Acousto-optic composite detection system

Abstract

The invention discloses an integrated acousto-optic composite system which comprises system hardware and a detection method, and the system hardware comprises a sensor module, a communication module, a power supply module and a voltage stabilization chip, is responsible for data acquisition and transmission of a bottom layer, and provides basic guarantee for subsequent reliable operation of the system. During detection, firstly, a space detection point is determined in a space, an acousto-optic modulation input module is used for converting an input acousto-optic signal into an electric signal for processing, and an analog electric signal is converted into a digital signal through an analog-to-digital converter. The digital signals are further processed through a computer, and acquired acousto-optic data are fitted through MATLAB software by using a neural network algorithm to obtain sound intensity and illumination intensity. According to the method, an indoor sound and illumination system is reconstructed, and according to the sound and illumination intensity of a finite point, a neural network is used for approximation and the system is reconstructed.

Description

Technical Field

[0001] The invention relates to the technical field of photoelectric detection, and in particular to an acoustic-optical composite detection system. Background Art

[0002] Photoacoustic testing is a technology that uses the acoustic effect caused by the material absorbing light energy with intensity varying with time and being heated to perform nondestructive testing on the specimen. The photoacoustic signal is usually a weak signal masked by noise. In order to extract the photoacoustic signal from the noise and improve the signal-to-noise ratio of the detection, relevant technologies and multiple averaging of the signal can be used. According to the basic theories of photoacoustics and photothermal, such as RG theory, McDonald-Wetse theory or photothermal deflection theory, the photoacoustic and photothermal signals generated by the sample absorbing modulated light energy can be quantitatively described by a set of linear differential equations and corresponding boundary conditions in the case of non-saturated absorption. A large number of experiments have also confirmed the correctness of these theories. This shows that when the sample is in non-saturated absorption, the system under study can be considered as a linear time-invariant system, and the photoacoustic signal can be directly obtained by Fourier transform. At the same time, weak signal detectors such as phase-locked analyzers are essentially analog correlators, and Boxcar integrators are essentially cumulative averagers. Therefore, the use of digital signal processing techniques such as digital correlation and digital sampling averaging, and the application of microcomputers for photoacoustic detection is not only theoretically valid, but also completely feasible in practice.

[0003] Then, the above-mentioned photoacoustic detection, as well as the detection and fitting of sound and light, has a wide range of application backgrounds, involving multiple fields, such as medical treatment, security, communications, environmental monitoring, etc.

[0004] Sound detection mainly captures sound waves through sensors, converts them into electrical signals or digital signals, and then analyzes and processes them. Sound is a mechanical wave generated by the vibration of an object, which propagates through a medium (air, liquid or solid). The detection of sound usually relies on the analysis of the frequency, amplitude and waveform of the sound wave. Common sound detection sensors include microphones, accelerometers and ultrasonic sensors. Microphones can capture sounds within the audible range of the human ear (20Hz-20kHz), while ultrasonic sensors are used to detect ultrasonic signals with higher frequencies. Through feature extraction and pattern recognition of sound, speech recognition technology is widely used in smart assistants, translation devices and control systems.

[0005] Light detection involves the capture, measurement, and analysis of light signals. Photon sensors or detectors can convert light signals into electrical signals for further processing. Light is part of the electromagnetic spectrum and has basic properties such as wavelength, frequency, and energy. Common light detection technologies include the detection of visible light, infrared light, and ultraviolet light. By converting light signals into electric current, it is used to measure the intensity of light. Sound and light detection technologies are driving development and change in many fields through continuous innovation and improvement.

[0006] In this application, it is the detection and fitting of sound and light. The fitting involves the neural network fitting process, which involves knowledge in multiple disciplines, including machine learning, statistics, mathematical optimization, etc. Neural network fitting refers to the process of adjusting the network weights to enable the model to approximate the target function.

[0007] A neural network is a machine learning model inspired by biological nervous systems and consists of multiple layers of artificial neurons. Each neuron receives multiple inputs, adds them up and applies an activation function, and then passes the output value to the next neuron. The basic components of a network include input layer, hidden layer, and output layer. The fitting process of a neural network involves neural network weights and biases, action functions, and weight learning algorithms. Each connection in a neural network has a weight, or weight coefficient, which is used to adjust the impact of the input. The bias term helps adjust the flexibility of the output. The essence of the fitting process is to find the right weight and bias values ​​so that the network output is close to the true target value. Summary of the invention

[0008] The object of the present invention is to provide an integrated acousto-optical composite system.

[0009] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is:

[0010] The integrated acoustic-optical composite system includes system hardware and acoustic-optical composite detection methods. The system hardware includes sensor modules, communication modules, power modules, and voltage regulator chips, which are responsible for underlying data collection and transmission, providing basic guarantees for the subsequent reliable operation of the system.

[0011] The acoustic-optical composite detection method includes the following:

[0012] Step S1, determining the acoustic and optical detection points through the sensor module, and collecting the acoustic signals and optical signals of the determined acoustic and optical detection points in space;

[0013] Step S2, using an acousto-optic modulation input module to convert the input acousto-optic signal into an electrical signal for processing, and converting the analog electrical signal into a digital signal through an analog-to-digital converter ADC;

[0014] Step S3, determining a fitting algorithm for a neural network, and constructing a neural network optoelectronic fitting learning and training model in the algorithm;

[0015] Step S4, fitting the collected sound and light data using the training model through MATLAB sound and light control software;

[0016] Step S4, reconstructing the sound and light system after fitting the sound and light data according to the sound and light intensities of the finite points.

[0017] Preferably, the MATLAB sound and light control software adjusts the magnitude of the current input to the lamp, controls the brightness of the lamp, adjusts the position and size of the input speaker, and controls the intensity of the sound according to the light intensity and sound intensity set by the user.

[0018] Preferably, the space in step 1 is an indoor enclosed space, and at least two sensor modules are provided, and the sensor modules are connected to the acousto-optic modulation input module via wireless signals or lines; the spacing between the detection points is 15 cm-25 cm.

[0019] Preferably, the fitting algorithm of the neural network includes the following:

[0020] Step A1, determine the neural network structure, the neural network structure is a multi-layer neural network, including an input layer, a hidden layer and an output layer, wherein the input layer includes 5 neurons, the hidden layer includes 8 neurons, and the output layer includes 2 neurons.

[0021] Step A2, data processing: Select the position information of the detection point as (σ x ,σ y ,σ z ) data, select the light intensity of the light source as E data, select the sound intensity of the sound source as I data, the above 5 data are the 5 neurons of the input layer, the data of the 5 neurons of the input layer are normalized, and the normalized data of the neural network output is obtained, and marked as in is x The normalized result is is y The normalized result is is z The normalized result is is the result of E normalization, is the result of normalization of I;

[0022] Step A3, determining two neuron data of the output layer, the two neuron data are the light intensity and the sound intensity of the detection point respectively;

[0023] Step A4, determine the data of hidden layer neurons and output them; the hidden layer neuron data is The hidden layer neuron data includes input X I , weight W IH , bias b H , activation function f;

[0024] The factors that affect the output of the hidden layer neurons include the weights from the input layer to the hidden layer, the activation values ​​of the input layer neurons and the hidden layer neurons, and the output σ of the neuron H and activation value X H It can be obtained by the following formula:

[0025] The calculation formula for the hidden layer weighted input is:

[0026] σ H =W IH X I +b H ;

[0027] Using the sigmoid activation function, the calculation formula for the hidden layer activation value is:

[0028]

[0029] Step 5: Determine the data of the output layer neurons and output them. The output layer neuron data is The output layer neuron data includes the input X H , weight W HO , bias b O , activation function f;

[0030] The output of the output layer neurons includes the weight from the hidden layer to the output layer, the activation value of the neurons from the hidden layer to the output layer, and the output σ of the neurons. O and activation value X O It can be obtained by the following formula:

[0031] The formula for calculating the weighted input of the output layer is:

[0032] σ O =W HO X H +b O ;

[0033] Using the sigmoid activation function, the calculation formula for the output layer activation value is:

[0034]

[0035] Step A6, based on the above output layer output and the actual measured data, the data loss is calculated using the mean square error of the loss function, the mean square error is:

[0036]

[0037] in is the output value of the neural network, is the actual value, that is, the ideal value;

[0038] Step 7: Train the weight matrix W by gradient descent of the loss function HO and W IH , after obtaining the ideal weight matrix, the final acousto-optic fitting is achieved.

[0039] 4. The integrated acousto-optic hybrid system according to claim 3, characterized in that in the loss function in step 6, back propagation is used to calculate the gradient of the loss function to the weights of each layer in the network, and the weights are updated by gradient descent;

[0040] Assume that the loss function is Γ, and the weights and biases of the lth layer are W [l] and b [l] ;

[0041] Use the gradient descent algorithm to update the weights and biases of the neural network. Assuming the learning rate is α, the weights and biases from the hidden layer to the output layer are updated using the following formula;

[0042] Weight update:

[0043]

[0044] Bias Update:

[0045]

[0046] Updates on the Law of Learning:

[0047] α:=α0exp(-k)+α1;

[0048] Where α0>0, α1>0, and k is the number of training times.

[0049] The input to the hidden layer weights and biases are updated by the following formula:

[0050] Weight update:

[0051]

[0052] Bias Update:

[0053]

[0054] Updates on the Law of Learning:

[0055] α:=α0exp(-k)+α1;

[0056] Where α0>0, α1>0, and k is the number of training times.

[0057] The beneficial effects of the present invention are:

[0058] In the present invention, innovative designs are made for the detection technology of sound and light, neural network fitting technology, and sound and light control technology in the field of electronic information. Backpropagation is the core algorithm of the neural network fitting process. It calculates the gradient of the loss function for each network parameter, propagates errors layer by layer, and adjusts the parameters so that the network output gradually approaches the target value. The background technology of neural network fitting covers aspects such as model structure, backpropagation, optimization algorithm, regularization, and data processing. With the continuous advancement of computing power and algorithms, neural networks have shown powerful capabilities in solving complex fitting problems. The present invention reconstructs the indoor sound and lighting system, and uses a neural network to approximate and reconstruct the system according to the sound and light intensity at a finite point. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of the acousto-optic composite detection method.

[0060] Figure 2 It is a flowchart of the neural network fitting algorithm. DETAILED DESCRIPTION

[0061] The present invention is described in detail below in conjunction with the accompanying drawings:

[0062] Example 1

[0063] The integrated acoustic-optical composite system includes system hardware and acoustic-optical composite detection methods. The system hardware includes sensor modules, communication modules, power modules, and voltage regulator chips, which are responsible for underlying data collection and transmission, providing basic guarantees for the subsequent reliable operation of the system.

[0064] The acoustic-optical composite detection method includes the following:

[0065] Step S1, determining the acoustic and optical detection points through the sensor module, and collecting the acoustic signals and optical signals of the determined acoustic and optical detection points in space;

[0066] Step S2, using an acousto-optic modulation input module to convert the input acousto-optic signal into an electrical signal for processing, and converting the analog electrical signal into a digital signal through an analog-to-digital converter ADC;

[0067] Step S3, determining a fitting algorithm for a neural network, and constructing a neural network optoelectronic fitting learning and training model in the algorithm;

[0068] Step S4, fitting the collected sound and light data using the training model through MATLAB sound and light control software;

[0069] Step S4, reconstructing the sound and light system after fitting the sound and light data according to the sound and light intensities of the finite points.

[0070] Example 2

[0071] The MATLAB sound and light control software adjusts the current of the control input lamp according to the light intensity and sound intensity set by the user, controls the brightness of the lamp, adjusts the position and size of the control input speaker, and controls the intensity of the sound.

[0072] The space in step 1 is an indoor enclosed space, and at least two sensor modules are provided. The sensor modules are connected to the acousto-optic modulation input module via wireless signals or lines; the spacing between the detection points is 15 cm-25 cm.

[0073] Example 3

[0074] The neural network fitting algorithm includes the following:

[0075] Step A1, determine the neural network structure, the neural network structure is a multi-layer neural network, including an input layer, a hidden layer and an output layer, wherein the input layer includes 5 neurons, the hidden layer includes 8 neurons, and the output layer includes 2 neurons.

[0076] Step A2, data processing: Select the position information of the detection point as (σ x ,σ y ,σ z ) data, select the light intensity of the light source as E data, select the sound intensity of the sound source as I data, the above 5 data are the 5 neurons of the input layer, the data of the 5 neurons of the input layer are normalized, and the normalized data of the neural network output is obtained, and marked as in is x The normalized result is is y The normalized result is is z The normalized result is is the result of E normalization, is the result of normalization of I;

[0077] Step A3, determining two neuron data of the output layer, the two neuron data are the light intensity and the sound intensity of the detection point respectively;

[0078] Step A4, determine the data of hidden layer neurons and output them; the hidden layer neuron data is The hidden layer neuron data includes input X I , weight W IH , bias b H , activation function f;

[0079] The factors that affect the output of the hidden layer neurons include the weights from the input layer to the hidden layer, the activation values ​​of the input layer neurons and the hidden layer neurons, and the output σ of the neuron H and activation value X H It can be obtained by the following formula:

[0080] The calculation formula for the weighted input of the hidden layer is:

[0081] σ H =W IH X I +b H ;

[0082] Using the sigmoid activation function, the calculation formula for the hidden layer activation value is:

[0083]

[0084] Step 5: Determine the data of the output layer neurons and output them. The output layer neuron data is The output layer neuron data includes the input X H , weight W HO , bias b O , activation function f;

[0085] The output of the output layer neurons includes the weight from the hidden layer to the output layer, the activation value of the neurons from the hidden layer to the output layer, and the output σ of the neurons. O and activation value X O It can be obtained by the following formula:

[0086] The formula for calculating the weighted input of the output layer is:

[0087] σ O =W HO X H +b O ;

[0088] Using the sigmoid activation function, the calculation formula for the output layer activation value is:

[0089]

[0090] Step A6, based on the above output layer output and the actual measured data, the data loss is calculated using the mean square error of the loss function, the mean square error is:

[0091]

[0092] in is the output value of the neural network, is the actual value, that is, the ideal value;

[0093] Step 7: Train the weight matrix W by gradient descent of the loss function HO and W IH , after obtaining the ideal weight matrix, the final acousto-optic fitting is achieved.

[0094] Example 4

[0095] On the basis of the disclosure of the above embodiment, the present embodiment further discloses as follows: according to the above output layer output and the actual measurement data, the mean square error of the loss function is used to calculate the data loss amount, and the loss function adopts back propagation to calculate the gradient of the loss function to the weights of each layer in the network, and the weights are updated by gradient descent;

[0096] Assume that the loss function is Γ, and the weights and biases of the lth layer are W [l] and b [l] ;

[0097] Use the gradient descent algorithm to update the weights and biases of the neural network. Assuming the learning rate is α, the weights and biases from the hidden layer to the output layer are updated using the following formula;

[0098] Weight update:

[0099]

[0100] Bias Update:

[0101]

[0102] Updates on the Law of Learning:

[0103] α:=α0exp(-k)+α1;

[0104] Among them, α0>0, α1>0, and k is the number of training times.

[0105] The input to the hidden layer weights and biases are updated by the following formula:

[0106] Weight update:

[0107]

[0108] Bias Update:

[0109]

[0110] Updates on the Law of Learning:

[0111] α:=α0exp(-k)+α1;

[0112] Where α0>0, α1>0, and k is the number of training times.

[0113] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. An integrated sound and light composite system, characterized in that: Including system hardware and acoustic-optical composite detection methods. The system hardware includes sensor modules, communication modules, power modules, and voltage regulator chips, which are responsible for underlying data collection and transmission, providing basic guarantees for the subsequent reliable operation of the system. The acoustic-optical composite detection method includes the following: Step S1, determining the acoustic and optical detection points through the sensor module, and collecting the acoustic signals and optical signals of the determined acoustic and optical detection points in space; Step S2, using an acousto-optic modulation input module to convert the input acousto-optic signal into an electrical signal for processing, and converting the analog electrical signal into a digital signal through an analog-to-digital converter ADC; Step S3, determining a fitting algorithm for a neural network, and constructing a neural network optoelectronic fitting learning and training model in the algorithm; Step S4, fitting the collected sound and light data using the training model through MATLAB sound and light control software; Step S4, reconstructing the sound and light system after fitting the sound and light data according to the sound and light intensities of the finite points.

2. The integrated acousto-optic composite system according to claim 1, characterized in that: The MATLAB sound and light control software adjusts the current of the control input lamp according to the light intensity and sound intensity set by the user, controls the brightness of the lamp, adjusts the position and size of the control input speaker, and controls the intensity of the sound.

3. The integrated acousto-optic composite system according to claim 1, characterized in that: The space in step 1 is an indoor enclosed space, and at least two sensor modules are provided. The sensor modules are connected to the acousto-optic modulation input module via wireless signals or lines; the spacing between the detection points is 15 cm-25 cm.

4. The integrated acoustic-optical composite detection method according to claim 1, characterized in that: The fitting algorithm of the neural network includes the following: Step A1, determine the neural network structure, the neural network structure is a multi-layer neural network, including an input layer, a hidden layer and an output layer, wherein the input layer includes 5 neurons, the hidden layer includes 8 neurons, and the output layer includes 2 neurons. Step A2, data processing: Select the position information of the detection point as (σ x ,σ y ,σ z ) data, select the light intensity of the light source as E data, select the sound intensity of the sound source as I data, and the above 5 data are 5 neurons in the input layer; Normalize the data of the five neurons in the input layer to obtain the normalized output data of the neural network and mark it as in is x The normalized result is is y The normalized result is is z The normalized result is is the result of E normalization, is the result of normalization of I; Step A3, determining two neuron data of the output layer, the two neuron data are the light intensity and the sound intensity of the detection point respectively; Step A4, determine the data of hidden layer neurons and output them; the hidden layer neuron data is The hidden layer neuron data includes input X I , weight W IH , bias b H , activation function f; The factors that affect the output of the hidden layer neurons include the weights from the input layer to the hidden layer, the activation values ​​of the input layer neurons and the hidden layer neurons, and the output σ of the neuron H and activation value X H It can be obtained by the following formula: The calculation formula for the hidden layer weighted input is: s H =W IH X I +b H ; Using the sigmoid activation function, the calculation formula for the hidden layer activation value is: Step 5: Determine the data of the output layer neurons and output them. The output layer neuron data is The output layer neuron data includes the input X H , weight W HO , bias b O , activation function f; The output of the output layer neurons includes the weight from the hidden layer to the output layer, the activation value of the neurons from the hidden layer to the output layer, and the output σ of the neurons. O and activation value X O It can be obtained by the following formula: The formula for calculating the weighted input of the output layer is: s O =W HO X H +b O ; Using the sigmoid activation function, the calculation formula for the output layer activation value is: Step A6, based on the above output layer output and the actual measured data, the data loss is calculated using the mean square error of the loss function, the mean square error is: in is the output value of the neural network, is the actual value, that is, the ideal value; Step 7: Train the weight matrix W by gradient descent of the loss function HO and W IH , after obtaining the ideal weight matrix, the final acousto-optic fitting is achieved.

5. The integrated acousto-optic composite system according to claim 4, characterized in that: In the loss function in step 6, back propagation is used to calculate the gradient of the loss function to the weights of each layer in the network, and the weights are updated by gradient descent; Assume that the loss function is Γ, and the weights and biases of the lth layer are W [l] and b [l] ; Use the gradient descent algorithm to update the weights and biases of the neural network. Assuming the learning rate is α, the weights and biases from the hidden layer to the output layer are updated using the following formula; Weight update: Bias Update: Updates on the Law of Learning: α:=α0exp(-k)+α1; Where α0>0, α1>0, and k is the number of training times. The input to the hidden layer weights and biases are updated by the following formula: Weight update: Bias Update: Updates on the Law of Learning: α:=α0exp(-k)+α1; Where α0>0, α1>0, and k is the number of training times.

Images