Distributed acoustic sensing system ambient noise monitoring and event identification method, system and medium

Abstract

The application relates to an environment noise monitoring and event identification method, system and medium of a distributed acoustic sensing system, which comprises the following steps: obtaining real-time environment sound signals f(t) at different distances from the distributed acoustic sensing system; obtaining a frequency domain signal S in the sound signal through Fourier transform; extracting characteristic values corresponding to each event based on the frequency domain signal by adopting a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling; classifying the sound signal characteristics through a convolutional neural network machine learning and determining the corresponding event; and reporting the noise determination result. The application can effectively process diversified noise types and realize real-time monitoring of key infrastructures.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical fiber sensing, in particular to a distributed acoustic sensing system environmental noise monitoring and event identification method, system and medium. BACKGROUND

[0002] The existing monitoring technology mainly faces the following limitations: in the field of environmental pollution control, traditional noise monitoring equipment mainly relies on sound level meters, wireless sensor networks or Internet of Things node networks for data acquisition. Although such devices can achieve basic noise intensity detection, they have significant response delay, high equipment deployment cost, difficult power consumption management, and poor terrain adaptability, etc. Technical bottlenecks. Especially in the face of traffic noise sources (such as heavy vehicles), construction vibration sources and complex noise fields caused by industrial equipment operation, the low-frequency vibration energy will be conducted along the ground medium to the underground pipe network system, causing safety hazards such as cable sheath structure fatigue damage and optical fiber transmission performance degradation. The traditional monitoring system is difficult to synchronously capture the spatio-temporal correlation characteristics of the sound wave signal and the conducted vibration, which restricts the risk early warning capability of the underground infrastructure. SUMMARY

[0003] The purpose of the embodiments of the present application is to overcome the deficiencies of the prior art, and provide a distributed acoustic sensing system environmental noise monitoring and event identification method, system and medium, which does not require tedious signal preprocessing, reduces the computational complexity, and is particularly suitable for safety monitoring applications that require fast response.

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] In a first aspect, the embodiments of the present application provide a distributed acoustic sensing system environmental noise monitoring and event identification method, comprising the following steps:

[0006] Obtain real-time environmental sound signals f(t) at different distances from the distributed acoustic sensing system;

[0007] Obtain the frequency domain signal in the sound signal through Fourier transform That is:

[0008] Based on the frequency domain signal, a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling is used to extract the characteristic values corresponding to each event;

[0009] Classify the sound signal features through convolutional neural network machine learning to determine the corresponding event;

[0010] Report the noise determination result.

[0011] The frequency domain signal is extracted by using a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling, and the characteristic values corresponding to each event are extracted, specifically,

[0012] Step 1: Dynamic variable-scale morphological base separation;

[0013] Step 2: Differential-energy dual-domain enhancement;

[0014] Step 3: Closed-loop morphological peak positioning.

[0015] The dynamic variable-scale morphological base separation is specifically,

[0016] Step 1.1: Structure element exponential increment sequence generation

[0017] ① Define the upper limit of iteration times Niter;

[0018] ② Calculate the initial structure element length:

[0019] Where fs is the sampling rate, and k represents the iteration number;

[0020] Step 1.2: Multi-scale morphological opening operation

[0021] ① Create a horizontal structure element: SE k = strel('rectangle', [1, L se (k)])

[0022] ② Perform morphological opening operation: Opened k = imopen(S-Trend k-1 , SE k )

[0023] Where S is the original signal, and Trendk is the trend estimate of the kth iteration

[0024] Step 1.3: Savitzky-Golay trend prediction correction

[0025] ① Set filter parameters:

[0026] Polynomial order d = 2 (quadratic polynomial)

[0027] Window length W k = 2L se (k) + 1

[0028] ② Trend correction update:

[0029] Treand k = α·Treand k-1 +(1-α)·sgolayfilt(Opened k , d, W k )

[0030] where, a e [0.6, 0.8] is the smoothing coefficient

[0031] Step 1.4: Trend convergence judgment

[0032] Calculate the relative change rate:

[0033]

[0034] When Δ k < ε (threshold ε = 0.01) stop iteration, output the final trend base. The differential-energy dual-domain enhancement is specifically,

[0035] Step 2.1: Second-order differential sharpening

[0036] Calculate the residual signal:

[0037] Residual = S-Trend final

[0038] Perform central difference:

[0039]

[0040] Enhancement signal construction:

[0041]

[0042] Step 2.2: Dynamic energy window statistics

[0043] ① Define the length of the sliding window:

[0044]

[0045] ② Calculate the moving average energy:

[0046]

[0047] Step 2.3: Dual threshold generation

[0048] ① Calculate the background statistics:

[0049]

[0050] IQR energy = iqr(Energy)

[0051] ② Generate adaptive threshold:

[0052] Thresh peak = Median energy + 3 x IQR energy .

[0053] The closed loop morphological peak positioning is specifically,

[0054] Step 3.1: Binary mask generation

[0055] ① Initial labeling:

[0056]

[0057] ② Closed operation noise reduction:

[0058]

[0059] Step 3.2: Precise peak positioning

[0060] ① Extract candidate regions:

[0061] Candidate = Residual ⊙ Mask clean

[0062] ② Peak identification condition setting:

[0063] Minimum prominence P min

[0064] Minimum peak distance

[0065]

[0066] ③ Apply peak detection algorithm:

[0067] PeakLocations = findpeaks(Candidate, 'MinPeakProminence', P min , 'MinPeakDistance', I.

[0068] Classify the features of the sound signal through convolutional neural network machine learning to determine the corresponding event, specifically,

[0069] Data preprocessing and feature structuring;

[0070] Select a 2D CNN architecture suitable for spectral graph image features:

[0071] Training and optimization;

[0072] Evaluation and tuning;

[0073] Expand the optimization strategy to accurately associate feature values with target event categories.

[0074] In a second aspect, the embodiments of the present application provide a distributed acoustic sensing system ambient noise monitoring and event identification system, comprising a memory and a processor, wherein the memory comprises a program of a distributed acoustic sensing system ambient noise monitoring and event identification method, and the program of the distributed acoustic sensing system ambient noise monitoring and event identification method is executed by the processor to implement the steps as described above.

[0075] In a third aspect, the embodiments of the present application provide a computer readable storage medium, which stores a program code, and the program code is executed by a processor to implement the steps of the distributed acoustic sensing system ambient noise monitoring and event identification method as described above.

[0076] Compared with the prior art, the present application has the following beneficial effects:

[0077] The noise monitoring and identification method of the present application improves the adaptability and robustness of the distributed acoustic sensing (DAS) system in complex and dynamic environments.

[0078] The hybrid model combining 1D-CNN and 2D-CNN architectures can effectively process diversified noise types and realize real-time monitoring of critical infrastructure.

[0079] The method of the present application does not require tedious signal preprocessing, reduces the computational complexity, and is particularly suitable for security monitoring applications that require fast response. BRIEF DESCRIPTION OF DRAWINGS

[0080] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0081] Figure 1 Time domain signal diagram of single distance sound;

[0082] Figure 2 Frequency domain signal diagram of sound at abnormal distance point;

[0083] Figure 3 Effect diagram of dynamic basis signal separation;

[0084] Figure 4 Residual signal and accurate peak detection diagram;

[0085] Figure 5 Flowchart of the method of the present application. DETAILED DESCRIPTION

[0086] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0087] The terms “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0088] The terms “first,” “second,” etc., are used only to distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance, nor as requiring or implying any such actual relationship or order between these entities or operations.

[0089] Please see Figure 5 A method for environmental noise monitoring and event identification using a distributed acoustic sensing system includes the following steps:

[0090] Obtain real-time ambient sound signals at different distances from a distributed acoustic sensing system;

[0091] The frequency domain signal in the sound signal is obtained by Fourier transform;

[0092] Based on frequency domain signals, a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling is used to extract the feature values ​​corresponding to each event.

[0093] The system uses convolutional neural networks to machine learn and classify the features of sound signals to determine the corresponding events.

[0094] Report the noise assessment results.

[0095] The method for extracting feature values ​​corresponding to each event using a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling, based on frequency domain signals, specifically involves:

[0096] Step 1: Dynamic variable-scale morphological basis separation;

[0097] Step 2: Differential-Energy Dual-Domain Enhancement;

[0098] Step 3: Closed-loop morphological peak localization.

[0099] The dynamic variable-scale morphological basis separation specifically refers to...

[0100] Step 1.1: Generation of exponentially increasing sequence of structuring elements

[0101] ① Define an upper limit on the number of iterations, Niter;

[0102] ② Calculate the initial struct element length:

[0103] Where fs is the sampling rate and k represents the iteration number;

[0104] Step 1.2: Multiscale Morphological Opening Operation

[0105] ① Create horizontal structural elements: SE k =strel('rectangle',[1,L) se (k)])

[0106] ② Perform morphological opening operation: Opened k =imopen(S-Trend) k-1 SE k )

[0107] In the formula, S is the original signal, and Trendk is the trend estimate for the k-th iteration.

[0108] Step 1.3: Savitzky-Golay Trend Forecast Correction

[0109] ① Set filter parameters:

[0110] The order of the polynomial is d = 2 (quadratic polynomial).

[0111] Window length W k =2L se (k)+1

[0112] ② Trend correction update:

[0113] Treand k =α·Treand k-1 +(1-α)·sgolayfilt(Opened k ,d,W k )

[0114] Where α∈[0.6,0.8] is the smoothing coefficient.

[0115] Step 1.4: Trend Convergence Judgment

[0116] Calculate the relative rate of change:

[0117]

[0118] When Δ k Iteration stops when the threshold ε (threshold ε = 0.01) is reached, and the final trend basis is output. The differential-energy dual-domain enhancement specifically involves...

[0119] Step 2.1: Second-order differential sharpening

[0120] Calculate the residual signal:

[0121] Residual = S-Trend final

[0122] Execution Center Differential:

[0123]

[0124] Enhanced signal construction:

[0125]

[0126] Step 2.2: Dynamic Energy Window Statistics

[0127] ① Define the length of the sliding window:

[0128]

[0129] ② Calculate the moving average energy:

[0130]

[0131] Step 2.3: Dual Threshold Generation

[0132] ① Calculate the background statistics:

[0133]

[0134] IQR energy =iqr(Energy) ② Generates an adaptive threshold:

[0135] Thresh peak =Median energy +3×IQR energy .

[0136] The closed-loop morphological peak localization is specifically as follows:

[0137] Step 3.1: Binary Mask Generation

[0138] ① Initial marker:

[0139]

[0140] ② Noise reduction through closing operations:

[0141]

[0142] Step 3.2: Precise Peak Positioning

[0143] ① Extract candidate regions:

[0144] Candidate=Residual⊙Mask clean

[0145] ②Peak identification condition settings:

[0146] Minimum protrusion Pmin

[0147] Minimum peak spacing

[0148]

[0149] ③ Apply peak detection algorithm:

[0150] PeakLocations=findpeaks(Candidate,'MinPeakProminence',P min ,'MinPeakDistance',I.

[0151] By using convolutional neural networks for machine learning, the features of sound signals are classified to determine the specific events they correspond to.

[0152] (1) Data preprocessing and feature structuring

[0153] Input format:

[0154] The feature is that the time spectrum is regarded as a two-dimensional input (similar to a single-channel grayscale image of an RGB image), with dimensions of (height, width, 1).

[0155] standardization:

[0156] Normalize the features (e.g., scale to [0,1] or standardize to a mean of 0 and a variance of 1).

[0157] Annotation matching:

[0158] Ensure that each feature sample corresponds to the correct event category label (one-hot encoding is commonly used for classification tasks).

[0159] Select CNN architecture type

[0160] 2D CNN (suitable for spectrogram image features):

[0161] Input shape: (number of frequency bands, number of time frames, 1) (single channel) or (number of frequency bands, number of time frames, number of channels) (multi-feature fusion).

[0162] Hierarchical design: Conv2D layer extracts local frequency domain and time domain patterns; MaxPooling2D reduces dimensionality.

[0163] 1D CNN (suitable for time series features):

[0164] Input shape: (time step, number of features).

[0165] Hierarchical design: Conv1D layers capture local temporal patterns (such as the temporal correlation of speech segments). MaxPooling1D compresses the temporal dimension.

[0166] (3) Training and optimization

[0167] Loss functions: Multi-class: Categorical Crosssentropy (labels must be one-hot). Integer labels: Sparse Categorical Crosssentropy.

[0168] Optimizer: Use Adam (default learning rate 1e-3) or adjust the learning rate (e.g., 1e-4).

[0169] Regularization: Add a Dropout layer (e.g., 0.5 dropout rate) or L2 regularization to prevent overfitting.

[0170] Data augmentation: Adding noise, time shift, and speed adjustment to the original audio signal; if the feature is in image form, slight deformation and occlusion can be applied to the spectrogram for enhancement.

[0171] (4) Evaluation and optimization

[0172] Validation metrics: Accuracy, F1-score (especially when classes are imbalanced). Error patterns are analyzed using a confusion matrix.

[0173] Early Stopping: Monitor the validation set loss and terminate training if it does not decrease for several consecutive rounds.

[0174] Hyperparameter tuning: Adjusting the kernel size, number of layers, number of filters, learning rate, etc.

[0175] (5) Extended optimization strategy

[0176] Pre-trained model transfer: If the amount of data is small, the feature extraction layer of a pre-trained audio model (such as VGGish, pre-trained ResNet) can be reused.

[0177] Hybrid models: These combine RNNs (such as LSTMs) to handle long-term temporal dependencies, for example, by following an LSTM layer after a CNN. Through this process, the CNN can automatically learn discriminative patterns from sound features and accurately associate feature values ​​with target event categories.

[0178] In a DAS system, time-domain signals of sound at different distances can be obtained. Even under a single distance condition, the time-domain signal of the sound can be obtained. For example... Figure 1 As shown.

[0179] Once the monitoring terminal detects an anomaly in the sound signal at a certain distance, it can determine that a special event has occurred at that location. Based on the time-domain signal of the sound at this distance, feature extraction will be used to qualitatively determine the type of event.

[0180] First, the time-domain signal is transformed into a frequency-domain signal using a Fourier transform. For example... Figure 2 As shown.

[0181] As can be seen, the signal exhibits a downward trend similar to a negative exponential function. The next step is to suppress noise in the signal and remove the base signal.

[0182] like Figure 3 The image shows the base signal after noise reduction of the frequency domain signal. This base signal is then removed from the original frequency domain signal to obtain the residual signal. The residual signal is shown below. Figure 4 As shown.

[0183] like Figure 4 The image shows the signal after the spectral audio signal has been processed by the algorithm. The peaks of the signal are the feature values ​​used to determine the event category. The positions are determined by... Figure 4 The triangular symbol in the middle.

[0184] Data on various known event categories is collected and regions are divided according to the location and magnitude of feature peaks. A 2D CNN network can then be used to determine the type of abnormal event. This allows for the identification of sound events occurring at specific times and distances, and the monitoring of abnormal environmental noise.

[0185] This application provides a distributed acoustic sensing system for environmental noise monitoring and event recognition, including a memory and a processor. The memory includes a program for a distributed acoustic sensing system for environmental noise monitoring and event recognition. When the program for the distributed acoustic sensing system for environmental noise monitoring and event recognition is executed by the processor, it implements the steps described above.

[0186] This application provides a computer-readable storage medium storing program code. When the program code is executed by a processor, it implements the steps of the distributed acoustic sensing system environmental noise monitoring and event recognition method as described above.

[0187] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for environmental noise monitoring and event recognition in a distributed acoustic sensing system, characterized in that, Includes the following steps: Obtain real-time ambient sound signals at different distances from a distributed acoustic sensing system. ; The frequency domain signal in the sound signal is obtained by Fourier transform. Abbreviated as S; Based on frequency domain signals, a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling is used to extract the feature values ​​corresponding to each event. The system uses convolutional neural networks to machine learn and classify the features of sound signals to determine the corresponding events. Report the noise assessment results; The method for extracting feature values ​​corresponding to each event using a dual-domain adaptive signal peak extraction method based on dynamic morphological coupling, based on frequency domain signals, specifically involves: Step 1: Dynamic variable-scale morphological basis separation; Step 2: Differential-Energy Dual-Domain Enhancement; Step 3: Closed-loop morphological peak localization; The dynamic variable-scale morphological basis separation specifically refers to... Step 1.1: Generating an exponentially increasing sequence of structuring elements ① Define an upper limit for the number of iterations, Niter; ② Calculate the initial struct element length: , Where fs is the sampling rate and k represents the iteration number; Step 1.2: Multi-scale morphological opening operation ① Create horizontal structuring elements: , ② Perform morphological opening operation: , In the formula, S is the original signal, and Trendk is the trend estimate for the k-th iteration. Step 1.3: Savitzky-Golay Trend Prediction Correction ① Set filter parameters: Polynomial order d=2 Window length ② Trend correction update: ， in, Smoothing coefficient Step 1.4: Calculate the relative rate of change to determine trend convergence: ， when Stop iterating when the time is right, output the final trend base, and set the threshold ε = 0.01; The differential-energy dual-domain enhancement specifically refers to... Step 2.1: Calculate the residual signal using second-order differential sharpening: ， Execution Center Differential: ， Enhanced signal construction: ， Step 2.2: Dynamic Energy Window Statistics ① Define the sliding window length: ， ② Calculate the moving average energy: ， Step 2.3: Dual Threshold Generation ① Calculate background statistics: ， ， ② Generate adaptive threshold: ； The specific method for locating the closed-loop morphological peak is as follows: Step 3.1: Binary Mask Generation ① Initial Marking: ， ② Noise reduction through closing operation: ， Step 3.2: Precise Peak Location ① Extract Candidate Region: ， ② Peak identification condition settings: Minimum prominence Pmin Minimum peak spacing ③ Apply peak detection algorithm: 。 2. The method for environmental noise monitoring and event recognition in a distributed acoustic sensing system according to claim 1, characterized in that, By using convolutional neural networks for machine learning, the features of sound signals are classified to determine the specific events they correspond to. Data preprocessing and feature structuring; Choose a 2D CNN architecture suitable for spectrogram image features: Training and optimization; Evaluation and optimization; Extend the optimization strategy to accurately correlate feature values ​​with target event categories.

3. A distributed acoustic sensing system for environmental noise monitoring and event recognition, characterized in that, The system includes a memory and a processor. The memory contains a program for a distributed acoustic sensing system environmental noise monitoring and event recognition method. When the program for the distributed acoustic sensing system environmental noise monitoring and event recognition method is executed by the processor, it implements the steps described in any one of claims 1-2.

4. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code, which, when executed by a processor, implements the steps of the distributed acoustic sensing system environmental noise monitoring and event recognition method as described in any one of claims 1 to 2.

Images