Fish Recognition Method for Underwater Video Based on Image Enhancement

Wenjing Yang; Ming Chen; Guofu Feng

doi:10.3788/LOP202158.2211002

Laser & Optoelectronics Progress, Volume. 58, Issue 22, 2211002(2021)

Fish Recognition Method for Underwater Video Based on Image Enhancement

Wenjing Yang, Ming Chen, and Guofu Feng^*

College of Information Technology, Key Laboratory of Fisheries Information, Ministry of Agriculture and Rural Affairs, Shanghai Ocean University, Shanghai 201306, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Flowchart of proposed method

Download full size

Fig. 2. Main flowchart of FUnIE-GAN

Download full size

Fig. 3. Network architecture of FUnIE-GAN. (a) Generator (five encoders-decoders designed according to principle of U-Net); (b) discriminator (Markov discriminator model)

Download full size

Fig. 4. YOLOv4 network structure

Download full size

Fig. 5. Renderings of different underwater image enhancement methods. (a) Original image; (b) method proposed by Ancuti et al.^[8]; (c) CLAHE^[19]; (d) GUM^[7]; (e) MSRCR^[16]; (f) UDCP^[20]; (g) DCP^[10]; (h) FUnIE-GAN; (i) proposed method

Download full size

Fig. 6. Variations in IOU and mAP with iterations. (a) Variation in IOU with iterations; (b) variation in mAP with iterations

Download full size

Fig. 7. Detection results. (a) Zaccop; (b) zaccop and opsarichthys; (c) pseudorasbora; (d) zaccop and pseudorasbora

Download full size

Table 1. Comparison of AG and UIQM of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

View table

Table 1. Comparison of AG and UIQM of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

Image	Fig. 5(b)		Fig. 5(c)		Fig. 5(d)		Fig. 5(e)
Image	AG	UIQM	AG	UIQM	AG	UIQM	AG	UIQM
1	2.517	3.585	4.369	3.635	3.636	3.721	3.988	3.912
2	3.413	3.575	3.611	3.507	2.785	3.533	3.315	4.546
3	4.211	3.562	3.759	3.552	2.971	3.601	3.968	4.950
4	4.130	2.290	3.550	2.121	2.764	3.349	3.125	3.042
Average	3.568	3.253	3.822	3.204	3.039	3.551	3.599	4.113

Table 2. Comparison of AG and UIQM of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

View table

Table 2. Comparison of AG and UIQM of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

Image	Fig. 5(f)		Fig. 5(g)		Fig. 5(h)		Fig. 5(l)
Image	AG	UIQM	AG	UIQM	AG	UIQM	AG	UIQM
1	2.917	4.199	3.096	3.752	1.839	2.832	3.944	4.490
2	2.034	4.110	2.271	3.511	2.992	2.843	3.792	4.467
3	2.861	4.284	2.821	3.595	2.946	3.069	3.883	4.501
4	2.545	2.548	2.609	2.008	1.864	1.536	3.768	4.317
Average	2.589	3.785	2.699	3.217	2.410	2.570	3.847	4.444

Table 3. Comparison of information entropy and UCIQE of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

View table

Table 3. Comparison of information entropy and UCIQE of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

Image	Fig. 5(b)		Fig. 5(c)		Fig. 5(d)		Fig. 5(e)
Image	Information entropy	UCIQE	Information entropy	UCIQE	Information entropy	UCIQE	Information entropy	UCIQE
1	6.996	0.774	6.358	0.791	5.862	0.707	7.063	0.311
2	6.851	0.364	5.736	0.385	5.371	0.372	6.684	0.304
3	6.935	0.487	6.044	0.502	5.609	0.434	6.513	0.289
4	6.751	0.682	5.941	0.688	5.426	0.655	6.071	0.326
Average	6.883	0.577	6.019	0.591	5.567	0.542	6.583	0.308

Table 4. Comparison of information entropy and UCIQE of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

View table

Table 4. Comparison of information entropy and UCIQE of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

Image	Fig. 5(f)		Fig. 5(g)		Fig. 5(h)		Fig. 5(l)
Image	Information entropy	UCIQE	Information entropy	UCIQE	Information entropy	UCIQE	Information entropy	UCIQE
1	6.901	0.328	6.925	0.218	6.975	0.275	7.431	0.793
2	6.938	0.325	6.930	0.233	7.166	0.399	7.117	0.440
3	7.128	0.345	7.063	0.270	7.083	0.236	7.219	0.511
4	7.007	0.312	6.976	0.212	7.096	0.310	7.177	0.694
Average	6.994	0.328	6.974	0.233	7.080	0.305	7.236	0.609

Table 5. Comparison of mAP of different target detection and image enhancement methods%

View table

Table 5. Comparison of mAP of different target detection and image enhancement methods%

Method	Fig. 5(a)	Fig. 5(b)	Fig. 5(c)	Fig. 5(d)	Fig. 5(e)	Fig. 5(f)	Fig. 5(g)	Fig. 5(h)	Fig. 5(l)
Faster-RCNN	65.08	67.86	68.57	69.78	68.28	68.36	69.65	68.34	71.47
YOLOv3	75.38	75.72	75.84	74.53	73.69	72.75	74.38	73.56	82.13
YOLOv4	82.56	84.33	84.69	84.12	83.71	83.78	84.54	83.89	89.59

Table 6. Comparison of different detection algorithms for targets in turbid waters
View table
Table 6. Comparison of different detection algorithms for targets in turbid waters
Method Precision Recall FPS /(frame·s^-1) Training time /h
Faster-RCNN 0.95 0.79 12 40
YOLOv3 0.90 0.76 79 37
YOLOv4 0.93 0.91 90 25

Table 7. mAP varying with iterations
View table
Table 7. mAP varying with iterations
Iterations mAP /%
1000 1.54
10000 75.70
20000 78.01
30000 80.23
40000 84.78
50000 89.59
60000 89.43

Tools

Get Citation

Copy Citation Text

Wenjing Yang, Ming Chen, Guofu Feng. Fish Recognition Method for Underwater Video Based on Image Enhancement[J]. Laser & Optoelectronics Progress, 2021, 58(22): 2211002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Imaging Systems

Received: Dec. 1, 2020

Accepted: Jan. 21, 2021

Published Online: Nov. 5, 2021

The Author Email: Guofu Feng (gffeng@shou.edu.cn)

DOI:10.3788/LOP202158.2211002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of AG and UIQM of method proposed by Ancuti et al.[8], CLAHE[19], GUM[7], and MSRCR[16]

Table 1. Comparison of AG and UIQM of method proposed by Ancuti et al.[8], CLAHE[19], GUM[7], and MSRCR[16]

Table 2. Comparison of AG and UIQM of UDCP[20],DCP[10],FUnIE-GAN, and proposed method

Table 2. Comparison of AG and UIQM of UDCP[20],DCP[10],FUnIE-GAN, and proposed method

Table 3. Comparison of information entropy and UCIQE of method proposed by Ancuti et al.[8], CLAHE[19], GUM[7], and MSRCR[16]

Table 3. Comparison of information entropy and UCIQE of method proposed by Ancuti et al.[8], CLAHE[19], GUM[7], and MSRCR[16]

Table 4. Comparison of information entropy and UCIQE of UDCP[20],DCP[10],FUnIE-GAN, and proposed method

Table 4. Comparison of information entropy and UCIQE of UDCP[20],DCP[10],FUnIE-GAN, and proposed method

Table 5. Comparison of mAP of different target detection and image enhancement methods%

Table 5. Comparison of mAP of different target detection and image enhancement methods%

Table 6. Comparison of different detection algorithms for targets in turbid waters

Table 6. Comparison of different detection algorithms for targets in turbid waters

Table 7. mAP varying with iterations

Table 7. mAP varying with iterations

Table 1. Comparison of AG and UIQM of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

Table 1. Comparison of AG and UIQM of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

Table 2. Comparison of AG and UIQM of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

Table 2. Comparison of AG and UIQM of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

Table 3. Comparison of information entropy and UCIQE of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

Table 3. Comparison of information entropy and UCIQE of method proposed by Ancuti et al.^[8], CLAHE^[19], GUM^[7], and MSRCR^[16]

Table 4. Comparison of information entropy and UCIQE of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method

Table 4. Comparison of information entropy and UCIQE of UDCP^[20],DCP^[10],FUnIE-GAN, and proposed method