Share
Export Citation
Optimization of the YOLOv4 Algorithm with MFNMS for Defect Detection in Arabica Coffee Beans
Kasman M.F.
Proceeding 2025 IEEE 11th Information Technology International Seminar Itis 2025
Abstract
Coffee is one of Indonesia’s leading commodities with high economic value, especially the Arabica variety because it contains higher levels of natural sugars and acids, compared to other coffee varieties. However, the quality of Arabica beans is greatly influenced by their physical and visual conditions, especially the presence of defects. Defect detection in coffee beans is crucial for maintaining the quality and efficiency of postharvest processing in the coffee industry. YOLOv4 is a fast object detection algorithm, but its performance decreases when detecting small objects that are close together. To improve defect detection in Arabica coffee beans, this study proposes an improvement in YOLOv4 using the proposed More Focused Non Maximum Suppression (MF-NMS) method, which normalizes the coordinates of the lower left and upper right corners of the predicted bounding box and the ground truth bounding box. This is done to reduce the impact of small errors in object position prediction. The data used in this study consists of green Arabica coffee beans categorized as normal, hollow, broken, and black. The results show that the proposed model achieves a 7.50% increase in $\mathbf{m A P} \text{@} \mathbf{0. 5}$ compared to the baseline model.