Environmental noise source positioning and classifying method based on Bayesian reasoning

Abstract

The invention discloses an environmental noise source positioning and classifying method based on Bayesian reasoning, which belongs to the technical field of environmental noise detection, and comprises the following steps: setting a dynamic noise threshold value, and detecting whether the noise exceeds the standard or not through noise time-frequency domain joint detection; carrying out noise source positioning by adopting a spherical acoustic array; identifying and classifying noise sources based on the noise target classification model to obtain the probability of each noise category; and fusing the positioning result and the classification result based on Bayesian reasoning to obtain a final noise classification result. According to the method, the accuracy of noise monitoring is improved, accurate noise source positioning is realized, the reliability of noise classification is improved, the robustness and adaptability of the system are enhanced, and the application prospect is wide.

Description

Technical Field

[0001] The invention belongs to the field of environmental noise monitoring, and in particular relates to an environmental noise source positioning and classification method based on Bayesian reasoning. Background Art

[0002] At present, as the complexity of environmental noise increases year by year, the existing noise source positioning and identification and classification technologies may misjudge that the noise exceeds the standard when faced with complex environmental noise, or may only provide noise intensity but cannot further classify the specific type and location of the noise source; in addition, the existing noise positioning and identification and classification technologies are mostly independent, that is, there is no correlation between positioning and identification and classification, which to some extent will lead to a low accuracy rate in noise source classification. Summary of the invention

[0003] In view of the above problems existing in the prior art, the present invention proposes an environmental noise source positioning and classification method based on Bayesian reasoning, which has a reasonable design, solves the shortcomings of the prior art and has good effects.

[0004] The technical solution of the present invention is as follows:

[0005] The method for locating and classifying environmental noise sources based on Bayesian reasoning includes the following steps:

[0006] S1. Setting dynamic noise threshold;

[0007] S2, jointly detect whether the noise exceeds the standard through the noise time and frequency domain;

[0008] S3, using spherical acoustic array to locate noise sources;

[0009] S4, identifying and classifying noise sources based on the noise target classification network to obtain the probability of each noise category;

[0010] S5: Based on Bayesian reasoning, the positioning result of S3 and the classification result of S4 are integrated to obtain the final noise classification result.

[0011] Furthermore, the S1 is specifically:

[0012] First, in the time window [t, t+Δt], the noise signal x(t) is collected; the noise signal is divided into N frames, each frame length is L;

[0013] The energy value of each frame is calculated as:

[0014]

[0015] Among them, E n is the energy value of the nth frame, x n(i) is the value of the i-th sampling point in the n-th frame, n = 1, 2, ..., N, i = 1, 2, ..., L;

[0016] The average energy value E is calculated as:

[0017]

[0018] Calculate the standard deviation σ E for:

[0019]

[0020] Set the dynamic noise threshold T to:

[0021]

[0022] Among them, K is the adjustment coefficient.

[0023] Furthermore, the S2 comprises the following sub-steps:

[0024] S2.1. First, perform a preliminary test and use time domain energy detection to filter out possible audio frames that exceed the standard. Specifically:

[0025] The continuous acoustic signal b(t) is divided into a series of frames b by frame length L and frame shift R. n (i), where n is the frame number, b n (i) is the value of the i-th sampling point in the n-th frame;

[0026] For each frame b n (i) is multiplied by the window function ω(i), and the expression is:

[0027] c n (i) = b n (i)*ω(i); (5)

[0028] Among them, c n (i) is the windowed signal;

[0029] Calculate the short-time energy E′ n for:

[0030]

[0031] The short-time energy is smoothed by first-order recursion, and the expression is:

[0032]

[0033] in, is the smoothed short-time energy, α is a constant, 0<α<1;

[0034] like Then continue to perform frequency domain energy detection. If The noise is judged to be normal;

[0035] S2.2, frequency domain energy detection is specifically:

[0036] To meet Perform fast Fourier transform on each frame to get the spectrum X n (k), the spectrum amplitude is calculated as:

[0037]

[0038] Among them, X n (k) is the spectrum of the nth frame, |X n (k)| is the spectrum amplitude of the nth frame;

[0039] Calculate the spectrum energy and accumulate the energy of the noise target frequency range [k1, k2]. The expression is:

[0040]

[0041] in, is the spectrum energy of the nth frame;

[0042] like Then the current noise exceeds the standard. The noise is judged to be normal.

[0043] Furthermore, S3 includes the following sub-steps:

[0044] S31. M microphones are evenly distributed on a spherical surface with a radius of r to form a spherical acoustic array structure. Taking the center of the array as the origin, a three-dimensional rectangular coordinate system (x, y, z) and a corresponding spherical coordinate system are established. Where r is the radial distance; θ is the azimuth angle, and its value range is [0,2π); is the pitch angle, and its value range is [0,π);

[0045] Assume the sound source is at an unknown position P s =(x s ,y s ,z s ), the position of the mth microphone is P m =(x m ,y m ,z m ), then the distance d from the sound source to the mth microphone m for:

[0046] d m =‖P s -P m ‖; (10)

[0047] The signal x received by the mth microphone m (t) is:

[0048] x m (t) = a m s(t-τ m )+n m (t); (11)

[0049] Among them, a m is the attenuation coefficient, d m is the distance between the sound source signal and the microphone; s(t-τ m ) is the propagation delay τ of the source signal m The signal that reaches the mth microphone is c is the speed of sound, n m (t) is the noise of the mth microphone;

[0050] S32, perform time difference estimation, first select a microphone as the reference microphone, record the reference microphone as the m0th microphone, calculate the cross-correlation function of the signal received by the mth microphone and the reference microphone for:

[0051]

[0052] Calculate the time delay corresponding to the maximum value of the cross-correlation function, which is the relative time difference The expression is:

[0053]

[0054] Construct a time difference matrix. Each microphone and the reference microphone form a microphone pair. Collect the time differences of all microphone pairs to form a time difference matrix Δτ.

[0055] S33. Time difference and distance difference The relationship between them is:

[0056]

[0057] For each microphone pair, the following equation is established:

[0058]

[0059] Wherein, m=1,2,…,M and m≠m0; construct a nonlinear equation system;

[0060] Finally, the linearization method is used for position estimation. For the far-field approximation, it is assumed that the sound source distance is much larger than the diameter of the spherical acoustic array structure. The nonlinear equations are linearized and the sound source position P is solved using the least squares method. s ;

[0061] The sound source position vector is:

[0062] d=P s -P0; (17)

[0063] Among them, P0 is the center position of the array, usually the origin;

[0064] The azimuth angle θ is: θ = arctan2(d y ,d x ), where d x ,d y are the x and y components of d respectively;

[0065] Pitch Angle for: where d z is the z component of d.

[0066] Furthermore, a training set containing various noise categories is constructed, and the noise target recognition and classification network is trained and tested using the training set.

[0067] The noise target recognition and classification network includes an input layer, a preprocessing layer, a feature extraction layer, a sequence modeling layer, a fully connected unit and an output layer;

[0068] The input layer input is a signal q(t) that is determined to have excessive noise after noise time-frequency domain joint detection. The length size is (FN, 1), where FN is the number of sampling points of the audio signal.

[0069] The preprocessing layer performs denoising and normalization processing on the audio signal, and converts the time domain signal into cepstral features through Mel frequency cepstral coefficients;

[0070] The feature extraction layer includes convolution layer 1, maximum pooling layer 1, convolution layer 2, maximum pooling layer 2, residual block 1 and residual block 2 in sequence;

[0071] The convolution layer 1 is a 1D convolution layer, including 64 filters, a convolution kernel size of 3, and an activation function of ReLU;

[0072] The sizes of the maximum pooling layer 1 and the maximum pooling layer 2 are both 2;

[0073] The convolution layer 2 is a 1D convolution layer, including 128 filters, a convolution kernel size of 3, and an activation function of ReLU;

[0074] The residual block 1 includes a convolution layer 3 and a convolution layer 4, wherein the convolution layer 3 and the convolution layer 4 are 1D convolution layers, including 256 filters, a convolution kernel size of 3, and an activation function of ReLU;

[0075] The residual block 2 includes a convolution layer 5 and a convolution layer 6, wherein the convolution layer 5 and the convolution layer 6 are 1D convolution layers, including 512 filters, a convolution kernel size of 3, and an activation function of ReLU;

[0076] The sequence modeling layer includes a bidirectional LSTM layer 1 and a bidirectional LSTM layer 2, wherein the bidirectional LSTM layer 1 includes 128 units, and the bidirectional LSTM layer 2 includes 256 units;

[0077] The fully connected unit includes a flattening layer, a fully connected layer 1, a Dropout layer 1, a fully connected layer 2, and a Dropout layer 2;

[0078] The flattening layer expands the three-dimensional feature map into a one-dimensional vector;

[0079] The fully connected layer 1 includes 512 neurons, and the activation function is ReLU;

[0080] The Dropout rate of the Dropout layer 1 is 0.5;

[0081] The fully connected layer 2 includes 256 neurons, and the activation function is ReLU;

[0082] The Dropout rate of the Dropout layer 2 is 0.5;

[0083] The output layer is a fully connected layer 3, the number of output neurons is the number of categories, and the activation function is Softmax.

[0084] Furthermore, the S5 comprises the following sub-steps:

[0085] S51, use the noise target recognition and classification network to process q(t) and output the prior probability of each noise category, denoted as P(Class i |NetworkOut,Loc);

[0086] S52. Assume that in each noise category, the positioning feature obeys Gaussian distribution:

[0087]

[0088] Among them, μ i is the pitch angle under the i-th noise category The average value of i is the pitch angle under the i-th noise category The standard deviation of

[0089] S53, for the observed pitch angle Calculate the probability of observing the pitch angle under each noise category, denoted as The expression is:

[0090]

[0091] S54. Apply the Bayesian formula to calculate the posterior probability of each noise category:

[0092]

[0093] Wherein, j represents the index when traversing all categories, that is, there are C types of noise in the output category of the neural network, then j = 1, 2, ... C;

[0094] S55. According to the calculated posterior probability, select the noise category with the largest probability as the final noise classification result.

[0095] Beneficial technical effects brought by the present invention:

[0096] The present invention combines dynamic threshold setting, time-frequency domain joint detection, spherical acoustic array positioning, neural network classification, and a fusion method based on a probability model to achieve high-precision detection, positioning, and classification of environmental noise. Its beneficial technical effects are mainly reflected in the following aspects:

[0097] Improve the accuracy of noise detection: 1. Dynamic threshold setting: According to the real-time collected environmental noise data, dynamically adjust the noise threshold to adapt to the complex noise environment changes, reduce the false detection rate and missed detection rate. 2. Joint detection in time and frequency domains: Through dual detection in time and frequency domains, accurate judgment of possible excessive noise is made, further improving the accuracy of detection.

[0098] Achieve accurate noise source positioning: Utilize spherical acoustic array and time difference estimation technology to accurately obtain the coordinates, azimuth, elevation angle and distance of the noise source relative to the center of the array. By effectively locating the noise source in the environment, reliable data support is provided for subsequent noise source management and control.

[0099] Improve the reliability of noise classification: 1. Use deep learning models to identify and classify noise, which can accurately distinguish different types of noise sources. 2. Bayesian reasoning fusion: Combine the positioning results with the classification results, and use Bayesian reasoning to correct the classification probability, which improves the accuracy and reliability of classification.

[0100] Enhance the robustness and adaptability of the system: 1. Multi-information fusion: Combining acoustic features and positioning features, the system can maintain high performance in complex and changeable noise environments. 2. Reduction of false detection and missed detection: Through comprehensive judgment and fusion algorithms, the false detection rate and missed detection rate in the noise detection and classification process are reduced.

[0101] Broad application prospects: Mainly used in the field of environmental monitoring, suitable for urban noise monitoring, traffic noise management, industrial noise control and other fields.

[0102] The present invention realizes high-precision detection, positioning and classification of environmental noise, significantly improving the accuracy, reliability and intelligence level of the system. Its beneficial technical effects include reducing the false detection and missed detection rate, improving positioning and classification accuracy, and enhancing the robustness and adaptability of the system, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 It is a flow chart of the method for locating and classifying environmental noise sources based on Bayesian reasoning in the present invention;

[0104] Figure 2 This is a flow chart of the invention for detecting whether the noise exceeds the standard in the time-frequency domain jointly;

[0105] Figure 3 An installation diagram of an environmental monitoring device used in the method of the present invention;

[0106] Figure 4 Output images for cameras of environmental monitoring equipment;

[0107] Among them, (a) is the image before the noise source is detected; (b) is the image after the noise is detected;

[0108] Figure 5 The noise classification result diagram output by the environmental monitoring equipment; DETAILED DESCRIPTION

[0109] The specific implementation of the present invention is further described below in conjunction with specific embodiments:

[0110] Ambient noise source location and classification methods based on Bayesian reasoning, such as Figure 1 and 2 As shown, it includes the following sub-steps:

[0111] S1. Set dynamic noise threshold;

[0112] First, in the time window [t, t+Δt], the noise signal x(t) is collected; the noise signal is divided into N frames, each frame length is L;

[0113] The energy value of each frame is calculated as:

[0114]

[0115] Among them, E n is the energy value of the nth frame, x n (i) is the value of the i-th sampling point in the n-th frame, n = 1, 2, ..., N, i = 1, 2, ..., L;

[0116] The average energy value E is calculated as:

[0117]

[0118] Calculate the standard deviation σ E for:

[0119]

[0120] Set the dynamic noise threshold T to:

[0121]

[0122] Wherein, K is the adjustment coefficient, which can be set to 2 or 3 according to actual conditions.

[0123] According to the real-time collected environmental noise data, the noise threshold is dynamically adjusted to adapt to the complex noise environment changes and improve the accuracy of detection.

[0124] S2. In order to improve the accuracy of detection and reduce the false detection rate and missed detection rate, the noise time-frequency domain is used to jointly detect whether the noise exceeds the standard, including the following sub-steps:

[0125] S2.1. First, perform a preliminary test and use time domain energy detection to quickly filter out possible audio frames that exceed the standard. Specifically:

[0126] The continuous acoustic signal b(t) is divided into a series of frames b by frame length L and frame shift R. n (i), where n is the frame number, b n (i) is the value of the i-th sampling point in the n-th frame;

[0127] For each frame b n (i) is multiplied by the window function ω(i), and the expression is:

[0128] c n (i) = b n (i)*ω(i); (5)

[0129] Among them, c n (i) is the signal after windowing. This process is used to reduce the "leakage effect" that may occur when the signal is analyzed in the frequency domain and helps to obtain a more accurate spectrum estimation;

[0130] Calculate the short-time energy E′ n for:

[0131]

[0132] The short-time energy is recursively smoothed to reduce fluctuations. The expression is:

[0133]

[0134] in, is the short-time energy after smoothing, α is a constant, 0<α<1, and is usually taken as 0.9;

[0135] like Then continue to perform frequency domain energy detection. If The noise is judged to be normal;

[0136] S2.2, frequency domain energy detection is specifically:

[0137] To meet Perform fast Fourier transform on each frame to get the spectrum X n (k), the spectrum amplitude is calculated as:

[0138]

[0139] Among them, X n (k) is the spectrum of the nth frame, |X n (k)| is the spectrum amplitude of the nth frame;

[0140] Calculate the spectrum energy and accumulate the energy of the noise target frequency range [k1, k2]. The expression is:

[0141]

[0142] in, is the spectrum energy of the nth frame;

[0143] like Then the current noise exceeds the standard. The noise is judged to be normal.

[0144] S3, using a spherical acoustic array to locate the noise source, including the following sub-steps:

[0145] S31. M microphones are evenly distributed on a sphere with a radius of r. The center of the array is taken as the origin. The center of the array is the center of the sphere. A three-dimensional rectangular coordinate system (x, y, z) and a corresponding spherical coordinate system are established. Where r is the radial distance; θ is the azimuth angle, and its value range is [0,2π); is the pitch angle, and its value range is [0,π);

[0146] Assume the sound source is at an unknown position P s =(x s ,y s ,z s ), the position of the mth microphone is P m =(x m ,ym ,z m ), then the distance from the sound source to the mth microphone is:

[0147] d m =‖P s -P m ‖; (10)

[0148] The signal received by the mth microphone is:

[0149] x m (t) = a m s(t-τ m )+n m (t); (11)

[0150] Among them, a m is the attenuation coefficient, d m is the distance between the sound source signal and the microphone; τ m is the propagation delay, c is the speed of sound, about 343m / s, n m (t) is the noise of the mth microphone;

[0151] S32, perform time difference estimation, first select a microphone as a reference microphone, usually select the first microphone m0, and calculate the cross-correlation function of the signal received by the mth microphone and the reference microphone as follows:

[0152]

[0153] Calculate the time delay corresponding to the maximum value of the cross-correlation function, which is the relative time difference The expression is:

[0154]

[0155] Construct a time difference matrix, collect the time differences of all microphone pairs, and form a time difference matrix Δτ;

[0156] S33, establishing a geometric relationship, wherein the relationship between the time difference and the distance difference is:

[0157]

[0158] For each microphone pair, the following equation is established:

[0159]

[0160] Wherein, m=1,2,…,M, and m≠m0; construct a nonlinear equation system;

[0161] Finally, the linearization method or iterative optimization method is used to estimate the position.

[0162] In the linearization method, for the far-field approximation, it is assumed that the sound source distance is much larger than the array size, the nonlinear equations are linearized, and the sound source position P is solved using the least squares method. s .

[0163] In the iterative optimization method, an initial position is randomly set first, and the Levenberg-Marquardt algorithm is used for iterative optimization to minimize the residual error:

[0164]

[0165] The sound source position vector is:

[0166] d=P s -P0; (17)

[0167] Among them, P0 is the center position of the array, usually the origin;

[0168] The azimuth angle θ is: θ = arctan2(d y ,d x ), where d x ,d y are the x and y components of d respectively;

[0169] Pitch Angle for: where d z is the z component of d.

[0170] S4, identifying and classifying noise sources based on the noise target classification network to obtain the probability of each noise category;

[0171] The noise target recognition and classification network includes an input layer, a preprocessing layer, a feature extraction layer, a sequence modeling layer, a fully connected unit, and an output layer;

[0172] The input layer inputs the audio signal, and the length size is set to (FN, 1), where FN is the number of sampling points of the audio signal;

[0173] The preprocessing layer performs denoising and normalization on the audio signal, and converts the time domain signal into cepstrum features through Mel-frequency cepstrum coefficients (MFCC). For example, the feature dimension obtained by calculating the MFCC feature is FM (for example, a 13-dimensional vector);

[0174] The feature extraction layer includes convolution layer 1, maximum pooling layer 1, convolution layer 2, maximum pooling layer 2, residual block 1 and residual block 2 in sequence;

[0175] Convolutional layer 1 is a 1D convolutional layer with 64 filters, a convolution kernel size of 3, and an activation function of ReLU;

[0176] The size of max pooling layer 1 and max pooling layer 2 are both 2;

[0177] Convolutional layer 2 is a 1D convolutional layer with 128 filters, a convolution kernel size of 3, and an activation function of ReLU;

[0178] Residual block 1 includes convolutional layer 3 and convolutional layer 4. Convolutional layer 3 and convolutional layer 4 are 1D convolutional layers, containing 256 filters, the convolution kernel size is 3, and the activation function is ReLU;

[0179] Residual block 2 includes convolutional layer 5 and convolutional layer 6. Convolutional layer 5 and convolutional layer 6 are 1D convolutional layers, containing 512 filters, the convolution kernel size is 3, and the activation function is ReLU;

[0180] The sequence modeling layer includes bidirectional LSTM layer 1 and bidirectional LSTM layer 2. Bidirectional LSTM layer 1 includes 128 units and bidirectional LSTM layer 2 includes 256 units.

[0181] The fully connected unit includes a flattening layer, a fully connected layer 1, a Dropout layer 1, a fully connected layer 2, and a Dropout layer 2;

[0182] The flattening layer expands the three-dimensional feature map into a one-dimensional vector;

[0183] The fully connected layer 1 includes 512 neurons, and the activation function is ReLU;

[0184] The dropout rate of Dropout layer 1 is 0.5;

[0185] The fully connected layer 2 includes 256 neurons, and the activation function is ReLU;

[0186] The Dropout rate of Dropout layer 2 is 0.5;

[0187] The output layer is the fully connected layer 3, the number of output neurons is the number of categories, and the activation function is Softmax.

[0188] S5: Based on Bayesian reasoning, the positioning result of S3 and the classification result of S4 are integrated to obtain the final noise classification result.

[0189] Bayes' theorem describes the relationship between the posterior probability P(A|B) of event A, the prior probability P(A), and the likelihood function P(B|A) when condition B is known:

[0190]

[0191] Where P(A) is the prior probability of event A; P(B|A) is the probability of observing B given the occurrence of event A (likelihood function); P(B) is the marginal probability of event B, which can be calculated using the total probability formula:

[0192] P(B)=∑ i P(B|A i )P(A i );(19)

[0193] Among them, event A is the noise source category Class i , where i represents the category number, such as "car horn", "drone buzz", etc.

[0194] The observation data B includes the classification results and positioning results (such as pitch angle and azimuth angle) of the neural network.

[0195] S5 includes the following sub-steps:

[0196] S51, use the noise target recognition and classification network to process the acoustic signal and output the prior probability of each noise category, denoted as P (Class i |NetworkOut,Loc), where NetworkOut represents the classification result of the neural network;

[0197] S52, establishing a likelihood function of the positioning feature. The positioning result is associated with the noise category. It is necessary to establish the probability of observing a specific positioning feature under each category, that is, the likelihood function;

[0198] Assume that under each noise category, the positioning feature follows a Gaussian distribution:

[0199]

[0200] Among them, μ i is the pitch angle under the i-th noise category The average value of i is the pitch angle under the i-th noise category The standard deviation of

[0201] For example, in parameter setting, the μ of the car horn Car = 0 degrees, σ Car =10 degrees; Drone Buzzing μ Drone =75 degrees, σ Drone =10 degrees.

[0202] S53, calculate the likelihood value. For the observed pitch angle, calculate the probability of observing the pitch angle under each noise category, which is recorded as The expression is:

[0203]

[0204] S54. Apply the Bayesian formula to calculate the posterior probability of each noise category:

[0205]

[0206] Wherein, j represents the index when traversing all categories, that is, there are C types of noise in the output category of the neural network, then j = 1, 2, ... C;

[0207] S55. According to the calculated posterior probability, select the noise category with the largest probability as the final noise classification result.

[0208] The method of the present invention is applied to and tested in an environmental noise monitoring device, wherein the device is integrated with a camera, such as Figure 3 As shown in the figure; when the spherical acoustic array detects that the noise exceeds the standard, the azimuth, pitch angle and distance of the noise source are calculated, and the camera rotates to the position according to the positioning information to take photos of the noise exceeding the standard behavior for evidence collection. The noise source target is located in the middle of the video image, represented by a white bar, as shown in the figure. Figure 4 (a) and (b), the corresponding noise source classification results are shown in Figure 5 shown.

[0209] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for locating and classifying environmental noise sources based on Bayesian reasoning, characterized in that: The following steps are involved: S1. Set dynamic noise threshold; S2, jointly detect whether the noise exceeds the standard through the noise time and frequency domain; S3, using spherical acoustic array to locate noise sources; S4, identifying and classifying noise sources based on the noise target classification network to obtain the probability of each noise category; S5: Based on Bayesian reasoning, the positioning result of S3 and the classification result of S4 are integrated to obtain the final noise classification result.

2. The method for locating and classifying environmental noise sources based on Bayesian reasoning according to claim 1, characterized in that: The S1 is specifically: First, in the time window [t, t+Δt], the noise signal x(t) is collected; the noise signal is divided into N frames, each frame length is L; The energy value of each frame is calculated as: Among them, E n is the energy value of the nth frame, x n (i) is the value of the i-th sampling point in the n-th frame, n = 1, 2, ..., N, i = 1, 2, ..., L; The average energy value E is calculated as: Calculate the standard deviation σ E for: Set the dynamic noise threshold T to: Among them, K is the adjustment coefficient.

3. The method for locating and classifying environmental noise sources based on Bayesian reasoning according to claim 2, characterized in that: The S2 comprises the following sub-steps: S2.

1. First, perform a preliminary test and use time domain energy detection to filter out possible audio frames that exceed the standard. Specifically: The continuous acoustic signal b(t) is divided into a series of frames b by frame length L and frame shift R. n (i), where n is the frame number, b n (i) is the value of the i-th sampling point in the n-th frame; For each frame b n (i) is multiplied by the window function ω(i), and the expression is: c n (i)=b n (i)*ω(i); (5) Among them, c n (i) is the windowed signal; Calculate the short-time energy E′ n for: The short-time energy is smoothed by first-order recursion, and the expression is: in, is the smoothed short-time energy, α is a constant, 0<α<1; like Then continue to perform frequency domain energy detection. If The noise is judged to be normal; S2.2, frequency domain energy detection is specifically: To meet Perform fast Fourier transform on each frame to get the spectrum X n (k), the spectrum amplitude is calculated as: Among them, X n (k) is the spectrum of the nth frame, |X n (k)| is the spectrum amplitude of the nth frame; Calculate the spectrum energy and accumulate the energy of the noise target frequency range [k1, k2]. The expression is: in, is the spectrum energy of the nth frame; like Then the current noise exceeds the standard. The noise is judged to be normal.

4. The method for locating and classifying environmental noise sources based on Bayesian reasoning according to claim 3, characterized in that: The S3 comprises the following sub-steps: S31. M microphones are evenly distributed on a spherical surface with a radius of r to form a spherical acoustic array structure. Taking the center of the array as the origin, a three-dimensional rectangular coordinate system (x, y, z) and a corresponding spherical coordinate system are established. Where r is the radial distance; θ is the azimuth angle, and its value range is [0,2π); is the pitch angle, and its value range is [0,π); Assume the sound source is at an unknown position P S =(x s ,y s ,z s ), the position of the mth microphone is P m =(x m ,y m ,z m ), then the distance d from the sound source to the mth microphone m for: d m =‖P s -P m ‖;(10) The signal x received by the mth microphone m (t) is: x m (t)=a m s(t-τ m )+n m (t);(11) Among them, a m is the attenuation coefficient, d m is the distance between the sound source signal and the microphone; s(t-τ m ) is the propagation delay τ of the source signal m The signal that reaches the mth microphone is c is the speed of sound, n m (t) is the noise of the mth microphone; S32, perform time difference estimation, first select a microphone as the reference microphone, record the reference microphone as the m0th microphone, calculate the cross-correlation function of the signal received by the mth microphone and the reference microphone for: Calculate the time delay corresponding to the maximum value of the cross-correlation function, which is the relative time difference The expression is: Construct a time difference matrix. Each microphone and the reference microphone form a microphone pair. Collect the time differences of all microphone pairs to form a time difference matrix Δτ. S33. Time difference and distance difference The relationship between them is: For each microphone pair, the following equation is established: Wherein, m=1,2,…,M and m≠m0; construct a nonlinear equation system; Finally, the linearization method is used for position estimation. For the far-field approximation, it is assumed that the sound source distance is much larger than the diameter of the spherical acoustic array structure. The nonlinear equations are linearized and the sound source position P is solved using the least squares method. s ; The sound source position vector is: d=P s -P0;(17) Among them, P0 is the center position of the array, usually the origin; The azimuth angle θ is: θ = arctan2(d y ,d x ), where d x ,d y are the x and y components of d respectively; Pitch Angle for: where d z is the z component of d.

5. The method for locating and classifying environmental noise sources based on Bayesian reasoning according to claim 4, characterized in that: Construct a training set containing various noise categories, and use the training set to train and test the noise target recognition and classification network. The noise target recognition and classification network includes an input layer, a preprocessing layer, a feature extraction layer, a sequence modeling layer, a fully connected unit and an output layer; The input layer input is a signal q(t) that is determined to have excessive noise after noise time-frequency domain joint detection. The length size is (FN, 1), where FN is the number of sampling points of the audio signal. The preprocessing layer performs denoising and normalization processing on the audio signal, and converts the time domain signal into cepstral features through Mel frequency cepstral coefficients; The feature extraction layer includes convolution layer 1, maximum pooling layer 1, convolution layer 2, maximum pooling layer 2, residual block 1 and residual block 2 in sequence; The convolution layer 1 is a 1D convolution layer, including 64 filters, a convolution kernel size of 3, and an activation function of ReLU; The sizes of the maximum pooling layer 1 and the maximum pooling layer 2 are both 2; The convolution layer 2 is a 1D convolution layer, including 128 filters, a convolution kernel size of 3, and an activation function of ReLU; The residual block 1 includes a convolution layer 3 and a convolution layer 4, wherein the convolution layer 3 and the convolution layer 4 are 1D convolution layers, including 256 filters, a convolution kernel size of 3, and an activation function of ReLU; The residual block 2 includes a convolution layer 5 and a convolution layer 6, wherein the convolution layer 5 and the convolution layer 6 are 1D convolution layers, including 512 filters, a convolution kernel size of 3, and an activation function of ReLU; The sequence modeling layer includes a bidirectional LSTM layer 1 and a bidirectional LSTM layer 2, wherein the bidirectional LSTM layer 1 includes 128 units, and the bidirectional LSTM layer 2 includes 256 units; The fully connected unit includes a flattening layer, a fully connected layer 1, a Dropout layer 1, a fully connected layer 2, and a Dropout layer 2; The flattening layer expands the three-dimensional feature map into a one-dimensional vector; The fully connected layer 1 includes 512 neurons, and the activation function is ReLU; The Dropout rate of the Dropout layer 1 is 0.5; The fully connected layer 2 includes 256 neurons, and the activation function is ReLU; The Dropout rate of the Dropout layer 2 is 0.5; The output layer is a fully connected layer 3, the number of output neurons is the number of categories, and the activation function is Softmax.

6. The method for locating and classifying environmental noise sources based on Bayesian reasoning according to claim 5, characterized in that: The S5 comprises the following sub-steps: S51, use the noise target recognition and classification network to process q(t) and output the prior probability of each noise category, denoted as P(Class i |NetworkOut,Loc); S52. Assume that in each noise category, the positioning feature obeys Gaussian distribution: Among them, μ i is the pitch angle under the i-th noise category The average value of i is the pitch angle under the i-th noise category The standard deviation of S53, for the observed pitch angle Calculate the probability of observing the pitch angle under each noise category, denoted as The expression is: S54. Apply the Bayesian formula to calculate the posterior probability of each noise category: Wherein, j represents the index when traversing all categories, that is, there are C types of noise in the output category of the neural network, then j = 1, 2, ... C; S55. According to the calculated posterior probability, select the noise category with the largest probability as the final noise classification result.

Images