Download PDFOpen PDF in browser

Deep Learning Architecture TLU-Net for Detecting Steel Surface Defects

EasyChair Preprint 14121

11 pagesDate: July 25, 2024

Abstract

The detection of surface defects in steel manufacturing is a critical quality assurance process that ensures the integrity and performance of steel products. Traditional methods often rely on manual inspection or basic image processing techniques, which can be time-consuming and prone to human error. In this study, we propose TLU-Net, a novel deep learning architecture designed specifically for detecting steel surface defects with high accuracy and efficiency. TLU-Net leverages the strengths of Convolutional Neural Networks (CNNs) and Transformer layers to capture both local and global features of steel surface images.

 

Our model incorporates a tailored feature extraction process, combining traditional CNN layers for localized feature detection and Transformer layers to model long-range dependencies and contextual information. This hybrid approach allows TLU-Net to achieve superior performance in identifying various types of defects, such as scratches, dents, and inclusions, which are challenging to detect with conventional methods.

 

We validate the effectiveness of TLU-Net using a comprehensive dataset of steel surface images, demonstrating its ability to outperform existing state-of-the-art models in terms of accuracy, precision, and recall. The results indicate that TLU-Net not only improves defect detection rates but also reduces false positives, thus enhancing overall inspection reliability.

 

In conclusion, TLU-Net represents a significant advancement in the application of deep learning for industrial quality control, offering a robust and scalable solution for real-time steel surface defect detection. Future work will focus on optimizing the model for deployment in production environments and exploring its applicability to other materials and defect types.

Keyphrases: Convolutional Neural Networks, Steel surface defect detection, TLU-Net, Transformer Networks, deep learning

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:14121,
  author    = {Oluwaseun Abiade},
  title     = {Deep Learning Architecture TLU-Net for Detecting Steel Surface Defects},
  howpublished = {EasyChair Preprint 14121},
  year      = {EasyChair, 2024}}
Download PDFOpen PDF in browser