Marine Creature Detection Based on Sample Iterative Fusion

Fig. 7. Detection results of the proposed method on Taiwan fish datase. (a) Abudefduf lorenzi; (b) Acanthurus blochii; (c) Abudefduf bengalensis; (d) Acanthurus nigricauda; (e) Acanthurus japonicus; (f) Acanthuridae

Download full size

Fig. 8. Detection result comparison of different methods. (a) Original image; (b) YOLOv4; (c) RFB; (d) proposed method

Download full size

Table 1. Number of samples on URPC2018 dataset
View table
Table 1. Number of samples on URPC2018 dataset
Sample Total Train Test
Echinus 453 2672 622
Scallop 343 783 232
Starfish 463 9271 2337

Table 2. Number of samples on Taiwan fish dataset

View table

Table 2. Number of samples on Taiwan fish dataset

Sample	Total	Train	Test
Abudefduf bengalensis	453	317	136
Abudefduf lorenzi	463	324	139
Abudefduf septemfasciatus	343	240	103
Abudefduf sexfasciatus	208	146	62
Abudefduf vaigiensis	469	328	141
Acanthuridae	284	199	85
Acanthurus blochii	250	175	75
Acanthurus dussumieri	358	251	107
Acanthurus japonicus	349	244	105
Acanthurus lineatus	207	145	62
Acanthurus nigricauda	470	329	141
Acanthurus nigrofuscus	271	190	81
Acanthurus olivaceus	249	174	75
Acanthurus olivaceus	366	256	110
Acanthurus pyroferus	469	328	141
Acanthurus thompsoni	285	200	86
Acanthurus xanthopterus	493	345	148
Aeoliscus strigatus	320	224	96

Table 3. Ablation test

View table

Table 3. Ablation test

Iterative fusion	L_water	Dilated_Block	Deconv_Block	AP /%			mAP /%	Speed /（frame·s^-1）
Iterative fusion	L_water	Dilated_Block	Deconv_Block	Echinus	Scallop	Starfish	mAP /%	Speed /（frame·s^-1）
				81.08	89.65	89.35	86.69	36.33
√				86.30	90.24	89.82	88.79	49.31
	√			90.35	90.84	89.65	90.28	39.97
		√		84.71	88.63	90.21	87.85	35.89
			√	83.66	89.01	89.38	87.35	32.38
		√	√	85.42	88.34	90.17	88.31	30.15
√	√	√	√	92.24	90.37	91.46	91.36	49.30

Table 4. Detection accuracy of different methods on URPC dataset

View table

Table 4. Detection accuracy of different methods on URPC dataset

Sample	Faster R-CNN	SSD	YOLOv3	YOLOv4	RoIMix	RFB	Efficientdet	Proposed method
mAP	66.24	80.79	81.66	91.06	92.48	72.84	86.19	91.36
Echinus	47.32	76.23	85.39	91.77	93.73	53.71	87.21	92.24
Scallop	75.23	79.77	85.82	92.82	93.67	80.39	81.01	91.46
Starfish	76.16	86.38	73.76	88.58	90.04	84.42	90.36	90.37

Table 5. Detection accuracy of different methods on Taiwan fish dataset

View table

Table 5. Detection accuracy of different methods on Taiwan fish dataset

Sample	Faster R-CNN	SSD	YOLOv3	YOLOv4	RoIMix	RFB	Efficientdet	Proposed method
mAP	25.99	73.54	73.02	85.55	88.93	92.62	72.15	90.27
Abudefduf bengalensis	57.99	73.71	52.90	87.38	99.13	98.20	72.53	99.36
Abudefduf lorenzi	57.99	99.29	94.74	99.95	99.98	99.60	62.02	100.00
Abudefduf septemfasciatus	2.71	88.48	86.62	90.75	96.43	99.19	79.52	100.00
Abudefduf sexfasciatus	55.42	77.05	88.23	93.18	92.72	99.25	83.36	84.94
Abudefduf vaigiensis	19.61	56.37	72.01	67.02	80.98	90.42	77.43	85.54
Acanthuridae	1.42	43.24	60.00	62.55	88.13	76.84	44.91	46.97
Acanthurus blochii	6.29	88.12	58.63	82.62	99.95	98.42	89.46	100.00
Acanthurus dussumieri	0.17	41.92	38.81	89.03	36.49	82.72	64.31	85.97
Acanthurus japonicus	4.30	62.00	73.54	90.65	95.46	94.97	75.96	96.15
Acanthurus lineatus	0.68	82.03	67.44	45.57	86.65	95.55	62.82	90.97
Acanthurus nigricauda	4.80	93.78	71.37	99.09	98.67	95.77	94.41	100.00
Acanthurus nigrofuscus	8.83	60.78	63.18	82.07	85.69	86.38	41.43	69.47
Acanthurus olivaceus	13.49	74.77	73.31	86.43	84.98	95.23	58.84	81.56
Acanthurus olivaceus	44.30	85.18	81.39	99.39	99.97	94.65	80.20	100.00
Acanthurus pyroferus	58.40	82.82	85.16	94.89	94.61	97.34	89.49	96.56
Acanthurus thompsoni	35.01	81.22	83.61	89.19	100.00	97.00	67.69	100.00
Acanthurus xanthopterus	55.29	45.50	74.88	88.12	69.78	66.18	72.94	87.42
Aeoliscus strigatus	41.19	87.42	88.54	91.98	91.13	99.46	81.37	100.00

Table 6. Comparison of detection speed of different methods
View table
Table 6. Comparison of detection speed of different methods
Parameter Faster R-CNN SSD YOLOv3 YOLOv4 RoIMix RFB Efficientdet Proposed method
Speed /（frame·s^-1） 23.64 40.80 30.23 36.33 23.70 10.38 18.54 49.30

Tools

Get Citation

Copy Citation Text

Lidong Wu, Zongju Peng, Xin Li, Tao Su, Fen Chen, Xiaodong Wang. Marine Creature Detection Based on Sample Iterative Fusion[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0220001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites