Metal surface defect detection based on improved YOLOv5

Abstract

During the production of metal material, various complex defects may come into being on the surface, together with large amount of background texture information, causing false or missing detection in the process of small defect detection. To resolve those problems, this paper introduces a new model which combines the advantages of CSPlayer module and Global Attention Enhancement Mechanism based on the YOLOv5s model. First of all, we replace C3 module with CSPlayer module to augment the neural network model, so as to improve its flexibility and adaptability. Then, we introduce the Global Attention Mechanism (GAM) and build the generalized additive model. In the meanwhile, the attention weights of all dimensions are weighted and averaged as output to promote the detection speed and accuracy. The results of the experiment in which the GC10-DET augmented dataset is involved, show that the improved algorithm model performs better than YOLOv5s in precision, mAP@0.5 and mAP@0.5: 0.95 by 5.3%, 1.4% and 1.7% respectively, and it also has a higher reasoning speed.

Evaluating Generalization, Bias, and Fairness in Deep Learning for Metal Surface Defect Detection: A Comparative Study

“…Manas Mehta's model based on YOLOv5 [10], incorporating ECA-Net and BiFPN, showcased enhanced real-time steel surface defect detection. Lastly, Chuande Zhou et al [8] improved YOLOv5s with CSPLayer and GAMAttention and effectively detected small metal surface defects, marking a leap in detection sensitivity.…”

“…Studies, including those by Drury and Fox (1975), show human error contributes substantially to inspection inaccuracies, ranging between 20% and 30% [2]. The evolution of deep learning models for defect detection has marked a transformative shift in this domain, with recent deep learning models [3][4][5][6][7][8] increasingly used for industrial inspection to automate the identification of specific defect classes, provide valuable insights into production process issues, and allow for timely interventions that can prevent costly machine or process shutdowns.…”

“…This is likely to lead to overconfident predictions, biased and unfair performance and lack of generalizability, even when otherwise high-performing optimized models are deployed. For example, 30% of the widely used GC10-DET dataset [3][4][5]8], also examined in this work, comprises multi-class images with varying co-occurrence defects. This disparity raises a crucial question about the adequacy of current augmentation methods used in recent papers and the potential need for more tailored approaches to enhance dataset stratification and thereby improve model performance.…”

“…As the focus of this work is not the development of a new defect detection model, it uses a variant of the commonly deployed YOLOv5 model. A large number of recent works use YOLOv5 as their basis [3,[5][6][7][8][9][10], and more recently YOLOv7 [4]. Although YOLOv7 outperforms YOLOv5 in terms of performance, this is only possible on high-performance GPUs, which may not be easily accessible to industry.…”

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

“…This helps to better capture both detail and semantic information within the images, thereby enhancing model performance. Though most existing residual blocks have achieved success in deep learning 30 – 33 , optimization is needed for practical industrial defect detection applications, especially for small-scale object detection and fine-grained feature extraction. Introducing residual blocks can address some of the information loss issues, but an emphasis on detailed information is still lacking, with insufficient fine-grained feature extraction capabilities.…”