Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 61, Issue 14, 1437001(2024)

Microexpression Recognition Algorithm Based on a Two-Branch Lightweight Network

Bo Zhang and Yufan Wu*
Author Affiliations
  • College of Computer Science and Technology, Shenyang University of Chemical Technology, Shenyang 110142, Liaoning , China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (15)
    Equations (10)
    References (35)
    Tools
    Paper Information
    Topics
    Figures & Tables(15)
    Flowchart of the proposed algorithm

    Fig. 1. Flowchart of the proposed algorithm

    Download full size
    Start frame, peak frame, and optical flow characteristics corresponding to the TV-L1 optical flow method

    Fig. 2. Start frame, peak frame, and optical flow characteristics corresponding to the TV-L1 optical flow method

    Download full size
    Overall architecture of the MobileViT model

    Fig. 3. Overall architecture of the MobileViT model

    Download full size
    MI module

    Fig. 4. MI module

    Download full size
    MobileViT modular structure

    Fig. 5. MobileViT modular structure

    Download full size
    Structure of the attention module

    Fig. 6. Structure of the attention module

    Download full size
    Channel attention module

    Fig. 7. Channel attention module

    Download full size
    Spatial attention module

    Fig. 8. Spatial attention module

    Download full size
    Experimental result. (a) Confusion matrix of composite dataset; (b) UAR line charts of training and verification

    Fig. 9. Experimental result. (a) Confusion matrix of composite dataset; (b) UAR line charts of training and verification

    Download full size
    Visualization of experimental results

    Fig. 10. Visualization of experimental results

    Download full size

    • Table 1. MISEViTNet model structure

      View table

      Table 1. MISEViTNet model structure

      LayerOutput sizeOperatorNumber of output channelsStride
      Image256×2563
      Conv 3×3128×128Conv2d162
      Layer 1128×128MI321
      Layer 264×64MI642
      64×64MI641
      64×64MI641
      Layer 332×32MI962
      32×32MobileViT961
      Layer 416×16MI1282
      16×16MobileViT1281
      Layer 58×8MI1602
      8×8MobileViT1601
      8×8Conv2d6401
      8×8Avgpool 8×8640
      1×1FC640
      Global pool linear1×1Conv2d1000
      Number of parameters1.43×106

    • Table 2. Microexpression data set information

      View table

      Table 2. Microexpression data set information

      DatasetSMICCASME IISAMM
      Release time201320142018
      Type of expression377
      Positive sample513326
      Negative sample708792
      Surprise sample432515
      Other sample010226
      Sample count164247159

    • Table 3. Result comparison of different feature extraction methods

      View table

      Table 3. Result comparison of different feature extraction methods

      MethodComposite datasetSMICCASME IISAMM
      UF1UARUF1UARUF1UARUF1UAR
      LBP-TOP［1］0.58820.57850.20000.52800.70260.74290.39540.4102
      MDMO［4］0.56350.51250.49260.48120.54920.53820.50210.5108
      CNN-LSTM［9］0.38520.39430.41500.42760.41130.41250.30200.3086
      ATNet［30］0.6310.6310.5530.5430.7980.7750.4960.482
      OFF-ApexNet［31］0.71960.70960.68170.66950.87640.86810.54090.5392
      MobileNetV2［32］0.66520.64250.65890.63680.63280.61250.66140.6236
      DeiT［33］0.67310.68790.69700.68810.69940.68140.70280.7052
      DeepViT［34］0.71580.70250.73690.71520.69820.36280.69280.7114
      ECANet34-DA［35］0.70730.72220.84090.85430.66510.7143
      Proposed model0.75870.77360.73720.71390.85210.84610.72160.6781

    • Table 4. Comparison of the number of parameters corresponding to different models

      View table

      Table 4. Comparison of the number of parameters corresponding to different models

      ModelNumber of parameters /106Processing time for individual sample /ms
      EDSMISEViTNet3.971.8
      ResNet1811.782.3
      ResNet5025105.9
      DeiT［33］2853.7
      MobileNetV2［32］1724

    • Table 5. Ablation experiment

      View table

      Table 5. Ablation experiment

      Experiment No.UARUF1Number of parameters /106Processing time for individual sample /ms
      10.75870.77363.9071.8
      20.72580.71911.9635.3
      30.69080.66221.9435.1
      40.75140.72053.6372.3
      50.75520.75683.9372.6
    Tools

    Get Citation

    Copy Citation Text

    Bo Zhang, Yufan Wu. Microexpression Recognition Algorithm Based on a Two-Branch Lightweight Network[J]. Laser & Optoelectronics Progress, 2024, 61(14): 1437001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Category: Digital Image Processing

    Received: Jul. 13, 2023

    Accepted: Oct. 30, 2023

    Published Online: Jul. 8, 2024

    The Author Email: Yufan Wu (z2021360@stu.syuct.edu.cn)

    DOI:10.3788/LOP231714

    CSTR:32186.14.LOP231714

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology