A lightweight transformer with linear self‐attention for defect recognition

Abstract Visual defect recognition techniques based on deep learning models are crucial for modern industrial quality inspection. The backbone, serving as the primary feature extraction component of the defect recognition model, has not been thoroughly exploited. High‐performance vision transformer...

Full description

Saved in:
Bibliographic Details
Main Authors: Yuwen Zhai, Xinyu Li, Liang Gao, Yiping Gao
Format: Article
Language:English
Published: Wiley 2024-09-01
Series:Electronics Letters
Subjects:
Online Access:https://doi.org/10.1049/ell2.13292
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850193007566913536
author Yuwen Zhai
Xinyu Li
Liang Gao
Yiping Gao
author_facet Yuwen Zhai
Xinyu Li
Liang Gao
Yiping Gao
author_sort Yuwen Zhai
collection DOAJ
description Abstract Visual defect recognition techniques based on deep learning models are crucial for modern industrial quality inspection. The backbone, serving as the primary feature extraction component of the defect recognition model, has not been thoroughly exploited. High‐performance vision transformer (ViT) is less adopted due to high computational complexity and limitations of computational resources and storage hardware in industrial scenarios. This paper presents LSA‐Former, a lightweight transformer architectural backbone that integrates the benefits of convolution and ViT. LSA‐Former proposes a novel self‐attention with linear computational complexity, enabling it to capture local and global semantic features with fewer parameters. LSA‐Former is pre‐trained on ImageNet‐1K and surpasses state‐of‐the‐art methods. LSA‐Former is employed as the backbone for various detectors, evaluated specifically on the PCB defect detection task. The proposed method reduces at least 18M parameters and exceeds the baseline by more than 2.2 mAP.
format Article
id doaj-art-2af147328379430db0eede2338fccd5f
institution OA Journals
issn 0013-5194
1350-911X
language English
publishDate 2024-09-01
publisher Wiley
record_format Article
series Electronics Letters
spelling doaj-art-2af147328379430db0eede2338fccd5f2025-08-20T02:14:22ZengWileyElectronics Letters0013-51941350-911X2024-09-016017n/an/a10.1049/ell2.13292A lightweight transformer with linear self‐attention for defect recognitionYuwen Zhai0Xinyu Li1Liang Gao2Yiping Gao3School of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan ChinaSchool of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan ChinaSchool of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan ChinaSchool of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan ChinaAbstract Visual defect recognition techniques based on deep learning models are crucial for modern industrial quality inspection. The backbone, serving as the primary feature extraction component of the defect recognition model, has not been thoroughly exploited. High‐performance vision transformer (ViT) is less adopted due to high computational complexity and limitations of computational resources and storage hardware in industrial scenarios. This paper presents LSA‐Former, a lightweight transformer architectural backbone that integrates the benefits of convolution and ViT. LSA‐Former proposes a novel self‐attention with linear computational complexity, enabling it to capture local and global semantic features with fewer parameters. LSA‐Former is pre‐trained on ImageNet‐1K and surpasses state‐of‐the‐art methods. LSA‐Former is employed as the backbone for various detectors, evaluated specifically on the PCB defect detection task. The proposed method reduces at least 18M parameters and exceeds the baseline by more than 2.2 mAP.https://doi.org/10.1049/ell2.13292automatic optical inspectionconvolutional neural netsimage recognitionobject detection
spellingShingle Yuwen Zhai
Xinyu Li
Liang Gao
Yiping Gao
A lightweight transformer with linear self‐attention for defect recognition
Electronics Letters
automatic optical inspection
convolutional neural nets
image recognition
object detection
title A lightweight transformer with linear self‐attention for defect recognition
title_full A lightweight transformer with linear self‐attention for defect recognition
title_fullStr A lightweight transformer with linear self‐attention for defect recognition
title_full_unstemmed A lightweight transformer with linear self‐attention for defect recognition
title_short A lightweight transformer with linear self‐attention for defect recognition
title_sort lightweight transformer with linear self attention for defect recognition
topic automatic optical inspection
convolutional neural nets
image recognition
object detection
url https://doi.org/10.1049/ell2.13292
work_keys_str_mv AT yuwenzhai alightweighttransformerwithlinearselfattentionfordefectrecognition
AT xinyuli alightweighttransformerwithlinearselfattentionfordefectrecognition
AT lianggao alightweighttransformerwithlinearselfattentionfordefectrecognition
AT yipinggao alightweighttransformerwithlinearselfattentionfordefectrecognition
AT yuwenzhai lightweighttransformerwithlinearselfattentionfordefectrecognition
AT xinyuli lightweighttransformerwithlinearselfattentionfordefectrecognition
AT lianggao lightweighttransformerwithlinearselfattentionfordefectrecognition
AT yipinggao lightweighttransformerwithlinearselfattentionfordefectrecognition