that image-based deep learning can significantly enhance low-cost postharvest inspection, while also identifying some categories, primarily pest-infested groundnuts, where the model's discriminative ability is relatively poorer and necessitates further methodological development.