Air drum structure and low-frequency sound signal detection device and method based on air drum structure

Abstract

The invention relates to the field of optical fiber sensing and the field of deep learning, in particular to an air drum structure and a low-frequency acoustic signal detection device and method based on the air drum structure, which combine the advantages of an F-P cavity formed by double-fiber Bragg gratings, a conical fiber and the air drum structure and fully play the advantages of simple structure, high sensitivity, easy multiplexing and the like of the optical F-P cavity. Round diaphragms with different areas and different corresponding resonant frequencies are prepared by considering the characteristic that the fiber Bragg grating only reflects specific wavelength and the natural high sensitivity characteristic of the conical fiber and combining the narrow-band filtering advantage of the round diaphragms, so that the empty drum structure detects sound waves with different frequencies, and the detection accuracy is improved. The frequency of the external sound wave can be directly obtained without additional Fourier transform and filtering operation, and intelligent autonomous identification of the target sound signal is realized; compared with a traditional low-frequency sound signal sensing system, the system is more sensitive, compact and efficient, and has the advantages of narrow-band filtering, real-time spectrum extraction, autonomous target recognition and the like.

Description

Technical Field

[0001] The present invention relates to the fields of optical fiber sensing and deep learning, and in particular to a hollow drum structure, and a low-frequency sound signal detection device and method based on the structure. The device uses the hollow drum structure as a primitive, is equipped with a fiber Bragg grating, a circular diaphragm and combines a neural network algorithm. The device is particularly suitable for early warning detection and intelligent identification of low-frequency sound signals generated by low-altitude drones and natural disasters. Background Art

[0002] With the development of drone technology and the frequent occurrence of natural disasters, how to detect the existence of low-altitude drones or the occurrence of natural disasters is an urgent problem to be solved. Traditional low-frequency detection technology faces many difficulties, so it is of great significance to develop an acoustic signal detection device that can be applied to low-frequency bands and has extremely high sensitivity and intelligent recognition capabilities.

[0003] Compared with traditional electrical sensing technology, optical fiber sensing has the advantages of anti-electromagnetic interference, corrosion resistance, small size and light weight. It has gradually been widely used in the field of low-frequency acoustic signal detection. At present, the sensing technology used to detect low frequencies is mainly distributed optical fiber acoustic wave sensing technology (Distributed Acoustic Sensing, DAS). Representative achievements in the literature reports on DAS technology include: Chinese invention patent application "A method and system for real-time detection and positioning of low-altitude unmanned aerial vehicles" (application number CN202310503374.4, published on 2023-10-31), " Distributed Fiber Optic Sensing Seismograph ( Distributed fiber optic sensing seismograph and its application [J], Petroleum Geophysical Prospecting, 2022, 61(1): 41-49), Yu Yongshuang proposed "Distributed fiber optic acoustic sensing seismic detector based on DAS" (Design of distributed fiber optic acoustic sensing seismic detector based on DAS [D], Jilin: Changchun, Changchun University of Science and Technology, 2022).

[0004] The above-mentioned existing technology adopts the DAS technology path based on optical fiber sensing to realize the detection of low-frequency sound signals, which can improve the sound wave detection efficiency to a certain extent, but it needs to be further improved in terms of sensitivity improvement, demodulation algorithm optimization, preparation cost reduction, large-scale array multiplexing, and target autonomous identification. Summary of the invention

[0005] The present invention aims to solve the defects of the prior art and proposes a hollow drum structure, a low-frequency sound signal detection device and method based on the structure, so as to realize acoustic target detection with compact structure, extremely high sensitivity and autonomous target identification, which is particularly suitable for the detection and intelligent identification of low-frequency sound signals generated by low-altitude drones and natural disasters.

[0006] The technical solution adopted by the present invention is:

[0007] A hollow drum structure, the hollow drum structure is a flat drum-shaped cavity with a hollow interior, and the surface structure is composed of a plurality of sound tongues 601 of different sizes, the sound tongues 601 and the drum-shaped cavity form a resonance cavity to achieve acoustic and structural sensitivity enhancement, the center of each sound tongue 601 is hollowed out and embedded with a circular diaphragm 604, a tapered optical fiber 603 is combined with the circular diaphragm 604 by gluing, and fiber Bragg gratings 602 are distributed on both sides of the tapered optical fiber 603, and the circular diaphragm 604 is used to pick up incident sound wave signals, convert the intensity of the incident sound wave signals into the vibration amplitude of the diaphragm, and then improve the acoustic and structural sensitivity. The length of the tapered optical fiber 603 on the variable diaphragm, the tapered optical fiber 603 is used to convert the vibration amplitude of the diaphragm into the phase change of the optical signal, thereby changing the intensity of the interference light output; there are N of the above structures, N is a positive integer, and the N sound tongues 601 are of different sizes, and the corresponding N circular diaphragms 604 are also of different sizes. The two fiber Bragg gratings 602 on both sides of the tapered optical fiber 603 have the same properties and constitute an FP cavity. The reflection spectra of the fiber Bragg gratings 602 corresponding to different FP cavities are different, so that the reflection center wavelength of each FP cavity is different.

[0008] Preferably, the circular diaphragm 604 adopts nonlinear amplification based on the resonant working mode. When the sound wave frequency is close to the resonant frequency of a certain diaphragm and acts on the hollow drum structure, the diaphragm can achieve more obvious vibration compared with other diaphragms, thereby causing a significant change in the length of its corresponding FP cavity. This property can be used to effectively obtain the sound wave frequency, thereby achieving high sound pressure sensitivity.

[0009] Preferably, the circular diaphragm 604 adopts narrow-band filtering based on the control of the resonance working mode, that is, the resonance frequencies corresponding to the diaphragms of different sizes in the hollow drum structure are not consistent, so that each diaphragm will only produce a significant vibration effect when it is affected by the corresponding resonance frequency. This method realizes the recognition of the sound wave spectrum characteristics;

[0010] Preferably, the sensing structure formed by the combination of the sound tongue 601, the circular diaphragm 604, and the FP cavity composed of the adjacent fiber Bragg grating 602 and the tapered optical fiber 603 can be prepared according to the needs, thereby realizing the resonant absorption of multi-frequency sound waves.

[0011] Preferably, the FP cavity formed by the two fiber Bragg gratings 602 and the tapered optical fiber 603 can be more affected by the circular diaphragm while reflecting the specific Bragg wavelength optical signal, further improving the sound pressure sensitivity;

[0012] The present invention also proposes a low-frequency sound signal detection device based on the above hollow drum structure, comprising a broadband laser 1, a first 1×2 coupler 21, a time-delay optical fiber 3, a second 1×2 coupler 22, a piezoelectric ceramic 4, a circulator 5, a hollow drum structure 6, an optical isolator 7, a demultiplexer group 8, a photodetector group 9, an acquisition card 10 and a host computer 11;

[0013] The laser emitted from the broadband laser 1 is first divided into two optical signals of equal intensity through the first 1×2 coupler 21, one of which has a delay optical fiber 3 so that there is a phase difference between the optical signal of the optical signal and the optical signal of the other path, and the other has a piezoelectric ceramic 4, which is controlled by the host computer 11 to keep the phase difference between the two paths stable, and then the two optical signals are emitted through the second 1×2 coupler 22 and successively enter one port of the circulator 5 and are emitted from the second port of the circulator 5, and then the two optical signals successively enter the hollow drum structure 6, and successively pass through the N hollow drum structure 6. The FP cavity is composed of a tapered optical fiber and fiber Bragg gratings distributed on both sides of the tapered optical fiber. There are two beams of reflected light inside each FP cavity. One beam is reflected by the first fiber Bragg grating in the FP cavity, and the other beam is reflected by the second fiber Bragg grating after passing through the first grating. Since there are two optical signals entering the hollow drum structure 6 successively, there are four types of reflected light in total. According to the matching interference theory, since the length of the delay fiber 3 is equal to the length of the FP cavity (that is, the sum of the length of a single fiber Bragg grating area and the tapered optical fiber), the phase difference generated at the delay fiber 3 is equal to the F The difference between the phase differences generated at the P cavity meets the coherence condition of light, so an interference light signal will eventually be generated at each FP cavity. The corresponding wavelengths of the interference light signals generated by each FP cavity are different. Subsequently, all interference light signals return to the second port of the circulator 5 and are emitted from the third port. The remaining unreflected light signals are emitted to the optical isolator 7 after passing through the Nth FP cavity to prevent them from being reflected back to the original optical path; then, the interference light signals emitted from different FP cavities enter the de-wavelength division multiplexer group 8 composed of N de-wavelength division multiplexers respectively, and the de-wavelength division multiplexer group 8 separates the interference light signals of different wavelengths. The interference light signal is separated from the interference light signal of other wavelengths, and then the interference light signal is converted into an electrical signal by a photoelectric detector group 9 composed of N photoelectric detectors and enters the acquisition card 10 to obtain the interference light signal changes of each group of FP cavities. Finally, the obtained data is transmitted to the host computer 11 and the PGC algorithm (Li Heping, Research on Fiber Optic Vector Hydrophone Based on FBG-FP Cavity [D], Sichuan: Chengdu, University of Electronic Science and Technology of China, 2017) is used to demodulate the data. The demodulated target signal data is sent to the trained neural network model to successfully identify the type of external sound source.

[0014] The present invention also proposes a low-frequency sound signal detection method based on the above-mentioned low-frequency sound signal detection device, and the method is divided into the following steps:

[0015] S1. Modulation and transmission of optical signals between two 1×2 couplers:

[0016] S1.1 The broadband laser 1 is used as a light source to emit an initial optical signal, which then enters the first 1×2 coupler 21 with a splitting ratio of 50:50 and is divided into two optical signals: one optical signal passes through the delay optical fiber 3 to generate a phase difference with the other optical signal, and the phase difference is used to alleviate the situation where the subsequent FP cavity length is large different from the coherence length of the light and cannot interfere with each other, and thus causes the two optical signals to enter the FP cavity in a sequence, wherein a piezoelectric ceramic 4 is added to the second optical signal to maintain the stability of the phase difference, and the piezoelectric ceramic 4 is controlled by the host computer 11;

[0017] S1.2 The two optical signals successively enter the second 1×2 coupler 22 with the same performance as the first 1×2 coupler 21, then enter from one port of the circulator 5 and output from the second port, and successively enter the hollow drum structure 6; there are N series-connected FP cavities composed of two fiber Bragg gratings and a tapered optical fiber in the hollow drum structure 6, N is a positive integer, and each FP cavity can reflect a different central wavelength of the optical signal, these FP cavities are respectively close to N circular diaphragms of different sizes, and the vibration of the circular diaphragms is used to cause the length of the FP cavity to change. When the sound wave is at the resonant frequency of a certain circular diaphragm, the diaphragm will produce significant deformation, thereby leading to a significant change in the length of the FP cavity, and then the phase of the light transmitted in the fiber Bragg grating will change at the same time, affecting the intensity of the final interference light; in addition, the interior of the FP cavity is composed of a tapered optical fiber, and the tapered optical fiber is more sensitive to pressure than ordinary optical fibers, and is more affected by the action of the circular diaphragm, so the change in cavity length is more obvious.

[0018] S2. Propagation and matching interference of optical signals in the FP cavity composed of two fiber Bragg gratings and a tapered fiber:

[0019] After the two optical signals pass through N FP cavities in sequence, each time the optical signals pass through an FP cavity, the optical signals that meet the Bragg wavelength condition of the fiber Bragg grating will be reflected inside the corresponding FP cavity, and two types of reflected light will be generated at this time: one is the optical signal that is directly reflected after passing through the first fiber Bragg grating in the FP cavity, and the other is the optical signal that passes through the first fiber Bragg grating and is reflected by the second fiber Bragg grating and then re-transmitted back to the first fiber Bragg grating to return to the original path. Since the distance between the two fiber Bragg gratings is relatively far (usually several decimeters), there is a large phase difference between the two types of reflected light signals; since there are two optical signals that enter the hollow drum structure one after another, there are a total of four types of reflected light between the two optical signals: the first is the reflected light that is reflected by the first fiber Bragg grating in the FP cavity after passing through the delayed optical fiber; the second is the reflected light that is reflected by the FP cavity after passing through the delayed optical fiber. The first type is the reflected light reflected by the second fiber Bragg grating in the FP cavity; the third type is the reflected light reflected by the first fiber Bragg grating in the FP cavity without passing through the delayed fiber; the fourth type is the reflected light reflected by the second fiber Bragg grating in the FP cavity without passing through the delayed fiber; according to the matched interference theory (Lin Huizu, Research on Key Technologies of Fiber Bragg Grating Hydrophone Array Based on Matched Interference [D], Hunan: Changsha, National University of Defense Technology, 2013), there is only one situation in which these four reflected lights will interfere with each other, that is, the second reflected light and the third reflected light interfere with each other, and the remaining optical signal that does not interfere but is reflected back by the fiber Bragg grating is retained in the output signal in the form of a direct current. Finally, N interference lights of different wavelengths are generated in the N FP cavities and return to the two ports of the circulator. The remaining optical signal that is not reflected is emitted to the optical isolator 7 after passing through the Nth FP cavity to prevent it from being reflected back to the original optical path;

[0020] S3. Detection of interference light signals and target recognition:

[0021] S3.1 Detection:

[0022] N interference light signals of different wavelengths are output from the third port after returning to the second port of the circulator. Considering that the wavelengths of the N interference lights are inconsistent, a de-wavelength division multiplexer group 8 composed of N de-wavelength division multiplexers of different wavelengths is set to separate the wavelengths of the interference light signals. Then, the N optical signals are respectively converted into electrical signals by a photoelectric detector group 9 composed of N photoelectric detectors, and then the signals are uniformly collected by an acquisition card 10. The acquired data can be demodulated by PGC to obtain the target acoustic wave signal.

[0023] S3.2 Identification

[0024] Inputting the obtained target sound wave signal into the trained neural network model can identify the type of external sound source. Since the area of ​​each circular diaphragm is different, the corresponding resonant frequency is also different. Therefore, when the resonant frequency of a diaphragm is consistent with the external sound wave frequency, its vibration amplitude will be more significant than other diaphragms, and the corresponding FP cavity length will change more, which will greatly affect the output intensity of the interference light on the FP cavity. Therefore, the sound wave frequency can be used as the actual label to classify these data results to achieve early warning detection and intelligent identification of low-altitude drones and natural disasters;

[0025] The demodulated target signal data is input into the trained neural network model for identifying external sound sources. According to the existing neural network models, common models include BP neural network, convolutional neural network, generative adversarial network, recurrent neural network, deep belief network, deep residual network, long short-term memory network, etc. Among them, the long short-term memory network (LSTM for short) was first proposed by Hochreiter and Schmidhuber in 1997 (Long short-term memory [J], Neural Computation, 1997, 9 (8): 1735-1780). Compared with other models, this neural network model has the characteristics of being able to handle long-term dependencies, having dynamic memory and forgetting capabilities, strong generalization capabilities and wide applicability. Therefore, the long short-term memory network is selected as the main body to construct the neural network model. The following are the steps of constructing and training the LSTM neural network model:

[0026] S3.2.1 After using the acoustic characteristic data of the known sound source obtained in the actual test as the driving signal of the transducer, the transducer is placed near the hollow drum structure, so that the circular diaphragm in the hollow drum structure vibrates and thus changes the length of the FP cavity. Then, the acoustic signal data generated by the known sound source are collected and demodulated, and the demodulated data obtained are divided into a training set and a validation set in a quantity ratio of 7:3: the training set is used to train and preliminarily establish a neural network model, and the validation set is used to evaluate the performance of the neural network model obtained in each training; each demodulated data in the training set and the validation set is classified according to the sound wave frequency they represent, and each category is used as a true label, that is, as a basis for judging the prediction effect of the model during subsequent training, validation and testing. A total of N categories are set, representing the resonant frequencies of N diaphragms of different areas;

[0027] S3.2.2 After the input layer of the neural network model receives the training set data, the weights and biases of the input layer neurons are randomly initialized, and the iteration cycle is set to C times, the learning rate S and the regularization parameter σ. Among them, the learning rate S is the amplitude of updating the model parameters (such as weights or biases) at each iteration, and the regularization parameter σ is used to avoid overfitting of the model. The iteration cycle C is usually set to 600 times, the learning rate S is 0.001, and the regularization parameter σ is 0.01. According to S1, 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 light signal at a wavelength.

[0028] S3.2.3 The training set data propagates forward in the network into the hidden layer. There are M hidden layers in total, where M is a positive integer. Each layer needs to be weighted summed and activated to enable the model to cope with complex training tasks. Commonly used activation functions include Sigmoid, ReLU, Softmax, etc. Softmax is better at multi-target classification and recognition than other types of activation functions, so Softmax is preferred as the activation function of the neural network model.

[0029] S3.2.4 After the training set data passes through the hidden layer and enters the output layer, the Hinge loss function is preferably used as the loss function to calculate the loss between the predicted output and the true label, and the gradient information of the function is combined with the back propagation algorithm (Beyond regression: New tools for prediction and analysis in the behavioral sciences [D], 1974, Harvard University) to calculate the gradient corresponding to the model weight (i.e., the weight coefficient of each output value in each layer of the model) and bias (i.e., a constant term added to the output of each layer of the model). According to the gradient feedback, the gradient descent algorithm Adam (Adam: A Method for Stochastic Optimization [C], International Conference on Learning Representations, 2015, USA) is used to update the weight and bias in the model in real time;

[0030] S3.2.5 Repeat the above method for C iterations until the performance of the neural network model on the training set reaches the convergence condition or reaches the predetermined number of iterations C. In the process of building the neural network model, the validation set is needed to evaluate the model performance after each iteration, including accuracy, recall rate and F1 score, and use the above indicators to optimize the learning rate S and the size of the regularization parameter σ for the next iteration; after the iteration process is completed, the model with the best performance on the validation set will be selected as the final neural network model;

[0031] S3.2.6 places the hollow drum structure in an environment with an unknown sound source, and uses the host computer 11 to collect the interference light signals generated by the sound source; after the host computer 11 demodulates and processes these signal data, it inputs them into the neural network model for prediction and classification, so as to realize intelligent recognition of low-frequency sound targets.

[0032] The present invention innovatively combines the advantages of the FP cavity composed of dual fiber Bragg gratings, tapered optical fiber and hollow drum structure, and proposes a low-frequency sound signal detection device based on the hollow drum structure, which gives full play to the advantages of the optical FP cavity structure, such as simple structure, high sensitivity, and easy multiplexing. Considering that the fiber Bragg grating only reflects specific wavelengths and the natural high sensitivity of the tapered optical fiber, and combining the narrowband filtering advantages of the circular diaphragm, circular diaphragms of different sizes are prepared, and the corresponding resonant frequencies are different, so that the hollow drum structure can detect sound waves of different frequencies, and the frequency composition of external sound waves can be directly obtained without additional Fourier transform and filtering operations, realizing intelligent and autonomous recognition of target sound signals. Compared with traditional low-frequency sound signal sensing systems, this 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.

[0033] The invention provides a high-sensitivity detection and intelligent identification device for low-frequency sound signals based on a multi-frequency resonance mechanism. Compared with the prior art, the invention has the following beneficial technical effects: firstly, a hollow drum structure is innovatively introduced, and an FP cavity composed of a double fiber grating is introduced on the outer surface. The center of the sound tongue of the hollow drum is hollowed out and a circular diaphragm is embedded. The reflection of the sound wave inside the hollow drum is continuously applied to the diaphragm, thereby realizing high-sensitivity detection of low-frequency disturbance signals and a multiplexed array in a compact structure. Secondly, when a single diaphragm is acted upon by a resonant frequency sound wave, it has a nonlinear amplification effect on the signal, and thus is comparable to other diaphragms. The vibration amplitude of the diaphragm is more obvious than that of the diaphragm, which can achieve extremely high sensitivity detection; furthermore, by controlling the area of ​​the diaphragm, the optimal design of the resonant frequency 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 intelligent target recognition; in addition, the introduction of tapered optical fiber as the FP cavity can further enhance the effect of the external environment on the FP cavity; finally, each diaphragm can be simulated as a neuron cell, and the demodulated target sound signal is used as the input of the neural network to achieve intelligent recognition of low-frequency sound targets, which can maximize the detection and early warning of low-altitude drones and natural disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] 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:

[0035] Figure 1 It is a schematic diagram of the assembly structure of a low-frequency sound signal high-sensitivity detection and intelligent identification device based on a multi-frequency resonance mechanism according to the present invention;

[0036] Figure 2 It is a schematic diagram of the three-dimensional structure of the hollow drum structure 6 of the present invention;

[0037] Figure 3 is a schematic cross-sectional view of the three-dimensional structure of the hollow drum structure 6 of the present invention;

[0038] Figure 4 It is a schematic diagram of the three-dimensional structure when the FP cavity composed of the tapered optical fiber 603 and the fiber Bragg gratings 602 on both sides of the present invention is mounted on the circular diaphragm 604 in the middle of the embedded sound tongue 601;

[0039] Figure 5 is a schematic diagram of the deformation of the circular diaphragm 604 of the present invention when it is acted upon by sound waves;

[0040] Figure 6 It is a schematic diagram of collecting external sound wave signals and then importing them into a neural network model for real-time target recognition according to the present invention;

[0041] Description of reference numerals:

[0042] 1: broadband laser; 2: 1×2 coupler; 3: time-delay optical fiber; 4: piezoelectric ceramics; 5: circulator; 6: hollow drum structure; 7: optical isolator; 8: demultiplexer group; 9: photodetector group; 10: acquisition card; 11: host computer.

[0043] 601: sound tongue; 602: fiber Bragg grating; 603: tapered optical fiber; 604: circular diaphragm. DETAILED DESCRIPTION

[0044] 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.

[0045] Figure 1 is a structural schematic diagram of an embodiment of a low-frequency sound signal high-sensitivity detection and intelligent identification device based on a multi-frequency resonance mechanism according to the present invention, Figure 2 is a schematic diagram of the three-dimensional structure of the hollow drum structure 6 of the present invention, Figure 3 is a schematic cross-sectional view of the three-dimensional structure of the hollow drum structure 6 of the present invention, Figure 4 6 is a schematic diagram of a three-dimensional structure when the FP cavity composed of the tapered optical fiber 603 and the fiber Bragg gratings 602 on both sides of the present invention is mounted on the central circular diaphragm 604 embedded in the sound tongue 601, Figure 5 is a schematic diagram of the deformation of the circular diaphragm 604 of the present invention when it is acted upon by sound waves, Figure 6 It is a schematic diagram of collecting external sound wave signals and then importing them into a neural network model for real-time target recognition as described in the present invention.

[0046] The present invention discloses a low-frequency sound signal detection and intelligent identification device based on a hollow drum structure, the device structure comprising: a broadband laser 1, a 1×2 coupler 2, a time-delay optical fiber 3, a piezoelectric ceramic 4, a circulator 5, a hollow drum structure 6, an optical isolator 7, a demultiplexer group 8, a photodetector group 9, a collection card 10 and a host computer 11;

[0047] The hollow drum structure 6 is a hollow structure inside, and the surface structure is composed of a sound tongue 601. A circular diaphragm 604 is embedded in the middle of the sound tongue 601, and a tapered optical fiber 603 is covered on the top and fiber Bragg gratings 602 are distributed on both sides of the optical fiber. There are four of the above structures, and the four sound tongues 601 are of different sizes. The sizes of the four circular diaphragms 604 corresponding to each other are also different. The fiber Bragg gratings 602 on both sides of a single tapered optical fiber are consistent and form an FP cavity. The refractive index modulation depth and grating period of the fiber Bragg grating 602 of each group of FP cavities are different, and thus the reflection center wavelengths of each FP cavity are different. In order to achieve the effect of low-frequency sound signal sensing, nitrile rubber is preferably used as the raw material when designing the four circular diaphragms, and its Young's modulus is 2.3×10 9 Pa, Poisson's ratio is 0.49, and density is 1300 kg / m 3 The designed radius is 0.05m, 0.1m, 0.3m, and 0.5m from small to large, and the corresponding thickness is 80μm. According to formula (4), the resonant frequencies of the four diaphragms are 143.3489Hz, 35.8372Hz, 3.9819Hz, and 1.4335Hz, respectively, which are consistent with the frequency band of low-frequency sound signals and can be used for detection and identification of this frequency band. When a 25mW laser is emitted from a broadband laser 1 with a bandwidth of 1520nm to 1580nm, the light source is first divided into two optical signals through a 1×2 coupler 2, with an intensity ratio of 50:50. One of the optical signals has a delay fiber 3, which causes a phase difference between the optical signal of the other optical signal. The other path has a piezoelectric ceramic 4 which can be controlled by the host computer 11 to make the phase difference between the two paths Keep stable, and then the two optical signals successively enter another 1×2 coupler 2 and enter one port of the circulator 5, and then enter the hollow drum structure 6 after being emitted from the second port of the circulator 5. Then the two optical signals successively pass through four FP cavities composed of adjacent fiber Bragg gratings 602 and tapered optical fibers 603, wherein the reflective central wavelengths between the four FP cavities are 1535nm, 1550nm, 1565nm, and 1580nm respectively. After passing through the fourth FP cavity, the remaining optical signal is emitted to the optical isolator 7 along with the optical fiber. There are two beams of reflected light inside each FP cavity, one beam is obtained after being reflected by the first grating, and the other beam is obtained after being reflected by the second grating after passing through the first grating. A phase difference will be generated between the two beams of reflected light. According to the matching interference theory, the phase difference Δφ generated at the delayed fiber and the phase difference generated at the FP cavity The difference between When the coherence condition of light is met, interference light signals will eventually be generated at each FP cavity, and the corresponding wavelengths of the respective interference light signals are different. Subsequently, all interference light signals return to the second port of the circulator 5 to be coupled into a total interference light signal and emitted from the third port. After that, light signals of different wavelengths enter the demultiplexer group 8 of the corresponding wavelength to be separated from light signals of other wavelengths, and the light signals are converted into electrical signals by the photoelectric detector group 9 and then enter the acquisition card 10 to obtain the light signal changes of each wavelength. Finally, the obtained signal data is imported into the host computer 11 for demodulation, wherein the normalized interference light intensity change of the FP cavity composed of a pair of fiber gratings is between 0.4 and 0.7. Finally, the obtained demodulated data is input into the trained long short-term memory network model to identify the type of external target sound signal.

[0048] The present invention is based on the following principles:

[0049] There is a phase difference between the optical signal corresponding to the optical path passing through the delay optical fiber and the other optical signal, and the process is as follows:

[0050] Assuming that the wavelengths of the optical signals in the two optical paths output from the coupler are both λ, the length of the delay fiber is D, and the refractive index of the fiber is n, a stable phase difference Δφ can be generated between the two optical signals, and its magnitude is:

[0051]

[0052] The coherence length l of light can be expressed as:

[0053]

[0054] Among them, λ is the central wavelength of the broadband light source, Δλ is the spectral width of the light source, and the length D of the delay fiber is much larger than the coherence length l. Therefore, the optical signals on the two optical paths have a sequence when passing through the second coupler, and no interference occurs.

[0055] When the sound wave frequency is at the resonant frequency of a circular diaphragm, the circular diaphragm will produce a more significant deformation, thereby driving the change of the FP cavity length. The deformation process of the circular diaphragm is as follows:

[0056] For a circular diaphragm, the deformation Δh at its center can be expressed as:

[0057]

[0058] Where μ is Poisson's ratio, r is the diaphragm radius, t is the diaphragm thickness, E is Young's modulus, P is the sound pressure, f is the sound wave frequency, f0 is the resonant frequency, and ξ is the damping coefficient. From formula (3), we can see that the closer the sound wave frequency is to the resonant frequency, the greater the deformation at the center of the diaphragm. The resonant frequency f0 of the diaphragm is:

[0059]

[0060] Where ρ is the density of the diaphragm. It can be seen from formula (3) that by changing the radius of the diaphragm, i.e. the area, the resonant frequency of the diaphragm can be effectively changed, which is helpful for the identification of different external sound wave signals. At the same time, since the tapered optical fiber and the diaphragm where the FP cavity is located are bonded together by gluing, the distance between the two can be regarded as zero. At this time, the tapered optical fiber is subjected to pressure and strain through the deformation of the circular diaphragm, which in turn causes the length of the FP cavity to change, thereby changing the intensity of the interference light.

[0061] The optical signal band that meets the Bragg wavelength condition of the fiber Bragg grating will be reflected inside the corresponding FP cavity, and the process is as follows:

[0062] According to the characteristics of fiber grating, when a beam of light passes through the grating, if the wavelength of the light satisfies:

[0063] λ m =2nΛ (5)

[0064] This light beam is called the Bragg wavelength and will be reflected by the grating. Where n is the fiber refractive index, Λ is the fiber grating period, and considering that the dual fiber grating needs to form an FP cavity structure, the grating used is a weak reflection grating to ensure that the subsequent optical signal can interfere normally.

[0065] The optical signal mainly generates two types of reflected light when passing through each FP cavity, and there are two optical signals so that there are four types of reflected light in total. According to the matching interference effect, there is interference between the four types of reflected light, and the process is as follows:

[0066] It is known that there is a phase difference of Δφ between the two optical signals, and the length of the delay fiber is D. Select one of the FP cavities as the object of discussion, and divide the incident optical signal into two cases for discussion. The first case is the optical signal that has not passed through the delay fiber: Assume that the amplitude of the optical signal is A, and the initial phase is δ; the length of the tapered fiber is d, and its refractive index is the same as that of the ordinary fiber, both are n; the reflectivity of the fiber gratings on both sides of the tapered fiber are R, the transmittance are T, and the grating area length is L. Then the reflected light can be expressed as:

[0067] The expression of reflected light directly reflected from the first fiber Bragg grating surface is:

[0068] E1=ARe iδ (6)

[0069] The expression of the reflected light that passes through the first fiber grating and then reflects on the second fiber grating surface and then passes through the first fiber grating again is:

[0070]

[0071] in, is the phase difference generated when passing through the FP cavity, which can be expressed as:

[0072]

[0073] The second case is the optical signal after the delayed optical fiber. The initial phase of the optical signal is δ+Δφ. Ignoring the attenuation of the optical signal during propagation, the two reflected lights are expressed as:

[0074] Similarly, the expression of the reflected light directly reflected on the first fiber grating surface is:

[0075] E3=ARe i(δ+Δφ) (9)

[0076] Similarly, the expression of the reflected light that passes through the first fiber grating, reflects on the second fiber grating surface, and then passes through the first fiber grating again is:

[0077]

[0078] By comparison, it is not difficult to find that among the four reflected lights, only the phase difference between E2 and E3 has the possibility of interference, that is, the phase difference compensation can be satisfied by adjusting the length of the delay fiber. Then the phase difference satisfies the coherence condition, and finally produces an interference effect. And through subsequent derivation, it can be known that the intensity I of the interference light in the FP cavity is:

[0079]

[0080] This indicates that the interference effect is best when the length of the delay fiber is closer to the sum of the lengths of a single fiber Bragg grating region and the tapered fiber.

Claims

1. A hollow drum structure, characterized in that: The hollow drum structure is a flat drum-shaped cavity with a hollow interior, and the surface structure is composed of a plurality of sound tongues (601) of different sizes. The sound tongues (601) and the drum-shaped cavity form a resonance cavity to achieve acoustic and structural sensitivity enhancement. The center of each sound tongue (601) is hollowed out and embedded with a circular diaphragm (604). The tapered optical fiber (603) is combined with the circular diaphragm 604, and the fiber Bragg gratings (602) are distributed on both sides of the tapered optical fiber (603). The circular diaphragm (604) is used to pick up the incident sound wave signal, convert the intensity of the incident sound wave signal into the vibration amplitude of the diaphragm, and then change the tapered shape on the diaphragm. The length of the optical fiber (603), the tapered optical fiber (603) is used to convert the vibration amplitude of the diaphragm into the phase change of the optical signal, thereby changing the intensity of the interference light output; there are N of the above structures, N is a positive integer, and the sizes of the N sound tongues (601) are different from each other, and the sizes of the N circular diaphragms (604) corresponding to each other are also different from each other, the two fiber Bragg gratings (602) on both sides of the tapered optical fiber (603) have the same properties and constitute an FP cavity, and the reflection spectra of the fiber Bragg gratings (602) corresponding to different FP cavities are different, so that the reflection center wavelength of each FP cavity is different.

2. A hollow drum structure according to claim 1, characterized in that: The circular diaphragm (604) is made of nitrile rubber.

3. A hollow drum structure according to claim 1 or 2, characterized in that: The circular diaphragm (604) adopts nonlinear amplification based on the resonant working mode. When the sound wave frequency is close to the resonant frequency of a certain diaphragm and acts on the hollow drum structure, the diaphragm can achieve more obvious vibration than other diaphragms, thereby causing a significant change in the length of the corresponding FP cavity. This property can be used to effectively obtain the sound wave frequency, thereby achieving high sound pressure sensitivity.

4. A hollow drum structure according to claim 1 or 2, characterized in that: The circular diaphragm (604) adopts narrow-band filtering based on the control of the resonant working mode, that is, the resonant frequencies corresponding to diaphragms of different sizes in the hollow drum structure are not consistent, so that each diaphragm will only produce a significant vibration effect when it is affected by the corresponding resonant frequency. This method realizes the recognition of the sound wave spectrum characteristics.

5. A hollow drum structure according to claim 1 or 2, characterized in that: The tapered optical fiber (603) and the circular diaphragm (604) are combined by gluing.

6. A hollow drum structure according to claim 1, characterized in that: The sensing structure is formed by combining the sound tongue (601), the circular diaphragm (604), and the FP cavity composed of the adjacent fiber Bragg grating (602) and the tapered optical fiber (603). The number N of the sensing structure can be prepared according to the needs, thereby realizing the resonant absorption of multi-frequency sound waves.

7. A hollow drum structure according to claim 1, characterized in that: The FP cavity formed by the two fiber Bragg gratings (602) and the tapered optical fiber (603) can be more significantly affected by the circular diaphragm while reflecting a specific Bragg wavelength optical signal, thereby further improving the acoustic pressure sensitivity.

8. A low-frequency sound signal detection device based on the hollow drum structure of claim 1, characterized in that: It includes a broadband laser (1), a first 1×2 coupler (21), a time-delay optical fiber (3), a second 1×2 coupler (22), a piezoelectric ceramic (4), a circulator (5), a hollow drum structure (6), an optical isolator (7), a demultiplexer group (8), a photodetector group (9), an acquisition card (10) and a host computer (11); Laser light emitted from a broadband laser (1) is first divided into two optical signals of equal intensity through a first 1×2 coupler (21), one of which has a time-delay optical fiber (3) so that there is a phase difference between the optical signal of the optical signal and the optical signal of the other, and the other has a piezoelectric ceramic (4). The piezoelectric ceramic (4) is controlled by an upper computer (11) to keep the phase difference between the two paths stable. Then, the two optical signals are emitted through a second 1×2 coupler (22) and successively enter one port of a circulator (5) and are emitted at a second port of the circulator (5). Then, the two optical signals are first Then it enters the hollow drum structure (6), and successively passes through N FP cavities in the hollow drum structure (6) which are composed of tapered optical fibers and fiber Bragg gratings distributed on both sides of the tapered optical fibers. There are two beams of reflected light inside each FP cavity, one beam is reflected by the first fiber Bragg grating in the FP cavity, and the other beam is reflected by the second fiber Bragg grating after passing through the first grating. Since there are two optical signals entering the hollow drum structure (6) successively, there are four types of reflected light in total. According to the matching interference theory, since the length of the delay optical fiber (3) is equal to the length of the FP cavity , that is, the difference between the phase difference generated at the delay optical fiber (3) and the phase difference generated at the FP cavity satisfies the coherence condition of light, so an interference optical signal will eventually be generated at each FP cavity, and the corresponding wavelengths of the interference optical signals generated by each FP cavity are different from each other. Subsequently, all interference optical signals return to the second port of the circulator (5) and are emitted from the third port. The remaining optical signals that are not reflected are emitted to the optical isolator (7) after passing through the Nth FP cavity to prevent them from being reflected back to the original optical path; then, the interference optical signals emitted from different FP cavities enter the N wavelength division multiplexing devices respectively. The demultiplexer group (8) is composed of a plurality of devices, the demultiplexer group (8) separates interference light signals of different wavelengths from interference light signals of other wavelengths, and then converts the interference light signals into electrical signals through a photoelectric detector group (9) composed of N photoelectric detectors, and then enters an acquisition card (10) to obtain the interference light signal changes of each group of FP cavities, and finally transmits the obtained data to a host computer (11) and demodulates the data using a PGC algorithm, and the demodulated target signal data is sent to a trained neural network model to successfully identify the type of external sound source.

9. A low-frequency sound signal detection method based on the low-frequency sound signal detection device according to claim 8, characterized in that: The method is divided into the following steps: S1. Modulation and transmission of optical signals between two 1×2 couplers: S1.1 A broadband laser (1) is used as a light source to emit an initial optical signal, which then enters a first 1×2 coupler (21) with a splitting ratio of 50:50 and is divided into two optical signals: one optical signal passes through a delay optical fiber (3) to generate a phase difference with the other optical signal, and the phase difference is used to alleviate the situation where interference cannot be caused by a large difference between the cavity length of the subsequent FP cavity and the coherence length of the light, and thus causes a sequence of the two optical signals when entering the FP cavity, wherein a piezoelectric ceramic (4) is added to the second optical signal to maintain the stability of the phase difference, and the piezoelectric ceramic (4) is controlled by a host computer (11); S1.2 The two optical signals successively enter the second 1×2 coupler (22) having the same performance as the first 1×2 coupler (21), then enter from one port of the circulator (5) and output from the second port, and successively enter the hollow drum structure (6); there are N FP cavities in series in the hollow drum structure (6) consisting of two fiber Bragg gratings and a tapered optical fiber, N is a positive integer, and each FP cavity can reflect different central wavelengths of optical signals, these FP cavities are respectively close to N circular diaphragms of different sizes, and the vibration of the circular diaphragms is used to cause the length of the FP cavity to change. When the sound wave is at the resonant frequency of a certain circular diaphragm, the diaphragm will produce significant deformation, thereby driving the length of the FP cavity to change significantly, and then the phase of the light transmitted in the fiber Bragg grating will change at the same time, affecting the intensity of the final interference light; in addition, the interior of the FP cavity is composed of a tapered optical fiber, which is more sensitive to pressure than ordinary optical fibers and is more affected by the action of the circular diaphragm, so the change in cavity length is more obvious; S2. Propagation and matching interference of optical signals in the FP cavity composed of two fiber Bragg gratings and a tapered fiber: After the two optical signals pass through N FP cavities in sequence, each time the optical signals pass through an FP cavity, the optical signals that meet the Bragg wavelength conditions of the fiber Bragg grating will be reflected inside the corresponding FP cavity, and two types of reflected light will be generated at this time: one is the optical signal that is directly reflected after passing through the first fiber Bragg grating in the FP cavity, and the other is the optical signal that passes through the first fiber Bragg grating and is reflected by the second fiber Bragg grating and then re-transmitted back to the first fiber Bragg grating to return to the original path. Since the distance between the two fiber Bragg gratings is several decimeters, there is a large phase difference between the two types of reflected light signals; since there are two optical signals that enter the hollow drum structure one after another, there are four types of reflected light between the two optical signals: the first is the reflected light reflected by the first fiber Bragg grating in the FP cavity after passing through the delayed optical fiber; the second is the reflected light reflected by the second fiber Bragg grating in the FP cavity after passing through the delayed optical fiber; The third type is the reflected light that is reflected by the first fiber Bragg grating in the FP cavity without passing through the delay fiber; the fourth type is the reflected light that is reflected by the second fiber Bragg grating in the FP cavity without passing through the delay fiber; according to the matched interference theory, there is only one situation in which these four reflected lights will interfere with each other, that is, the second reflected light and the third reflected light interfere with each other, and the remaining optical signal that does not interfere but is reflected back by the fiber Bragg grating is retained in the output signal in the form of a direct current, and finally N interference lights of different wavelengths are generated in the N FP cavities and return to the two ports of the circulator, and the remaining optical signal that is not reflected is emitted to the optical isolator (7) after passing through the Nth FP cavity to prevent it from being reflected back to the original optical path; S3. Detection of interference light signals and target recognition: S3.1 Detection: N interference light signals of different wavelengths are output from the third port after returning to the second port of the circulator. Considering that the wavelengths of the N interference lights are inconsistent, a de-wavelength division multiplexer group (8) composed of N de-wavelength division multiplexers of different wavelengths is set to separate the wavelength bands of the interference light signals. Then, the N optical signals are respectively converted into electrical signals by a photoelectric detector group (9) composed of N photoelectric detectors, and then the signals are uniformly collected by a collection card (10). The acquired data is input into a host computer (11) and demodulated by a PGC to obtain a target acoustic wave signal. S3.2 Identification The demodulated target signal data is input into the trained neural network model to identify the external sound source. The long short-term memory network is selected as the main body to build the neural network model. The following are the steps for building and training the LSTM neural network model: S3.2.1 After using the acoustic characteristic data of the known sound source obtained in the actual test as the driving signal of the transducer, the transducer is placed near the hollow drum structure, so that the circular diaphragm in the hollow drum structure vibrates and thus changes the length of the FP cavity. Then, the acoustic signal data generated by the known sound source are collected and demodulated, and the demodulated data obtained are divided into a training set and a validation set in a quantity ratio of 7:3: the training set is used to train and preliminarily establish a neural network model, and the validation set is used to evaluate the performance of the neural network model obtained in each training; each demodulated data in the training set and the validation set is classified according to the sound wave frequency they represent, and each category is used as a true label, that is, as a basis for judging the prediction effect of the model during subsequent training, validation and testing. A total of N categories are set, representing the resonant frequencies of N diaphragms of different areas; S3.2.2 After the input layer of the neural network model receives the training set data, the weights and biases of the input layer neurons are randomly initialized, and the iteration cycle is set to C times, the learning rate S and the regularization parameter σ, where the learning rate S is the amplitude of updating the model parameters at each iteration, and the regularization parameter σ is used to avoid overfitting of the model; according to S1, there are N interference light signals transmitted back, and the number of neurons in the input layer is N, and each neuron corresponds to an interference light signal at a wavelength; S3.2.3 The training set data propagates forward in the network into the hidden layer. The number of hidden layers is M, where M is a positive integer. Each layer needs to be weighted summed and activated to enable the model to cope with complex training tasks. Softmax is selected as the activation function of the neural network model. S3.2.4 After the training set data passes through the hidden layer and enters the output layer, the Hinge loss function is selected as the loss function to calculate the loss between the predicted output and the true label, and the gradient information of the function is combined with the back propagation algorithm to calculate the gradient corresponding to the weight and bias of the model. According to the gradient feedback, the gradient descent algorithm Adam is used to update the weight and bias in the model in real time; S3.2.5 Repeat the above method for C iterations until the performance of the neural network model on the training set reaches the convergence condition or reaches the predetermined number of iterations C; in the process of building the neural network model, it is necessary to use the validation set to evaluate the model performance after each round of iteration, including accuracy, recall rate and F1 score, and use the above indicators to optimize the learning rate S and the size of the regularization parameter σ for the next round of iteration; after the iteration process is completed, the model with the best performance on the validation set will be selected as the final neural network model; S3.2.6 The hollow drum structure is placed in an environment with an unknown sound source, and the upper computer (11) is used to collect the interference light signals generated by the sound source; after the upper computer (11) demodulates and processes these signal data, it is input into the neural network model for prediction and classification, thereby realizing intelligent recognition of low-frequency sound targets.

10. A low-frequency sound signal detection method according to claim 9, characterized in that: Set the iteration cycle C to 600 times, the learning rate S to 0.001, and the regularization parameter σ to 0.01.

Images