Cross-Modality Self-Attention and Fusion-Based Neural Network for Lower Limb Locomotion Mode Recognition

Although there are many wearable sensors that make the acquisition of multi-modality data easier, effective feature extraction and fusion of the data is still challenging for lower limb locomotion mode recognition. In this article, a novel neural network is proposed for accurate prediction of five c...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE transactions on automation science and engineering Ročník 22; s. 5411 - 5424
Hlavní autori: Zhao, Changchen, Liu, Kai, Zheng, Hao, Song, Wenbo, Pei, Zhongcai, Chen, Weihai
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: IEEE 2025
Predmet:
ISSN:1545-5955, 1558-3783
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Abstract Although there are many wearable sensors that make the acquisition of multi-modality data easier, effective feature extraction and fusion of the data is still challenging for lower limb locomotion mode recognition. In this article, a novel neural network is proposed for accurate prediction of five common lower limb locomotion modes including level walking, ramp ascent, ramp descent, stair ascent, and stair descent. First, the encoder-decoder structure is employed to enrich the channel diversity for the separation of the useful patterns from combined patterns. Second, a self-attention based cross-modality interaction module is proposed, which enables bilateral information flow between two encoding paths to fully exploit the interdependencies and to find complementary information between modalities. Third, a multi-modality fusion module is designed where the complementary features are fused by a channel-wise weighted summation whose coefficients are learned end-to-end. A benchmark dataset is collected from 10 health subjects containing EMG and IMU signals and five locomotion modes. Extensive experiments are conducted on one publicly available dataset ENABL3S and one self-collected dataset. The results show that the proposed method outperforms the compared methods with higher classification accuracy. The proposed method achieves a classification accuracy of 98.25% on ENABL3S dataset and 95.51% on the self-collected dataset. Note to Practitioners-This article aims to solve the real challenges encountered when intelligent recognition algorithms are applied in wearable robots: how to effectively and efficiently fuse the multi-modality data for better decision-making. First, most existing methods directly concatenate the multi-modality data, which increases the data dimensionality and brings computational burden. Second, existing recognition neural networks continuously compress the feature size such that the discriminative patterns are submerged in the noise and thus difficult to be identified. This research decomposes the mixed input signals on the channel dimension such that the useful patterns can be separated. Moreover, this research employs self-attention mechanism to associate correlations between two modalities and use this correlation as a new feature for subsequent representation learning, generating new, compact, and complementary features for classification. We demonstrate that the proposed network achieves 98.25% accuracy and 3.5 ms prediction time. We anticipate that the proposed network could be a general scientific and practical methodology of multi-modality signal fusion and feature learning for intelligent systems.
AbstractList Although there are many wearable sensors that make the acquisition of multi-modality data easier, effective feature extraction and fusion of the data is still challenging for lower limb locomotion mode recognition. In this article, a novel neural network is proposed for accurate prediction of five common lower limb locomotion modes including level walking, ramp ascent, ramp descent, stair ascent, and stair descent. First, the encoder-decoder structure is employed to enrich the channel diversity for the separation of the useful patterns from combined patterns. Second, a self-attention based cross-modality interaction module is proposed, which enables bilateral information flow between two encoding paths to fully exploit the interdependencies and to find complementary information between modalities. Third, a multi-modality fusion module is designed where the complementary features are fused by a channel-wise weighted summation whose coefficients are learned end-to-end. A benchmark dataset is collected from 10 health subjects containing EMG and IMU signals and five locomotion modes. Extensive experiments are conducted on one publicly available dataset ENABL3S and one self-collected dataset. The results show that the proposed method outperforms the compared methods with higher classification accuracy. The proposed method achieves a classification accuracy of 98.25% on ENABL3S dataset and 95.51% on the self-collected dataset. Note to Practitioners-This article aims to solve the real challenges encountered when intelligent recognition algorithms are applied in wearable robots: how to effectively and efficiently fuse the multi-modality data for better decision-making. First, most existing methods directly concatenate the multi-modality data, which increases the data dimensionality and brings computational burden. Second, existing recognition neural networks continuously compress the feature size such that the discriminative patterns are submerged in the noise and thus difficult to be identified. This research decomposes the mixed input signals on the channel dimension such that the useful patterns can be separated. Moreover, this research employs self-attention mechanism to associate correlations between two modalities and use this correlation as a new feature for subsequent representation learning, generating new, compact, and complementary features for classification. We demonstrate that the proposed network achieves 98.25% accuracy and 3.5 ms prediction time. We anticipate that the proposed network could be a general scientific and practical methodology of multi-modality signal fusion and feature learning for intelligent systems.
Author Song, Wenbo
Zheng, Hao
Pei, Zhongcai
Liu, Kai
Zhao, Changchen
Chen, Weihai
Author_xml – sequence: 1
  givenname: Changchen
  orcidid: 0000-0002-0546-6016
  surname: Zhao
  fullname: Zhao, Changchen
  organization: School of Computer Science, Hangzhou Dianzi University, Hangzhou, China
– sequence: 2
  givenname: Kai
  surname: Liu
  fullname: Liu, Kai
  organization: College of Information Engineering, Zhejiang University of Technology, Hangzhou, China
– sequence: 3
  givenname: Hao
  surname: Zheng
  fullname: Zheng, Hao
  organization: Hangzhou Innovation Institute, Beihang University, Hangzhou, China
– sequence: 4
  givenname: Wenbo
  surname: Song
  fullname: Song, Wenbo
  organization: College of Physical Education, Jilin Normal University, Siping, China
– sequence: 5
  givenname: Zhongcai
  orcidid: 0000-0001-7748-8591
  surname: Pei
  fullname: Pei, Zhongcai
  organization: Hangzhou Innovation Institute, Beihang University, Hangzhou, China
– sequence: 6
  givenname: Weihai
  orcidid: 0000-0001-7912-4505
  surname: Chen
  fullname: Chen, Weihai
  email: whchen@buaa.edu.cn
  organization: School of Electrical Engineering and Automation, Anhui University, Hefei, China
BookMark eNp9kE1OwzAQhS1UJNrCAZBY5AIu_onjeFmqtiAVkGhZR44zAUMaI8dV1dvj0C4QCzYzT6P5nmbeCA1a1wJC15RMKCXqdjNdzyeMsHTCU0aZzM7QkAqRYy5zPuh1KrBQQlygUdd9kLiZKzJE7zPvug4_uko3NhySNTQ1noYAbbCuTXRbJYtdFyW-0x1UyRPsvG5iC3vnP5Pa-WTl9hCr3ZZRGrd1P2R0hOQFjHtrbT-4ROe1bjq4OvUxel3MN7N7vHpePsymK2xYlgUsyloppoDpsqw4JVABrQijHPKqlGXKFRimcqmoFpk0lOo8FVLWpGY6Ghg-RvLoa_rHPNSFsUH3FwSvbVNQUvSBFX1gRR9YcQoskvQP-eXtVvvDv8zNkbEA8GtfxKtEyr8BmJh54A
CODEN ITASC7
CitedBy_id crossref_primary_10_3390_technologies13090399
crossref_primary_10_1109_LRA_2025_3597482
crossref_primary_10_1109_TII_2025_3545108
crossref_primary_10_1080_00051144_2025_2513734
Cites_doi 10.1109/EMBC.2017.8037600
10.1109/ICASSP40776.2020.9053270
10.3390/mi13081205
10.1016/j.bspc.2022.103693
10.1007/s12541-017-0081-9
10.1109/SAS48726.2020.9220037
10.1016/j.inffus.2021.11.006
10.1109/TBME.2011.2161671
10.1109/JBHI.2023.3252091
10.3389/frobt.2018.00127
10.48550/ARXIV.1706.03762
10.3389/frobt.2018.00078
10.1109/LRA.2024.3379800
10.1109/LRA.2020.3007455
10.1109/CVPRW.2018.00177
10.1109/THMS.2021.3107256
10.1109/CVPR.2019.01233
10.1038/s41598-023-28390-w
10.1109/IROS45743.2020.9341649
10.3390/s19194242
10.1109/JBHI.2023.3238406
10.1109/tnsre.2021.3087135
10.1109/tbme.2009.2034734
10.3390/s19204447
10.1109/TNSRE.2020.2987155
10.1109/JSEN.2022.3167686
10.1109/jas.2017.7510619
10.1109/tase.2023.3276856
10.1016/j.patcog.2021.108332
10.1109/TASE.2023.3250240
10.1109/JSEN.2022.3146446
10.1109/TNSRE.2022.3149793
10.1007/s11517-021-02335-9
10.1109/LRA.2021.3062003
10.1007/s42235-023-00419-w
10.1007/3-540-57868-4_57
10.1109/IROS45743.2020.9341183
10.1109/TFUZZ.2024.3364382
10.1109/RCAR54675.2022.9872282
10.1109/TBME.2008.2003293
10.1109/TBME.2017.2750139
10.1007/s11227-021-03768-7
10.1016/j.bspc.2007.07.009
10.1109/CVPR52688.2022.00520
10.1109/TNSRE.2019.2909585
10.1109/jsen.2021.3077698
10.1109/ACCESS.2020.2971552
10.1109/ICINFA.2010.5512456
10.1016/j.inffus.2022.12.023
10.1016/j.bspc.2024.106105
ContentType Journal Article
DBID 97E
RIA
RIE
AAYXX
CITATION
DOI 10.1109/TASE.2024.3421276
DatabaseName IEEE All-Society Periodicals Package (ASPP) 2005–Present
IEEE All-Society Periodicals Package (ASPP) 1998–Present
IEEE/IET Electronic Library
CrossRef
DatabaseTitle CrossRef
DatabaseTitleList
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Electronic Library (IEL)
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
Discipline Engineering
EISSN 1558-3783
EndPage 5424
ExternalDocumentID 10_1109_TASE_2024_3421276
10584554
Genre orig-research
GrantInformation_xml – fundername: Natural Science Foundation of Zhejiang Province
  grantid: LGG21E050003
  funderid: 10.13039/501100004731
– fundername: Key Research and Development Program of Zhejiang Province
  grantid: 2024C03042; 2021C03050
  funderid: 10.13039/501100019033
GroupedDBID -~X
0R~
29I
4.4
5GY
5VS
6IK
97E
AAJGR
AARMG
AASAJ
AAWTH
ABAZT
ABQJQ
ABVLG
ACGFO
ACGFS
ACIWK
AENEX
AETIX
AGQYO
AGSQL
AHBIQ
AIBXA
AKJIK
AKQYR
ALMA_UNASSIGNED_HOLDINGS
ATWAV
BEFXN
BFFAM
BGNUA
BKEBE
BPEOZ
CS3
DU5
EBS
EJD
F5P
HZ~
H~9
IFIPE
IPLJI
JAVBF
LAI
M43
O9-
OCL
PQQKQ
RIA
RIE
RNS
AAYXX
CITATION
ID FETCH-LOGICAL-c266t-5bf9929e2abbd310ede1d0213e8db7b439ec298791a567c11a84577f0f2a266c3
IEDL.DBID RIE
ISICitedReferencesCount 7
ISICitedReferencesURI http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=001263415300001&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
ISSN 1545-5955
IngestDate Tue Nov 18 21:37:43 EST 2025
Sat Nov 29 08:13:27 EST 2025
Wed Aug 27 01:43:35 EDT 2025
IsPeerReviewed false
IsScholarly true
Language English
License https://ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html
https://doi.org/10.15223/policy-029
https://doi.org/10.15223/policy-037
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c266t-5bf9929e2abbd310ede1d0213e8db7b439ec298791a567c11a84577f0f2a266c3
ORCID 0000-0001-7912-4505
0000-0002-0546-6016
0000-0001-7748-8591
PageCount 14
ParticipantIDs crossref_citationtrail_10_1109_TASE_2024_3421276
crossref_primary_10_1109_TASE_2024_3421276
ieee_primary_10584554
PublicationCentury 2000
PublicationDate 20250000
2025-00-00
PublicationDateYYYYMMDD 2025-01-01
PublicationDate_xml – year: 2025
  text: 20250000
PublicationDecade 2020
PublicationTitle IEEE transactions on automation science and engineering
PublicationTitleAbbrev TASE
PublicationYear 2025
Publisher IEEE
Publisher_xml – name: IEEE
References ref13
ref12
ref15
ref14
ref11
ref10
ref17
ref16
ref19
ref18
ref50
ref46
ref45
ref48
ref47
ref42
ref41
ref44
ref43
ref49
ref8
ref7
ref9
ref4
ref3
ref6
ref5
ref40
ref35
ref34
ref37
ref36
ref31
ref30
ref33
ref32
ref2
ref1
ref39
ref38
ref24
ref23
ref26
ref25
ref20
ref22
ref21
ref28
ref27
ref29
References_xml – ident: ref42
  doi: 10.1109/EMBC.2017.8037600
– ident: ref14
  doi: 10.1109/ICASSP40776.2020.9053270
– ident: ref38
  doi: 10.3390/mi13081205
– ident: ref27
  doi: 10.1016/j.bspc.2022.103693
– ident: ref1
  doi: 10.1007/s12541-017-0081-9
– ident: ref25
  doi: 10.1109/SAS48726.2020.9220037
– ident: ref10
  doi: 10.1016/j.inffus.2021.11.006
– ident: ref28
  doi: 10.1109/TBME.2011.2161671
– ident: ref47
  doi: 10.1109/JBHI.2023.3252091
– ident: ref19
  doi: 10.3389/frobt.2018.00127
– ident: ref35
  doi: 10.48550/ARXIV.1706.03762
– ident: ref29
  doi: 10.3389/frobt.2018.00078
– ident: ref40
  doi: 10.1109/LRA.2024.3379800
– ident: ref20
  doi: 10.1109/LRA.2020.3007455
– ident: ref44
  doi: 10.1109/CVPRW.2018.00177
– ident: ref21
  doi: 10.1109/THMS.2021.3107256
– ident: ref45
  doi: 10.1109/CVPR.2019.01233
– ident: ref41
  doi: 10.1038/s41598-023-28390-w
– ident: ref48
  doi: 10.1109/IROS45743.2020.9341649
– ident: ref3
  doi: 10.3390/s19194242
– ident: ref5
  doi: 10.1109/JBHI.2023.3238406
– ident: ref31
  doi: 10.1109/tnsre.2021.3087135
– ident: ref17
  doi: 10.1109/tbme.2009.2034734
– ident: ref24
  doi: 10.3390/s19204447
– ident: ref11
  doi: 10.1109/TNSRE.2020.2987155
– ident: ref32
  doi: 10.1109/JSEN.2022.3167686
– ident: ref43
  doi: 10.1109/jas.2017.7510619
– ident: ref6
  doi: 10.1109/tase.2023.3276856
– ident: ref49
  doi: 10.1016/j.patcog.2021.108332
– ident: ref7
  doi: 10.1109/TASE.2023.3250240
– ident: ref18
  doi: 10.1109/JSEN.2022.3146446
– ident: ref23
  doi: 10.1109/TNSRE.2022.3149793
– ident: ref30
  doi: 10.1007/s11517-021-02335-9
– ident: ref4
  doi: 10.1109/LRA.2021.3062003
– ident: ref33
  doi: 10.1007/s42235-023-00419-w
– ident: ref39
  doi: 10.1007/3-540-57868-4_57
– ident: ref8
  doi: 10.1109/IROS45743.2020.9341183
– ident: ref36
  doi: 10.1109/TFUZZ.2024.3364382
– ident: ref16
  doi: 10.1109/RCAR54675.2022.9872282
– ident: ref12
  doi: 10.1109/TBME.2008.2003293
– ident: ref22
  doi: 10.1109/TBME.2017.2750139
– ident: ref26
  doi: 10.1007/s11227-021-03768-7
– ident: ref9
  doi: 10.1016/j.bspc.2007.07.009
– ident: ref46
  doi: 10.1109/CVPR52688.2022.00520
– ident: ref2
  doi: 10.1109/TNSRE.2019.2909585
– ident: ref37
  doi: 10.1109/jsen.2021.3077698
– ident: ref15
  doi: 10.1109/ACCESS.2020.2971552
– ident: ref13
  doi: 10.1109/ICINFA.2010.5512456
– ident: ref50
  doi: 10.1016/j.inffus.2022.12.023
– ident: ref34
  doi: 10.1016/j.bspc.2024.106105
SSID ssj0024890
Score 2.4298325
Snippet Although there are many wearable sensors that make the acquisition of multi-modality data easier, effective feature extraction and fusion of the data is still...
SourceID crossref
ieee
SourceType Enrichment Source
Index Database
Publisher
StartPage 5411
SubjectTerms Accuracy
Cross-modality interaction
Electromyography
Feature extraction
Fuses
locomotion mode recognition
lower limb
neural network
Neural networks
Representation learning
self-attention
Stairs
Title Cross-Modality Self-Attention and Fusion-Based Neural Network for Lower Limb Locomotion Mode Recognition
URI https://ieeexplore.ieee.org/document/10584554
Volume 22
WOSCitedRecordID wos001263415300001&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVIEE
  databaseName: IEEE Electronic Library (IEL)
  customDbUrl:
  eissn: 1558-3783
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0024890
  issn: 1545-5955
  databaseCode: RIE
  dateStart: 20040101
  isFulltext: true
  titleUrlDefault: https://ieeexplore.ieee.org/
  providerName: IEEE
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV07T8MwELZoxQADzyLKSx6YkFych-t6LFUrhlIhWqRuUWxfRCVIUEn5_ZydFLqAxGZFjvO4RN9357v7CLkWscyQZQMDnqUsFsIybTVnBtmGMjyyXWu82IScTHrzuXqsi9V9LQwA-OQz6Lih38u3hVm5UBn-4QiXiH8N0pBSVsVaP431ej6g4igBE0qIegsz4Op21p8O0RUM407kNkBdf5ENENpQVfGgMtr_5-0ckL2aPdJ-Ze5DsgX5Ednd6Cl4TF4G7pLsobCeYdMpvGasX5ZVWiNNc0tHKxciY3cIYJa67hy45KRKB6fIYenYKafR8eJN49AUldAPdapp9Gmdb1TkLfI8Gs4G96yWU2AGUbhkQmcKyRCEqdYWWR1YCCxCfAQ9q6VGZgImVD2pglR0pQmCFJ9OyoxnYYoLmOiENPMih1NCQxXbDB1B3bUQR6BUDN1Yp5anoY24Fm3C1-83MXWvcSd58Zp4n4OrxJkkcSZJapO0yc33Ke9Vo42_JrecOTYmVpY4--X4OdkJnW6vD51ckGa5XMEl2Taf5eJjeeW_oy9AKsX2
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3JTsMwELXYJODAWkRZfeCE5OIkdlMfC6IqolQIitRbFNsTUakkCFK-n7EToBeQuFmR4yyT6L0Zz8wj5EyKOEOWDQx4ljIhpWXaas4Msg1leGTb1nixiXg47IzH6r4uVve1MADgk8-g5YZ-L98WZuZCZfiHI1wi_i2SZSlEGFTlWj-t9To-pOJIAZNKynoTM-DqYtR9vEZnMBStyG2Bug4jczA0p6viYaW3-c8b2iIbNX-k3crg22QB8h2yPtdVcJc8X7lLsrvCeo5NH2GasW5ZVomNNM0t7c1ckIxdIoRZ6vpz4JLDKiGcIoulA6edRgeTF41DU1RSP9TpptGHr4yjIm-Qp9716KrPakEFZhCHSyZ1ppAOQZhqbZHXgYXAIshH0LE61shNwISqE6sgle3YBEGKTxfHGc_CFBcw0R5Zyosc9gkNlbAZuoK6bUFEoJSAttCp5WloI65lk_Cv95uYutu4E72YJt7r4CpxJkmcSZLaJE1y_n3Ka9Vq46_JDWeOuYmVJQ5-OX5KVvuju0EyuBneHpK10Kn4-kDKEVkq32ZwTFbMRzl5fzvx39Qnt-TJPQ
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Cross-Modality+Self-Attention+and+Fusion-Based+Neural+Network+for+Lower+Limb+Locomotion+Mode+Recognition&rft.jtitle=IEEE+transactions+on+automation+science+and+engineering&rft.au=Zhao%2C+Changchen&rft.au=Liu%2C+Kai&rft.au=Zheng%2C+Hao&rft.au=Song%2C+Wenbo&rft.date=2025&rft.issn=1545-5955&rft.eissn=1558-3783&rft.volume=22&rft.spage=5411&rft.epage=5424&rft_id=info:doi/10.1109%2FTASE.2024.3421276&rft.externalDBID=n%2Fa&rft.externalDocID=10_1109_TASE_2024_3421276
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=1545-5955&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=1545-5955&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=1545-5955&client=summon