Infrared and Laser Engineering, Volume. 50, Issue 11, 20210075(2021)

Few-shot prohibited item segmentation algorithm based on graph matching network

Zhenyue Zhu1... Shujing Lv1,2,*, and Yue Lv12 |Show fewer author(s)
Author Affiliations
  • 1Department of Computer Science and Technology, East China Normal University, Shanghai 200062, China
  • 2Shanghai Key Laboratory of Multidimensional Information Processing, Shanghai 200241, China
  • show less
    Figures & Tables(8)
    Overview of proposed framework in the 1-shot prohibited item segmentation
    Structure diagram of the node attention module
    Experimental effect results on SIXray dataset. (a) Support image; (b) Support image mask; (c) Query image; (d) Segmentation result, of which red region is the predicted prohibited item region
    Experimental effect results on Xray-PI dataset. (a) Support image; (b) Support image mask; (c) Query image; (d) Segmentation result, of which red region is the predicted prohibited item region
    • Table 1. Parameter setting of prohibited item segmentation model based on graph matching network

      View table
      View in Article

      Table 1. Parameter setting of prohibited item segmentation model based on graph matching network

      Operational layerConfiguration
      Graph embeddingInput image321×321×3
      Convolution layer#maps: 64,k: 7×7,s: 2×2
      Maxpool layerw: 3×3,s: 2×2
      Convolution layer$\left[ #maps:64,k:1×1s:1×1#maps:64,k:3×3s:1×1#maps:256,k:1×1s:1×1 \right]{\rm{ \times 3} }$
      Convolution layer$\left[ #maps:128,k:1×1s:2×2#maps:128,k:3×3s:2×2#maps:512,k:1×1s:2×2 \right]{\rm{ \times 4} }$
      Convolution layer$\left[ #maps:256,k:1×1s:1×1#maps:256,k:3×3s:1×1#maps:1024,k:1×1s:1×1 \right]{\rm{ \times 6} }$
      Convolution layer#maps: 256,k: 1×1,s: 1×1
      Graph matchingConvolution layer#maps: 256,k: 1×1,s: 1×1
      Avgpool layerw: 11×11,s: 1×1
      Convolution layer#maps: 256,k: 1×1,s: 1×1
      Convolution layer#maps: 256,k: 1×1,s: 1×1
      Maxpool layerw: 10×10,s: 1×1
      SegmentationConvolution layer#maps: 256,k: 1×1,s: 1×1
      Convolution layer$\left[ #maps:256,k:3×3s:1×1#maps:256,k:3×3s:1×1 \right]{\rm{ \times 3} }$
      Convolution layer#maps: 1,k: 1×1,s: 1×1
    • Table 2. Segmentation performance of model with different filters and length of subgraphs

      View table
      View in Article

      Table 2. Segmentation performance of model with different filters and length of subgraphs

      FilterSizemeanIoU
      Xray-PISIXray
      Average filter548.8%34.0%
      1050.4%35.8%
      1549.7%35.5%
      Maximum filter550.1%34.8%
      1051.2%36.4%
      1550.4%35.3%
    • Table 3. Segmentation performance of 1-shot task and 5-shot task on SIXray dataset

      View table
      View in Article

      Table 3. Segmentation performance of 1-shot task and 5-shot task on SIXray dataset

      MethodsGunKnifeWrenchPliersScissorsmeanIoU
      1-shotCANet[13]40.341.635.233.918.333.9%
      PGNet[16]38.941.537.433.218.033.8%
      Ours41.442.135.634.028.536.4%
      5-shotCANet[13]43.043.236.835.418.935.5%
      PGNet[16]41.142.937.035.719.135.2%
      Ours43.743.436.335.929.337.7%
    • Table 4. Segmentation performance of 1-shot task and 5-shot task on Xray-PI dataset

      View table
      View in Article

      Table 4. Segmentation performance of 1-shot task and 5-shot task on Xray-PI dataset

      MethodsFireworkFirecrackerBottleGunWrenchPliersBlademeanIoU
      1-shotCANet[13]51.445.542.248.134.753.067.948.9%
      PGNet[16]44.141.931.847.136.351.166.445.5%
      Proposed52.945.747.451.237.555.568.751.2%
      5-shotCANet[13]54.847.645.249.535.656.168.051.0%
      PGNet[16]46.043.035.148.537.155.668.347.7%
      Proposed55.449.148.753.638.556.468.952.9%
    Tools

    Get Citation

    Copy Citation Text

    Zhenyue Zhu, Shujing Lv, Yue Lv. Few-shot prohibited item segmentation algorithm based on graph matching network[J]. Infrared and Laser Engineering, 2021, 50(11): 20210075

    Download Citation

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

    Category: Image processing

    Received: Jan. 27, 2021

    Accepted: --

    Published Online: Dec. 7, 2021

    The Author Email:

    DOI:10.3788/IRLA20210075

    Topics