A Compact Convolutional Neural Network for Surface Defect Inspection
The advent of convolutional neural networks (CNNs) has accelerated the progress of computer vision from many aspects. However, the majority of the existing CNNs heavily rely on expensive GPUs (graphics processing units). to support large computations. Therefore, CNNs have not been widely used to ins...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2020-04-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/20/7/1974 |
| _version_ | 1797571724764512256 |
|---|---|
| author | Yibin Huang Congying Qiu Xiaonan Wang Shijun Wang Kui Yuan |
| author_facet | Yibin Huang Congying Qiu Xiaonan Wang Shijun Wang Kui Yuan |
| author_sort | Yibin Huang |
| collection | DOAJ |
| description | The advent of convolutional neural networks (CNNs) has accelerated the progress of computer vision from many aspects. However, the majority of the existing CNNs heavily rely on expensive GPUs (graphics processing units). to support large computations. Therefore, CNNs have not been widely used to inspect surface defects in the manufacturing field yet. In this paper, we develop a compact CNN-based model that not only achieves high performance on tiny defect inspection but can be run on low-frequency CPUs (central processing units). Our model consists of a light-weight (LW) bottleneck and a decoder. By a pyramid of lightweight kernels, the LW bottleneck provides rich features with less computational cost. The decoder is also built in a lightweight way, which consists of an atrous spatial pyramid pooling (ASPP) and depthwise separable convolution layers. These lightweight designs reduce the redundant weights and computation greatly. We train our models on groups of surface datasets. The model can successfully classify/segment surface defects with an Intel i3-4010U CPU within 30 ms. Our model obtains similar accuracy with MobileNetV2 while only has less than its 1/3 FLOPs (floating-point operations per second) and 1/8 weights. Our experiments indicate CNNs can be compact and hardware-friendly for future applications in the automated surface inspection (ASI). |
| first_indexed | 2024-03-10T20:44:34Z |
| format | Article |
| id | doaj.art-1610ac042fd24fc1bacab64a5c1128cc |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-10T20:44:34Z |
| publishDate | 2020-04-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-1610ac042fd24fc1bacab64a5c1128cc2023-11-19T20:24:03ZengMDPI AGSensors1424-82202020-04-01207197410.3390/s20071974A Compact Convolutional Neural Network for Surface Defect InspectionYibin Huang0Congying Qiu1Xiaonan Wang2Shijun Wang3Kui Yuan4Institute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing 100190, ChinaCivil Engineering & Engineering Mechanics Department, Columbia University, New York, NY 10024, USAInstitute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing 100190, ChinaInstitute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing 100190, ChinaInstitute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing 100190, ChinaThe advent of convolutional neural networks (CNNs) has accelerated the progress of computer vision from many aspects. However, the majority of the existing CNNs heavily rely on expensive GPUs (graphics processing units). to support large computations. Therefore, CNNs have not been widely used to inspect surface defects in the manufacturing field yet. In this paper, we develop a compact CNN-based model that not only achieves high performance on tiny defect inspection but can be run on low-frequency CPUs (central processing units). Our model consists of a light-weight (LW) bottleneck and a decoder. By a pyramid of lightweight kernels, the LW bottleneck provides rich features with less computational cost. The decoder is also built in a lightweight way, which consists of an atrous spatial pyramid pooling (ASPP) and depthwise separable convolution layers. These lightweight designs reduce the redundant weights and computation greatly. We train our models on groups of surface datasets. The model can successfully classify/segment surface defects with an Intel i3-4010U CPU within 30 ms. Our model obtains similar accuracy with MobileNetV2 while only has less than its 1/3 FLOPs (floating-point operations per second) and 1/8 weights. Our experiments indicate CNNs can be compact and hardware-friendly for future applications in the automated surface inspection (ASI).https://www.mdpi.com/1424-8220/20/7/1974surface defect inspectionconvolutional neural networkmachine vision |
| spellingShingle | Yibin Huang Congying Qiu Xiaonan Wang Shijun Wang Kui Yuan A Compact Convolutional Neural Network for Surface Defect Inspection Sensors surface defect inspection convolutional neural network machine vision |
| title | A Compact Convolutional Neural Network for Surface Defect Inspection |
| title_full | A Compact Convolutional Neural Network for Surface Defect Inspection |
| title_fullStr | A Compact Convolutional Neural Network for Surface Defect Inspection |
| title_full_unstemmed | A Compact Convolutional Neural Network for Surface Defect Inspection |
| title_short | A Compact Convolutional Neural Network for Surface Defect Inspection |
| title_sort | compact convolutional neural network for surface defect inspection |
| topic | surface defect inspection convolutional neural network machine vision |
| url | https://www.mdpi.com/1424-8220/20/7/1974 |
| work_keys_str_mv | AT yibinhuang acompactconvolutionalneuralnetworkforsurfacedefectinspection AT congyingqiu acompactconvolutionalneuralnetworkforsurfacedefectinspection AT xiaonanwang acompactconvolutionalneuralnetworkforsurfacedefectinspection AT shijunwang acompactconvolutionalneuralnetworkforsurfacedefectinspection AT kuiyuan acompactconvolutionalneuralnetworkforsurfacedefectinspection AT yibinhuang compactconvolutionalneuralnetworkforsurfacedefectinspection AT congyingqiu compactconvolutionalneuralnetworkforsurfacedefectinspection AT xiaonanwang compactconvolutionalneuralnetworkforsurfacedefectinspection AT shijunwang compactconvolutionalneuralnetworkforsurfacedefectinspection AT kuiyuan compactconvolutionalneuralnetworkforsurfacedefectinspection |