Opto-Electronic Engineering, Volume. 52, Issue 6, 250048(2025)
Ship fine-grained classification of ship targets driven by data and knowledge
Figures & Tables(12)
Fig. 4. Visualization results of selected channels in the shared feature map
Fig. 5. Examples of ship graph structures and adjacency matrix constructions. (a) Visualization of detection boxes for Arleigh Burke-class destroyer; (b) Visualization of detection boxes for Ticonderoga-class cruiser; (c) Graph structure and adjacency matrix for Arleigh Burke-class destroyer; (d) Graph structure and adjacency matrix for Ticonderoga-class cruiser
Fig. 6. Schematic diagram of graph convolution feature propagation integrating residual connections
Table 1. Forward propagation steps of graph convolutional network
View tableView in ArticleTable 1. Forward propagation steps of graph convolutional network
Step Operation 1 GCNConv1 2 ReLU 3 GCNConv2 4 MeanPooling 5 Softmax
Table 2. FGSCR-9 dataset division and class distribution
View tableView in ArticleTable 2. FGSCR-9 dataset division and class distribution
No. Name Train Validation Test 1 Nimitz-class aircraft carrier 84 26 281 2 Arleigh Burke-class destroyer 83 32 220 3 Ticonderoga-class cruiser 45 15 130 4 Type 45 destroyer 35 8 101 5 Osumi-class amphibious ship 27 3 82 6 Independence-class littoral combat ship 49 10 143 7 Medical ship 57 7 145 8 San Antonio-class amphibious ship 63 13 166 9 Wasp-class amphibious assault ship 22 3 90
Table 3. Comparison of accuracy of different models in ship fine-grained classification tasks
View tableView in ArticleTable 3. Comparison of accuracy of different models in ship fine-grained classification tasks
Model Accuracy/% ResNet50 85.61 VGG16 90.01 VIT 90.68 B-CNN 92.49 RA-CNN 94.48 DKSCN 97.28
Table 4. Comparison of accuracy and recall rates of different models in ship classification tasks
View tableView in ArticleTable 4. Comparison of accuracy and recall rates of different models in ship classification tasks
Ship FRCNN26 FRCNN9 DKSCN Pr/% Re/% Pr/% Re/% Pr/% Re/% Nimitz 81.21 100.00 83.38 100.00 97.91 100.00 Arleigh Burke 71.05 98.18 73.08 99.13 93.45 97.27 Ticonderoga 69.68 83.08 74.83 86.92 91.87 87.60 Type45 99.01 99.01 98.10 97.17 100.00 97.03 Osumi 98.80 100.00 98.75 100.00 98.80 100.00 Independence 87.73 100.00 94.44 100.00 100.00 99.30 Medical ship 70.30 97.93 75.38 99.34 100.00 100.00 San Antonio 81.41 97.59 84.13 96.36 97.60 98.19 Wasp 84.16 94.44 83.18 98.89 100.00 93.22
Table 5. List of detected object location and category information
View tableView in ArticleTable 5. List of detected object location and category information
Index (category) BBox (x1,y1,x2,y2) Confidence Class probabilities (first 7 dimensions) 1 (Arleigh Burke) [181.51, 13.52, 288.78, 321.61] 0.68 [0.04481, 0.0, 0.00005, 0.00006, 0.0, 0.0, 0.0…] 2 (Ticonderoga cruiser) [180.88, 10.81, 290.45, 320.48] 0.77 [0.00782, 0.0, 0.00023, 0.00030, 0.00001, 0.0, 0.0…] 3 (Bow2) [242.10, 15.85, 277.29, 61.80] 0.77 [0.09948, 0.00031, 0.0, 0.0, 0.1063, 0.00001, 0.0…] 4 (VLS_front) [251.40, 84.27, 264.39, 99.47] 0.52 [0.44401, 0.0, 0.0, 0.0, 0.00269, 0.52351, 0.01966…] 5 (VLS_rear) [206.43, 221.27, 221.18, 239.50] 0.73 [0.09814, 0.0, 0.0, 0.0, 0.00002, 0.73007, 0.17059…] 5 (VLS_rear) [0.05, 150.56, 17.60, 194.93] 0.63 [0.25756, 0.0, 0.0, 0.0, 0.0, 0.63424, 0.00803...] 6 (Helideck) [183.44, 257.51, 226.35, 306.75] 0.91 [0.07547, 0.00001, 0.0, 0.0, 0.00117, 0.00001, 0.00001…]
Tools
Get Citation
Copy Citation Text
Jiasheng Guo, Jun Liu, Lan He, Pan Jiang, Anke Xue, Yu Gu, Li Han, Jie Zhang. Ship fine-grained classification of ship targets driven by data and knowledge[J]. Opto-Electronic Engineering, 2025, 52(6): 250048
Paper Information
Category: Article
Received: Feb. 22, 2025
Accepted: May. 22, 2025
Published Online: Sep. 3, 2025
The Author Email: Jun Liu (刘俊)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20