Development of a machine vision system based on convolutional neural networks to detect open and closed tips in asparagus

Abstract

The project aims to develop a computer vision system based on convolutional neural networks (CNNs) for the automatic, eal-time detection of asparagus tip status (open or closed), in order to improve efficiency and accuracy in agroindustrial sorting and classification. The YOLOv11 model was employed, training on a set of 1,663 annotated asparagus images captured under varying lighting conditions and from multiple angles. Using the COCO Annotator platform for labeling and Google Colab with advanced computational esources for training, the system achieved 96% accuracy, with a 94.3% mAP and an F1-score of 0.95, standing out for its low false-positive and false-negative rates. Experimental results demonstrate that the system overcomes the limitations of manual inspection by delivering higher precision and faster processing, making it suitable for automating classification in industrial production. In conclusion, the YOLOv11-based system provides an effective and accurate solution for real-time asparagus tip detection, representing a significant step oward modernizing the agroindustrial sector. In future work, the system will be expanded to detect additional defects and further optimized for deployment in real industrial environments.