High-Accuracy Real-Time Defect Classification for Coffee Beans Using Deep Learning

Abstract

Consistent quality inspection of coffee beans remains a critical requirement in the agricultural supply chain, as manual sorting is slow, inconsistent, and dependent on worker experience. With recent advancements in deep learning, real-time object recognition models provide an opportunity to automate quality assessment with high accuracy and minimal intervention. This study presents a YOLOv8-based computer vision system designed to detect and classify defective coffee beans, including broken, discolored, insect-damaged, and mold-affected samples. The proposed model is trained on an annotated dataset of roasted and unroasted beans under varying illumination and background conditions. The system demonstrates strong generalization capability, outperforming traditional CNN classification systems in speed and reliability, achieving up to 97.8% detection accuracy at 60 FPS on GPU inference. The study concludes that YOLOv8 can significantly improve the speed and precision of coffee quality analysis, enabling fully automated grading lines for industrial deployment.[2]

Abstract from the paper’s OpenAlex record, shown for indexing and research. Copyright remains with the authors or the publisher. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.