Research on Textile Defect Detection Based on Improved Cascade R-CNN

• Published in: 2021 International Conference on Artificial Intelligence and Electromechanical Automation (AIEA) pp. 43 - 46
• Main Authors: Li, Yu, Wang, ZiWei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2021
• Subjects: Automation; Cascade R-CNN; component; Data models; Deep learning; Defect detection; Fabrics; FPN; Manuals; Production; Real-time systems
• DOI: 10.1109/AIEA53260.2021.00017

Textile fabric defects are an important factor affecting the quality of textile fabrics. Traditional manual-based textile defect detection methods are slow and inefficient, and cannot meet the production needs of the modern textile industry. Aiming at the process and technical difficulties of textile defect detection, this paper analyzes different defect detection algorithms based on deep learning through theoretical research and experimental comparison, and finally designs and implements a defect detection algorithm based on improved Cascade R-CNN. Through the introduction of FPN on the basic structure of Cascade R-CNN, the defect detection experiment based on improved Cascade R-CNN is designed and realized. This paper discusses the defect detection model design and network structure based on the improved Cascade R-CNN, and uses the multi-scale detection characteristics of FPN to detect textile defects. The experimental results show that, compared with the defect detection based on Cascade R-CNN, the accuracy of the improved Cascade R-CNN model defect detection is increased by 4.09% to 95.43%, and the improved model has better detection of multi-scale defects effect.