Surface defect detection of hot rolled steel based on multi-scale feature fusion and attention mechanism residual block

Hongkai Zhang,Suqiang Li,Qiqi Miao,Ruidi Fang,Song Xue,Qianchuan Hu,Jie Hu,Sixian Chan

DOI: https://doi.org/10.1038/s41598-024-57990-3

IF: 4.6

2024-04-02

Scientific Reports

Abstract:To improve the precision of defect categorization and localization in images, this paper proposes an approach for detecting surface defects in hot-rolled steel strips. The approach uses an improved YOLOv5 network model to overcome the issues of inadequate feature extraction capacity and suboptimal feature integration when identifying surface defects on steel strips. The proposed method achieves higher detection accuracy and localization precision, making it more competitive and applicable in real production. Firstly, the multi-scale feature fusion (MSF) strategy is utilized to fuse shallow and deep features effectively and enrich detailed information relevant to target defects. Secondly, the CSPLayer Res2Attention block (CRA block) residual module is introduced to reduce the loss of defect information during hierarchical transmission, thereby enhancing the extraction of fine-grained features and improving the perception of details and global features. Finally, the experimental results indicate that the mAP on the NEU-DET and GC10-DET datasets approaches 78.5% and 67.3%, respectively, which is 4.9% and 2.1% higher than that of the baseline. Meanwhile, it has higher precision and more precise localization capabilities than other methods. Furthermore, it also achieves 59.2% mAP on the APDDD dataset, indicating its potential for growth in further domains.

multidisciplinary sciences

What problem does this paper attempt to address?

The paper aims to address the issues of accuracy and localization in the detection of surface defects on hot-rolled steel strips. Specifically, the study proposes an improved method to overcome the shortcomings of existing methods in feature extraction capability and feature fusion. The main problems the paper attempts to solve are as follows:

1. **Insufficient feature extraction capability**: Existing deep learning models have issues with insufficient feature extraction capability when identifying surface defects on hot-rolled steel strips.

2. **Poor feature fusion**: Although multi-scale feature fusion can improve detection performance, existing methods still suffer from information loss and inaccurate localization.

To address these issues, the study proposes the following methods:

1. **Multi-Scale Feature Fusion (MSF) Strategy**: By fusing shallow and deep features, it effectively enriches the detailed information related to the target defects, thereby improving detection accuracy.

2. **CSPLayer Res2Attention Block (CRA Block)**: By introducing an attention mechanism, it reduces information loss during feature transmission, enhances fine-grained feature extraction, and improves the perception of both detailed and global features.

Experimental results show that the improved YOLOv5 network model achieved a mean Average Precision (mAP) of 78.5% and 67.3% on the NEU-DET and GC10-DET datasets, respectively, which is an improvement of 4.9% and 2.1% over the baseline model. Additionally, it achieved a mAP of 59.2% on the APDDD dataset, demonstrating the potential of this method in other fields. In summary, the study significantly improves the accuracy and localization capability of surface defect detection on hot-rolled steel strips by improving feature fusion and introducing an attention mechanism.