Acta Photonica Sinica, Volume. 52, Issue 1, 0110001(2023)

AF-ICNet Semantic Segmentation Method for Unstructured Scenes Based on Small Target Category Attention Mechanism and Feature Fusion

Qinglin AI1、*, Junrui ZHANG1, and Feiqing WU2
Author Affiliations
  • 1Key Laboratory of Special Purpose Equipment and Advanced Manufacturing Technology,Ministry of Education & Zhejiang Province,Zhejiang University of Technology,Hangzhou,310023,China
  • 2School of Information Science and Engineering,Ningbotech University,Ningbo 315100,China
  • show less
    Figures & Tables(20)
    Pyramid pooling module structure
    Atrous spatial pyramid pooling structure
    Coordinate attention structure
    Attention and feature fusion structure
    Sample distribution of Cityscapes dataset and IDD dataset
    Parameter sensitivity distribution of λ and c
    The structure of AF-ICNet
    Cityscapes dataset training sample image
    IDD dataset training sample image
    Cityscapes dataset training process curve
    Test result of Cityscapes dataset
    IDD dataset training process curve
    Test result of unstructured road segmentation of IDD dataset
    The experimental testing system
    The image of field work on unstructured roads
    Segmentation results of unstructured roads tested in the field
    • Table 1. The meaning of TP,FP,FN,FP

      View table
      View in Article

      Table 1. The meaning of TP,FP,FN,FP

      Real valuePredictive value
      PositiveNegative
      PositiveTP(True positive)FN(False negative)
      NegativeFP(False positive)TN(True negative)
    • Table 2. Test results of AF-ICNet on Cityscapes dataset

      View table
      View in Article

      Table 2. Test results of AF-ICNet on Cityscapes dataset

      ModelMIoU/%PixAcc/%FPSParam/(×106
      SegNet1757.913.329.5
      ENet1858.390.90.4
      ICNet69.780.945.526.5
      DeepLabV3Plus-MobileNet72.183.810.441.3
      Proposed AF-ICNet71.581.343.527.9
    • Table 3. Test results of AF-ICNet on IDD dataset

      View table
      View in Article

      Table 3. Test results of AF-ICNet on IDD dataset

      ModelMIoU/%PixAcc/%FPSParam/(×106
      SegNet31.913.029.5
      ENet48.383.30.4
      ICNet60.989.447.626.5
      DeepLabV3Plus-MobileNet61.289.611.241.3
      Proposed AF-ICNet62.589.841.727.9
    • Table 4. Ablation experiment

      View table
      View in Article

      Table 4. Ablation experiment

      GroupW-CEASPPCACityscapesIDD
      MIoU/%PixAcc/%FPSMIoU/%PixAcc/%FPS
      Exp 1×××69.780.945.560.989.447.6
      Exp 2××70.881.145.261.589.045.4
      Exp 3×71.081.144.462.189.043.4
      Exp 471.581.343.562.589.841.7
    Tools

    Get Citation

    Copy Citation Text

    Qinglin AI, Junrui ZHANG, Feiqing WU. AF-ICNet Semantic Segmentation Method for Unstructured Scenes Based on Small Target Category Attention Mechanism and Feature Fusion[J]. Acta Photonica Sinica, 2023, 52(1): 0110001

    Download Citation

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

    Category:

    Received: Jun. 29, 2022

    Accepted: Aug. 18, 2022

    Published Online: Feb. 27, 2023

    The Author Email: AI Qinglin (aqlaql@163.com)

    DOI:10.3788/gzxb20235201.0110001

    Topics