Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 7, 900(2022)

Fire smoke detection combined with detailed features and hybrid attention mechanism

Rui-qing WANG, Hui-qin WANG*, and Ke WANG
Author Affiliations
  • College of Information and Control Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China
  • show less
    Figures & Tables(16)
    Structure of YOLOv4
    Detailed feature fusion module
    Process of feature fusion
    Diagram of CBAM
    Detailed feature fusion module with CBAM(DFF-CBAM)
    Structure of the proposed network
    Examples of detection results.(a)Indoor smoke;(b)Indoor smoke under light interference;(c)Outdoor small target smoke;(d)Outdoor smoke in complex background;(e)Forest fire smoke;(f)Night fire smoke.
    Image of failed to detection
    PR curves.(a)PR curve of YOLOv4;(b)PR curve of proposed algorithm.
    Precision curves
    Recall curves
    Visualization of feature map
    Comparison of heatmaps
    • Table 1. Parameters setting of DFF modules

      View table
      View in Article

      Table 1. Parameters setting of DFF modules

      DFF模块1DFF模块2
      自底向上支路

      输入(104,104,128)

      最大池化层,s=2(52,52,128)

      卷积层,c=128(52,52,128)

      输入(208,208,64)

      最大池化层,s=2(104,104,64)

      卷积层,c=128(104,104,128)

      最大池化层,s=2(52,52,128)

      卷积层,c=128(52,52,128)

      自顶向下支路

      输入(26,26,512)

      双线性插值,r=2(52,52,512)

      卷积层,c=256(52,52,256)

      卷积层,c=128(52,52,128)

      输入(13,13,1 024)

      双线性插值,r=2(26,26,1 024)

      卷积层,c=512(26,26,512)

      双线性插值,r=2(52,52,512)

      卷积层,c=256(52,52,256)

      卷积层,c=128(52,52,128)

      拼接通道维度拼接(52,52,256)通道维度拼接(52,52,256)
    • Table 2. Experimental results of ablation study

      View table
      View in Article

      Table 2. Experimental results of ablation study

      精确率/%召回率/%平均精确率/%检测速率/(frame·s-1
      YOLOv496.1681.5991.9151.9
      YOLOv4+DFF94.5288.7992.8751.2
      YOLOv4+CBAM97.3285.3193.4752.7
      本文算法97.3791.4596.2250.8
    • Table 3. Performance comparison with other algorithms

      View table
      View in Article

      Table 3. Performance comparison with other algorithms

      精确率/%召回率/%平均精确率/%检测速率/(frame·s-1
      Faster-RCNN73.8281.7282.5520.3
      SSD65.3171.4669.8972.8
      RetinaNet91.4979.5689.9352.7
      YOLOv389.8979.3486.7767.3
      EfficientNet2882.9183.7187.6649.5
      文献[2974.8282.2180.3170.5
      YOLOv496.1681.5991.9151.9
      SE-YOLOv43093.3286.4793.6448.5
      本文算法97.3791.4596.2250.8
    Tools

    Get Citation

    Copy Citation Text

    Rui-qing WANG, Hui-qin WANG, Ke WANG. Fire smoke detection combined with detailed features and hybrid attention mechanism[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(7): 900

    Download Citation

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

    Category:

    Received: Dec. 14, 2021

    Accepted: --

    Published Online: Jul. 7, 2022

    The Author Email: Hui-qin WANG (hqwang@xauat.edu.cn)

    DOI:10.37188/CJLCD.2021-0325

    Topics