Lightweight Ship Detection Based on Optical Remote Sensing Images for Embedded Platform

Huiying Wang; Chunping Wang; Qiang Fu; Zishuo Han; Dongdong Zhang

doi:10.3788/AOS221689

Acta Optica Sinica, Volume. 43, Issue 12, 1212001(2023)

Lightweight Ship Detection Based on Optical Remote Sensing Images for Embedded Platform

Huiying Wang¹, Chunping Wang¹, Qiang Fu^1、*, Zishuo Han², and Dongdong Zhang¹

¹Department of Electronic and Optical Engineering, Shijiazhuang Campus of Army Engineering University, Shijiazhuang 050003, Hebei, China

²32356 Troops of the Chinese People's Liberation Army, Xining 710003, Qinghai, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. Structure diagram of STYOLO

Download full size

Fig. 2. Structure diagrams of ShuffleNet. (a)(b) ShuffleNet v1 building block; (c)(d) ShuffleNet v2 building block

Download full size

Fig. 3. Backbone network

Download full size

Fig. 4. GSConv module structure diagram

Download full size

Fig. 5. Characteristic diagrams of 10th layer of YOLOv5s. (a) Characteristic diagram generated by SC operation; (b) characteristic diagram generated by DSC operation; (c) characteristic diagram generated by GSConv operation

Download full size

Fig. 6. Feature enhancement network

Download full size

Fig. 7. Coordinate attention mechanism

Download full size

Fig. 8. Slim-neck structure diagram with coordinate attention mechanism

Download full size

Fig. 9. Clustering result of HRSC2016 dataset

Download full size

Fig. 10. Data enhancement process

Download full size

Fig. 11. Training curves of five detection models. (a) Training set box loss; (b) validation set box loss; (c) training object loss; (d) validation object loss; (e) precision; (f) recall; (g) mAP@0.5; (h) mAP@0.25

Download full size

Fig. 12. Comparison of detection effects with different detection models

Download full size

Table 1. Number of self-built dataset
View table
Table 1. Number of self-built dataset
Dataset Image Ship
Training set 880 2392
Test set 110 345
Validation set 110 250

Table 2. Effect of network lightweighting on the model
View table
Table 2. Effect of network lightweighting on the model
Model P /% R /% mAP@0.5 /% FLOPs/B Parameter/MB
YOLOv5s 90.25 76.80 87.64 15.8 7.01
STYOLO 84.12 78.36 85.52 6.0 2.66

Table 3. Comparison of detection capabilities of CA applied at different positions
View table
Table 3. Comparison of detection capabilities of CA applied at different positions
Application position mAP@0.5/% Parameter/MB
Baseline 85.52 2.66
Position 1 80.05 2.67
Position 2 90.46 2.68
Position 3 85.66 2.67

Table 4. Comparison of detection capabilities with different attention mechanisms
View table
Table 4. Comparison of detection capabilities with different attention mechanisms
Attention mechanism r
mAP@
0.5 /%
Parameter /MB
Baseline - 85.52 2.66
SE 8 86.01 2.72
16 86.25 2.68
32 87.46 2.67
CBAM 8 86.32 2.72
16 86.88 2.68
32 87.26 2.67
CA 8 89.32 2.75
16 88.68 2.71
32 90.46 2.68

Table 5. Comparison of detection capabilities with different transfer learning methods

View table

Table 5. Comparison of detection capabilities with different transfer learning methods

Training method	Dataset	mAP@0.5 /%
Cross-domain	COCO dataset+self-built dataset	90.46
In-domain	HRSC2016 dataset+self-built dataset	80.16
Cross-domain+in-domain	COCO dataset+HRSC2016 dataset+self-built dataset	94.33

Table 6. Comparison of detection performance with different models

View table

Table 6. Comparison of detection performance with different models

Model	mAP@0.5 /%	Parameter /MB	FLOPs /B	Desktop computer（FPS）/ （frame·s^-1）	Jetson Nano（FPS）/ （frame·s^-1）
YOLOv5n	69.72	1.77	4.1	136.7	68.4
YOLOv5s	91.63	7.01	15.8	95.4	46.5
ShuffleNetv2-YOLOv5s	90.14	5.54	11.4	150.6	75.5
MobileNetv3-YOLOv5s	87.06	3.54	6.3	120.8	60.4
STYOLO	94.33	2.68	6.2	200.6	102.8

Tools

Get Citation

Copy Citation Text

Huiying Wang, Chunping Wang, Qiang Fu, Zishuo Han, Dongdong Zhang. Lightweight Ship Detection Based on Optical Remote Sensing Images for Embedded Platform[J]. Acta Optica Sinica, 2023, 43(12): 1212001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Instrumentation, Measurement and Metrology

Received: Sep. 7, 2022

Accepted: Oct. 27, 2022

Published Online: Apr. 25, 2023

The Author Email: Fu Qiang (1245316750@qq.com)

DOI:10.3788/AOS221689

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Number of self-built dataset

Table 1. Number of self-built dataset

Table 2. Effect of network lightweighting on the model

Table 2. Effect of network lightweighting on the model

Table 3. Comparison of detection capabilities of CA applied at different positions

Table 3. Comparison of detection capabilities of CA applied at different positions

Table 4. Comparison of detection capabilities with different attention mechanisms

Table 4. Comparison of detection capabilities with different attention mechanisms

Table 5. Comparison of detection capabilities with different transfer learning methods

Table 5. Comparison of detection capabilities with different transfer learning methods

Table 6. Comparison of detection performance with different models

Table 6. Comparison of detection performance with different models