• Laser & Optoelectronics Progress
  • Vol. 57, Issue 4, 041015 (2020)
Hang Liu and Xili Wang*
Author Affiliations
  • School of Computer Science, Shaanxi Normal University, Xi'an, Shaanxi 710119, China
  • show less
    DOI: 10.3788/LOP57.041015 Cite this Article Set citation alerts
    Hang Liu, Xili Wang. Remote Sensing Image Segmentation Model Based on Attention Mechanism[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041015 Copy Citation Text show less
    Comparison between RSANet and LinkNet. (a) RSANet; (b) LinkNet
    Fig. 1. Comparison between RSANet and LinkNet. (a) RSANet; (b) LinkNet
    Structure of encoder block
    Fig. 2. Structure of encoder block
    Structure of decoder block
    Fig. 3. Structure of decoder block
    Position attention module
    Fig. 4. Position attention module
    Channel attention module
    Fig. 5. Channel attention module
    Remote sensing images and road labels in Massachusetts Roads dataset. (a) Part of the remote sensing images in the training set; (b) road labels corresponding to images in the training set; (c) some remote sensing images in the test set; (d) road labels corresponding to images in the test set
    Fig. 6. Remote sensing images and road labels in Massachusetts Roads dataset. (a) Part of the remote sensing images in the training set; (b) road labels corresponding to images in the training set; (c) some remote sensing images in the test set; (d) road labels corresponding to images in the test set
    Remote sensing images and road labels in DeepGlobe dataset. (a) Part of the remote sensing images in the training set; (b) road labels corresponding to images in the training set; (c) some remote sensing images in the test set; (d) road labels corresponding to images in the test set
    Fig. 7. Remote sensing images and road labels in DeepGlobe dataset. (a) Part of the remote sensing images in the training set; (b) road labels corresponding to images in the training set; (c) some remote sensing images in the test set; (d) road labels corresponding to images in the test set
    Segmentation results of different depth models in the Massachusetts Roads test set. (a) Original images; (b) labels; (c) segmentation results of LinkNet; (d) segmentation results of RSANet
    Fig. 8. Segmentation results of different depth models in the Massachusetts Roads test set. (a) Original images; (b) labels; (c) segmentation results of LinkNet; (d) segmentation results of RSANet
    Segmentation results of different depth models on the DeepGlobe Road Extraction test set. (a) Original images; (b) labels; (c) segmentation results of LinkNet; (d) segmentation results of RSANet
    Fig. 9. Segmentation results of different depth models on the DeepGlobe Road Extraction test set. (a) Original images; (b) labels; (c) segmentation results of LinkNet; (d) segmentation results of RSANet
    ModelImage 1Image 2Image 3Image 4Image 5
    PrF1-scorePrF1-scorePrF1-scorePrF1-scorePrF1-score
    LinkNet[3]0.8390.8590.9130.8810.8210.8320.8000.8170.7470.779
    RSANet0.9790.9760.9230.8920.8290.8350.8220.8890.8180.827
    Table 1. Segmentation results of depth models on each test image in Fig. 8
    ModelPrF1-scoreTraining time /hInference time per picture /s
    LinkNet[3](baseline)0.8020.815310.31
    LinkNet-PAM0.8120.823320.32
    LinkNet-CAM0.8180.828400.40
    RSANet0.8270.843410.41
    Table 2. Segmentation results of different depth models on the Massachusetts Roads test set
    ModelImage 6Image 7Image 8Image 9Image 10
    PrPIOUPrPIOUPrPIOUPrPIOUPrPIOU
    LinkNet[3]0.7700.7120.5970.5150.7330.6310.8860.7220.7290.704
    RSANet0.7920.7350.8240.6960.8350.6940.9070.7900.7730.723
    Table 3. Segmentation results of different depth models on each test image in Fig. 9
    ModelPPPIOUTraining time /hInference time per picture /s
    LinkNet[3](baseline)0.7850.598310.31
    LinkNet-PAM0.7910.610320.32
    LinkNet-CAM0.7960.616400.40
    RSANet0.8110.624410.41
    Table 4. Segmentation results of different depth models on the DeepGlobe Road Extraction test set
    ModelPrPPF1-score
    FCN-4s[2]0.6600.7100.684
    SegNet[18]0.7650.7730.768
    ELU-SegNet[18]0.7730.8520.788
    ELU-SegNet-R[18]0.7800.8470.812
    DCED[19]0.8390.8250.829
    RSANet0.8270.8590.843
    Table 5. Segmentation results of different depth models on the Massachusetts Roads test set
    ModelPPF1-scoreTraining time /hInference time per picture /s
    U-Net[1]0.7640.597700.94
    SegNet[20]0.7740.602981.8
    LinkNet[3](baseline)0.7850.598310.31
    RSANet0.8110.624410.41
    Table 6. Segmentation results of different depth models on the DeepGlobe Road Extraction test set
    Hang Liu, Xili Wang. Remote Sensing Image Segmentation Model Based on Attention Mechanism[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041015
    Download Citation