Laser & Optoelectronics Progress, Volume. 60, Issue 6, 0610005(2023)
Application of a Lightweight Convolutional Neural Network in Ship Classification
Figures & Tables(13)
Fig. 2. Ghost bottleneck module. (a) Bottleneck module with a step size of 1; (b) bottleneck module with a step size of 2
Fig. 5. Asymmetric Ghost bottleneck module. (a) Bottleneck module with a step size of 1; (b) bottleneck module with a step size of 2
Table 1. Detail parameters of AGNet
View tableTable 1. Detail parameters of AGNet
Input Operator Exp Out SE Stride 2242×3 Conv 2d 3×3 - 8 - 2 1122×8 AG-bneck 8 8 - 1 1122×8 AG-bneck 24 12 - 2 562×12 AG-bneck 36 12 - 1 562×12 AG-bneck 36 20 1 2 282×20 AG-bneck 60 20 1 1 282×20 AG-bneck 120 40 - 2 142×40 AG-bneck 100 40 - 1 142×40 AG-bneck 92 40 - 1 142×40 AG-bneck 92 40 - 1 142×40 AG-bneck 240 56 1 1 142×56 AG-bneck 336 56 1 1 142×56 AG-bneck 336 80 1 2 72×80 AG-bneck 480 80 - 1 72×80 AG-bneck 480 80 1 1 72×80 AG-bneck 480 80 - 1 72×80 AG-bneck 480 80 1 1 72×80 Conv 2d 1×1 - 480 - 1 72×480 AvgPool 7×7 - - - - 12×480 FC - classes - -
Table 2. Information table of sample set
View tableTable 2. Information table of sample set
ID Class Train_set Test_set 1 5 470 104 2 33 2938 636 3 59 4408 986
Table 3. Comparison of each network evaluation indicators
View tableTable 3. Comparison of each network evaluation indicators
Model Multiply-accumulate operations /106 Parameters /106 Accuracy /% Speed /（frame·s-1） AGNet 49.35 0.72 93.87 47.76 AGNet-large 49.97 1.36 92.50 41.79 GhostNet-50 45.84 1.35 90.94 43.61 GhostNet-50-small 45.22 0.71 92.03 50.13
Table 4. Comparison of each network performance
View tableTable 4. Comparison of each network performance
Model Multiply-accumulate operations /106 Parameters /106 Accuracy /% Speed /（frame·s-1） AGNet 49.35 0.73 93.55 46.93 AGNet-large 49.97 1.39 90.32 41.24 GhostNet-50 45.84 1.38 87.60 42.97 GhostNet-50-small 45.22 0.72 92.94 49.64
Table 5. Comparison results of different classification networks
View tableTable 5. Comparison results of different classification networks
Model Multiply-accumulate operations /106 Parameters /106 Accuracy /% MobileNetv3-small-100［ 24 ］70.06 2.54 87.81 ResNet18［ 1 ］1821.66 11.69 89.84 DPN-68b［ 26 ］2338.55 12.61 91.87 RegNetx-002［ 27 ］200.69 2.68 89.53 GhostNet-50［ 23 ］45.84 1.35 92.50 AGNet 49.35 0.72 93.87
Tools
Get Citation
Copy Citation Text
Wenliang Wang, Xiaodi Yang, Boya Zhang, Jishun Ma, Peng Zeng, Peng Han. Application of a Lightweight Convolutional Neural Network in Ship Classification[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0610005
Paper Information
Category: Image Processing
Received: Nov. 23, 2021
Accepted: Jan. 17, 2022
Published Online: Mar. 7, 2023
The Author Email: Xiaodi Yang (1755018902@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20