Underwater Target Detection Algorithm Based on Automatic Color Level and Bidirectional Feature Fusion

Ting Yang; Wuqi Gao; Peng Wang; Xiaoyan Li; Lü Zhigang; Ruohai Di

doi:10.3788/LOP213139

Laser & Optoelectronics Progress, Volume. 60, Issue 6, 0610012(2023)

Underwater Target Detection Algorithm Based on Automatic Color Level and Bidirectional Feature Fusion

Ting Yang¹, Wuqi Gao², Peng Wang^3、*, Xiaoyan Li⁴, Lü Zhigang⁴, and Ruohai Di⁴

Author Affiliations

¹School of Ordnance Science and Technology, Xi'an Technological University, Xi'an 710021, Shaanxi, China

²School of Computer Science and Technology, Xi'an Technological University, Xi'an 710021, Shaanxi, China

³Development Planning Office, Xi'an Technological University, Xi'an 710021, Shaanxi, China

⁴School of Electronic Information Engineering, Xi'an Technological University, Xi'an 710021, Shaanxi, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Basic structure of Faster R-CNN

Download full size

Fig. 2. Overall network structure of the proposed algorithm

Download full size

Fig. 3. Histograms of original image and enhanced image. (a) Original image; (b) enhanced image

Download full size

Fig. 4. Structure of PAFPN

Download full size

Fig. 5. Structure of bottom-up module

Download full size

Fig. 6. Partial data samples

Download full size

Fig. 7. Dataset analysis. (a) Aspect ratio of the real box; (b) size ratio of the real box in the original image

Download full size

Fig. 8. Results of image enhancement obtained by different algorithms

Download full size

Fig. 9. PR curves of different improved algorithms

Download full size

Fig. 10. Loss function curve of the proposed network

Download full size

Fig. 11. Comparison of detection results of proposed algorithm before and after improvement

Download full size

Fig. 12. Comparison of detection effects for blurred image and clear image

Download full size

Table 1. Evaluation of image enhancement results

View table

Table 1. Evaluation of image enhancement results

Image	Algorithm	NIQE	ILNIQE	UCIQE
1	Src	7.1606	38.2495	0.3061
	MSR	6.8799	39.6676	0.2722
	MSRCP	3.7418	27.1790	0.4390
	Proposed algorithm	6.3834	29.2617	0.2722
2	Src	12.9428	42.7309	0.2785
	MSR	6.8799	39.6676	0.3636
	MSRCP	8.0962	46.2339	0.4708
	Proposed algorithm	10.3499	32.5586	0.4609
3	Src	10.4168	33.5792	0.3585
	MSR	9.1082	46.0194	0.2029
	MSRCP	7.0049	30.098	0.4275
	Proposed algorithm	8.8039	27.8706	0.4276
4	Src	5.7975	31.2044	0.2955
	MSR	5.6905	38.2397	0.2566
	MSRCP	4.1026	34.4918	0.4275
	Proposed algorithm	3.8041	27.3969	0.4668

Table 2. Comparison of defferent improved detection algorithms

View table

Table 2. Comparison of defferent improved detection algorithms

Algorithm	Input image	Backbone	Feature enhancement	ROI	Bbox	Loss function	Epoch
Faster R-CNN	Src	VGG16	None	ROI pooling	NMS（T+V）	L1Loss	None
Baseline	Src	ResNet50	FPN	ROI align	NMS（T+V）	L1Loss	60
Method 1	Src	ResNet50	FPN	ROI align	NMS（T+V）	IoULoss	60
Method 2	Src	ResNet50	FPN	ROI align	NMS（T+V）	FocalLoss	60
Method 3	Src	ResNet50	FPN	ROI align	NMS（T+V）	Gaussian FocalLoss	60
Our-1	IE	ResNet50	FPN	ROI align	NMS（T+V）	FocalLoss	60
Our-2	IE	ResNet50	FPN	ROI align	Soft-NMS（V）	FocalLoss	60
Our-3	IE	ResNet50	FPN	ROI align	Soft-NMS（T+V）	FocalLoss	60
Our-4	IE	ResNet50	PAFPN	ROI align	Soft-NMS（T+V）	FocalLoss	60

Table 3. AP and AR of different improved detection algorithms

View table

Table 3. AP and AR of different improved detection algorithms

Method	AP /%	AP^0.50 /%	AP^0.75 /%	AP^S /%	AP^M /%	AP^L /%	AR /%	AR^S /%	AR^M /%	AR^L /%
Baseline	54.2	88.8	59.5	27.1	48.2	60.1	62.1	40.6	57.2	67.0
Method 1	54.9	88.7	61.1	27.4	49.0	60.8	62.9	40.3	58.0	67.8
Method 2	56.0	90.2	62.3	28.5	50.0	61.9	63.4	41.1	58.3	68.3
Method 3	55.5	90.0	61.7	28.9	49.4	61.5	63.0	40.7	58.0	68.1
Our-1	56.7	90.7	63.1	29.0	50.5	62.6	64.0	41.4	58.8	69.1
Our-2	57.9	90.8	66.1	30.6	51.7	63.6	70.5	49.4	65.5	74.8
Our-3	59.1	90.8	67.9	32.4	52.7	65.1	70.3	49.4	65.0	75.0
Our-4	59.7	91.2	68.6	33.3	53.1	67.5	70.5	49.7	65.1	75.3

Tools

Get Citation

Copy Citation Text

Ting Yang, Wuqi Gao, Peng Wang, Xiaoyan Li, Lü Zhigang, Ruohai Di. Underwater Target Detection Algorithm Based on Automatic Color Level and Bidirectional Feature Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0610012

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Dec. 2, 2021

Accepted: Jan. 21, 2022

Published Online: Mar. 16, 2023

The Author Email: Peng Wang (wp_xatu@163.com)

DOI:10.3788/LOP213139

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology