• Journal of Infrared and Millimeter Waves
  • Vol. 40, Issue 6, 798 (2021)
Fu-Qiang ZHENG1、2, Ding-Bo KUANG2, Yong HU2、*, Cai-Lan GONG2, Shuo HUANG2, Lan LI1、2, and Zhi-Jie HE1、2
Author Affiliations
  • 1University of Chinese Academy of Sciences,Beijing 100049China
  • 2Key Laboratory of Infrared System Detection and Imaging Technology,Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China
  • show less
    DOI: 10.11972/j.issn.1001-9014.2021.06.014 Cite this Article
    Fu-Qiang ZHENG, Ding-Bo KUANG, Yong HU, Cai-Lan GONG, Shuo HUANG, Lan LI, Zhi-Jie HE. Refined segmentation method based on U-ASPP-Net for Arctic independent sea ice[J]. Journal of Infrared and Millimeter Waves, 2021, 40(6): 798 Copy Citation Text show less
    Location of research distribution (a)Independent sea ice distribution between narrow waterways bordering land,(b)independent sea ice distribution under thin ice disturbance,(c)normal independent sea ice distribution,(d)independent sea ice distribution under thin cloud disturbance
    Fig. 1. Location of research distribution (a)Independent sea ice distribution between narrow waterways bordering land,(b)independent sea ice distribution under thin ice disturbance,(c)normal independent sea ice distribution,(d)independent sea ice distribution under thin cloud disturbance
    Architecture of U-ASPP-Net
    Fig. 2. Architecture of U-ASPP-Net
    Architecture of ASPP
    Fig. 3. Architecture of ASPP
    Convolution process of atrous depthwise separable convolution
    Fig. 4. Convolution process of atrous depthwise separable convolution
    Schematic diagram of overlapping edge elimination strategy
    Fig. 5. Schematic diagram of overlapping edge elimination strategy
    Schematic diagram of experimental data (a)schematic of data,(b)schematic of label
    Fig. 6. Schematic diagram of experimental data (a)schematic of data,(b)schematic of label
    Thumbnails of test set (a)Normal independent sea ice distribution, (b)independent sea ice distribution under thin ice disturbance, (c-d)independent sea ice distribution under thin cloud disturbance, (e)independent sea ice distribution between narrow waterways bordering land
    Fig. 7. Thumbnails of test set (a)Normal independent sea ice distribution, (b)independent sea ice distribution under thin ice disturbance, (c-d)independent sea ice distribution under thin cloud disturbance, (e)independent sea ice distribution between narrow waterways bordering land
    The changing trend of each indicator under different overlaps
    Fig. 8. The changing trend of each indicator under different overlaps
    Independent sea ice segmentation results by different methods (a)Images to be segmented, (b)manually interpreted images, (c)U-ASPP-Net, (d)U-Net, (e)Deeplab v3+, (f)partition gradient difference and bimodal threshold segmentation algorithm
    Fig. 9. Independent sea ice segmentation results by different methods (a)Images to be segmented, (b)manually interpreted images, (c)U-ASPP-Net, (d)U-Net, (e)Deeplab v3+, (f)partition gradient difference and bimodal threshold segmentation algorithm
    Comparison of ability to resist interference of thin cloud and sea ice (a)Images to be segmented, (b)Manually interpreted images, (c)U-ASPP-Net, (d)U-Net, (e)Deeplab v3+, (f)partition gradient difference and bimodal threshold segmentation algorithm
    Fig. 10. Comparison of ability to resist interference of thin cloud and sea ice (a)Images to be segmented, (b)Manually interpreted images, (c)U-ASPP-Net, (d)U-Net, (e)Deeplab v3+, (f)partition gradient difference and bimodal threshold segmentation algorithm
    性能参数指标
    量化等级12 bit
    扫描范围±55.1°±0.1°
    每条扫描线采样点数2 048(1 000 m),8 192(250 m)
    波段范围0.4∼12.5 µm
    通道间像元配准<0.3个像元
    定标精度

    可见光和近红外通道:5%(反射率),星上定标器实现可见光星上定标(相对和绝对辐射)。

    红外通道(星上黑体):0.5 K(270 K),分裂窗两个通道定标误差不一致性<小于0.5 K

    Table 1. Performance indicators of MERSI-Ⅱ
    通道号对应传感器通道号中心波长/μm光谱带宽/μm空间分辨率/m
    110.4700.05250
    220.5500.05250
    330.6500.05250
    440.8650.05250
    561.6400.02250(重采样)
    Table 2. Channel characteristics of sea ice datasets
    样本大小192×192
    训练样本数量1622幅
    验证样本数量302幅
    独立海冰占比0.19
    Table 3. Distribution of independent sea ice datasets
    混淆矩阵真实值
    PositiveNegative
    预测值PositiveTPFP
    NegativeFNTN
    Table 4. Confusion matrix
    损失函数OAKappa系数IOUDice系数
    FDWloss93.61%0.740.660.78
    Diceloss89.69%0.550.450.61
    Focalloss91.43%0.620.520.67
    Table 5. Comparison of independent sea ice segmentation accuracies of U-ASPP-NET model based on different loss function
    准确率KappaIOUDice系数
    U-ASPP-Net93.61%0.740.660.78
    U-Net91.20%0.680.580.73
    DeeplabV3+90.65%0.590.480.65
    分区梯度差分与双峰阈值分割法86.07%0.400.310.47
    Table 6. Comparison of accuracies for segmentation of independent sea ice with different methods
    Fu-Qiang ZHENG, Ding-Bo KUANG, Yong HU, Cai-Lan GONG, Shuo HUANG, Lan LI, Zhi-Jie HE. Refined segmentation method based on U-ASPP-Net for Arctic independent sea ice[J]. Journal of Infrared and Millimeter Waves, 2021, 40(6): 798
    Download Citation