Surface defect detection of steel strips based on improved YOLOv4
Graphical abstract
Introduction
Strip steel is an indispensable material for national construction and relates to the safety of urban infrastructure. During the production and processing of steel strips, owing to the influence of modern production technology and external factors, their surfaces may have various defects, such as plaques, inclusions, and scratches. These defects compromise their integrity and functionality, leading to construction accidents and endangering personal safety. Therefore, a speedy and accurate detection of surface defects in steel strips is critical. The traditional defect detection method involves random manual visual inspections. This method has the disadvantages of slow speed, low efficiency, high cost, poor detection environment, and potential safety hazards. Manual sampling is overly subjective, which leads to the low credibility of the results. Therefore, surface defect detection based on computer vision and deep learning is increasingly being used in modern production to solve these problems.
This paper is organised as follows. Section 2 presents the related work on computer vision and deep learning in the field of steel defect detection. In Section 3, information regarding the original you-only-look-once YOLO version 4 (YOLOv4) algorithm and the loss function is described. Detailed algorithm improvements are provided in Section 4. Section 5 presents the experimental results, and the conclusions are presented in Section 6.
Our contributions are as follows:
- •
We propose a modified YOLOv4 algorithm for defect detection of industrial strip steel.
- •
We design a convolutional block attention module (CBAM) for the backbone network and a receptive field block (RFB)-like structure replaces an enhance path aggregation network (PANet) to empower the information acquisition and feature extraction capability of the network.
- •
The experimental results verify the superiority of the proposed algorithm.
Section snippets
Related work
As traditional defect detection methods have numerous drawbacks, computer vision-based methods are gradually gaining traction. Ref. [1] proposed a global adaptive percentile threshold method based on a gradient image. This method selectively segments the defect area and preserves defect features regardless of its size. Ref. [2] proposed a new algorithm based on wavelet anisotropic diffusion filtering. This method is effective at filtering out unwanted textural backgrounds in cold-rolled strips
YOLOV4 network structure
YOLOv4 is an updated and reinforced version of the YOLOv3 target-detection algorithm. The network structure is displayed in Fig. 1. First, the backbone network applies the cross-stage partial network (CSPnet) [16] to Darknet53, a 53-layer-deep CNN [17], and adds a CSPnet [16] structure to each group of resblock_body, continuously carrying out downsampling to obtain higher-level semantic information, which not only reduces the amount of computation but also enhances the gradient information.
A model's robustness improvement of YOLOv4
The improved YOLOv4 algorithm network architecture proposed in this study is marked in red boxes in Fig. 2, and the details are given in the subsequent two sections.
Experiment and result analysis
In this section, we describe the dataset, evaluation indicators, comparison objects, and methods, and finally dissect the experimental results to confirm the practicality of the improved model.
Conclusions
In response to the multiple difficulties faced in the task of steel strip defect detection, this study proposed a detection algorithm based on an improved YOLOv4. The value of this algorithm lies in the following:
- •
Its YOLOv4 backbone network is embedded with CBAM.
- •
Its YOLOv4 feature extraction network SPP module is improved to the customised RFB structure to deepen the network.
- •
Among the four defect types in the NEU-DET dataset, our model achieved a 3.87% improvement in mAP over the original
CRediT authorship contribution statement
Mengjiao Li: Conceptualization, Methodology, Software, Investigation, Formal analysis, Writing – original draft. Hao Wang: Data curation, Writing – original draft. Zhibo Wan: Funding acquisition, Supervision, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (29)
- et al.
Defect detection of nickel-plated punched steel strip based on improved least square method[J]
Opt. Int. J. Light Electron Opt.
(2020) - et al.
Development of a YOLO-V3-based model for detecting defects on steel strip surface[J]
Measurement
(2021) - et al.
Defect detection of steel surfaces with global adaptive percentile thresholding of gradient image[J]
J. Inst. Eng.
(2017) - et al.
An effective method of weld defect detection and classification based on machine vision[J]
IEEE Trans. Ind. Inf.
(2019) - et al.
SSD: Single Shot MultiBox Detector[J]
(2016) - et al.
Focal loss for dense object detection[J]
IEEE Trans. Pattern Anal. Mach. Intell.
(2017) - et al.
You Only Look Once: Unified, Real-Time Object Detection[J]
(2016) - et al.
YOLO9000: Better, Faster, Stronger
(2017) - Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767,...
- Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of object detection[J]. arXiv preprint...
Faster R-CNN: towards real-time object detection with region proposal networks[J]
IEEE Trans. Pattern Anal. Mach. Intell.
Mask R-CNN [J]
IEEE Trans. Pattern Anal. Mach. Intell.
Cited by (46)
A detection network for small defects of steel surface based on YOLOv7
2024, Digital Signal Processing: A Review JournalDepth feature fusion based surface defect region identification method for steel plate manufacturing
2024, Computers and Electrical EngineeringLDD-Net: Lightweight printed circuit board defect detection network fusing multi-scale features
2024, Engineering Applications of Artificial IntelligenceRevolutionising the Sustainability of Steel Manufacturing Using Computer Vision
2024, Procedia Computer ScienceHypertuned-YOLO for interpretable distribution power grid fault location based on EigenCAM
2024, Ain Shams Engineering JournalAn automated defect detection method for optimizing industrial quality inspection
2024, Engineering Applications of Artificial Intelligence