Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Surface Defect Detection Model Combining Multi-scale Feature Fusion and Attention Mechanism
1School of Artificial Intelligence, Henan University, China.
2School of Automation, Guangdong University of Petrochemical Technology, China.
3School of Computer and Information Engineering, Henan University.
ABSTRACT
Surface defect detection technology is considered one of the key supporting technologies for quality inspection and equipment maintenance. In this paper, a model for surface defect detection is proposed, which combines multiscale feature fusion and attention mechanism. The objective is to tackle the issue of low accuracy in simultaneously detecting multiple types of surface defects using existing methods. A small object detection layer is added to the YOLOv8 model in order to raise the detection accuracy for small objects. Furthermore, attention modules are incorporated into the backbone network to extract feature information, expand the receptive field, and inhibit the entry of interfering information into the feature fusion module. Subsequently, an upsampling operation is conducted in the feature fusion module based on the feature content to prioritize local information. Lastly, the model utilizes ECIoU as the bounding box regression loss function to attain improved convergence speed and localization results. The experiments indicates that on the NEU dataset, the overall detection accuracy of the improved model in this study increased by 4.0% compared to YOLOv8. It also outperforms other detection models, demonstrating its superior performance in terms of detection effectiveness. Keywords: Surface defect detection, YOLOv8, Multi-scale feature fusion, Attention mechanism.
Download PDF
1School of Artificial Intelligence, Henan University, China.
2School of Automation, Guangdong University of Petrochemical Technology, China.
3School of Computer and Information Engineering, Henan University.
ABSTRACT
Surface defect detection technology is considered one of the key supporting technologies for quality inspection and equipment maintenance. In this paper, a model for surface defect detection is proposed, which combines multiscale feature fusion and attention mechanism. The objective is to tackle the issue of low accuracy in simultaneously detecting multiple types of surface defects using existing methods. A small object detection layer is added to the YOLOv8 model in order to raise the detection accuracy for small objects. Furthermore, attention modules are incorporated into the backbone network to extract feature information, expand the receptive field, and inhibit the entry of interfering information into the feature fusion module. Subsequently, an upsampling operation is conducted in the feature fusion module based on the feature content to prioritize local information. Lastly, the model utilizes ECIoU as the bounding box regression loss function to attain improved convergence speed and localization results. The experiments indicates that on the NEU dataset, the overall detection accuracy of the improved model in this study increased by 4.0% compared to YOLOv8. It also outperforms other detection models, demonstrating its superior performance in terms of detection effectiveness. Keywords: Surface defect detection, YOLOv8, Multi-scale feature fusion, Attention mechanism.
Download PDF