Laser & Optoelectronics Progress, Volume. 57, Issue 20, 201021(2020)

Fire Detection Method Based on Localization Confidence and Region-Based Fully Convolutional Network

Hong Zhang1,2, Yunyang Yan1,2、*, Yian Liu1, and Shangbing Gao2
Author Affiliations
  • 1School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • 2Faculty of Computer & Software Engineering, Huaiyin Institute of Technology, Huaian, Jiangsu 223003, China;
  • show less
    Figures & Tables(18)
    Detection diagram of R-FCN
    Separable convolution
    Incomplete information marked by anchors
    Change of translation and scaling
    Comparison of iterations accuracy changes[11]. (a)FPN iterations accuracy changes; (b)R-CNN iterations accuracy changes
    Comparison of pooling. (a) ROI pooling; (b) precise ROI pooling
    Non-maximum suppression
    Lack of localization confidence of NMS
    Prediction of IOU
    Flow chart of LOF-FCN
    Comparison of fire missed detection rate
    Comparison of fire detection accuracy
    Experimental video. (a) Video 1; (b) video 2; (c) video 3; (d) video 4; (e) video 5; (f) video 6
    • Table 1. Comparison of the speed of the detection

      View table

      Table 1. Comparison of the speed of the detection

      MethodKernelSpeed /(frame/s)Accuracy /%
      R-FCN13×131596.9
      Separableconvolution13×1,1×132296.4
    • Table 2. Training data

      View table

      Table 2. Training data

      CategoryPictureNumber
      L-Fire12232104
      S-Fire35125921
      Total47358025
    • Table 3. Comparison of algorithm performance

      View table

      Table 3. Comparison of algorithm performance

      ModelYOLOv3Fast-RCNNFaster-RCNNR-FCNLOF-FCN
      Speed /(frame/s)340.671317
      mAP /%76.254.083.980.381.7
    • Table 4. Detection results of flame

      View table

      Table 4. Detection results of flame

      VideoVideo frameFire frameMethod in Ref. [6]Method in Ref. [8]Method in Ref. [9]Proposed method
      TP /%FP /%TP /%FP /%TP /%FP /%TP /%FP /%
      117015696.23.897.62.498.61.499.20.8
      223425794.55.593.26.892.77.395.44.6
      320018299.30.799.80.210001000
      41249399.70.31000100099.01.0
      Average182.0172.097.42.697.72.397.82.298.41.6
    • Table 5. Detection results of non-flame

      View table

      Table 5. Detection results of non-flame

      VideoVideo frameFire frameMethod in Ref. [6]Method in Ref. [8]Method in Ref. [9]Proposed method
      TN /%FN /%TN /%FN /%TN /%FN /%TN /%FN /%
      5420098.81.297.32.795.44.698.31.7
      6376097.22.896.23.896.73.397.62.4
      Average398.00.098.02.096.83.296.13.998.02.0
    Tools

    Get Citation

    Copy Citation Text

    Hong Zhang, Yunyang Yan, Yian Liu, Shangbing Gao. Fire Detection Method Based on Localization Confidence and Region-Based Fully Convolutional Network[J]. Laser & Optoelectronics Progress, 2020, 57(20): 201021

    Download Citation

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

    Category: Image Processing

    Received: Feb. 12, 2020

    Accepted: Mar. 12, 2020

    Published Online: Oct. 13, 2020

    The Author Email: Yan Yunyang (yunyang@hyit.edu.cn)

    DOI:10.3788/LOP57.201021

    Topics