COMPUTER VISION IN INDUSTRIAL QUALITY CONTROL: A YOLO-BASED DEFECT DETECTION SYSTEM

This paper presents an intelligent system for automatic detection and classification of industrial production defects based on the YOLO (You Only Look Once) neural network architecture. The developed system provides real-time image analysis, defect localization using bounding boxes, and multi-class defect classification. In contrast to existing solutions, the approach proposed in the article combines data augmentation methods specific to industrial images (simulation of lighting changes, addition of industrial noise, synthetic generation of rare defect types) with the technique of progressive learning from general to specific features. The system is implemented as an embedded module for industrial quality control lines with a REST API. A comparative evaluation of YOLOv5, YOLOv8 and YOLOv11 versions on industry datasets including surface defects of metal, textile and plastic products was conducted. The high efficiency of the proposed approach is demonstrated: mAP@0.5 reaches 0.934 with a single frame processing time of 12–18 ms on GPU. The system architecture includes image preprocessing, defect detection, post-processing and visualization modules. A methodology for forming and annotating training datasets for industrial quality control tasks is proposed.

computer vision, defect recognition, quality control, deep learning, object detection, neural networks, image processing, industrial automation.

1. Bergmann P. et al. MVTec AD — A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection. CVPR, 2025, pp. 9592–9600.

2. Goodfellow I., Bengio Y., Courville A. Deep Learning. MIT Press, 2021, 800 p.

3. He K., Zhang X., Ren S., Sun J. Deep Residual Learning for Image Recognition. CVPR, 2023, pp. 770–778.

4. Jocher G. et al. Ultralytics YOLOv8. 2023. URL:https://github.com/ultralytics/ultralytics.

5. Lin T.-Y. et al. Feature Pyramid Networks for Object Detection. CVPR, 2017, pp. 2117–2125.

6. Redmon J., Divvala S., Girshick R., Farhadi A. You Only Look Once: Unified, Real-Time Object Detection. CVPR, 2022, pp. 779–788.

7. Song K., Yan Y. A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects. Applied Surface Science, 2021, vol. 285, pp. 858–864.

8. Tan M., Le Q. EfficientDet: Scalable and Efficient Object Detection. CVPR, 2024, pp.10781–10790.

9. Ultralytics YOLOv11. 2024. URL: https://docs.ultralytics.com/models/yolo11.

10. Yakimov P.Yu., Nagibin V.N. Surface defect detection using convolutional neural networks. Computer Optics, 2021, vol. 45, no.6, pp. 918–929.

2026-08-24