Improved YOLOv7-based steel surface defect detection algorithm

Yinghong Xie ¹, Biao Yin ¹, Xiaowei Han ²

⁰School of Information Engineering, Shenyang University, Shenyang 110003, China.

Mathematical Biosciences and Engineering : Mbe +

|

February 2, 2024

View abstract on PubMed

Summary

This study enhances the YOLOv7 algorithm for steel surface defect detection, improving small target identification. The novel approach boosts mean Average Precision (mAP) by up to 6% on benchmark datasets.

Area Of Science

Materials Science
Computer Vision
Artificial Intelligence

Background

Steel surface defect detection is crucial for quality control.
Existing YOLOv7 algorithm has limitations in detecting small targets and generalizing.
Accurate detection of surface defects is essential for industrial applications.

Purpose Of The Study

To improve the detection ability and model generalization of the YOLOv7 algorithm for steel surface defects.
To enhance the performance of small defect detection using an improved YOLOv7 model.
To address limitations in current algorithms for identifying subtle flaws on steel surfaces.

Main Methods

Designed a Transformer-InceptionDWConvolution (TI) module to enhance small object detection.
Introduced Spatial Pyramid Pooling Fast Cross-Stage Partial Channel (SPPFCSPC) for improved training.
Incorporated a Global Attention Mechanism (GAM) to focus on relevant defect features.
Utilized the Mish activation function for better generalization and feature extraction.
Developed a Minimum Partial Distance Intersection over Union (MPDIoU) loss function for accurate localization.

Main Results

The improved YOLOv7 model achieved a 6% increase in mean Average Precision (mAP) on the NEU-DET dataset.
A 2.6% mAP improvement was observed on the VOC2012 dataset.
The proposed algorithm demonstrated enhanced performance in detecting small steel surface defects.

Conclusions

The enhanced YOLOv7 algorithm effectively improves small defect detection on steel surfaces.
The integration of TI module, SPPFCSPC, GAM, Mish function, and MPDIoU contributes to superior performance.
The developed model shows significant potential for industrial application in steel quality inspection.

Keywords:

SPPFCSPC YOLOv7 attention mechanism defect detection transformer