Lightweight Underwater Optical Image Recognition Algorithm Based on YOLOv8

Algorithm	Information entropy	Mean	Standard deviation	Average gradient
ACE	7.52	0.36	0.19	459
DCP	7.51	0.53	0.34	419
HE	7.97	0.50	0.28	378
AHE	7.29	0.35	0.16	386
MSR	7.52	0.63	0.19	196
SSR	5.54	0.80	0.06	69
Laplace enhancement	6.92	0.29	0.13	226
Gamma transform	6.44	0.12	0.12	138

Table 2. Enhancement effect by ACE
View table
Table 2. Enhancement effect by ACE
Method AP mAP50 mAP50‒90
Echinus Starfish Holothurian Scallop
URPC 69.7 90.6 82.4 63.4 76.5 39.9
URPC+ACE enhanced 70.6 89.9 83.2 65.1 77.2 42.9

Table 3. Results of parameters for different detection models

View table

Table 3. Results of parameters for different detection models

Model	Parameters /10⁶	mAP50 /%	mAP50‒90 /%	FLOPs /10⁹	Model size /MB
YOLOv3-tiny	8.67	71.0	34.6	12.9	17.4
YOLOv5n	7.02	79.2	44.9	15.8	14.4
YOLOv7-tiny	6.02	74.9	38.4	13.2	12.3
YOLOv8n	3.00	76.7	42.9	8.1	6.3
YOLOv9c	25.30	79.9	45.6	102.3	51.6
RT-DETR	15.50	79.8	44.5	39.7	31.5
SCDDI-YOLOv8	2.38	77.3	41.4	7.5	5.1

Table 4. Detection effect of different models on the UWG dataset
View table
Table 4. Detection effect of different models on the UWG dataset
Model Parameters /10⁶ mAP50 /% mAP50‒90 /% FLOPs /10⁹ Model size /MB
YOLOv8n 3.00 69.7 44.3 8.1 6.3
RT-DETR 15.59 64.5 41.6 37.6 31.5
SCDDI-YOLOv8 2.38 71.5 44.5 7.5 5.1

Table 5. Performance comparison of different backbone networks

View table

Table 5. Performance comparison of different backbone networks

Backbone network	Parameters /10⁶	mAP50 /%	mAP50‒90 /%	FLOPs /10⁹	Model size /MB
SENetV2	3.00	76.6	43.5	12.9	6.3
MobileNetV3	5.65	77.3	43.6	10.7	11.7
ShuffleNetV2	2.79	75.4	41.7	7.4	5.9
EfficientNetV2	21.75	80.3	45.7	55.1	44.3
GhostNetV2	6.30	74.7	41.6	8.7	13.3
EfficientViT2	4.01	75.3	42.0	9.5	8.7
RepViT	4.12	76.8	42.9	11.8	8.7
SwinTransformer	29.90	73.4	39.8	402.1	60.5
PPHGNetV2	12.00	78.3	44.4	29.2	24.3

Table 6. Performance comparison of different neck networks
View table
Table 6. Performance comparison of different neck networks
Neck network Parameters /10⁶ mAP50 /% FLOPs /10⁹ Model size /MB
CCFM 1.96 75.6 6.7 4.2
Slim-Neck 2.79 76.3 7.4 5.9
BiFPN 2.78 77.4 8.1 5.8
RepGFPN 3.28 75.6 8.5 6.8
Gold-YOLO 8.06 76.4 17.6 16.6
ASF-YOLO 3.05 76.3 8.6 6.4

Table 7. Performance comparison of different samplers
View table
Table 7. Performance comparison of different samplers
Sampler Parameters /10⁶ mAP50 /% FLOPs /10⁹ Model size /MB
DySample 3.01 76.7 12.9 6.3
CARAFE 3.14 76.3 8.4 6.5

Table 8. Performance comparison of different detection heads
View table
Table 8. Performance comparison of different detection heads
Detection head Parameters /10⁶ mAP50 /% FLOPs /10⁹ Model size /MB
DyHead 2.50 77.0 7.8 5.3
Detect_FRM 87.90 75.8 76.3 176.2
Detect_FASFF 4.31 75.1 15.4 9.1
Detect_ASFF 4.37 76.1 10.4 9.0
RT-DETR_Decoder 9.48 71.8 16.7 19.2

Table 9. Comparison of loss functions

View table

Table 9. Comparison of loss functions

Loss function	AP				mAP50	mAP50‒90
Loss function	Echinus	Starfish	Holothurian	Scallop	mAP50	mAP50‒90
InnerMPDIoU	70.9	90.6	82.0	64.7	77.1	40.6
MPDIoU	71.8	90.3	81.1	63.3	76.6	40.5
InnerCIoU	71.5	90.0	82.6	63.7	77.0	40.6
InnerWIoU	58.9	77.6	63.2	39.9	59.9	28.3
InnerSIoU	70.3	90.5	81.9	63.2	76.5	40.2

Table 10. Results of the ablation experiments on the URPC2020 dataset

View table

Table 10. Results of the ablation experiments on the URPC2020 dataset

Method	SENetV2	CCFM	DyHead	DySample	InnerMPDIoU	Parameters /10⁶	mAP50 /%	mAP50‒90 /%	FLOPs /10⁹	Model size /MB
a						3.00	76.7	42.9	8.1	6.3
b	√					3.00	76.6	43.5	12.9	6.3
c		√				1.96	75.6	42.5	6.7	4.2
d			√			2.50	77.0	43.4	7.8	5.3
e				√		3.01	76.7	43.0	8.2	6.3
f					√	3.00	77.1	40.6	8.1	6.3
g	√	√				2.78	76.5	43.2	7.8	5.9
h	√	√	√			2.30	77.2	43.5	7.5	5.1
i	√	√	√	√		2.38	75.6	42.7	7.5	5.1
j	√	√	√	√	√	2.38	77.3	41.4	7.5	5.1

Tools

Get Citation

Copy Citation Text

Shun Cheng, Jianrong Li, Zhiqian Wang, Shaojin Liu, Muyuan Wang. Lightweight Underwater Optical Image Recognition Algorithm Based on YOLOv8[J]. Laser & Optoelectronics Progress, 2025, 62(4): 0437008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites