Title: YOLO-C: a new surface defect detection model of steel plate based on YOLO optimisation

Authors: Lingyun Zhu; Chongliu Jia; Juan Zhang; Lu Wang; YueMou Jian

Addresses: Liangjiang International College, Chongqing University of Technology, Chongqing, Banan District, China ' Liangjiang International College, Chongqing University of Technology, Chongqing, Banan District, China ' Liangjiang International College, Chongqing University of Technology, Chongqing, Banan District, China ' Liangjiang International College, Chongqing University of Technology, Chongqing, Banan District, China ' Liangjiang International College, Chongqing University of Technology, Chongqing, Banan District, China

Abstract: It is necessary to detect steel defects in the manufacturing industry. Proper quality control can reduce problems arising from steel defects implemented by traditional digital image processing or labour sight in the past. However, it cannot meet the requirements of accuracy and real-time for intelligent manufacturing lines. With the extensional combination of IT and manufacture, related scholars used deep learning to detect the steel defects, despite the limitations of data set and application scenarios. We propose a detection model called YOLO-C, an optimisation model that combines Convolutional Neural Networks (CNN) with attention mechanisms. The proposed model can further improve the detection accuracy, which adds several attention mechanism modules in the convolutional neural network structure, and the attention mechanism module is used to enhance the performance of feature extraction. The experimental results show that the YOLO-C model proposed is superior to other models, and the precision reaches 86%.

Keywords: YOLO; attention mechanism; defect detection; convolutional neural network.

DOI: 10.1504/IJWMC.2022.125535

International Journal of Wireless and Mobile Computing, 2022 Vol.23 No.1, pp.46 - 56

Received: 29 Sep 2021
Received in revised form: 16 Mar 2022
Accepted: 17 Mar 2022

Published online: 13 Sep 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article