An adaptive deep-sea environment learning four-quadrant azimuth measurement system based on artificial neural network and training method

Abstract

The application belongs to the underwater detection field, and particularly relates to a self-adaptive deep-sea environment learning four-quadrant azimuth measurement system and a training method based on an artificial neural network, the system comprising: a four-quadrant detection module, a deep-sea environment sensing module, and an information processing module; the four-quadrant detection module converts the energy distribution of the projection light spot of the guide light field on the surface of the detector into an analog voltage signal, the deep-sea environment sensing module converts the deep-sea water quality, flow rate, and flow direction into a voltage signal, the information processing module converts the voltage signal into a digital signal, calculates the azimuth angle through an algorithm, and transmits the azimuth angle to a control system. The application solves the influence of the non-uniform, nonlinear, and unstable factors of the underwater environment on the accuracy of the four-quadrant azimuth measurement, and improves the accuracy of the deep-sea optical guidance based on the four-quadrant azimuth measurement.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of underwater detection, and particularly relates to a four-quadrant azimuth measurement system and a training method based on an artificial neural network and adaptive deep-sea environment learning. BACKGROUND

[0002] An underwater unmanned vehicle (UUV) is an unmanned and autonomously navigated and controlled underwater vehicle, which is widely applied to the fields of marine information sensing, automatic cruising, underwater intelligent weapons, etc. In order to improve the working efficiency and the range of the UUV and meet the needs of marine exploration and reconnaissance, the UUV autonomous docking for energy supply and data interaction becomes an important research topic.

[0003] Guiding technology is the key to autonomous docking, and the optical guiding technology based on four-quadrant azimuth measurement has the advantages of fast speed, high precision and good concealment. However, unlike on land, in the complex and changeable deep-sea environment, the guiding light field is non-uniform, nonlinear and unstable, which reduces the measurement precision and deteriorates the stability. SUMMARY

[0004] To solve the above technical problems, the application provides a four-quadrant azimuth measurement system and a training method based on an artificial neural network and adaptive deep-sea environment learning.

[0005] The four-quadrant azimuth measurement system based on an artificial neural network and adaptive deep-sea environment learning comprises a four-quadrant detection module, a deep-sea environment perception module and an information processing module.

[0006] The four-quadrant detection module comprises a four-quadrant photoelectric detector and an analog circuit. The guiding light field is received by the detector receiving lens to form a light spot with a certain size on the photosensitive surface of the photoelectric detector, and the analog circuit converts the energy distribution of the light spot on the photosensitive surface into a voltage signal.

[0007] The deep-sea environment perception module comprises a water quality sensor, a flow rate sensor and a flow direction sensor. The deep-sea water quality is converted into a voltage signal by the water quality sensor, the deep-sea water flow is converted into a voltage signal by the flow rate sensor, and the flow direction of the deep-sea water flow is converted into a voltage signal by the flow direction sensor.

[0008] The information processing module collects the voltage signals generated by the four-quadrant detection module and the deep-sea environment perception module through AD acquisition and converts them into digital signals. The digital signals are processed by using an artificial neural network algorithm to obtain the azimuth data of the current light field, and the obtained azimuth data are transmitted to the control system of the vehicle through a communication interface.

[0009] Preferably, the voltage signal is converted into a digital signal, and the conversion specifically comprises:

[0010] Data=V / Vref *(2 N -1)

[0011] wherein, Data represents the digital signal obtained by conversion, V represents the voltage signal to be converted, V ref represents the reference voltage of the analog-digital converter, and N represents the conversion bit number of the analog-digital converter.

[0012] Preferably, the artificial neural network algorithm is adopted to process the digital signal to obtain the azimuth angle data of the current light field, which is represented as:

[0013] Step one: a three-layer neural network structure is constructed by Matlab, the three-layer neural network structure comprises an input layer, a hidden layer and an output layer; the input layer is composed of seven neuron nodes corresponding to the four-way voltage signals of the four-quadrant photoelectric detector of the measurement system and the digital signals converted from the voltage signals of the water quality sensor, the flow rate sensor and the flow direction sensor; the hidden layer corresponds to the seven neuron nodes; and the output layer comprises one neuron node corresponding to the azimuth angle data of the light field.

[0014] Step two: the seven neuron nodes of the input layer process the digital signals converted from the voltage signals of the four-quadrant photoelectric detector, the water quality sensor, the flow rate sensor and the flow direction sensor to obtain digital signals with weights; each neuron is connected with the next layer of neurons and signal transmission is performed.

[0015] Step three: the seven neuron nodes of the hidden layer generate bias and sum for the digital signals transmitted by the input layer and transmit the signals through an activation function.

[0016] Step four: the neuron node of the output layer obtains the azimuth angle data of the light field according to the weights and bias of the digital signals.

[0017] Further, the azimuth angle data of the light field is obtained according to the weights and bias of the digital signals, which is represented as:

[0018]

[0019] wherein, y represents the azimuth angle data, w ij represents the weight of the neuron connection between the input layer and the hidden layer, g() represents the activation function of the hidden layer, X i represents the four-way voltage signals of the four-quadrant photoelectric detector of the measurement system and the digital signals converted from the voltage signals of the water quality sensor and the flow rate sensor, b j represents the bias of the hidden layer neuron, N represents the number of input neurons, M represents the number of hidden layer neurons, W j represents the weight of the neuron connection between the hidden layer and the output layer, and f() represents the activation function of the output layer.

[0020] A training method of an adaptive deep-sea environment learning four-quadrant azimuth angle measurement system based on an artificial neural network, comprising the following steps:

[0021] S1: in the deep-sea environment, using blue-green LED light source to generate target guide light field, fixing the system on a two-dimensional angular displacement platform, connecting all power supply and signal cables;

[0022] S2: adjusting the two-dimensional angular displacement platform to make the system face the light source, saving the 7-way sensor data uploaded by the test system through the upper computer software;

[0023] S3: continuously deflecting the displacement platform by a certain angle, saving each group of data and recording the angle value;

[0024] S4: changing the water quality and the fluctuation amplitude of water, repeating S3 to obtain a certain number of sample data;

[0025] S5: dividing the sample data into a training set and a test set according to a 9:1 ratio, inputting the sample data in the training set into the system for system parameter adjustment;

[0026] S6: inputting the sample data in the test set into the system after parameter adjustment, testing the accuracy of the system, if the measurement accuracy of the system is less than 95%, modifying the system parameters through a genetic optimization algorithm until the measurement accuracy of the system is greater than 95%, and obtaining the optimal system.

[0027] The beneficial effects of the present application are:

[0028] 1. The voltage signal generated by simulating the deep-sea environment guide light field is realized by the artificial neural network algorithm to realize the angle measurement of the light field, and the present application solves the influence of the non-uniform, nonlinear and unstable factors of the underwater environment on the accuracy of the four-quadrant azimuth angle measurement, and improves the accuracy of the deep-sea four-quadrant azimuth angle measurement optical guide.

[0029] 2. The sample data is obtained by building a data acquisition platform in the deep-sea environment, and the system is adaptively learned through the sample data, so that the system has the deep-sea environment adaptive learning ability. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 is the measurement system structure diagram of the present application;

[0031] Figure 2 is the artificial neural network structure diagram of the present application;

[0032] Figure 3 is the building schematic diagram of the underwater test system of the present application. DETAILED DESCRIPTION

[0033] Clearly, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present application.

[0034] An adaptive deep-sea environment learning four-quadrant azimuth measurement system based on artificial neural network, as shown in Figure 1 The four-quadrant detection module, the deep-sea environment perception module, and the information processing module are included.

[0035] The four-quadrant detection module includes a four-quadrant photodetector and an analog circuit. The four-quadrant photodetector guides a light field to form a light spot with a certain size through a detector receiving lens and irradiate the light spot on the photosensitive surface of the photodetector. The analog circuit converts the energy distribution of the light spot on the photosensitive surface into a voltage signal.

[0036] The deep-sea environment perception module includes a water quality sensor, a flow rate sensor, and a flow direction sensor. The deep-sea water quality is converted into a voltage signal by the water quality sensor, the deep-sea water flow is converted into a voltage signal by the flow rate sensor, and the flow direction of the deep-sea water flow is converted into a voltage signal by the flow direction sensor.

[0037] The information processing module collects the voltage signals generated by the four-quadrant detection module and the deep-sea environment perception module through an AD acquisition circuit and converts the voltage signals into digital signals. The artificial neural network algorithm is used to process the digital signals to obtain the azimuth data of the current light field. The obtained azimuth data is transmitted to the control system of the vehicle through a communication interface.

[0038] Figure 1 The system block diagram of the design is shown in the figure, which records the composition of the design. The light power is converted into a voltage signal by the four-quadrant detector module and output to the information processing module. The water quality, speed, and direction of the seawater are converted into voltage signals by the deep-sea perception module and output to the information processing module. The analog signals are converted into digital signals by the information processing module and the AD acquisition. Then, the signal processing is performed through the artificial neural network algorithm. Finally, the results are output through the communication interface.

[0039] The four-quadrant detection module mainly completes photoelectric conversion function; the guide light field passes through a detector receiving lens to form a light spot with a certain size to irradiate on the photosensitive detector, and the area distribution of the light spot on the four quadrants is proportional to the output voltage of each quadrant; when the vehicle is directly opposite the guide light field, the projection area of the light spot on each quadrant is consistent, and the output voltages of the four quadrants are equal; when the vehicle has an angle with the guide light field, the projection area of the light spot on each quadrant changes, and the output voltages of the four quadrants are not equal. The voltages of the four quadrants are transmitted to the information processing system, and the azimuth of the vehicle and the light field is calculated.

[0040] The deep-sea environment sensing module mainly completes deep-sea water quality measurement, flow speed and direction measurement. Since the deep-sea guide light field is affected by the quality of seawater, flow speed and direction, by measuring the parameters of the deep-sea environment and transmitting the data to the information processing system for intelligent processing, the function of self-adaptive detection of the deep-sea environment is realized

[0041] Preferably, the voltage signal is converted into a digital signal, specifically including:

[0042] Data = V / V ref *(2 N -1)

[0043] Wherein, Data represents the digital signal obtained after conversion, V represents the voltage signal to be converted, V ref represents the reference voltage of the analog-to-digital converter, and N represents the conversion bit number of the analog-to-digital converter.

[0044] Preferably, the artificial neural network algorithm is used to process the digital signal to obtain the azimuth data of the current light field, which is represented as:

[0045] Step one: construct a three-layer neural network structure through Matlab, the three-layer neural network structure includes: input layer, hidden layer, output layer; the input layer is composed of seven neuron nodes, corresponding to the four-way voltage signals of the four-quadrant photoelectric detector of the measurement system and the digital signals converted from the voltage signals of the water quality sensor, flow speed sensor and flow direction sensor, the hidden layer corresponds to seven neuron nodes, and the output layer includes one neuron node, corresponding to the azimuth data of the light field;

[0046] Step two: the seven neuron nodes of the input layer process the digital signals converted from the voltage signals of the four-quadrant photoelectric detector, the water quality sensor, the flow speed sensor and the flow direction sensor, and obtain digital signals with weights, and each neuron is connected with the next layer of neurons and transmits signals;

[0047] Step three: the seven neuron nodes of the hidden layer generate bias to the digital signals transmitted by the input layer and sum them up, and transmit the signals through the activation function;

[0048] Step four: the neuron node of the output layer obtains the azimuth data of the light field according to the weight and bias of the digital signal.

[0049] An artificial neural network is a mathematical model algorithm that simulates the behavior characteristics of animal neural networks for distributed parallel information processing. The model is composed of a large number of nodes (or neurons) connected to each other, each node represents a specific output function (activation function), the connection between two nodes represents the weight Wi of the signal passing through the connection, and the output node has a bias Bi. The input signal is obtained by weighting Wi and bias Bi to get the input signal of the activation function, and then the signal is transmitted to the next neuron through the activation function.

[0050] According to the complexity of the system design, the neural network algorithm is designed as a 3-layer network structure, including an input layer, a hidden layer and an output layer. The input layer includes 7 neurons, corresponding to the four voltage signals of the four-quadrant detector, the water quality measurement signal, the water speed measurement signal and the direction measurement signal. The number of neuron nodes in the hidden layer has an important influence on the prediction ability of the model. If the number of nodes is too small, the model does not have generalization ability; if the number of nodes is too large, it is easy to fall into local optimization, and even overfitting. Therefore, this parameter is determined using a genetic optimization algorithm. The output layer only includes one neuron node, which is the predicted azimuth.

[0051] Further, the azimuth data of the light field is obtained according to the weight and bias of the digital signal, which is represented as:

[0052]

[0053] where y represents the azimuth data, w ij represents the weight of the connection between the input layer and the hidden layer neurons, g() represents the activation function of the hidden layer, X i represents the four voltage signals of the four-quadrant photodetector of the measurement system and the voltage signal converted from the water quality sensor and the water wave sensor, b j represents the bias of the hidden layer neurons, N represents the number of input neurons, M represents the number of hidden layer neurons, W j represents the weight of the connection between the hidden layer and the output layer neurons, f() represents the activation function of the output layer.

[0054] A training method of an adaptive deep-sea environment learning four-quadrant azimuth measurement system based on an artificial neural network, comprising the following steps:

[0055] S1: Use blue-green LED light source to generate target guide light field, fix the system on a two-dimensional angular displacement platform, and connect all power and signal cables;

[0056] S2: adjust the two-dimensional angular displacement platform to make the system directly face the light source, and save the 7-way sensor data uploaded by the test system through the upper computer software;

[0057] S3: continuously deflect the displacement platform by a certain angle, save each group of data and record the angle value;

[0058] S4: change the water quality and the fluctuation amplitude of the water, repeat S3, and obtain a certain number of sample data;

[0059] S5: divide the sample data into a training set and a test set according to a ratio of 9:1, input the sample data in the training set into the system to perform system parameter adjustment;

[0060] S6: input the sample data in the test set into the system after adjustment, perform system accuracy test, if the system measurement accuracy is less than 95%, modify the system parameters through genetic optimization algorithm until the system measurement accuracy is greater than 95%, and the optimal system is obtained.

[0061] Figure 3 is a data acquisition test diagram, which records the construction of the underwater test system, indicates that the 7-way sensor signals at different angles can be obtained by changing the mass and fluctuation characteristics of the water medium, and on the basis of the hardware system, the data in the deep water environment obtained through the underwater test is used as the training sample of the network. Then the network model is trained and learned through the sample, and the trained network is used as a prediction model with deep water environment self-adaptive learning. Finally, the model is programmed and written into the signal processing module of the hardware system, so that the system has the ability of deep sea environment self-adaptive learning.

[0062] Although the embodiments of the present application have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An artificial neural network based adaptive deep sea environment learning four quadrant azimuth measurement system characterized by, The application relates to a four-quadrant detection module, a deep-sea environment sensing module and an information processing module. The four-quadrant detection module comprises a four-quadrant photoelectric detector and an analog circuit, the four-quadrant photoelectric detector guides a light field to pass through a detector receiving lens to form a light spot with a certain size, and the analog circuit converts the energy distribution of the light spot on the photoelectric detector into a voltage signal. The deep-sea environment sensing module comprises a water quality sensor, a flow rate sensor and a flow direction sensor, the deep-sea water quality is converted into a voltage signal by the water quality sensor, the deep-sea water flow is converted into a voltage signal by the flow rate sensor, and the deep-sea water flow direction is converted into a voltage signal by the flow direction sensor. The information processing module collects the voltage signals generated by the four-quadrant detection module and the deep-sea environment sensing module through AD acquisition, converts the voltage signals into digital signals, processes the digital signals by using an artificial neural network algorithm, obtains the azimuth angle data of the current light field, and transmits the obtained azimuth angle data to the control system of the vehicle through a communication interface. The voltage signals are converted into digital signals, and the conversion specifically comprises the following steps.

2. The self-adapting deep-sea environment learning four-quadrant azimuth measurement system based on artificial neural network according to claim 1, characterized in that, The artificial neural network algorithm is used to process the digital signals, and the azimuth angle data of the current light field are obtained, and the conversion specifically comprises the following steps. Data = V / V ref *(2 N -1) Wherein, Data represents the digital signal obtained by conversion, V represents the voltage signal to be converted, V ref represents the reference voltage of the analog-to-digital converter, and N represents the conversion bit number of the analog-to-digital converter.

3. The self-adapting deep-sea environment learning four-quadrant azimuth measurement system based on artificial neural network according to claim 1, characterized in that, Step 1: a three-layer neural network structure is constructed through Matlab, the three-layer neural network structure comprises an input layer, a hidden layer and an output layer, the input layer is composed of seven neuron nodes and corresponds to the four-way voltage signals of the four-quadrant photoelectric detector of the measuring system and the digital signals converted from the voltage signals of the water quality sensor, the flow rate sensor and the flow direction sensor, the hidden layer corresponds to the seven neuron nodes, and the output layer comprises one neuron node and corresponds to the azimuth angle data of the light field. Step 2: the seven neuron nodes of the input layer process the digital signals converted from the voltage signals of the four-quadrant photoelectric detector, the water quality sensor, the flow rate sensor and the flow direction sensor, and digital signals with weights are obtained, each neuron is connected with the next layer of neurons and signal transmission is carried out. Step 3: the seven neuron nodes of the hidden layer generate bias for the digital signals transmitted by the input layer and carry out summation, and the signals are transmitted through an activation function. Step 4: the neuron node of the output layer obtains the azimuth angle data of the light field according to the weights and bias of the digital signals. The azimuth angle data of the light field are obtained according to the weights and bias of the digital signals, and the conversion is represented as follows.

4. The self-adapting deep-sea environment learning four-quadrant azimuth measurement system based on artificial neural network according to claim 3, characterized in that, The application further relates to a method for training the artificial neural network, and the method comprises the following steps. where y represents azimuth angle data, w ij represents the weight of the connection between the input layer and the hidden layer neurons, g() represents the hidden layer activation function, X i represents the four-way voltage signal of the four-quadrant photodetector of the measuring system and the digital signal converted from the voltage signals of the water quality sensor, flow rate sensor and flow direction sensor, b j represents the hidden layer neuron bias, N represents the number of input neurons, M represents the number of hidden layer neurons, W j represents the weight of the connection between the hidden layer and the output layer neurons, f() represents the output layer activation function.

5. A training method of an artificial neural network-based adaptive deep-sea environment learning four-quadrant azimuth measurement system according to claims 1-3, characterized in that, S1: under the deep-sea environment, a target guiding light field is generated by using a blue-green LED light source, the system is fixed on a two-dimensional angle displacement platform, and all power supply and signal cables are connected; S2: the two-dimensional angle displacement platform is adjusted so that the system faces the light source, and 7-way sensor data uploaded by the test system is saved through the upper computer software; S3: the displacement platform is continuously deflected by a certain angle, each group of data is saved, and the angle value is recorded; S4: the water quality and the fluctuation amplitude of the water are changed, S3 is repeated, and a certain number of sample data are obtained; S5: the sample data are divided into a training set and a test set according to a 9:1 ratio, and the sample data in the training set are input into the system to carry out system parameter adjustment. ​ S6: input the sample data in the test set into the system after parameter adjustment, and test the accuracy of the system; if the accuracy of the system is less than 95%, modify the system parameters by genetic optimization algorithm until the accuracy of the system is greater than 95%, and obtain the optimal system.

Images