Deep learning-based detection of wine bottle capsules for contamination prevention

• YOLOv8, YOLOv9, and YOLOv11 effectively detected the absence of capsules. • AI-based inspection systems are suitable for preventing aerial contamination. • Precision, recall, and mAP of over 96% confirm that DL can assist human inspection. Quality control of wine bottle packaging is crucial for preventing wine contamination and safety hazards, ensuring product integrity, and maintaining the overall quality of the product. The presence of metallic capsules has been demonstrated to significantly reduce airborne contamination, particularly from 2,4,6-trichloroanisol (TCA), a major contributor to cork taint. Traditional capsule detection in the wine industry primarily relies on visual inspection, thus prone to human error. Despite the progress of Industry 4.0, the application of artificial intelligence (AI) for automated inspection in the wine industry remains limited. This study evaluates the suitability of deep learning (DL) for inspecting capsules in wine bottles. A dataset of 12,050 images was used to train, test, and validate the performance of YOLOv8, YOLOv9, and YOLOv11 models. All three models demonstrated similar performance, achieving high precision (>96%), recall (>97%), and mAP (>98%). Notably, YOLOv11 exhibited the fastest inference speed, making it a strong candidate for real-time detection. These results demonstrated the feasibility of deep learning (DL) in assisting humans with wine packaging inspection to prevent further airborne contamination and enhance real-time quality control. The proposed approach provides a practical and affordable solution to reduce reliance on manual inspection, particularly for small and medium producers, making inspection processes smarter and more efficient.