Low-frequency underwater acoustic signal detection method and device based on bionic optical fiber cilia array

Abstract

The present invention relates to the fields of optical fiber sensing and deep learning, and specifically to a low-frequency underwater acoustic signal detection method and device based on a bionic optical fiber ciliary array; the advantages of the optical F-P cavity structure being simple, high sensitivity, and easy multiplexing are fully utilized, and the advantages of the ciliary array being extremely high sensitivity and narrowband filtering are combined to control the length of each ciliary inside the array, so that the cavity length of the F-P cavity formed by the end face of each ciliary and the reflector is different, and the corresponding resonant frequency is also inconsistent, so that the ciliary array can detect sound waves of different frequencies, and the frequency composition of the incoming wave can be directly obtained without complex Fourier transform and filtering operations, thereby realizing intelligent and autonomous recognition of target signals. Compared with the traditional hydrophone system, the device is more sensitive, compact, and efficient, and has the advantages of narrowband filtering, real-time spectrum extraction, and autonomous target recognition, and has broad application prospects.

Description

Technical Field

[0001] The present invention relates to the fields of optical fiber sensing and deep learning, and in particular to a method and device for detecting low-frequency underwater acoustic signals based on a bionic optical fiber ciliary array. The device uses a bionic ciliary vector hydrophone as a primitive and combines a neural network algorithm, and is particularly suitable for the detection and intelligent identification of low-frequency underwater acoustic signals. Background Art

[0002] With the development of material technology and modern manufacturing technology, the noise generated by underwater moving targets has been reduced to a level below the ocean environmental noise. Traditional underwater acoustic detection technology faces many difficulties. It is of great significance to develop underwater acoustic detection devices suitable for low-frequency bands, with extremely high sensitivity and autonomous signal processing capabilities.

[0003] Vector hydrophones can measure vector parameters in the sound field. They have great advantages in suppressing marine environmental noise, obtaining high-gain target radiation noise, and tracking and positioning targets. They are a type of underwater acoustic detection device with great development prospects in the field of underwater acoustic detection. According to different technical routes, vector hydrophones can generally be divided into piezoelectric, optical fiber, and MEMS types. Representative results reported in the literature include: Chinese invention patent "three-dimensional piezoelectric ceramic vector hydrophone" (ZL201210372979.6, "three-dimensional torus-shaped vector hydrophone" (ZL201310067898.X), "composite scalar vector hydrophone" (ZL201310392752.2), Chinese invention patent application "low-frequency acoustic vector hydrophone" (application number 201610795311.0, published on 2017-01-04), Chinese utility model patent "two-dimensional isochronous vector hydrophone" (ZL201820683164.2), etc. These patents all propose to use piezoelectric hydrophones to achieve isochronous or velocity-type vector hydroacoustic sensing. However, piezoelectric vector hydrophones have low sensitivity, poor anti-interference ability, limited signal transmission distance, and poor practical application effects. Chinese invention patents "Push-pull fiber laser vector hydrophone" (ZL201310035388.4), "A fiber laser vector hydrophone" (application number 201910032857.4, published on 2019-04-19), Chinese utility model patents "A three-dimensional pressure difference fiber vector hydrophone" (ZL201721863974.8), "A combined fiber vector hydrophone" (ZL201721878821.0), and Chinese invention patent application "An interferometric three-dimensional vector hydrophone based on fiber grating" (application number 201310711632.4, published on 2014-03-26) proposed different methods for realizing water acoustic vector field measurement based on fiber hydrophone, respectively using fiber laser or fiber interferometer to realize water acoustic sensing, and combining push-pull structure, pressure difference structure or three-component combined structure to realize the acquisition of water acoustic field vector information. Among them, the dual-arm interferometric fiber optic hydrophone solution is the most mature and the most widely used, but it has a complex structure, bulky size, and high cost, and faces great difficulties in frequency band expansion.Chinese invention patents such as "Differential vibration isolation MEMS vector hydrophone" (ZL201410626399.4), "MEMS three-dimensional isochronous vector hydrophone based on piezoresistive effect" (ZL201710450802.6), "High-sensitivity wide-band piezoelectric MEMS vector hydrophone" (ZL201810434347.5), and Chinese invention patent application "Wide-band MEMS vector hydrophone imitating seal whiskers" (application number 201510670121.1, published on December 23, 2015) all proposed schemes for realizing miniature hydrophones based on MEMS technology, mainly focusing on how to use ciliary microstructures combined with piezoelectric or piezoresistive effects to achieve better device performance. However, such sensors mainly use electrical signals for sensing, are easily affected by external interference, and have poor stability. China's invention patent "A MOMES vector hydrophone" (ZL201510411902.9) draws on the basic structure of ciliary bionic sensing and uses four symmetrically distributed FP interferometers to measure the diaphragm deformation caused by sound waves. The sensing structure is complex and difficult to assemble.

[0004] The above-mentioned patents use different technical paths to achieve vector detection of underwater acoustic signals, which can improve the efficiency of underwater acoustic detection to a certain extent, but there is still room for further improvement in terms of sensitivity enhancement, large-scale array multiplexing, and autonomous identification of underwater acoustic targets. Summary of the invention

[0005] The present invention aims to solve the defects of the prior art and proposes a low-frequency underwater acoustic signal detection method and device based on a bionic optical fiber cilia array, so as to achieve underwater acoustic target detection with compact structure, extremely high sensitivity and autonomous target identification, which is particularly suitable for applications in the field of low-frequency underwater acoustic signal detection and intelligent identification.

[0006] The technical solution adopted by the present invention is a low-frequency underwater acoustic signal detection method based on a bionic optical fiber cilia array, which is divided into the following steps:

[0007] Step 1: Modulation and transmission of optical signals:

[0008] The laser is used as the light source to emit the original light signal, and the optical attenuator is used to attenuate the intensity of the light signal to a level that will not cause damage to the detection instrument. After intensity modulation, the light signal enters from the first port of the circulator and outputs from the second port to enter the coupler. The coupler divides the single light path into N light paths with the same properties. Among them, N is a natural number, and N ≥ 2. Then the light signals corresponding to these N light paths simultaneously enter N delay optical fibers of different lengths, so that a phase difference is generated between each signal. After that, these light signals successively enter the ciliary array composed of N end-face coated optical fibers of the same length but different ends. A reflector is placed at the end of the ciliary array so that the end faces of all coated optical fibers and the reflector form an FP cavity;

[0009] Step 2: Propagation of optical signal in FP cavity:

[0010] The optical signals corresponding to the N optical paths are output from the ciliary array. After being modulated by the FP cavity, the reflected lights corresponding to the N optical paths are obtained. There are mainly three types of reflected lights in each optical path, which interfere with each other. Among them, these three types of reflected lights include the reflected light directly reflected back to the fiber core from the lower end face of the coating, the reflected light that passes through the lower end face of the coating and is reflected by the upper end face and then re-transmitted back to the fiber core, and the reflected light that successively passes through the lower end face and the upper end face of the membrane and enters the external environment, then is reflected by the reflector and re-transmits through the coating to enter the fiber core. In addition to the above three types of reflected light, there are other types of reflected light, but these reflected lights have too much reflection loss and transmission loss during the round-trip propagation process, resulting in very weak intensity when they are transmitted back to the fiber core. Therefore, the existence of these reflected lights can be approximately ignored. After that, these optical signals pass through the coupler again to obtain the total interference optical signal;

[0011] Step 3: Detection and identification of optical signals:

[0012] The total interference light signal returns to the second port of the circulator and is output from the third port. The light signal is then converted into an electrical signal through a photodetector and collected using an oscilloscope. The collected data is demodulated to obtain the relationship between the light intensity and the cavity length of the FP cavity and the relationship between the cavity length and the sound pressure. Finally, the demodulated data is input into the trained neural network model to identify the type of external sound source. Since the lengths of the coated optical fibers at each end face in the ciliary array are different, the corresponding FP cavity lengths are inconsistent, and thus the resonant frequencies of each ciliary are different. Therefore, when the ciliary array is exposed to sound waves of different frequencies, the distribution of the response intensity with the frequency is also different. That is, the sound wave frequency can be used as the actual label to classify these data results to achieve detection and intelligent identification of underwater acoustic targets.

[0013] The optical signals corresponding to the N optical paths in step 1 need to generate phase differences through the delay optical fiber, and the process is as follows:

[0014] Assume that the wavelength of the optical signal in the N input optical paths is λ, and the length of the N delay optical fibers is d n (n=1,2,3,...,N), the length of each delay fiber is different, and the length difference between adjacent delay fibers is Δd=|d n -d n-1 |, and the corresponding optical signals can produce a stable phase difference Δφ, whose magnitude is:

[0015]

[0016] After the interference light is formed through the FP cavity, it will produce another phase difference after passing through the delay fiber, so the total phase difference is 2Δφ.

[0017] Therefore, when all the interference light signals return to the coupler input, due to the phase difference 2Δφ between the signals, after these signals are coupled into the same signal through the coupler, the original interference information of each optical path can be retained and will not interfere with each other.

[0018] The reflector described in step 1 and the end faces of all coated optical fibers form an FP cavity. The design process of the reflector is as follows:

[0019] The reflector is designed based on the dynamic range and sensitivity measured by the device. The theories involved in free-form surface design mainly include Gaussian surface, hyperbolic surface and XY polynomial surface, etc. The hyperbolic surface theory is preferred to design the reflector. The details are as follows:

[0020] In three-dimensional space, the free surface equation corresponding to the hyperbolic theory is as follows:

[0021]

[0022] Among them, a, b, and c are constants. By changing the values ​​of a, b, and c according to the distribution of the coated optical fiber positions in the ciliary array, a reflector that adapts to the position change of each coated optical fiber end face can be designed to ensure that each coated optical fiber can remain parallel to the reflector when subjected to external sound waves, thereby reducing signal loss and achieving the best interference effect of the FP cavity.

[0023] The optical signal on the single optical path described in step 2 mainly generates three types of reflected light when passing through the FP cavity and interferes with each other. The process is as follows:

[0024] Ignore the signal attenuation caused by the propagation of the optical signal in the end-face coated optical fiber, assume that the amplitude of these optical signals is A when they are just incident on the FP cavity, and the refractive indexes of the core inside the end-face coated optical fiber, the coating on the end face, the external environment medium and the reflector are n respectively. a 、n b 、n c 、n d , the thickness of the film is h, the length of the FP cavity is L, according to the Fresnel formula, a series of reflection coefficients and transmission coefficients involved in the propagation of the incident light can be obtained as follows:

[0025]

[0026] Among them, r ab is the reflection coefficient of light reflected from the lower end face of the coating back to the inside of the fiber core, r bais the reflection coefficient of light reflected from the upper end face of the fiber core back to the coating, r bc is the reflection coefficient of light incident on the contact surface between the coating and the external environment and then reflected back into the coating, r cb is the reflection coefficient of light incident on the contact surface between the external environment and the coating and then reflected back to the external environment, r cd is the reflection coefficient of light incident on the contact surface between the external environment and the reflector and then reflected back to the external environment, t ab is the transmission coefficient of light from the core to the coating, t bc is the transmission coefficient of light from the inside of the coating to the external environment, t cb is the transmission coefficient of light retransmitted from the external environment to the coating, t ba It is the transmission coefficient of light re-transmitted from the inside of the coating to the inside of the fiber core.

[0027] First, the optical signal is directly reflected back to the fiber core at the lower end face of the coating to form the first type of reflected light, and the corresponding amplitude is Ar ab , assuming the initial phase is δ, the plane wave expression is:

[0028] E1=Ar ab e iδ (4)

[0029] The second type of reflected light refers to the incident light that passes through the coated lower end face and is reflected by the upper end face and then re-transmitted back to the fiber core. After being reflected by the upper end face, there may be multiple reflections from the upper and lower end faces, so the amplitude obtained is At ab r bc (r ba r bc ) n t ba (n=0,1,...,∞), n is the number of times the upper and lower surfaces of the multiple coatings are reflected back and forth. Compared with the first type of reflected light, the second type of reflected light has a phase difference of Size:

[0030]

[0031] Therefore, the corresponding plane wave expression is:

[0032]

[0033] Further expanding and summing equation (6), we can obtain the total plane wave of the second type of reflected light as:

[0034]

[0035] Finally, the third type of reflected light refers to the incident light that successively passes through the lower and upper surfaces of the film and enters the external environment, then is reflected by the reflector and then re-transmits through the coating to enter the fiber core. There may be multiple reflections by the reflector surface and the light passes through the upper surface of the film and is reflected by the lower surface of the film before re-transmitting through the upper surface of the film and then returning to the reflector surface. Therefore, the amplitude obtained is summarized as At ab t bc r cd (t cb r ba t bc r cd ) m t cb t ba (m=0,1,...,∞), m is the number of times the reflection occurs between the reflector and the top surface of the coating. Similarly, compared with the first type of reflected light, the phase difference of the third type of reflected light is Size:

[0036]

[0037] The corresponding plane wave expression is:

[0038]

[0039] Similarly, by expanding and summing equation (9), we can obtain the total plane wave of the third type of reflected light:

[0040]

[0041] Based on the above derivation, there are mainly three types of plane waves in the FP cavity and the expressions can be seen in equations (4), (7), and (10). Assuming that the initial phase δ is 0, the interference theorem shows that the expression of the total interference light intensity is as follows:

[0042] I con =|E1+E2+E3| 2 (11)

[0043] The total interference light intensity I con It can be further expressed as:

[0044]

[0045] Where I0 represents the intensity of the optical signal initially incident on the FP cavity, and its relationship with the amplitude A is as follows:

[0046] I0=A 2 (13)

[0047] In step 3, the demodulated data is input into the trained neural network model for identifying the external sound source. According to the existing neural network models, the commonly used ones are BP neural network, convolutional neural network, generative adversarial network, recurrent neural network and long short-term memory network, etc. Among them, the long short-term memory network was first proposed by Hochreiter and Schmidhuber in 1997 (Long short-term memory [J], Neural Computation, 1997, 9 (8): 1735-1780). Compared with other types of neural networks, this model has the advantages of being able to process long sequence data, dynamic memory and forgetting, strong generalization ability and wide applicability. Therefore, the long short-term memory network is preferably used as the main body to construct the neural network model. The training process of the model is as follows:

[0048] First, the acoustic characteristic data of the measured known sound source is used as the driving signal of the transducer, and the transducer is placed near the ciliary array to cause the ciliary array to vibrate and change the FP cavity length corresponding to each end-face coated optical fiber. Then, the signal data generated by these known sound sources are collected and demodulated. Among them, the demodulated data obtained are divided into training set, validation set and test set according to the quantity ratio of 7:2:1. The training set is used to train the neural network model, the validation set is used to evaluate the performance of the neural network model obtained during the training process, and the test set is used to evaluate whether the neural network model after training can accurately predict the signal data category obtained in an unknown environment, that is, the generalization ability of the model. Then, each demodulated data in all sets is further classified according to the corresponding sound wave frequency when each is generated. Each category is counted as an actual label, that is, as the main basis for judging its performance during subsequent training, verification and testing of the neural network model. Here, a total of i categories (i = 1, 2, 3, ..., W) are set, and W is the total number of categories, which can be adjusted according to actual needs. After the training set data is imported into the input layer of the neural network model, the weight and bias of each neuron in the neural network model are randomly initialized, and the iteration cycle is set to P times, the learning rate S and the regularization parameter σ. Among them, the learning rate is the step size for updating model parameters such as weights and biases at each iteration, and the regularization parameter σ is the coefficient to prevent the model from overfitting. The iteration cycle P is usually set to 800 times, the learning rate S is 0.001, and the regularization parameter σ is 0.01. According to step 1, there are N interference light signals transmitted back, and a total of N neurons in the input layer are set, and each neuron corresponds to an interference signal of an optical path. Then, the training set data propagates forward in the network into the hidden layer. The number of hidden layers is set to M layers, where M is a natural number. Each layer needs to be weighted summed and activated. Function processing. Among them, the commonly used activation functions are mainly Sigmoid, ReLU, Tanh, Softmax, etc., and Softmax is better at classifying and identifying multiple targets than other types of activation functions. Therefore, Softmax is preferably used as the activation function of the neural network model, and its expression is:

[0049]

[0050] Among them, z i is the linear output of the i-th category, and W is the total number of categories.

[0051] When the training set data comes out of the hidden layer and enters the output layer, the cross entropy loss is used as the loss function at the output layer to calculate the loss between the predicted output and the actual label, and the gradient information of the function and the back propagation algorithm (Beyond regression: New tools for prediction and analysis in the behavioral sciences [D], 1974, Harvard University) are used to calculate the gradient corresponding to the weight and bias of each neuron. According to the gradient results, the gradient descent algorithm Adam (Adam: A Method for Stochastic Optimization [C], International Conference on Learning Representations, 2015, USA) is used to update the weights and biases in the model. The cross entropy loss function (CEL) is expressed as follows:

[0052]

[0053] Among them, W is also the total number of categories, y o,i is the judgment indicator, that is, category i is the correct classification, y o,i The value is 1, otherwise it is 0. o,i The model predicts the probability that the signal data belongs to category i.

[0054] After that, repeat the above method for P iterations until the performance of the neural network model on the training set reaches convergence or meets the predetermined iteration cycle. During the iteration process, the neural network model needs to use the validation set to evaluate the performance of the model after each iteration, so as to optimize the original model settings such as the number of hidden layers M, the size of the learning rate S or the size of the regularization parameter σ while updating the model weights and biases, and then select the model with the best performance as the final neural network model after the iteration. In addition, after the model is established, a test set is required to evaluate the prediction effect of the model to determine the generalization ability of the model. Finally, the ciliary array is placed in an environment with an unknown sound source to obtain the signal generated by the sound source, and these signal data are demodulated and substituted into the trained neural network model for prediction and classification, thereby realizing the detection and intelligent identification of underwater acoustic targets.

[0055] The present invention also proposes an underwater acoustic signal detection device based on the above method, comprising: a laser 1, an optical attenuator 2, a circulator 3, a coupler 4, an optical fiber distribution structure 5, a clamp 6, a reflector 7, a photodetector 8, an oscilloscope 9 and a host computer 10;

[0056] The optical fiber distribution structure 5 includes a transmission optical fiber 501, a delay optical fiber 502, a ciliary array 503 and a coating 504, which are fixed by a clamp 6 so that the optical fiber end face and the reflector 7 form an FP cavity, wherein the ciliary array 503 is composed of N end-face coated optical fibers, and the length of the delay optical fiber 502 corresponding to each optical path is different. After the laser is emitted from the laser 1, it first passes through the optical attenuator 2 to adjust its light intensity, then passes through the circulator 3 and enters the coupler 4 to be divided into N light paths and enter the optical fiber distribution structure 5, wherein the transmission optical fiber 501 is used for the propagation of the optical signal, the delay optical fiber 502 is used to control the transmission rate of the N light paths to achieve multiplexing and prevent interference between the signals, the ciliary array 503 as a sensing object will be affected by the external sound waves and vibrate, the coating 504 is used to enhance the interference effect of the FP cavity, and then the optical signal reflected from the FP cavity passes through the coupler 4 and the circulator 3 again and is incident on the photodetector 8 so that the optical signal is first converted into an electrical signal and the electrical signal is collected by the oscilloscope 9, and then the host computer 10 demodulates the signal to obtain the relationship between the light intensity and the cavity length of the FP cavity and the relationship between the cavity length and the sound pressure, and finally the demodulated data is input into the trained neural network model to identify the type of external sound source;

[0057] The ciliary array 503 adopts nonlinear amplification based on the resonance working mode, that is, when the FP cavity is formed between the end face of the ciliary and the reflector, the cavity length of the FP cavity is highly correlated with the resonance frequency. When the external sound wave acts on the ciliary array 503 at the resonance frequency of a certain ciliary, the corresponding ciliary can achieve more obvious vibration than other cilia, thereby causing a significant change in the cavity length of its FP cavity. By utilizing this property, the sound wave frequency can be effectively obtained, thereby achieving extremely high sensitivity.

[0058] The ciliary array 503 adopts narrowband filtering based on ciliary resonance mode control, that is, the resonance frequencies corresponding to the FP cavities of different lengths in the ciliary array 503 are inconsistent, so that each ciliary can only be significantly changed by the corresponding resonance frequency, thereby realizing the recognition of the acoustic wave spectrum characteristics;

[0059] The ciliary array 503 and the reflector 7 are combined to form an FP cavity structure based on end-face coated optical fiber, which is simple in design and convenient for the application of multiplexing technology, such as time division multiplexing or wavelength division multiplexing;

[0060] The number and length of the FP cavity formed by the combination of the ciliary array 503 and the reflector 7 can be adjusted according to the needs, so as to realize the resonant absorption and extraction of multi-frequency sound waves;

[0061] The mirror shape of the reflector 7 can be designed according to the free-form surface theory, so that when the coated optical fiber is subjected to external sound waves, the FP cavity formed between its end face and the reflector 7 can always maintain a state of parallel cavity surface, thereby reducing signal loss and achieving the best interference effect of the FP cavity;

[0062] The host computer 10 uses a neural network algorithm to perform real-time acoustic target recognition on the signal, thereby improving the accuracy of signal recognition.

[0063] The present invention innovatively combines the advantages of the optical fiber FP cavity structure and the sensor array, and proposes a high-sensitivity detection and intelligent identification device for low-frequency underwater acoustic signals based on an optical fiber ciliary array. It gives full play to the advantages of the optical FP cavity structure, high sensitivity, and easy multiplexing. Combined with the extremely high sensitivity and narrow-band filtering advantages of the ciliary array, the length of each ciliary inside the array is controlled, so that the cavity length of the FP cavity formed by the end face of each ciliary and the reflector is different, and the corresponding resonant frequency is also inconsistent, so that the ciliary array can detect sound waves of different frequencies, and can directly obtain the frequency composition of the incoming wave without complex Fourier transform and filtering operations, so as to realize intelligent and autonomous identification of the target signal. Compared with the traditional hydrophone system, the device is more sensitive, compact, and efficient, and has the advantages of narrow-band filtering, real-time spectrum extraction, and autonomous target identification, and has broad application prospects.

[0064] The present invention provides a low-frequency underwater acoustic signal detection and intelligent identification device based on a bionic optical fiber cilia array. Compared with the prior art, the present invention has the following beneficial technical effects: first, the cilia structure is innovatively introduced, that is, a plurality of end-face coated optical fibers are used as cilia to form an array and combined with a reflector to form a minimalist FP cavity, which can realize high-sensitivity detection of weak disturbance signals and large-scale array multiplexing under a compact structure; second, when a single cilia used to detect the incident sound wave is acted upon by a resonant frequency sound wave, it has a nonlinear amplification effect on the signal, so that the vibration amplitude of the cilia is more obvious than that of other cilia, and extremely high-sensitivity detection can be achieved; furthermore, by controlling the length of the optical fiber cilia and the cavity length of the FP cavity, the optimal design of the resonant frequency of the optical fiber cilia and the array combination can be achieved, so that the spectral characteristics of the incident sound wave signal can be extracted in real time, providing a basis for autonomous intelligent identification of the target; finally, each bionic cilia can simulate a neuron cell, and the output signal is used as the input of the neural network to realize autonomous intelligent identification of underwater acoustic targets, and the simulation of biological sensing and decision-making mechanisms can be realized to the greatest extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The present invention is further described with the aid of the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work. In the accompanying drawings:

[0066] Figure 1 It is a schematic diagram of the assembly structure of a low-frequency underwater acoustic signal detection and intelligent identification device based on a bionic optical fiber ciliary array according to the present invention;

[0067] Figure 2 is a schematic diagram of the two-dimensional structure of the optical fiber distribution structure 5 of the present invention;

[0068] Figure 3 It is a schematic diagram of the two-dimensional structure of the FP cavity composed of the reflector 7, the ciliary array 503 and the coating 504 of the present invention;

[0069] Figure 4 It is a schematic diagram of collecting external sound wave signals and then entering a neural network to perform real-time acoustic target recognition according to the present invention;

[0070] Description of reference numerals:

[0071] 1: Laser; 2: Optical attenuator; 3: Circulator; 4: Coupler; 5: Fiber distribution structure; 6: Clamp; 7: Reflector; 8: Photodetector; 9: Oscilloscope; 10: Host computer;

[0072] 501: transmission optical fiber; 502: delay optical fiber; 503: ciliary array; 504: coating. DETAILED DESCRIPTION

[0073] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0074] Figure 1 is a schematic structural diagram of an embodiment of a low-frequency underwater acoustic signal detection and intelligent identification device based on a bionic optical fiber ciliary array according to the present invention, Figure 2 is a schematic diagram of the two-dimensional structure of the optical fiber distribution structure 5 of the present invention, Figure 3 Schematic diagram of the two-dimensional structure of the FP cavity composed of the reflector 7, the ciliary array 503 and the coating 504 of the present invention. Figure 4 It is a schematic diagram of collecting external sound wave signals and entering a neural network to perform real-time acoustic target recognition as described in the present invention.

[0075] The present invention discloses a low-frequency underwater acoustic signal detection and intelligent identification device based on a bionic optical fiber ciliary array, the device structure includes: a laser 1, an optical attenuator 2, a circulator 3, a coupler 4, an optical fiber distribution structure 5 with a transmission optical fiber 501, a delay optical fiber 502, a ciliary array 503 and a coating 504, a fixture 6, a reflector 7, a photodetector 8, an oscilloscope 9 and a host computer 10;

[0076] The optical fiber distribution structure 5 is built into the box and fixed by the clamp 6, so that the end face of the cilia and the reflector 7 form an FP cavity to achieve the interference effect. Five cilia are used as the cilia array 503 as the FP cavity. When the laser with a power of 25mW is emitted from the laser 1 with a wavelength of 1550nm, the light source first passes through the optical attenuator 2 to adjust its power to 20μW, and then passes through the circulator 3 and enters the 1×5 coupler 4 to be divided into 5 light paths and enter the optical fiber distribution structure 5. The optical signal first passes through the 20% anti-reflection film, i.e., the coating 504, of which part of the light is directly reflected, and the other part of the light passes through the anti-reflection film and enters the FP cavity and is reflected by the reflector composed of the silicon chip. A small part of the light is reflected back and forth in the anti-reflection film and the FP cavity. These reflected lights produce interference after returning to the fiber core. These interference light signals After passing through the coupler 4 and the circulator 3 again and incident on the photodetector 8, the optical signal is first converted into an electrical signal and collected by the oscilloscope 9. Finally, the host computer 10 demodulates these signals. The normalized interference light intensity of one FP cavity varies between 0.433 and 0.708. Through demodulation, it can be obtained that the cavity length of the FP cavity varies between 1005.08μm and 1005.12μm. The corresponding sound pressure change can be obtained by using the cavity length change. Finally, the obtained data is input into the trained neural network model to identify the type of external sound source.

[0077] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber cilia array, characterized in that: The method is divided into the following steps: Step 1: Modulation and transmission of optical signals: The laser is used as a light source to emit the original optical signal, and the optical attenuator is used to attenuate the intensity of the optical signal to a level that will not cause damage to the detection instrument. After intensity modulation, the optical signal enters from the first port of the circulator and outputs from the second port to enter the coupler. The coupler divides a single optical path into N optical paths with the same properties, where N is a natural number and N≥2; then the optical signals corresponding to these N optical paths simultaneously enter N delay optical fibers of different lengths, so that a phase difference is generated between each signal, and then these optical signals successively enter the cilia array composed of N end-face coated optical fibers of the same length but different ends. A reflector is placed at the end of the cilia array so that the end faces of all coated optical fibers and the reflector form an FP cavity; Step 2: Propagation of optical signal in FP cavity: The optical signals corresponding to N optical paths are output from the ciliary array. After being modulated by the FP cavity, the reflected lights corresponding to N optical paths are obtained. There are mainly three types of reflected lights in each optical path, which interfere with each other: the three types of reflected lights include the reflected light directly reflected back to the fiber core from the lower end face of the coating, the reflected light that passes through the lower end face of the coating and is reflected by the upper end face and then re-transmitted back to the fiber core, and the reflected light that passes through the lower end face and the upper end face of the film successively and enters the external environment, then is reflected by the reflector and re-transmits through the coating into the fiber core; these optical signals then pass through the coupler again to obtain the total interference optical signal; Step 3: Detection and identification of optical signals: The total interference light signal returns to the second port of the circulator and is output from the third port. The light signal is then converted into an electrical signal through a photodetector and collected using an oscilloscope. The collected data is demodulated to obtain the relationship between the light intensity and the cavity length of the FP cavity, as well as the relationship between the cavity length and the sound pressure. Finally, the demodulated data is input into the trained neural network model to identify the type of external sound source. Since the lengths of the coated optical fibers at each end face in the ciliary array are different, the corresponding FP cavity lengths are inconsistent, and thus the resonant frequencies of each ciliary are different. Therefore, when the ciliary array is faced with sound waves of different frequencies, the distribution of the response intensity with the frequency is also different. That is, the sound wave frequency can be used as the actual label to classify these data results to achieve detection and intelligent identification of underwater acoustic targets.

2. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 1, characterized in that: The optical signals corresponding to the N optical paths in step 1 generate phase differences through the delay optical fiber, and the process is as follows: Assume that the wavelength of the optical signal in the N input optical paths is λ, and the length of the N delay optical fibers is d n , n=1,2,3,...,N, the length of each delay fiber is different, and the length difference of adjacent delay fibers is Δd=|d n -d n-1 |, a stable phase difference can be generated between the corresponding optical signals , whose size is: After the interference light is formed by the FP cavity, these interference lights will produce another phase difference after passing through the delay fiber, so the total phase difference is ; Therefore, when all the interfering light signals return to the coupler input, due to the phase difference between the signals After these signals are coupled into the same signal through the coupler, the original interference information of each optical path can be retained and will not interfere with each other.

3. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 1, characterized in that: The reflector described in step 1 and the end faces of all coated optical fibers form an FP cavity. The design process of the reflector is as follows: The reflector is designed based on the dynamic range and sensitivity measured by this method. The process of designing the reflector using the hyperbolic surface theory is as follows: In three-dimensional space, the free surface equation corresponding to the hyperbolic theory is as follows: Among them, a, b, c>1(2) where a, b, c are constants; the values ​​of a, b, c are changed according to the distribution of the positions of the coated optical fibers in the ciliary array, and a reflector is designed to adapt to the position change of the end faces of each coated optical fiber, ensuring that when each coated optical fiber is affected by external sound waves, the corresponding end face can still remain parallel to the reflector, thereby reducing signal loss and achieving the optimal interference effect of the FP cavity.

4. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 3, characterized in that: The reflector can also be designed using a Gaussian surface or an XY polynomial surface.

5. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 1, characterized in that: In step 2, the optical signal on the single optical path generates three types of reflected light when passing through the FP cavity and interferes with each other. The process is as follows: Ignore the signal attenuation caused by the propagation of the optical signal in the end-face coated optical fiber, assume that the amplitude of these optical signals is A when they are just incident on the FP cavity, and the refractive indexes of the core inside the end-face coated optical fiber, the coating on the end face, the external environment medium and the reflector are n respectively. a 、n b 、n c 、n d , the thickness of the film is h, the length of the FP cavity is L, and according to the Fresnel formula, a series of reflection coefficients and transmission coefficients involved in the propagation of the incident light are obtained as follows: Among them, r ab is the reflection coefficient of light reflected from the lower end face of the coating back to the inside of the fiber core, r ba is the reflection coefficient of light reflected from the upper end face of the fiber core back to the coating, r bc is the reflection coefficient of light incident on the contact surface between the coating and the external environment and then reflected back into the coating, r cb is the reflection coefficient of light incident on the contact surface between the external environment and the coating and then reflected back to the external environment, r cd is the reflection coefficient of light incident on the contact surface between the external environment and the reflector and then reflected back to the external environment, t ab is the transmission coefficient of light from the core to the coating, t bc is the transmission coefficient of light from the inside of the coating to the external environment, t cb is the transmission coefficient of light retransmitted from the external environment to the coating, t ba It is the transmission coefficient of light re-transmitted from the inside of the coating to the inside of the fiber core; First, the optical signal is directly reflected back to the fiber core at the lower end face of the coating to form the first type of reflected light, and the corresponding amplitude is Ar ab , assuming the initial phase is δ, the plane wave expression is: E1=Ar ab in iδ (4) The second type of reflected light refers to the incident light that passes through the coated lower end face and is reflected by the upper end face and then re-transmitted back to the fiber core. After being reflected by the upper end face, there may be multiple reflections from the upper and lower end faces, so the amplitude obtained is At ab r bc (r ba r bc ) n t ba , n=0,1,...,∞, n is the number of times the upper and lower surfaces of the multiple coatings are reflected back and forth; compared with the first type of reflected light, the second type of reflected light has a phase difference of Size: Therefore, the corresponding plane wave expression is: Further expanding and summing equation (6), we can obtain the total plane wave of the second type of reflected light as: Finally, the third type of reflected light refers to the incident light that successively passes through the lower and upper surfaces of the film and enters the external environment, then is reflected by the reflector and then re-transmits through the coating to enter the fiber core. There may be multiple reflections by the reflector surface and the light passes through the upper surface of the film and is reflected by the lower surface of the film before re-transmitting through the upper surface of the film and then returning to the reflector surface. Therefore, the amplitude obtained is summarized as At ab t bc r cd (t cb r ba t bc r cd ) m t cb t ba , where m = 0, 1, ..., ∞, and m is the number of times the reflection occurs between the reflector and the top surface of the coating. Similarly, compared with the first type of reflected light, the third type of reflected light has a phase difference of Size: The corresponding plane wave expression is: Similarly, by expanding and summing equation (9), we can obtain the total plane wave of the third type of reflected light: Based on the above derivation, there are mainly three types of plane waves in the FP cavity and the expressions can be seen in equations (4), (7), and (10). Assuming that the initial phase δ is 0, the interference theorem shows that the expression of the total interference light intensity is as follows: I con =|E1+E2+E3| 2 (11) The total interference light intensity I con It can be further expressed as: Where I0 represents the intensity of the optical signal initially incident on the FP cavity, and its relationship with the amplitude A is as follows: I0=A 2 (13)。 6. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 1, characterized in that: Step 3: Input the demodulated data into the trained neural network model for identifying external sound sources. The long short-term memory network is selected as the main body to construct the neural network model. The training process of the model is as follows: First, after using the measured acoustic characteristic data of the known sound source as the driving signal of the transducer, the transducer is placed near the ciliary array to cause the ciliary array to vibrate and change the FP cavity length corresponding to each end-face coated optical fiber. Then, the signal data generated by these known sound sources are collected and demodulated. Among them, the acquired demodulated data are divided into training set, validation set and test set in a quantity ratio of 7:2:

1. The training set is used to train the neural network model, the validation set is used to evaluate the performance of the neural network model obtained during the training process, and the test set is used to evaluate whether the neural network model after training can accurately predict the signal data category obtained in an unknown environment, that is, the generalization ability of the model. Subsequently, each demodulated data in all sets is further classified according to the corresponding sound wave frequency when each is generated, and each category is counted as an actual label, that is, used as a subsequent training, verification and testing neural network model to judge its performance. The main basis of the above method is that a total of i categories are set here, i = 1, 2, 3, ..., W, W is the total number of categories; after the training set data is imported into the input layer of the neural network model, the weight and bias of each neuron in the neural network model are randomly initialized, and the iteration cycle is set to P times, the learning rate S and the regularization parameter σ; where the learning rate is the step size for updating model parameters such as weights and biases at each iteration, and the regularization parameter σ is the coefficient to prevent the model from overfitting; according to the existence of N transmitted and returned interference light signals, a total of N neurons in the input layer are set, and each neuron corresponds to an interference signal of an optical path; then, the training set data propagates forward in the network into the hidden layer, and a total of M hidden layers are set, where M is a natural number, and each layer needs to be weighted summed and activated. Function processing; Softmax is selected as the activation function of the neural network model, and its expression is: Among them, z i is the linear output of the i-th category; When the training set data comes out of the hidden layer and enters the output layer, the cross entropy loss is used as the loss function at the output layer to calculate the loss between the predicted output and the actual label, and the gradient information of the function and the back propagation algorithm are used to calculate the gradient corresponding to the weight and bias of each neuron. According to the gradient results, the gradient descent algorithm is used to update the weights and biases in the model; the cross entropy loss function expression is as follows: Among them, y o,i is the judgment indicator, that is, category i is the correct classification, y o,i The value is 1, otherwise it is 0. o,i The model predicts the probability that the signal data belongs to category i; Afterwards, the above method is repeated for P iterations until the performance of the neural network model on the training set converges or meets the predetermined iteration cycle; during the iteration process, the neural network model needs to use the validation set to evaluate the performance of the model after each iteration, so as to optimize the original model settings such as the number of hidden layers M, the learning rate S or the regularization parameter σ while updating the model weights and biases, and then select the model with the best performance as the final neural network model after the iteration; in addition, after the model is established, a test set is required to evaluate the prediction effect of the model to determine the generalization ability of the model; finally, the ciliary array is placed in an environment with an unknown sound source to obtain the signal generated by the sound source, and these signal data are demodulated and substituted into the trained neural network model for prediction and classification, thereby realizing the detection and intelligent recognition of underwater acoustic targets.

7. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 6, characterized in that: You can also choose BP neural network, convolutional neural network, generative adversarial network, and recurrent neural network models.

8. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 6, characterized in that: Set the iteration period P to 800 times, the learning rate S to 0.001, and the regularization parameter σ to 0.

01.

9. A low-frequency underwater acoustic signal detection method based on a bionic optical fiber ciliary array according to claim 6, characterized in that: The activation function can also be ReLU, Tanh, or Softmax.

10. A hydroacoustic signal detection device based on the method according to any one of claims 1 to 9, comprising: Laser (1), optical attenuator (2), circulator (3), coupler (4), optical fiber distribution structure (5), fixture (6), reflector (7), photodetector (8), oscilloscope (9) and host computer (10); The optical fiber distribution structure (5) comprises a transmission optical fiber (501), a delay optical fiber (502), a ciliary array (503) and a coating (504), which are fixed by a clamp (6) so that the optical fiber end face and the reflector (7) form an FP cavity, wherein the ciliary array (503) is composed of N end-face coated optical fibers, and the lengths of the delay optical fibers (502) corresponding to each optical path are different; after the laser is emitted from the laser (1), it first passes through the optical attenuator (2) to adjust its light intensity, then passes through the circulator (3) and is incident on the coupler (4) to be divided into N beams of light paths and enter the optical fiber distribution structure (5), wherein the transmission optical fiber (501) is used for the propagation of optical signals, and the delay optical fiber (502) is used for controlling The transmission rates of N optical paths are multiplexed to prevent interference between signals. The ciliary array (503) as a sensing object will be affected by external sound waves and vibrate. The coating (504) is used to enhance the interference effect of the FP cavity. Then the optical signal reflected from the FP cavity passes through the coupler (4) and the circulator (3) again and is incident on the photodetector (8), so that the optical signal is first converted into an electrical signal and the electrical signal is collected using an oscilloscope (9). Then the host computer (10) demodulates the signal to obtain the relationship between the light intensity and the cavity length of the FP cavity and the relationship between the cavity length and the sound pressure. Finally, the demodulated data is input into the trained neural network model to identify the type of external sound source.

Images