Ship Classification and Detection Method for Optical Remote Sensing Images Based on Improved YOLOv5s

Fig. 3. Feature pyramid network (FPN).(a) Traditional feature pyramid network; (b) path aggregation network; (c) multi-stage weighted fusion pyramid network

Download full size

Fig. 4. Samples in dataset

Download full size

Fig. 5. Detection effect of different methods

Download full size

Table 1. Experimental results under different K
View table
Table 1. Experimental results under different K
K mAP AP₁（Aircraft_carrier） AP₂（Warship） AP₃（Civilian_ship） AP₄（Submarine）
1 85.8 95.0 95.8 68.3 84.0
3 87.4 99.2 94.4 69.1 86.8
5 84.4 94.6 95.1 66.4 81.4

Table 2. Experimental results under different number and location of modules

View table

Table 2. Experimental results under different number and location of modules

Experiment No.	Number of modules	mAP	AP₁（Aircraft_carrier）	AP₂（Warship）	AP₃（Civilian_ship）	AP₄（Submarine）
1	2	86.1	92.8	95.8	70.9	84.7
2	3	87.3	95.2	95.7	59.3	81.5
3	4（Neck）	81.8	90.0	94.2	64.8	78.0
4	4（Backbone）	87.4	99.2	94.4	69.1	86.8
5	5	87.4	98.2	95.9	68.8	86.6

Table 3. Performance comparison of different detection methods
View table
Table 3. Performance comparison of different detection methods
Method mAP /% t /s Weight /MB
YOLOv3 84.6 0.2870 235
SSD 89.7 1.622 92.1
YOLOv5s 83.9 0.1861 13.7
EB-YOLOv5s 89.2 0.2070 62.7

Table 4. Comparison for average accuracy

View table

Table 4. Comparison for average accuracy

Method	mAP	AP₁（Aircraft_carrier）	AP₂（Warship）	AP₃（Civilian_ship）	AP₄（Submarine）
YOLOv5s	83.9	97.3	94.9	62.8	80.8
YOLOv5s+SE	85.9	97.8	96.0	65.3	84.7
YOLOv5s+ECA	87.4	99.2	94.4	69.1	86.8
YOLOv5s+BIFPN	86.5	94.0	94.3	66.9	90.8
EB-YOLOv5s	89.2	98.9	95.8	71.6	90.3

Table 5. Comparison for detection speed and network complexity
View table
Table 5. Comparison for detection speed and network complexity
Method t /s Weight /MB Layer Parameters
YOLOv5s 0.1861 13.7 224 7062001
YOLOv5s+SE 0.2216 14.1 256 7237281
YOLOv5s+ECA 0.1998 13.7 240 7062013
YOLOv5s+BIFPN 0.1996 15.7 236 8128517
EB-YOLOv5s 0.2067 62.7 252 8128529

Tools

Get Citation

Copy Citation Text

Qikai Zhou, Wei Zhang, Dongjin Li, Fu Niu. Ship Classification and Detection Method for Optical Remote Sensing Images Based on Improved YOLOv5s[J]. Laser & Optoelectronics Progress, 2022, 59(16): 1628008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites