Acta Optica Sinica, Volume. 45, Issue 9, 0928001(2025)
SAR Ship Detection Method Based on Ladder Residual and Coordinate Information Recombination
Figures & Tables(22)
Fig. 2. Diagrams of conventional convolution, deep separable convolution, and partial convolution modules
Fig. 9. Visualization comparison of image features before and after outputting. (a) Before outputting image; (b) y direction coordinate recombination image; (c) x direction coordinate recombination image; (d) combined output image
Fig. 11. Time frequency 3D surface maps of CIRConv. (a) Small ship; (b) large ship; (c) moving ship
Fig. 12. Comparison of real detection effects between Ori and SRCDNet. (a) Real values; (b) detection of Ori model; (c) SRCDNet detection
Fig. 13. Comparison of algorithm detection results. (a1)‒(a5) True values; (b1)‒(b5) Center Net detection; (c1)‒(c5) Faster R-CNN detection; (d1)‒(d5) YOLOv8 detection; (e1)‒(e5) YOLOv7 detection; (f1)‒(f5) SRCDNet detection
Table 1. Parameter configuration of hardware and software platforms
View tableTable 1. Parameter configuration of hardware and software platforms
Configuration Model CPU 13rd Gen Intel Core i7-13700KF GPU NVIDIA GeForce RTX 4080 Operating system Windows 10 Running memory 32 GB Video memory 16 GB Language and framework Anconda3+Python3.8+Pytorch1.13.0
Table 2. Training parameter configuration
View tableTable 2. Training parameter configuration
Training parameter Value Momentum parameter 0.937 Initial learning rate 0.001 Batch size 8 Epochs 300 Optimizer SGD
Table 3. Comparison of improvement effects in ablation experiments
View tableTable 3. Comparison of improvement effects in ablation experiments
Model Dataset P R mAP0.5 mAP0.5∶0.95 FLOPs/109 Ori HRSID 0.876 0.786 0.863 0.602 105.1 Ori + SRPHRSID 0.915 0.783 0.874 0.618 85.7 Ori + SCAA-NetHRSID 0.900 0.796 0.881 0.614 101.1 Or i + SCAA-Net（Leaky ReLU）HRSID 0.892 0.798 0.877 0.608 101.1 Ori + CIRCovHRSID 0.897 0.788 0.874 0.615 107.1 Or i + SRP+ SCAA-Net（Leaky ReLU）HRSID 0.917 0.809 0.890 0.627 81.6 Ori + SRP+ SCAA-NetHRSID 0.901 0.782 0.875 0.611 81.6 Ori + SRP+ SCAA-Net（Leaky ReLU）+ CIRConvHRSID 0.919 0.812 0.893 0.628 83.6 Ori + SRP+ SCAA-Net（Leaky ReLU）+ CIRConv+ NWDHRSID 0.924 0.823 0.904 0.629 83.6 Ori SSDD 0.898 0.937 0.957 0.564 105.1 Ori + SRPSSDD 0.935 0.935 0.971 0.591 85.7 Ori + SCAA-NetSSDD 0.925 0.896 0.959 0.557 101.1 Ori + CIRCovSSDD 0.924 0.929 0.964 0.569 107.1 Or i + SCAA-Net（Leaky ReLU）SSDD 0.933 0.939 0.969 0.573 101.1 Or i + SRP+ SCAA-Net（Leaky ReLU）SSDD 0.946 0.911 0.965 0.576 81.6 Ori + SRP+ SCAA-NetSSDD 0.921 0.935 0.965 0.572 81.6 Ori + SRP+ SCAA-Net（Leaky ReLU）+ CIRConvSSDD 0.946 0.945 0.971 0.590 83.6 Ori + SRP+ SCAA-Net（Leaky ReLU）+ CIRConv+ NWDSSDD 0.947 0.950 0.985 0.592 83.6
Table 4. Ablation experiments of backbone network
View tableTable 4. Ablation experiments of backbone network
Backbone Dataset P R mAP0.5 mAP0.5∶0.95 FLOPs /109 Polymeric structure HRSID 0.876 0.786 0.863 0.602 105.1 Conventional residual structure HRSID 0.854 0.770 0.842 0.589 78.9 Stepwise residual structure HRSID 0.882 0.786 0.859 0.591 130.5 Stepwise residual structure+PConv HRSID 0.887 0.784 0.863 0.598 79.4 Stepwise residual structure+DWConv， k=7 HRSID 0.916 0.796 0.872 0.610 103.5 Stepwise residual structure+DWConv， k=5 HRSID 0.912 0.792 0.870 0.599 93.2 Stepwise residual structure+DWConv， k=3 HRSID 0.914 0.789 0.873 0.612 85.7 SRP HRSID 0.915 0.783 0.874 0.618 85.7 Polymeric structure SSDD 0.898 0.937 0.957 0.564 105.1 Conventional residual structure SSDD 0.880 0.912 0.932 0.558 78.9 Stepwise residual structure SSDD 0.914 0.928 0.963 0.567 130.5 Stepwise residual structure+PConv SSDD 0.918 0.924 0.965 0.568 79.4 Stepwise residual structure+DWConv， k=7 SSDD 0.930 0.932 0.969 0.574 103.5 Stepwise residual structure+DWConv， k=5 SSDD 0.924 0.928 0.965 0.572 93.2 Stepwise residual structure+DWConv， k=3 SSDD 0.934 0.931 0.968 0.577 85.7 SRP SSDD 0.935 0.935 0.971 0.591 85.7
Table 5. Ablation experiments of feature fusion parts
View tableTable 5. Ablation experiments of feature fusion parts
Neck Dataset P R mAP0.5 mAP0.5∶0.95 Polymeric structure HRSID 0.876 0.786 0.863 0.602 Polymeric structure +CA HRSID 0.875 0.789 0.861 0.599 Polymeric structure +SA HRSID 0.872 0.791 0.865 0.604 Polymeric structure +CA+SA HRSID 0.886 0.794 0.872 0.608 SCAA-Net HRSID 0.900 0.796 0.881 0.614 SCAA-Net （Leaky ReLU） HRSID 0.892 0.798 0.877 0.608 Polymeric structure SSDD 0.898 0.937 0.957 0.564 Polymeric structure +CA SSDD 0.892 0.933 0.954 0.569 Polymeric structure +SA SSDD 0.894 0.926 0.956 0.566 Polymeric structure +CA+SA SSDD 0.913 0.915 0.958 0.559 SCAA-Net SSDD 0.925 0.896 0.959 0.557 SCAA-Net （Leaky ReLU） SSDD 0.933 0.939 0.969 0.573
Table 6. Comparison of proposed model and mainstream models
View tableTable 6. Comparison of proposed model and mainstream models
Model P R mAP Center Net 91.2 70.62 87.43 SSD 80.1 83.34 82.01 Faster R-CNN 78.4 84.10 80.60 YOLOv8 89.3 85.20 87.50 Ori 87.6 78.60 86.30 SRCDNet 92.4 82.30 90.40
Tools
Get Citation
Copy Citation Text
Wenxing Liu, Huilin Shan, Xingtao Wang, Jieru Liu, Ge Chen, Mengjiao Shan. SAR Ship Detection Method Based on Ladder Residual and Coordinate Information Recombination[J]. Acta Optica Sinica, 2025, 45(9): 0928001
Paper Information
Category: Remote Sensing and Sensors
Received: Nov. 11, 2024
Accepted: Feb. 25, 2025
Published Online: May. 19, 2025
The Author Email: Huilin Shan (shanhuilin@nuist.edu.cn)
CSTR:32393.14.AOS241731
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20