Introduction
Textile manufacturing is a complex and intricate process, where even the slightest imperfections can have significant consequences on the final product quality. Textile defect detection plays a crucial role in ensuring the high standards demanded by today's consumers. Defects can lead to significant financial losses, as flawed products may need to be reworked or discarded entirely. In addition, accurate and timely identification of textile defects and streamline with production process is essential. By proactively detecting and addressing defects, manufacturers can improve their production efficiency, reduce waste, and deliver high-quality products that meet customer expectations.
The Complexity of Textile Defects
Textile products are inherently complex, with a wide range of potential defects that can occur during the manufacturing process. These defects can be caused by a variety of factors, such as machine malfunctions, human error, or the inherent variability of raw materials. Some common textile defects include holes, stains, foreign yarn, and knots, among others. Table 1 provides examples of nine common types of textile defects.
Examples of textile defects
Any kind of spot, dust, dirt, or oil spot. | A hole in the fabric. | Very small tightly gathered thread and fiber together. |
It appears in the fabric when warp yarns break during weaving. | Yarn having breakage in the fabric. | |
Any foreign yarn found that is inserted into the fabric from flying dust or another way. | One thread in fabrics that is thicker than others, counts as a thick yarn defect. | One thread is thinner than all others, counts as a thin yarn defect. |
Computer Vision Solution
Effectively identifying and addressing these diverse defects requires a comprehensive understanding of the production process and the ability to rapidly detect and classify a wide range of anomalies. To address the challenges of textile defect detection, we can leverage the power of computer vision and deep learning. One such solution is the YOLOv8 (You Only Look Once version 8) object detection algorithm, developed by Ultralytics.
YOLOv8 is an advanced object detection model that excels in quickly and accurately identifying and classifying objects in images, utilizing a single neural network for direct predictions without traditional methods like sliding windows. Its key benefits for textile defect detection include real-time performance, precise defect localization through bounding box predictions, versatile classification of various defect types, and ease of integration into existing manufacturing systems. By fine-tuning YOLOv8 with a dataset of textile images annotated with defects, the model can detect and classify defect types.
The YOLO model architecture consists of a backbone, neck, and head. The backbone is a pre-trained convolutional neural network that has been trained on a large, general dataset like COCO or ImageNet.
During the fine-tuning process, the weights and parameters of the pre-trained YOLO backbone are typically frozen and kept constant.
The YOLO neck and head components are the ones that are fine-tuned on the custom dataset.
- The input image is passed through a CNN backbone (e.g., pre-trained YOLOv8) to extract visual features.
The extracted features are then fed into the YOLO head, which is responsible for predicting the segmentation masks. - The YOLO head consists of convolutional and pooling layers that output a dense grid of segmentation predictions, including bounding boxes, objectless scores, and class labels.
- To adapt the YOLO model for segmentation, it is fine-tuned on a custom dataset that includes images and their corresponding segmentation masks.
- The fine-tuning process allows the model to learn the unique visual characteristics of the objects in the dataset, improving its segmentation performance.
- The output of the fine-tuned YOLO model is a detailed segmentation map, which can be used for various computer vision tasks, such as object counting, instance segmentation, or scene understanding.
Below showcases an experiment result of this solution, displaying the original textile image alongside the segmentation masks highlighting the identified defects of holes, stains, and foreign yarn.
Real defect showcases | |
| |
| |
By leveraging the power of computer vision, manufacturers can automate the defect detection process, improving accuracy, consistency, and speed compared to manual inspection methods. This can lead to significant cost savings, reduced waste, and enhanced product quality.
Conclusion
Textile defect detection is a critical component of the manufacturing process, ensuring the delivery of high-quality products that meet customer expectations. The complexity of textile defects, coupled with the need for rapid and accurate identification, makes this a challenging task. However, the integration of computer vision solutions, such as the YOLOv8 algorithm, offers a promising path forward. Computer vision models like YOLOv8, effectively trained to identify and classify a wide range of textile defects with real-time performance, accurate defect localization, and versatile defect classification capabilities, make them powerful tools for automating the textile inspection process. By integrating these advanced computer vision techniques into their manufacturing workflows, companies can enhance their efficiency, reduce costs, and maintain the integrity of their textile products.
Reference
R. Varghese and S. M., "YOLOv8: A Novel Object Detection Algorithm with Enhanced Performance and Robustness," 2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS), Chennai, India, 2024, pp. 1-6, doi: 10.1109/ADICS58448.2024.10533619.
Shakir, S., Topal, C. Unsupervised fabric defect detection with local spectra refinement (LSR). Neural Computing & Applications (2023).
if you want to know more defect detection of textile, please contact info@wisiotech.com for more information!