REAL-TIME MULTI-CAMERA TEXTILE DEFECT DETECTION USING YOLO-BASED DEEP LEARNING AND TENSORFLOW LITE ON CPU PLATFORMS
Abstract
Automated quality inspection is a fundamental requirement in modern textile manufacturing. Manual inspection techniques are increasingly unable to satisfy industrial demands due to subjectivity, low inspection speed, and high labor costs. This paper presents a complete end-to-end system for real-time textile defect detection based on deep learning, optimized for deployment on CPU-only platforms. A YOLO-based object detection model is trained on textile defect datasets and subsequently converted into TensorFlow Lite (TFLite) format for efficient inference. The system supports images, video streams, live camera feeds, and simultaneous multi-camera processing. A lightweight web-based interface is implemented using Streamlit, providing visualization, logging, statistical analysis, and CSV export functionalities.
Extensive optimization techniques, including input resolution reduction, inference rate limiting, and asynchronous multi-camera execution, are employed to achieve real-time performance without GPU acceleration. Experimental evaluation confirms the system’s efficiency, scalability, and practical applicability in industrial textile quality control.
Keywords
Textile defect detection, deep learning, YOLO, TensorFlow Lite, real-time inspection, multi-camera systems, computer vision, Streamlit.
References
- Redmon, J., et al., You Only Look Once: Unified, Real-Time Object Detection, CVPR, 2016.
- Bochkovskiy, A., Wang, C.-Y., Liao, H.-Y., YOLOv4, arXiv, 2020.
- Howard, A. et al., MobileNets, arXiv, 2017.
- TensorFlow Lite Documentation, Google, 2024.
- Xie, X., A Review of Defect Detection Methods for Textile Fabrics, Journal of Textile Research.
- Goodfellow, I., Bengio, Y., Courville, A., Deep Learning, MIT Press.