Related Experiment Video
Updated: Sep 5, 2025
05:40
Fabricating Cotton Analytical Devices
Published on: August 30, 2016
6.8K
Online Detection of Fabric Defects Based on Improved CenterNet with Deformable Convolution
Jun Xiang1, Ruru Pan1, Weidong Gao1
1School of Textile Science & Engineering, Jiangnan University, No. 1800, Lihu Avenue, Wuxi 214122, China.
Sensors (Basel, Switzerland)
|July 9, 2022
Summary
This study introduces an automated fabric defect detection system using machine vision. The developed framework enhances efficiency and accuracy, significantly improving upon traditional manual methods for textile quality control.
Area of Science:
- Textile Manufacturing
- Computer Vision
- Machine Learning
Background:
- Traditional manual fabric defect detection is inefficient, time-consuming, and labor-intensive.
- Automated solutions are needed to improve quality control in textile manufacturing.
Purpose of the Study:
- To propose an automatic detection framework for fabric defects.
- To enhance the efficiency and accuracy of fabric defect detection.
Main Methods:
- Developed a hardware system with cameras and lighting for high-quality image acquisition (up to 65 m/min).
- Treated fabric defect detection as an object detection task using a modified CenterNet algorithm.
- Incorporated deformable convolution for diverse defect shapes and i-FPN for varying defect sizes.
Main Results:
- The improved CenterNet achieved efficient fabric defect detection.
- Ablation studies confirmed the effectiveness of the proposed improvements.
- The system demonstrated a balance of accuracy and speed, reaching a maximum detection speed of 37.3 m/min.
Conclusions:
- The proposed automatic framework offers a superior solution for fabric defect detection.
- The system meets real-time requirements for industrial applications.
- The method provides a satisfactory balance between detection accuracy and speed.
