A sound source spatial positioning method, system, terminal and storage medium based on the fusion of artificial intelligence and optical fiber sensor

Abstract

The present invention belongs to the field of optical engineering technology and discloses a method, system, terminal, and storage medium for spatially locating a sound source based on the fusion of artificial intelligence and fiber optic sensors. The method comprises: obtaining the time difference between each element in an array of fiber optic sensor elements; using the time difference as the input of a sound source spatial localization model, and using the spatial position of the sound source in the array of fiber optic sensor elements as the output of the sound source spatial localization model; training and verifying the sound source spatial localization model using sound source signal samples at different spatial positions; and based on the trained sound source spatial localization model, utilizing the nonlinear relationship between time difference and spatial position to predict and output the spatial localization information of the sound source. The present invention proposes a spatial localization technology that integrates an artificial intelligence algorithm with an array of fiber optic vibration sensors. By establishing an accurate nonlinear relationship between time difference and spatial position, the method achieves high-precision spatial localization of the sound source.

Description

Technical Field

[0001] The present invention relates to the field of optical engineering technology, and in particular to a sound source spatial positioning method, system, terminal and storage medium based on the fusion of artificial intelligence and optical fiber sensors. Background Art

[0002] Locating and measuring vibration signals is crucial for improving marine safety, resource development efficiency, disaster warning capabilities, and structural health monitoring. Traditional positioning methods typically employ antenna arrays and microphone arrays, which typically use time differences to determine corresponding angles. However, in actual measurements, the speed of sound propagation is affected by environmental factors such as temperature, which can cause it to vary in speed. This results in a nonlinear relationship between the time difference and the sound source location, rather than a linear relationship directly derived from angle parameters. Using a simple linear relationship increases positioning errors and results in low positioning accuracy.

[0003] Therefore, the existing technology needs to be improved. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that, in response to the defects of the existing technology, the present invention provides a sound source spatial positioning method, system, terminal and storage medium based on the fusion of artificial intelligence and optical fiber sensors to solve the problem of low positioning accuracy of traditional positioning methods.

[0005] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0006] In a first aspect, the present invention provides a method for spatially localizing a sound source based on the fusion of artificial intelligence and optical fiber sensors, comprising:

[0007] Obtaining the time difference between each element in the optical fiber sensor array;

[0008] Using the time difference as input to a sound source spatial localization model, and using the sound source spatial position in the optical fiber sensor array as output of the sound source spatial localization model, the sound source spatial localization model is trained and verified using sound source signal samples at different spatial positions;

[0009] Based on the trained sound source spatial localization model, the nonlinear relationship between time difference and spatial position is used to predict and output the spatial localization information of the sound source.

[0010] In one implementation, obtaining the time difference between each element in the optical fiber sensor element array includes:

[0011] The acoustic signal / vibration signal caused by multiple vibration sources is decomposed by the variational mode decomposition algorithm, and the main components containing the vibration source information are determined by a preset threshold;

[0012] Different sound source information is identified through different frequencies, and the cross-correlation calculation is performed on the same-frequency signals detected by the four-point array elements to obtain the time difference between any two array elements.

[0013] In one implementation, decomposing the acoustic signal / vibration signal caused by multiple vibration sources using a variational mode decomposition algorithm and determining the main components containing vibration source information using a preset threshold includes:

[0014] The bandwidth of each modal component signal constituting the original signal in the frequency domain is estimated to be the minimum, forming a minimization problem;

[0015] Determining the constraints of the minimization problem and converting the minimization problem with the constraints into a minimization problem without constraints;

[0016] The optimization solution is obtained by the alternating direction multiplier method, and each submodal component signal is updated and optimized by the optimization algorithm;

[0017] The permutation entropy value is used as an effect indicator for decomposing the number of intrinsic mode functions, the preset threshold is determined by using an average value comparison method based on correlation analysis, and the intrinsic mode function is selected according to the preset threshold to obtain the main component.

[0018] In one implementation, performing cross-correlation calculation on the same-frequency signals detected by the four array elements includes:

[0019] Determine the distance between each array element and the sound source and the time it takes for the vibration of the sound source to propagate to each array element;

[0020] According to the time when the vibration of the sound source propagates to each array element, a cross-correlation maximum value is calculated using a cross-correlation algorithm, and the time difference between any two array elements is calculated based on the horizontal coordinate of the cross-correlation maximum value.

[0021] In one implementation, the sound source spatial localization model includes: a deep belief network and a back propagation neural network for integration;

[0022] Among them, the deep belief network includes: an input layer, a hidden layer and an output layer; the hidden layer is used to learn the characteristic information between time difference, signal amplitude and spatial position; the number of neurons corresponding to the input layer and the output layer corresponds to the number of characteristic quantities time difference, amplitude and spatial position respectively.

[0023] In one implementation, the training and verification of the sound source spatial localization model using sound source signal samples at different spatial positions includes:

[0024] Based on sound source signal samples at different spatial locations, multiple deep belief networks with different activation functions are used to perform ensemble learning in a back-propagation neural network to determine the number of hidden layers and corresponding numbers of neurons in the back-propagation neural network; wherein the input of the back-propagation neural network is the spatial positioning value output by the deep belief network under different activation functions, and the output is the spatial positioning calibration value under an ideal state;

[0025] The preprocessed data of the calibrated sound source position is used as the output of the sound source spatial localization model, and the acoustic data / vibration data corresponding to different positions collected are used as the input of the sound source spatial localization model to verify the sound source spatial localization model after integrated learning.

[0026] In one implementation, the method of performing integrated learning in a back-propagation neural network using multiple deep belief networks with different activation functions includes:

[0027] Based on the layer-by-layer unsupervised pre-training method and the global supervised fine-tuning method, multiple deep belief networks with different activation functions are trained. According to the accuracy of each deep belief network on the validation set, different weights are assigned to each deep belief network in a weighted manner to optimize the final accuracy.

[0028] In a second aspect, the present invention provides a sound source spatial positioning system based on the fusion of artificial intelligence and optical fiber sensors, comprising:

[0029] A time difference acquisition module is used to obtain the time difference between each array element in the optical fiber sensor array;

[0030] a sound source spatial localization model training and verification module, configured to use the time difference as the input of a sound source spatial localization model and the spatial position of the sound source in the optical fiber sensor array as the output of the sound source spatial localization model, and to train and verify the sound source spatial localization model using sound source signal samples at different spatial positions;

[0031] The sound source spatial positioning information output module is used to predict and output the spatial positioning information of the sound source based on the trained sound source spatial positioning model by utilizing the nonlinear relationship between time difference and spatial position.

[0032] In a third aspect, the present invention provides a terminal comprising: a processor and a memory, wherein the memory stores a sound source spatial positioning program based on the fusion of artificial intelligence and optical fiber sensors, and the sound source spatial positioning program based on the fusion of artificial intelligence and optical fiber sensors is used to implement the operation of the sound source spatial positioning method based on the fusion of artificial intelligence and optical fiber sensors as described in the first aspect when executed by the processor.

[0033] In a fourth aspect, the present invention also provides a storage medium, which is a computer-readable storage medium, and which stores a sound source spatial positioning program based on the fusion of artificial intelligence and optical fiber sensors. When the sound source spatial positioning program based on the fusion of artificial intelligence and optical fiber sensors is executed by a processor, it is used to implement the operation of the sound source spatial positioning method based on the fusion of artificial intelligence and optical fiber sensors as described in the first aspect.

[0034] The present invention adopts the above technical solution to achieve the following effects:

[0035] This invention obtains the time difference between each element in an optical fiber sensor array, uses the time difference as the input of a sound source spatial localization model, and uses the spatial position of the sound source in the optical fiber sensor array as the output of the sound source spatial localization model. The sound source spatial localization model is trained and verified using sound source signal samples at different spatial positions. Based on the trained sound source spatial localization model, the nonlinear relationship between time difference and spatial position is utilized to predict and output the spatial localization information of the sound source. This invention proposes a spatial localization technology that integrates an artificial intelligence algorithm with an optical fiber vibration sensor array. By establishing an accurate nonlinear relationship between time difference and spatial position, this technology achieves high-precision spatial localization of the sound source. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0037] Figure 1 It is a flow chart of the sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensor in the present invention.

[0038] Figure 2 It is a schematic diagram of the far-field sound source model in the present invention.

[0039] Figure 3 It is a schematic diagram of the positions of the sound source and eight sensors in the present invention.

[0040] Figure 4It is a schematic diagram of the sound source spatial positioning model in the present invention.

[0041] Figure 5 It is a functional principle diagram of a terminal in one implementation of the present invention.

[0042] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clear and distinct, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] Exemplary Methods

[0045] Locating and measuring vibration signals is crucial for improving marine safety, resource development efficiency, disaster warning capabilities, and structural health monitoring. Traditional positioning methods typically employ antenna arrays and microphone arrays, which typically use time differences to determine corresponding angles. However, in actual measurements, the speed of sound propagation is affected by environmental factors such as temperature, which can cause it to vary in speed. This results in a nonlinear relationship between the time difference and the sound source location, rather than a linear relationship directly derived from angle parameters. Using a simple linear relationship increases positioning errors and results in low positioning accuracy.

[0046] In response to the above technical problems, an embodiment of the present invention provides a sound source spatial positioning method based on the fusion of artificial intelligence and optical fiber sensors. This method obtains the time difference between each array element in the optical fiber sensor array element array, and can use the time difference as the input of the sound source spatial positioning model. The sound source spatial position in the optical fiber sensor array element array is used as the output of the sound source spatial positioning model. The sound source spatial positioning model is trained and verified through sound source signal samples at different spatial positions. Based on the trained sound source spatial positioning model, the nonlinear relationship between time difference and spatial position is used to predict and output the spatial positioning information of the sound source. An embodiment of the present invention proposes a spatial positioning technology that integrates an artificial intelligence algorithm with an optical fiber vibration sensor array. By constructing an accurate nonlinear relationship between time difference and spatial position, high-precision spatial positioning of the sound source is achieved.

[0047] Typically, the spacing between the array elements is much smaller than the distance between the sound source and the array. Therefore, in this embodiment, the sound source can be regarded as propagating in the form of a plane wave, such as Figure 2 shown. Figure 2Where P represents the far-field sound source signal, the array model has a total of M array elements, and d is the array element spacing. In this embodiment, the optical fiber sensor that can realize large-scale sound source positioning is studied. Therefore, the positioning of the near-field sound source signal is approximated as one-dimensional positioning, and the far-field sound source signal is mainly studied here. Figure 2 It can be seen that the angle between the line where the array is located and the direction of the sound source is a. The distance difference between the two array elements to the sound source can be represented by dcosa. Based on the cross-correlation operation, the time difference of the sound source arrival between the two array elements can be known, thereby obtaining the angle between the sound source and the array.

[0048] Based on the above theory, a four-element antenna array model structure is designed in this embodiment, such as Figure 3 As shown. Four sensors are placed in a three-dimensional Cartesian coordinate system, with sensor S0 as the origin. The other three sensors S1, S2, and S3 are located at a distance d from S0 on the x, y, and z axes, respectively. Sensors S0, S1, S2, and S3 form a set of rectangular tetrahedron-shaped first sensor clusters. The coordinates of the sound source P are (xp, yp, zp), and the coordinates of sensors S1, S2, S3, and S0 are (x1, y1, z1), (x2, y2, z2), (x3, y3, z3), and (x4, y4, z4), respectively. Figure 3 It can be seen from the equations x1 = x4 + d, y2 = y4 + d, and z3 = z4 + d. The propagation time difference from the sound source P to two different sensors is equal to the time difference between the sound waves reaching the corresponding two sensors. However, it is worth noting that the three angles (a, b, c) corresponding to these three time differences can only determine the straight line where the sound emission source is located. By adding a set of symmetrical (or identical) rectangular tetrahedral sensor clusters to another corner of the three-dimensional structure, and arranging the distance between the two sensor clusters so that the sound source is near-field compared to both, a second direction of the sound source can be obtained from the second sensor cluster. This second sensor cluster includes sensors S10, S12, S13, and S11. In this embodiment, these two sensor clusters are respectively encapsulated in two optical cables.

[0049] Traditional antenna array methods use time difference to determine angle for positioning. However, in actual measurements, the speed of sound propagation is affected by environmental factors such as temperature, which can cause its propagation speed to vary. This results in a nonlinear relationship between time difference and positioning. Using a simple linear relationship can increase positioning errors. Therefore, based on the proposed dot-matrix antenna array model, this embodiment proposes a sound source spatial localization method based on the fusion of artificial intelligence and fiber optic sensors. This method, a deep belief network localization method based on ensemble learning, constructs a nonlinear relationship between time difference and position, thereby achieving high-precision spatial localization of the sound source.

[0050] like Figure 1As shown, an embodiment of the present invention provides a sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors, comprising the following steps:

[0051] Step S100: obtaining the time difference between each element in the optical fiber sensor element array.

[0052] In this embodiment, the method is based on an artificial intelligence algorithm. As a tool with effective signal feature extraction and powerful nonlinear mapping capabilities, applying this algorithm to sensor positioning can alleviate the low positioning accuracy of traditional methods. Therefore, in this embodiment, the artificial intelligence algorithm is integrated with an optical fiber vibration sensor array to achieve spatial localization of sound sources.

[0053] It is understood that the artificial intelligence algorithm used in this embodiment is a sound source spatial localization algorithm, which is integrated with a fiber optic vibration sensor array to achieve spatial localization. First, an array of fiber optic sensor elements is deployed. The time difference between each element is then used as the input of a sound source spatial localization model, and the spatial position of the sound source is used as the output of the model. The model is trained and validated by collecting sound source signal samples at different spatial locations. By establishing an accurate nonlinear relationship between time difference and spatial position, high-precision spatial localization of the sound source is achieved.

[0054] Specifically, in one implementation of this embodiment, step S100 includes the following steps:

[0055] Step S101 : Decompose the acoustic signal / vibration signal caused by multiple vibration sources using a variational mode decomposition algorithm, and determine the main components containing vibration source information using a preset threshold.

[0056] In practical applications, it is unlikely that multiple wavefronts generated by different vibration sources will reach each array element sensor at the same frequency. Therefore, in this embodiment, the acoustic / vibration signals caused by multiple vibration sources can be decomposed using the Variational Mode Decomposition (VMD) algorithm. By setting a threshold, the main components containing the vibration source information are determined, so that different sound source information can be identified through different frequencies.

[0057] In one implementation of this embodiment, step S101 includes the following steps:

[0058] Step S101a, performing a minimum estimation on the bandwidth of each modal component signal constituting the original signal in the frequency domain, forming a minimization problem;

[0059] Step S101b, determining the constraints of the minimization problem, and converting the minimization problem with the constraints into a minimization problem without constraints;

[0060] Step S101c, performing optimization and solving by the alternating direction multiplier method, and updating and optimizing each submodal component signal by the optimization algorithm;

[0061] Step S101d: using the permutation entropy as an effect indicator for decomposing the number of intrinsic mode functions, determining the preset threshold using an average value comparison method based on correlation analysis, and selecting the intrinsic mode functions according to the preset threshold to obtain the main components.

[0062] In this embodiment, the VMD algorithm is a non-recursive adaptive signal processing method whose main goal is to decompose the original signal containing vibration information and noise into an integer number of sub-signals, namely, Intrinsic Mode Function (IMF). Each IMF component corresponds to the vibration signal and the noise signal respectively. Therefore, assuming that the original vibration signal consists of k different frequencies The signal is composed of , then the bandwidth of each modal component signal constituting the original signal in the frequency domain is estimated to be minimum, forming a minimization problem, which can be expressed as:

[0063] (1);

[0064] Where, and Represent the modal components contained in the original vibration sensing signal detected by the sensor, as well as the center frequency corresponding to each modal component. Therefore, taking the sum of the modal component signals equal to the original sensing signal as the constraint condition of the above minimization problem, it can be expressed as:

[0065] (2);

[0066] According to equations (1) and (2), the variational problem constructed above is a minimization problem with constraints.

[0067] To solve this problem, we can introduce the Lagrange multiplier method and convert it into an unconstrained minimization problem:

[0068] (3);

[0069] Where, Represents the Lagrange multiplier, whose main function is to represent the coefficients of each vector in the linear combination of the gradient in the constraint condition; Represents the quadratic penalty factor, which is mainly used to suppress the impact of noise on the signal and prevent distortion.

[0070] The convex optimization problem formed by the augmented Lagrangian function can be optimized and solved by the Alternating Direction Method of Multipliers (ADMM). This solution method mainly decomposes the original objective function into various sub-component minimization problems, solves the optimal solution of each sub-minimization problem through parallel operation, and finally obtains the optimal solution of the original objective function through comparative analysis. Among them, the corresponding variational mode function The update method can be expressed as:

[0071] (4);

[0072] In the formula The unilateral spectrum is the residual of the current signal spectrum The Wiener filter output. The bilateral spectrum can be obtained according to the symmetry of the real-valued signal, and then the time domain signal can be obtained by inverse Fourier transform.

[0073] According to the above formula, it can be found that the center frequency corresponding to each submodal component signal mainly exists in the estimated term. Therefore, the update optimization of each submodal component signal can be expressed as follows:

[0074] (5);

[0075] Similarly, transforming it into the frequency domain, we can get:

[0076] (6);

[0077] According to formula (6), the setting of the number of IMF components will have a significant impact on the center frequency estimation. An improper number of components will cause a deviation between the IMF components and the actual signal, resulting in modal aliasing. Generally, when the set number of decompositions is less than the actual number of decompositions, a single IMF will contain multiple frequency band components; when the set number is greater than the actual number, a single frequency band will be decomposed into multiple components. Therefore, selecting an appropriate number of decompositions is crucial to the decomposition effect of the VMD algorithm.

[0078] Permutation entropy (PE), a parameter used to measure the complexity of a one-dimensional time series, can be used to gauge the degree of disorder within a signal. A larger PE value indicates a more complex and random decomposed signal; conversely, a smaller PE value indicates a simpler and more regular decomposed signal. Therefore, to determine the number of IMFs in the VMD algorithm, this example uses the permutation entropy value as an indicator of the effectiveness of the decomposed IMFs.

[0079] After decomposing the original signal containing multi-point vibration information and noise by VMD algorithm, it is necessary to screen the obtained IMF components to filter out the noise signal and retain only the signal containing the main approximately same-frequency vibration information. This embodiment uses the average value comparison method based on correlation analysis to select the IMF components. Suppose that the original signal is decomposed into k IMF components, and the correlation coefficient between each component and the original signal is , ,..., , averaging the correlation coefficients of each IMF component yields:

[0080] (7);

[0081] The calculated average correlation coefficient value is set as the threshold (i.e., the preset threshold) for determining whether the component needs to be eliminated. IMF components greater than the threshold are retained; IMF components less than the threshold are filtered out.

[0082] In this embodiment, the main components are obtained by decomposing and filtering the original signal containing multi-point vibration information and noise through the above VMD algorithm.

[0083] Step S102 : identifying different sound source information by using different frequencies, and performing cross-correlation calculation on the same-frequency signals detected by the four array elements to obtain the time difference between any two array elements.

[0084] After determining the primary component containing the vibration source information, this embodiment can then identify different sound source information by frequency. Cross-correlation calculations are then performed on the same-frequency signals detected by the four-element array to obtain the time difference between each pair of elements. The time difference is calculated in the same manner for the other four-element sensor cluster.

[0085] In one implementation of this embodiment, step S102 includes the following steps:

[0086] Step S102a, determining the distance between each array element and the sound source and the time it takes for the vibration of the sound source to propagate to each array element;

[0087] Step S102b: Calculate the maximum value of the cross-correlation using a cross-correlation algorithm according to the time it takes for the vibration of the sound source to propagate to each array element, and calculate the time difference between any two array elements according to the horizontal coordinate of the maximum value of the cross-correlation.

[0088] In this embodiment, the method of calculating the time difference of arrival of the sound source between two array elements based on the cross-correlation operation is specifically as follows:

[0089] Assuming that the distances between the two array elements and the sound source are x1 and x2 respectively, the time it takes for the vibration of the sound source to propagate to the array elements is , , where v≈343m / s is the speed of sound. Assuming that the signal emitted by the sound source can be represented by φ(t), the signals received by the two array elements are , , the cross-correlation operation principle is as follows:

[0090] ;

[0091] By variable substitution , the integral becomes:

[0092] ;

[0093] Now consider the autocorrelation function , since the autocorrelation function is equivalent to and Cross-correlation, autocorrelation The maximum value is obtained at Equivalent to a function The signal with a time delay of Δt=t1-t2 For cross-correlation, consider that the convolution (cross-correlation) operation in the time domain corresponds to the conjugate product in the frequency domain, that is,

[0094] , yes The conjugate complex number of , and because of the time-shift property of Fourier transform, ;

[0095] Therefore:

[0096] ;

[0097] Then inverse Fourier transform, we have = , so according to The horizontal coordinate of the maximum value can be obtained , that is, to find the delay .

[0098] Since the antenna array proposed in this embodiment combines the propagation characteristics of far / near-field sound sources, in this embodiment, the time delay and signal phase amplitude calculated above are used as the input of the neural network, and the sound source position is used as the output of the neural network to train and verify the sound source spatial positioning model.

[0099] like Figure 1 As shown, an embodiment of the present invention provides a sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors, comprising the following steps:

[0100] In step S200, the time difference is used as the input of a sound source spatial localization model, and the spatial position of the sound source in the optical fiber sensor array is used as the output of the sound source spatial localization model. The sound source spatial localization model is trained and verified using sound source signal samples at different spatial positions.

[0101] like Figure 4 As shown, in this embodiment, the sound source spatial localization model includes: a deep belief network (DBN) and a back propagation neural network (BP) for integration; wherein the deep belief network includes: an input layer, a hidden layer and an output layer; the hidden layer is used to learn the feature information between time difference, signal amplitude and spatial position; the number of neurons corresponding to the input layer and the output layer respectively corresponds to the number of feature quantities time difference, amplitude and spatial position, and the number of hidden layers and the corresponding number of neurons are set through experience and specific parameter adjustment conditions.

[0102] Specifically, in one implementation of this embodiment, step S200 includes the following steps:

[0103] Step S201: Based on sound source signal samples at different spatial positions, multiple deep belief networks with different activation functions are used to perform integrated learning in a back propagation neural network to determine the number of hidden layers and corresponding number of neurons of the back propagation neural network; wherein the input of the back propagation neural network is the spatial positioning value output by the deep belief network under different activation functions, and the output is the spatial positioning calibration value under an ideal state.

[0104] In this embodiment, in order to verify the establishment of the model, data collection and preprocessing are required, wherein the data collection source is: setting up sound sources in different places in three-dimensional space at intervals of 1m, and then measuring the signals of six sensors at a certain position of the sound source, recording the coordinates of the sound source and the signal results of the sensor, demodulating the phase of the measured sensor signal, and finally obtaining six time delays and the amplitude of the signal based on the pairwise cross-correlation to complete the data preprocessing.

[0105] After completing data preprocessing (i.e., arranging the positions of signal-emitting sound sources and calibrating them), they are used as the output of the neural network. Then, acoustic / vibration data corresponding to different positions are collected as the input of the neural network, and 80% of the data set is used as training data to train the sound source spatial positioning model.

[0106] In order to improve the spatial positioning accuracy of the model, multiple DBN networks with different activation functions are used for integrated learning in the BP neural network. Figure 4The input of the BP neural network is the spatial positioning value output by the DBN network under different activation functions, and the output is the spatial positioning calibration value under ideal conditions. The number of hidden layers and the corresponding number of neurons of the BP neural network are also adjusted according to empirical formulas and actual conditions.

[0107] In one implementation of this embodiment, step S201 includes the following steps:

[0108] In step S201a, multiple deep belief networks with different activation functions are trained based on a layer-by-layer unsupervised pre-training method and a global supervised fine-tuning method, and different weights are assigned to each deep belief network in a weighted manner according to the accuracy of each deep belief network on the validation set to optimize the final accuracy.

[0109] In this embodiment, the DBN training process is divided into two stages: layer-by-layer unsupervised pre-training and global supervised fine-tuning. It is constructed by stacking multiple RBMs (restricted Boltzmann machines, or RBMs). The hidden layers of each RBM serve as the visible layers of the next RBM. Finally, a classification layer (such as a softmax layer) is added for supervised learning.

[0110] As an example, the structure of a 3-layer DBN is: input layer → RBM1 → RBM2 → RBM3 → Softmax layer.

[0111] First, layer-by-layer unsupervised pre-training is performed. The goal of pre-training is to initialize the DBN weight parameters by training the RBM layer by layer, thus avoiding the vanishing gradient problem caused by random initialization. The weights and biases are updated using the Contrastive Divergence (CD-k) algorithm. After training, the weights and biases of RBM1 are fixed. The hidden layer activation probabilities of the previous RBM are then used as the visible layer inputs of the next RBM. The parameters are trained layer by layer. After pre-training, all RBMs are stacked and a classification layer is added. Global optimization is performed using backpropagation. The pre-trained RBMs are stacked into a DBN, and a classification layer is added on top. The final output is calculated. The backpropagation algorithm selects a loss function based on the task type and calculates the gradients of the weights and biases of each layer using the chain rule. Gradient descent methods (such as SGD and Adam) are used to update all layer parameters to ultimately optimize classification accuracy.

[0112] Based on the above DBN network training process, the process of using multiple deep belief networks with different activation functions to perform ensemble learning in a back propagation neural network in this embodiment is as follows:

[0113] Three different DBN models are constructed, and then each DBN model is given a different weight by weighting according to the accuracy of each model on the validation set, so as to optimize the final accuracy.

[0114] In this embodiment, the activation function of the deep belief network includes:

[0115] ;

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] In this embodiment, the trained sound source spatial localization model is obtained through the above integrated learning method. In order to verify the feasibility and prediction accuracy of the model, it is also necessary to verify it in this embodiment.

[0121] In step S202, the preprocessed data for calibrating the sound source position is used as the output of the sound source spatial localization model, and the collected acoustic data / vibration data corresponding to different positions is used as the input of the sound source spatial localization model to verify the sound source spatial localization model after ensemble learning.

[0122] In this embodiment, the positions of the signal emitting sound sources are arranged and calibrated as the output of the neural network. Then, the acoustic / vibration data corresponding to different positions are collected as the input of the neural network, and 20% of the data set is used as verification data to verify the sound source spatial localization model after ensemble learning, thereby obtaining a model with a positioning accuracy that meets the requirements.

[0123] It is worth mentioning that in the process of training and verifying the model, if the final training result is overfitting, the regularization parameter is increased or the complexity of the model is reduced, such as reducing the 5-layer RBM to 3 layers, or reducing the number of neurons from 256 to 128, etc. If the result is underfitting, the regularization parameter is reduced accordingly and the complexity of the model is increased. In this way, the number of hidden layers and the corresponding number of neurons of the BP neural network are adjusted to optimize the model.

[0124] like Figure 1 As shown, an embodiment of the present invention provides a sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors, comprising the following steps:

[0125] Step S300 : Based on the trained sound source spatial localization model, the nonlinear relationship between the time difference and the spatial position is utilized to predict and output the spatial localization information of the sound source.

[0126] In this embodiment, the trained sound source spatial localization model can determine the accurate nonlinear relationship between the time difference and spatial position of each array element. Therefore, in an actual test scenario, the time delay and signal phase amplitude of the sensor array are used as inputs to the sound source spatial localization model to accurately predict the spatial localization information of the sound source.

[0127] It's worth noting that, for the space positioning model principle structure designed in this embodiment, even higher positioning accuracy can be achieved by increasing the number of array elements and utilizing more positioning feature parameters as neural network outputs, thereby achieving more precise positioning. Furthermore, constructing more antenna arrays on optical fiber cables could potentially achieve multi-point vibration spatial positioning over a wider range.

[0128] As a replacement for the above sound source spatial localization model in this embodiment, a long short-term memory network can be used to replace the deep belief network in this embodiment; and as a replacement for the optical fiber sensing structure, a forward transmission optical fiber sensor, an optical fiber sensor based on Rayli backscattering, and other structures can be used for replacement.

[0129] This embodiment achieves the following technical effects through the above technical solution:

[0130] This embodiment obtains the time difference between each element in the fiber optic sensor array, uses the time difference as the input of the sound source spatial localization model, and uses the spatial position of the sound source in the fiber optic sensor array as the output of the sound source spatial localization model. The sound source spatial localization model is trained and verified using sound source signal samples at different spatial positions. Based on the trained sound source spatial localization model, the nonlinear relationship between time difference and spatial position is utilized to predict and output the spatial localization information of the sound source. This embodiment proposes a spatial localization technology that integrates an artificial intelligence algorithm with a fiber optic vibration sensor array. By establishing an accurate nonlinear relationship between time difference and spatial position, it achieves high-precision spatial localization of the sound source.

[0131] Exemplary devices

[0132] Based on the above embodiments, the present invention further provides a sound source spatial positioning system based on the fusion of artificial intelligence and optical fiber sensors, comprising:

[0133] A time difference acquisition module is used to obtain the time difference between each array element in the optical fiber sensor array;

[0134] a sound source spatial localization model training and verification module, configured to use the time difference as the input of a sound source spatial localization model and the spatial position of the sound source in the optical fiber sensor array as the output of the sound source spatial localization model, and to train and verify the sound source spatial localization model using sound source signal samples at different spatial positions;

[0135] The sound source spatial positioning information output module is used to predict and output the spatial positioning information of the sound source based on the trained sound source spatial positioning model by utilizing the nonlinear relationship between time difference and spatial position.

[0136] This embodiment achieves the following technical effects through the above technical solution:

[0137] This embodiment obtains the time difference between each element in the fiber optic sensor array, uses the time difference as the input of the sound source spatial localization model, and uses the spatial position of the sound source in the fiber optic sensor array as the output of the sound source spatial localization model. The sound source spatial localization model is trained and verified using sound source signal samples at different spatial positions. Based on the trained sound source spatial localization model, the nonlinear relationship between time difference and spatial position is utilized to predict and output the spatial localization information of the sound source. This embodiment proposes a spatial localization technology that integrates an artificial intelligence algorithm with a fiber optic vibration sensor array. By establishing an accurate nonlinear relationship between time difference and spatial position, it achieves high-precision spatial localization of the sound source.

[0138] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 5 shown.

[0139] The terminal includes: a processor, memory, interface, display screen and communication module connected via a system bus; wherein the processor of the terminal is used to provide computing and control capabilities; the memory of the terminal includes a computer-readable storage medium and an internal memory; the computer-readable storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of the operating system and computer program in the computer-readable storage medium; the interface is used to connect to external devices; the display screen is used to display corresponding information; and the communication module is used to communicate with a cloud server or other devices.

[0140] When the computer program is executed by a processor, it is used to implement the operation of a sound source spatial positioning method based on the fusion of artificial intelligence and optical fiber sensors.

[0141] It will be understood by those skilled in the art that Figure 5The principle block diagram shown in the figure is only a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the terminal to which the solution of the present invention is applied. The specific terminal may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0142] In one embodiment, a terminal is provided, comprising: a processor and a memory, wherein the memory stores a sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors, and when the sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors is executed by the processor, it is used to implement the operation of the above-mentioned sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors.

[0143] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors. When the sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors is executed by a processor, it is used to implement the operation of the above-mentioned sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors.

[0144] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile storage medium. When executed, the computer program can include the processes in the above-described method embodiments. Any reference to memory, storage, database, or other media used in the various embodiments provided herein may include both non-volatile and volatile memory.

[0145] In summary, the present invention provides a method, system, terminal, and storage medium for spatially locating a sound source based on the fusion of artificial intelligence and fiber optic sensors, including: obtaining the time difference between each array element in an array of fiber optic sensor elements; using the time difference as the input of a sound source spatial positioning model, and using the spatial position of the sound source in the array of fiber optic sensor elements as the output of the sound source spatial positioning model, training and verifying the sound source spatial positioning model using sound source signal samples at different spatial positions; and based on the trained sound source spatial positioning model, utilizing the nonlinear relationship between time difference and spatial position to predict and output the spatial positioning information of the sound source. The present invention proposes a spatial positioning technology that integrates an artificial intelligence algorithm with an array of fiber optic vibration sensors, which achieves high-precision spatial positioning of the sound source by constructing an accurate nonlinear relationship between time difference and spatial position.

[0146] It should be understood that the application of the present invention is not limited to the above examples. For those skilled in the art, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors, characterized in that: include: Obtaining the time difference between each element in the optical fiber sensor array; Using the time difference as input to a sound source spatial localization model, and using the sound source spatial position in the optical fiber sensor array as output of the sound source spatial localization model, the sound source spatial localization model is trained and verified using sound source signal samples at different spatial positions; Based on the trained sound source spatial localization model, the nonlinear relationship between time difference and spatial position is used to predict and output the spatial localization information of the sound source; The sound source spatial localization model includes: a deep belief network and a back propagation neural network for integration; The deep belief network is composed of a stack of multiple multi-layer restricted Boltzmann machines; The deep belief network comprises an input layer, a hidden layer, and an output layer; the hidden layer is used to learn characteristic information between time difference, signal amplitude, and spatial position; the number of neurons corresponding to the input layer and the output layer corresponds to the number of characteristic quantities of time difference, amplitude, and spatial position, respectively; The training and verification of the sound source spatial localization model using sound source signal samples at different spatial positions includes: Based on sound source signal samples at different spatial locations, multiple deep belief networks with different activation functions are used to perform ensemble learning in a back-propagation neural network to determine the number of hidden layers and corresponding numbers of neurons in the back-propagation neural network; wherein the input of the back-propagation neural network is the spatial positioning value output by the deep belief network under different activation functions, and the output is the spatial positioning calibration value under an ideal state; Using the preprocessed data of the calibrated sound source position as the output of the sound source spatial localization model, and using the collected acoustic data / vibration data corresponding to different positions as the input of the sound source spatial localization model, and verifying the sound source spatial localization model after ensemble learning; The method of using multiple deep belief networks with different activation functions to perform integrated learning in a back-propagation neural network includes: Based on the layer-by-layer unsupervised pre-training method and the global supervised fine-tuning method, multiple deep belief networks with different activation functions are trained. According to the accuracy of each deep belief network on the validation set, different weights are assigned to each deep belief network in a weighted manner to optimize the final accuracy.

2. The sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensor according to claim 1 is characterized in that: The obtaining of the time difference between each element in the optical fiber sensor element array includes: The acoustic signal / vibration signal caused by multiple vibration sources is decomposed by the variational mode decomposition algorithm, and the main components containing the vibration source information are determined by a preset threshold; Different sound source information is identified through different frequencies, and the cross-correlation calculation is performed on the same-frequency signals detected by the four-point array elements to obtain the time difference between any two array elements.

3. The sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensor according to claim 2 is characterized in that: The method of decomposing the acoustic signal / vibration signal caused by multiple vibration sources by using a variational mode decomposition algorithm and determining the main components containing vibration source information by using a preset threshold value includes: The bandwidth of each modal component signal constituting the original signal in the frequency domain is estimated to be the minimum, forming a minimization problem; Determining the constraints of the minimization problem and converting the minimization problem with the constraints into a minimization problem without constraints; The optimization solution is obtained by the alternating direction multiplier method, and each submodal component signal is updated and optimized by the optimization algorithm; The permutation entropy value is used as an effect indicator for decomposing the number of intrinsic mode functions, the preset threshold is determined by using an average value comparison method based on correlation analysis, and the intrinsic mode function is selected according to the preset threshold to obtain the main component.

4. The sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensor according to claim 2 is characterized in that: The cross-correlation calculation of the same-frequency signals detected by the four array elements includes: Determine the distance between each array element and the sound source and the time it takes for the vibration of the sound source to propagate to each array element; According to the time it takes for the vibration of the sound source to propagate to each array element, a cross-correlation maximum value is calculated using a cross-correlation algorithm, and the time difference between any two array elements is calculated based on the horizontal coordinate of the cross-correlation maximum value.

5. A sound source spatial positioning system based on the fusion of artificial intelligence and optical fiber sensors, used to implement the sound source spatial positioning method based on the fusion of artificial intelligence and optical fiber sensors as described in any one of claims 1 to 4, characterized in that: include: A time difference acquisition module is used to obtain the time difference between each array element in the optical fiber sensor array; a sound source spatial localization model training and verification module, configured to use the time difference as the input of a sound source spatial localization model and the spatial position of the sound source in the optical fiber sensor array as the output of the sound source spatial localization model, and to train and verify the sound source spatial localization model using sound source signal samples at different spatial positions; The sound source spatial positioning information output module is used to predict and output the spatial positioning information of the sound source based on the trained sound source spatial positioning model by utilizing the nonlinear relationship between time difference and spatial position.

6. A terminal, characterized in that: include: A processor and a memory, wherein the memory stores a sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors, and when the sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors is executed by the processor, it is used to implement the operation of the sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors. When the sound source spatial localization program based on the fusion of artificial intelligence and optical fiber sensors is executed by the processor, it is used to implement the operation of the sound source spatial localization method based on the fusion of artificial intelligence and optical fiber sensors as described in any one of claims 1 to 4.

Images