Underwater bubbles recognition based on PCA feature extraction and elastic BP neural network

Yinbo Zhang; Sining Li; Peng Jiang; Jianfeng Sun

doi:10.3788/IRLA20200352

Journals >Infrared and Laser Engineering >Volume 50 >Issue 6 >Page 20200352 > Article

Infrared and Laser Engineering
Vol. 50, Issue 6, 20200352 (2021)

Underwater bubbles recognition based on PCA feature extraction and elastic BP neural network

Yinbo Zhang¹, Sining Li¹, Peng Jiang², and Jianfeng Sun^1、3

Author Affiliations

¹National Key Laboratory of Science and Technology on Tunable Laser, Institute of Opto-Electronic, Harbin Institute of Technology, Harbin 150001, China

²Science and Technology on Complex System Control and Intelligent Agent Cooperation Laboratory, Beijing 100074, China

³Harbin Institute of Technology (Beijing) Industrial Technology Innovation Research Institute Co., Ltd, Beijing 101312, China

show less

DOI: 10.3788/IRLA20200352 Cite this Article

Yinbo Zhang, Sining Li, Peng Jiang, Jianfeng Sun. Underwater bubbles recognition based on PCA feature extraction and elastic BP neural network[J]. Infrared and Laser Engineering, 2021, 50(6): 20200352 Copy Citation Text

show less

Experimental device for data acquisition of underwater bubbles; (b) Photo of the underwater lidar system (inside the metal box); (c) The laser runs through the bubbles without background light; (d) Air pump (voltage: 220 V, frequency: 50 Hz, power: 60 W, transmission volume: 50 L/min); (e) Bubbles plate (the diameter of the air bubbles is about 10-200 μm); (f) Valve: control airflow

Fig. 1. Experimental device for data acquisition of underwater bubbles; (b) Photo of the underwater lidar system (inside the metal box); (c) The laser runs through the bubbles without background light; (d) Air pump (voltage: 220 V, frequency: 50 Hz, power: 60 W, transmission volume: 50 L/min); (e) Bubbles plate (the diameter of the air bubbles is about 10-200 μm); (f) Valve: control airflow

Download full size | View in the Article

Fig. 2. Echo signals of bubbles

Download full size | View in the Article

Fig. 3. Score distribution of the first two principal components

Download full size | View in the Article

Fig. 4. Iterative convergence of different algorithms. (a) Elastic BP algorithm; (b) Adaptive and additional momentum BP algorithm

Download full size | View in the Article

Fig. 5. Underwater bubbles recognition process based on PCA and elastic BP neural network

Download full size | View in the Article

Fig. 6. Echo signals and recognition results under different conditions. (a) Different targets echo curves; (b) Low density bubbles recognition rate; (c) High density bubbles recognition rate

Download full size | View in the Article

Parameter	Value	Parameter	Value
Wavelength/nm	532	Receiving diameter/mm	80
Pulse width/ns	10	Emission diameter/mm	20
Pulse energy/mJ	10	Emission angle/mrad	1.7
Center distance/mm	100	Receiving angle/mrad	2.5

Table 1. System parameters

View in the Article

Number of hidden nodes	Training times	Convergence time/s
3	1451.71	3.43
6	1079.43	2.29
9	685.86	1.43
12	653.86	1.43

Table 2. Training of different hidden nodes

View in the Article

Number of eigenvalues	2			11			28
Cumulative contribution rate	39.7%			71.2%			90.6%
No bubbles(correct)	200	200	200	199	200	198	198	195	199
Bubbles(correct)	200	199	198	199	200	200	198	198	198
Glass plate(correct)	196	197	194	198	196	199	198	195	193
Recognition rate	99.3%	99.3%	98.7%	99.3%	99.3%	99.5%	99.0%	98.0%	99.0%
Average recognition rate	99.1%			99.4%			98.7%

Table 3. Recognition results under different cumulative contribution rates

View in the Article

Method	Recognition rate	Time(training+ recognition)/s
Adaptive and additional momentum BP	98.3%	48.94
PCA+Adaptive and additional momentum BP	96.6%	10.15
Elastic BP	98.8%	1.60
PCA+elastic BP	99.1%	1.36

Table 4. Recognition and contrast results of different algorithms

Download Citation

Save the article for my favorites

Paper Information