Laser & Optoelectronics Progress, Volume. 56, Issue 13, 131009(2019)

Real-Time Object Detection for Millimeter-Wave Images Based on Improved Faster Regions with Convolutional Neural Networks

Bingji Hou1,2,3, Minghui Yang1, and Xiaowei Sun1、*
Author Affiliations
  • 1 Key Laboratory of Terahertz Solid Technology, Shanghai Institute of Microsystem and Information Technology (SIMIT), Chinese Academy of Sciences, Shanghai 200050, China
  • 2 University of Chinese Academy of Sciences, Beijing 100049, China
  • 3 School of Information Science and Technology, ShanghaiTech University, Shanghai 201210, China
  • show less
    Figures & Tables(7)
    Diagram of millimeter wave imaging system
    Comparison between millimeter wave images and optical images. (a) Millimeter wave images, boxes in figure are real markers; (b) optical images, boxes in figure are the test results, and the numbers are the target confidence
    Deconvolution diagram
    Area statistics of label boxes in two data sets. (a) Statistical chart of millimeter wave image data sets; (b) statistical chart of VOC data sets
    Partial training results. (a1)(b1)(c1)(d1) real marks; (a2)(b2)(c2)(d2) test results of corresponding graph
    • Table 1. Feature extraction network structure

      View table

      Table 1. Feature extraction network structure

      TypeLayersFiltersSize/strideOutput
      Convolution1MaxPool2643×3/12×2/2(512,256)(256,128)
      Convolution2MaxPool21283×3/12×2/2(256,128)(128,64)
      Convolution3MaxPool32563×3/12×2/2(128,64)(64,32)
      Convolution4MaxPool35123×3/12×2/2(64,32)(32,16)
      Convolution535123×3/1(32,16)
      Deconvolution15122×2/2(64,32)
      Concateconvolution5(64,32)
      Convolution625123×3/1(64,32)
    • Table 2. Comparison of experimental results when the GPU environment for network testing is GTX 1080

      View table

      Table 2. Comparison of experimental results when the GPU environment for network testing is GTX 1080

      ModelInput size /(pixel×pixel)SamplingLayersWithout RCNNprerecF1 scoreTime /ms
      VGG-11000×5001616×0.8590.7440.797107
      VGG-21000×50016160.8640.7850.82297
      ZF-Net1000×5001650.8370.7570.79427
      ResNet1000×50016410.8670.7660.81366
      DenseNet1000×50016900.8500.5400.66094
      SSD300×3008160.8540.7630.80544
      YOLOv3416×4168590.8220.7840.80229
      Ours512×2568190.8760.8120.84336
    Tools

    Get Citation

    Copy Citation Text

    Bingji Hou, Minghui Yang, Xiaowei Sun. Real-Time Object Detection for Millimeter-Wave Images Based on Improved Faster Regions with Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2019, 56(13): 131009

    Download Citation

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

    Category: Image Processing

    Received: Dec. 17, 2018

    Accepted: Feb. 17, 2019

    Published Online: Jul. 11, 2019

    The Author Email: Sun Xiaowei (xwsun@mail.sim.ac.cn)

    DOI:10.3788/LOP56.131009

    Topics