Next Article in Journal
Use of GRP Pipe Waste Powder as a Filler Replacement in Hot-Mix Asphalt
Previous Article in Journal
Effect of Particle Specific Surface Area on the Rheology of Non-Brownian Silica Suspensions
Open AccessArticle

A Light-Weight Deep-Learning Model with Multi-Scale Features for Steel Surface Defect Classification

1
Bremen Institute for Mechanical Engineering-bime, University of Bremen, 28359 Bremen, Germany
2
Institute of Computer Science, University of Goettingen, 37077 Goettingen, Germany
3
Faculty of Mathematics, University of Goettingen, 37077 Goettingen, Germany
4
Key Laboratory of Materials Processing Engineering, School of Material Science and Engineering, Xi’an Shiyou University, Xi’an 710065, China
*
Authors to whom correspondence should be addressed.
Materials 2020, 13(20), 4629; https://doi.org/10.3390/ma13204629
Received: 7 September 2020 / Revised: 9 October 2020 / Accepted: 13 October 2020 / Published: 16 October 2020
(This article belongs to the Section Manufacturing Processes and Systems)
Automatic inspection of surface defects is crucial in industries for real-time applications. Nowadays, computer vision-based approaches have been successfully employed. However, most of the existing works need a large number of training samples to achieve satisfactory classification results, while collecting massive training datasets is labor-intensive and financially costly. Moreover, most of them obtain high accuracy at the expense of high latency, and are thus not suitable for real-time applications. In this work, a novel Concurrent Convolutional Neural Network (ConCNN) with different image scales is proposed, which is light-weighted and easy to deploy for real-time defect classification applications. To evaluate the performance of ConCNN, the NEU-CLS dataset is used in our experiments. Simulation results demonstrate that ConCNN performs better than other state-of-the-art approaches considering accuracy and latency for steel surface defect classification. Specifically, ConCNN achieves as high as 98.89% classification accuracy with only around 5.58 ms latency over low training cost. View Full-Text
Keywords: surface defect classification; multiple image scales; convolutional neural networks; classification accuracy; latency surface defect classification; multiple image scales; convolutional neural networks; classification accuracy; latency
Show Figures

Figure 1

MDPI and ACS Style

Liu, Y.; Yuan, Y.; Balta, C.; Liu, J. A Light-Weight Deep-Learning Model with Multi-Scale Features for Steel Surface Defect Classification. Materials 2020, 13, 4629.

Show more citation formats Show less citations formats
Note that from the first issue of 2016, MDPI journals use article numbers instead of page numbers. See further details here.

Article Access Map by Country/Region

1
Search more from Scilit
 
Search
Back to TopTop