Environmental noise detection device and method

Abstract

The invention relates to the field of environmental noise detection, and provides an environmental noise detection device and method, and the method comprises the following steps: collecting an environmental noise signal, and carrying out the preprocessing of the environmental noise signal, and obtaining a to-be-detected audio; adjusting a gain compensation coefficient according to the energy intensity of the to-be-tested audio to obtain a gain multi-source noise signal; separating the gain multi-source noise signal to obtain a plurality of separated noise signals, extracting key characteristics of the separated noise signals, and classifying the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals; performing feature extraction and time series data analysis on the classified noise signals to obtain abnormal noise data, identifying the abnormal noise data to obtain a noise abnormal level, and adopting a corresponding emergency solution; the detection data and the early warning record are stored in a cloud database; a visualization result is generated based on the stored data. According to the invention, efficient detection, classification, anomaly identification and data management of environmental noise are realized.

Description

Technical Field

[0001] The present invention relates to the field of environmental noise detection, and in particular to an environmental noise detection device and method. Background Art

[0002] With the acceleration of urbanization and the improvement of industrialization, environmental noise pollution has become increasingly serious and has become one of the important factors affecting the quality of human life and health. The sources of environmental noise are complex and diverse, including industrial noise, traffic noise and natural noise. These noise signals have multi-dimensional characteristics such as frequency, intensity and duration, and are often mixed together, making it difficult to effectively separate and classify them.

[0003] In the prior art, traditional sound pickup equipment is often unable to effectively identify and process invalid segments, such as background noise or interference signals, when collecting environmental noise signals, resulting in poor quality of the collected signals, containing a large amount of invalid data or distorted signals, and lack of comprehensive analysis of the energy intensity, spectral characteristics and time distribution of the noise signal, resulting in poor noise detection accuracy. Summary of the invention

[0004] In view of this, the present invention proposes an environmental noise detection device and method, which solves the problem that the prior art cannot effectively identify and process invalid sections, resulting in poor quality of the collected signal, containing a large amount of invalid data or distorted signals, and poor noise detection accuracy.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides an environmental noise detection method, comprising the following steps:

[0006] Collecting an environmental noise signal through a sound pickup device, wherein the sound pickup device includes a microphone array, and preprocessing the environmental noise signal to obtain an audio to be measured;

[0007] Sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and a second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal;

[0008] Using a microphone array to separate the gain multi-source noise signal to obtain a plurality of separated noise signals, using spectrum analysis and support vector machine classification algorithm to extract key characteristics of the separated noise signals, and classifying the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals;

[0009] Based on the convolutional neural network, the classified noise signal is feature extracted, and the time series data of the classified noise signal is analyzed in combination with the long short-term memory network model to obtain abnormal noise data, and the abnormal noise data is identified through the preset early warning mechanism to obtain the abnormal noise level, and the emergency solution corresponding to the abnormal noise level is adopted;

[0010] The characteristic data of separated noise signals, classified noise signal data, detection data of abnormal noise events and early warning records are stored in a cloud database;

[0011] The visualization results are generated based on the storage data of the cloud database, and the visualization results include a noise distribution heat map and a characteristic data generation time series change map.

[0012] On the basis of the above technical solution, preferably, the environmental noise signal is collected by a sound pickup device, the sound pickup device includes a microphone array, and the environmental noise signal is preprocessed to obtain the audio to be measured, specifically including:

[0013] Collecting an environmental noise signal through a distributed sound pickup device, and transmitting the collected environmental noise signal to a processor, wherein the processor preliminarily screens the environmental noise signal based on a preset first intensity threshold to obtain a preliminarily screened signal;

[0014] The processor performs spectrum analysis and time distribution analysis on the preliminary screening signal, identifies a human voice signal below the first intensity threshold as an invalid segment, and replaces the invalid segment with audio data of a preset frequency to obtain the audio to be tested.

[0015] On the basis of the above technical solution, preferably, the audio to be tested is sent to a sound card device, a first energy intensity of the audio to be tested at a preset frequency is detected, and a third energy intensity is calculated based on the first energy intensity and the second energy intensity of the audio to be tested, and a gain compensation coefficient of a sound pickup device is adjusted according to the third energy intensity to obtain a gain multi-source noise signal, wherein the second energy intensity is the overall energy intensity of the audio to be tested, specifically including:

[0016] The audio to be tested is sent to a sound card device, the sound card device detects a first energy intensity of the audio to be tested at a preset frequency, and calculates a third energy intensity based on a ratio of the first energy intensity to the second energy intensity;

[0017] According to the difference between the third energy intensity and the preset reference energy intensity, the gain compensation coefficient of the sound pickup device is dynamically adjusted, and the gain compensation coefficient is applied to the signal processing of the audio to be measured to obtain the gain multi-source noise signal.

[0018] On the basis of the above technical solution, preferably, the gain multi-source noise signal is separated by using a microphone array to obtain a plurality of separated noise signals, spectrum analysis and support vector machine classification algorithm are used to extract key characteristics of the separated noise signals, and the plurality of separated noise signals are classified based on the key characteristics to obtain different types of classified noise signals, specifically including:

[0019] Based on the spatial distribution characteristics of the microphone array, a spatial positioning model of the gain multi-source noise signal is constructed, and the time delay of the gain multi-source noise signal reaching different microphones is calculated using the time difference estimation algorithm. According to the time delay information and the geometric structure of the microphone array, the beamforming algorithm is used to spatially separate the gain multi-source noise signal, extract the signals of different noise sources, perform preliminary energy screening on the separated noise signals, remove low-intensity signals or invalid signals, and obtain multiple separated noise signals;

[0020] Perform fast Fourier transform on the separated noise signal, extract spectrum characteristic data, calculate the energy integral value of the separated noise signal, extract the intensity characteristic of the separated noise signal, analyze the duration characteristic of the separated noise signal based on time series, input the extracted frequency, intensity and duration characteristics into the support vector machine classification model, classify the separated noise signal in combination with the preset noise feature library, and obtain different types of classified noise signals;

[0021] The calculation formula of the classification decision function of the support vector machine classification algorithm is:

[0022]

[0023] Among them, f(x) is the output value of the classification decision function, sign(·) is the sign function, α i is the weight coefficient of the i-th support vector, y i is the category label of the i-th support vector, K(w·x i ,w·x) is the kernel function, w is the feature weight vector, x is the input feature, x i is the i-th support vector, N is the number of support vectors, and b is the input bias of the classification decision function.

[0024] On the basis of the above technical solution, preferably, the feature extraction of the classified noise signal based on the convolutional neural network is performed, and the time series data analysis of the classified noise signal is performed in combination with the long short-term memory network model to obtain abnormal noise data, and the abnormal noise data is identified through a preset early warning mechanism to obtain the abnormal noise level, and the emergency solution corresponding to the abnormal noise level is adopted, which specifically includes:

[0025] Based on the convolutional neural network, the classified noise signal is subjected to feature extraction, and a multi-dimensional feature vector including frequency feature, intensity feature and spatial distribution characteristic is extracted;

[0026] A convolutional neural network structure consisting of multiple convolutional layers and pooling layers is used to process the time-frequency images of classified noise signals. Features of different scales and directions are extracted through convolution operations. Multidimensional feature vectors are standardized and nonlinearly transformed using batch normalization and activation functions. A fully connected layer is added to the last layer of the convolutional neural network structure to map the extracted multidimensional feature vectors to a low-dimensional feature space, thus obtaining a multidimensional feature vector including frequency features, intensity features, and spatial distribution characteristics.

[0027] Combined with the long short-term memory network model, the time series data of the classified noise signal is analyzed to identify and extract abnormal noise data, which includes excessive noise and sudden noise. The abnormal noise data is monitored and warned in real time through a preset warning mechanism to determine the abnormal noise level, and take corresponding emergency solutions according to the abnormal noise level.

[0028] The bidirectional long short-term memory network model is used to perform temporal dependency analysis on the multidimensional feature vector to capture the long-term and short-term dependencies in the noise signal. The output of the bidirectional long short-term memory network model is weighted through the attention mechanism to obtain the context vector. Based on the context vector and the preset warning threshold, the noise anomaly level is obtained according to the attention-weighted temporal characteristics, and the emergency solution corresponding to the noise anomaly level is called for response processing.

[0029] The calculation formula for weighting the output of the bidirectional long short-term memory network model through the attention mechanism is:

[0030]

[0031] Among them, α t is the attention weight at time t, h t is the hidden state of the bidirectional long short-term memory network model at time t, h k is the hidden state of the bidirectional long short-term memory network model at time k, W h is the trainable matrix in the attention mechanism, b h is the bias vector in the attention mechanism, v is the trainable weight vector in the attention mechanism, tanh(·) is the hyperbolic tangent activation function, exp(·) is the exponential function, and c is the context vector.

[0032] On the basis of the above technical solution, preferably, the characteristic data of the separated noise signal, the classified noise signal data, the detection data of the abnormal noise event and the warning record are stored in the cloud database, specifically including:

[0033] Construct a hierarchical data storage structure in the cloud database, index and store the characteristic data of separated noise signals and classified noise signal data according to timestamp and spatial location information, and classify and store the detection data and warning records of abnormal noise events according to event type and severity;

[0034] Regularly back up and clean up data stored in the cloud database, compress and archive historical data that exceeds the preset storage period, and establish a data access permission control mechanism.

[0035] On the basis of the above technical solution, preferably, the storage data based on the cloud database generates visualization results, and the visualization results include a noise distribution heat map and a characteristic data generation time series change map, specifically including:

[0036] Retrieve noise monitoring data from the cloud database, obtain the spatial location information and intensity information of the noise signal based on the noise monitoring data to generate a noise distribution heat map, and generate a noise characteristic change trend map based on the time series data of the noise monitoring data;

[0037] The visualization results are dynamically updated and interactively displayed. The filtering method of the visualization results includes supporting user-defined time range, spatial range and noise type filtering. The visualization results are used for data export and report generation.

[0038] In a second aspect, the present invention further provides an environmental noise detection device, the device comprising:

[0039] A signal acquisition module is used to collect environmental noise signals through a sound pickup device, wherein the sound pickup device includes a microphone array, and pre-process the environmental noise signals to obtain audio to be measured;

[0040] A gain compensation module, used for sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and the second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal;

[0041] A signal separation module is used to separate the gain multi-source noise signal by using a microphone array to obtain a plurality of separated noise signals, extract key characteristics of the separated noise signals by using spectrum analysis and support vector machine classification algorithm, and classify the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals;

[0042] The abnormal warning module is used to extract features of classified noise signals based on convolutional neural networks, and to perform time series data analysis on classified noise signals in combination with long short-term memory network models to obtain abnormal noise data, and to identify the abnormal noise data through a preset early warning mechanism to obtain the abnormal noise level, and to adopt emergency solutions corresponding to the abnormal noise level;

[0043] A data storage module, used to store characteristic data of separated noise signals, classified noise signal data, detection data of abnormal noise events and early warning records in a cloud database;

[0044] The visual display module is used to generate visualization results based on the storage data of the cloud database, and the visualization results include a noise distribution heat map and a time series change map generated by characteristic data.

[0045] In a third aspect, the present invention further provides an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0046] The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps of an environmental noise detection method.

[0047] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement steps such as an environmental noise detection method.

[0048] The environmental noise detection device and method of the present invention have the following beneficial effects compared with the prior art:

[0049] (1) The environmental noise signal is collected and preprocessed through the sound pickup device, and the signal quality is optimized by combining the dynamic gain compensation mechanism. The microphone array is used to separate the multi-source noise signal, and the noise signal is classified through spectrum analysis and support vector machine classification algorithm. The convolutional neural network and long short-term memory network model are combined to extract features and perform time series analysis on the classified noise signal, accurately identify abnormal noise and issue early warnings, and finally store the noise data in the cloud database and generate visualization results, realizing efficient detection, classification, abnormality identification and data management of environmental noise, and improving the intelligence and accuracy of noise detection;

[0050] (2) The gain multi-source noise signals are separated by utilizing the spatial positioning capability of the microphone array, and the key characteristics of the separated noise signals such as frequency, intensity and duration are extracted by combining spectrum analysis. The separated noise signals are classified by using the support vector machine classification algorithm, which can accurately distinguish different types of classified noise signals such as industrial noise, traffic noise and natural noise.

[0051] (3) The convolutional neural network is used to extract features of the classified noise signal, and multi-dimensional feature vectors such as frequency features, intensity features and spatial distribution characteristics can be extracted from the time-frequency image. The design of multiple convolutional layers, pooling layers and fully connected layers enhances the feature expression ability and the generalization performance of the model. The long short-term memory network is combined to analyze the time series data of the classified noise signal, capture the long-term and short-term dependencies in the noise signal, and use the attention mechanism to highlight the abnormal features at critical moments, thereby improving the recognition sensitivity and accuracy of abnormal noise. The early warning mechanism determines the abnormal level of noise and takes corresponding emergency solutions, thus realizing real-time monitoring, accurate identification and efficient response to noise anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 these drawings without paying creative work.

[0053] Figure 1 This is a flow chart of an environmental noise detection method of the present invention;

[0054] Figure 2 This is a structural diagram of an environmental noise detection device of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See also Figure 1 The present invention provides an environmental noise detection method, comprising the following steps:

[0057] Collecting an environmental noise signal through a sound pickup device, wherein the sound pickup device includes a microphone array, and preprocessing the environmental noise signal to obtain an audio to be measured;

[0058] Sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and a second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal;

[0059] Using a microphone array to separate the gain multi-source noise signal to obtain a plurality of separated noise signals, using spectrum analysis and support vector machine classification algorithm to extract key characteristics of the separated noise signals, and classifying the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals;

[0060] Based on the convolutional neural network, the classified noise signal is feature extracted, and the time series data of the classified noise signal is analyzed in combination with the long short-term memory network model to obtain abnormal noise data, and the abnormal noise data is identified through the preset early warning mechanism to obtain the abnormal noise level, and the emergency solution corresponding to the abnormal noise level is adopted;

[0061] The characteristic data of separated noise signals, classified noise signal data, detection data of abnormal noise events and early warning records are stored in a cloud database;

[0062] The visualization results are generated based on the storage data of the cloud database, and the visualization results include a noise distribution heat map and a characteristic data generation time series change map.

[0063] Specifically, this embodiment collects environmental noise signals through sound pickup equipment and performs preprocessing, optimizes signal quality in combination with a dynamic gain compensation mechanism, uses a microphone array to separate multi-source noise signals, and classifies noise signals through spectrum analysis and support vector machine classification algorithms. It combines convolutional neural networks and long short-term memory network models to perform feature extraction and time series analysis on classified noise signals, accurately identifies abnormal noise and issues early warnings, and finally stores the noise data in a cloud database and generates visualization results, thereby achieving efficient detection, classification, abnormality identification and data management of environmental noise, and improving the intelligence and accuracy of noise detection.

[0064] The environmental noise signal is collected by a sound pickup device, the sound pickup device includes a microphone array, and the environmental noise signal is preprocessed to obtain the audio to be measured, specifically including:

[0065] Collecting an environmental noise signal through a distributed sound pickup device, and transmitting the collected environmental noise signal to a processor, wherein the processor preliminarily screens the environmental noise signal based on a preset first intensity threshold to obtain a preliminarily screened signal;

[0066] The distributed sound pickup device includes a plurality of microphone arrays, each of which is provided with a plurality of sound pickup units;

[0067] The plurality of sound pickup units are arranged according to a preset spatial distribution to form a ring-shaped or linear sound pickup array structure;

[0068] The processor collects omnidirectional environmental noise signals through the sound pickup array structure, and performs preliminary screening on the collected environmental noise signals based on the first intensity threshold;

[0069] The processor performs spectrum analysis and time distribution analysis on the preliminary screening signal, identifies a human voice signal below the first intensity threshold as an invalid section, and replaces the invalid section with audio data of a preset frequency to obtain the audio to be tested;

[0070] The processor uses fast Fourier transform to perform spectrum analysis on the preliminary screening signal to obtain spectrum characteristic data;

[0071] The processor identifies the frequency range and energy distribution characteristics of the human voice signal based on the frequency spectrum characteristic data;

[0072] The processor matches the recognized human voice signal with a preset audio feature library, determines the time range of the invalid section, and replaces it with audio data of a preset frequency.

[0073] Specifically, this embodiment collects environmental noise signals through distributed sound pickup equipment, and uses a processor to perform preliminary screening of the collected signals, which can effectively filter out low-intensity or invalid noise signals, reduce interference from irrelevant data, and improve the quality of the audio to be tested; at the same time, the frequency range and energy distribution characteristics of the human voice signal are identified based on the spectrum characteristic data, and the invalid segments are accurately identified and replaced in combination with a preset audio feature library, which further optimizes the effectiveness and accuracy of the signal and lays a high-quality foundation for noise signal processing and analysis.

[0074] The method of sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and a second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal, wherein the second energy intensity is the overall energy intensity of the audio to be tested, specifically includes:

[0075] The audio to be tested is sent to a sound card device, the sound card device detects a first energy intensity of the audio to be tested at a preset frequency, and calculates a third energy intensity based on a ratio of the first energy intensity to the second energy intensity;

[0076] The sound card device uses a bandpass filter to filter the audio to be tested, and extracts an audio signal within a preset frequency range;

[0077] The sound card device calculates the energy integral value of the audio signal within the preset frequency range as the first energy intensity;

[0078] The sound card device calculates the energy integral value of the audio to be tested within the full frequency band as the second energy intensity;

[0079] The third energy intensity is equal to the ratio of the first energy intensity to the second energy intensity;

[0080] According to the difference between the third energy intensity and the preset reference energy intensity, dynamically adjust the gain compensation coefficient of the sound pickup device, and apply the gain compensation coefficient to the signal processing of the audio to be measured to obtain the gain multi-source noise signal;

[0081] The preset reference energy intensity is determined based on the noise signal acquisition result under the standard test environment;

[0082] When the difference between the third energy intensity and the reference energy intensity is greater than a preset threshold, adjusting the gain compensation coefficient according to the positive or negative sign of the difference;

[0083] The adjustment range of the gain compensation coefficient is limited between a preset maximum value and a preset minimum value to avoid signal distortion.

[0084] Specifically, this embodiment performs bandpass filtering on the audio to be tested through a sound card device, accurately extracts the audio signal within a preset frequency range, and obtains a third energy intensity reflecting the gain characteristics of the sound pickup device by calculating the ratio of the energy integral value (first energy intensity) within the preset frequency range to the energy integral value (second energy intensity) within the full frequency band; further, based on the difference between the third energy intensity and the preset reference energy intensity, the gain compensation coefficient is dynamically adjusted within the preset maximum and minimum value range, thereby achieving accurate compensation for the gain characteristics of the sound pickup device and effectively avoiding signal distortion problems.

[0085] The method of using a microphone array to separate the gain multi-source noise signal to obtain a plurality of separated noise signals, using spectrum analysis and support vector machine classification algorithm to extract key characteristics of the separated noise signals, and classifying the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals, specifically includes:

[0086] Based on the spatial distribution characteristics of the microphone array, a spatial positioning model of the gain multi-source noise signal is constructed, and the time delay of the gain multi-source noise signal reaching different microphones is calculated using the time difference estimation algorithm. According to the time delay information and the geometric structure of the microphone array, the beamforming algorithm is used to spatially separate the gain multi-source noise signal, extract the signals of different noise sources, perform preliminary energy screening on the separated noise signals, remove low-intensity signals or invalid signals, and obtain multiple separated noise signals;

[0087] Perform fast Fourier transform on the separated noise signal, extract spectrum characteristic data, including main frequency, bandwidth and frequency distribution, calculate the energy integral value of the separated noise signal, extract the intensity characteristic of the separated noise signal, analyze the duration characteristic of the separated noise signal based on time series, input the extracted frequency, intensity and duration characteristics into the support vector machine classification model, classify the separated noise signal in combination with the preset noise feature library, and obtain different types of classified noise signals such as industrial noise, traffic noise and natural noise;

[0088] The calculation formula of the classification decision function of the support vector machine classification algorithm is:

[0089]

[0090] Among them, f(x) is the output value of the classification decision function, sign(·) is the sign function, α i is the weight coefficient of the i-th support vector, y i is the category label of the i-th support vector, K(w·x i ,w·x) is the kernel function, w is the feature weight vector, x is the input feature, x i is the i-th support vector, N is the number of support vectors, and b is the input bias of the classification decision function.

[0091] Specifically, this embodiment separates the gain multi-source noise signal by utilizing the spatial positioning capability of the microphone array, and extracts key characteristics such as frequency, intensity and duration of the separated noise signal in combination with spectrum analysis, thereby being able to effectively capture the multi-dimensional characteristics of the noise signal.

[0092] The support vector machine classification algorithm is used to classify the separated noise signals through the classification decision function. Combined with the preset noise feature library, it can accurately distinguish different types of noise signals such as industrial noise, traffic noise and natural noise. The kernel function and weight optimization mechanism of the support vector machine ensure the efficiency and robustness of the classification, especially when processing complex noise signals, which can significantly improve the classification accuracy.

[0093] The extracted characteristics such as frequency, intensity and duration comprehensively reflect the physical and temporal characteristics of the noise signal, and can provide rich feature information for anomaly detection and timing analysis. By combining the spatial positioning capability of the microphone array and the support vector machine classification algorithm, it can effectively separate and classify multi-source noise signals in complex environments, adapt to scenarios with mixed noise sources, and improve the applicability and reliability of noise detection methods.

[0094] The feature extraction of the classified noise signal based on the convolutional neural network and the time series data analysis of the classified noise signal in combination with the long short-term memory network model are performed to obtain abnormal noise data, and the abnormal noise data is identified through a preset early warning mechanism to obtain the abnormal noise level, and the emergency solution corresponding to the abnormal noise level is adopted, which specifically includes:

[0095] Based on the convolutional neural network, the classified noise signal is subjected to feature extraction, and a multi-dimensional feature vector including frequency feature, intensity feature and spatial distribution characteristic is extracted;

[0096] A convolutional neural network structure consisting of multiple convolutional layers and pooling layers is used to process the time-frequency images of classified noise signals. Features of different scales and directions are extracted through convolution operations. Batch normalization and activation functions are used to standardize and nonlinearly transform multidimensional feature vectors to enhance the expressiveness of features and the generalization performance of the model. A fully connected layer is added to the last layer of the convolutional neural network structure to map the extracted multidimensional feature vectors to a low-dimensional feature space for subsequent time series analysis and processing, thereby obtaining a multidimensional feature vector including frequency features, intensity features, and spatial distribution characteristics.

[0097] Combined with the long short-term memory network model, the time series data of the classified noise signal is analyzed to identify and extract abnormal noise data, which includes excessive noise and sudden noise. The abnormal noise data is monitored and warned in real time through a preset warning mechanism to determine the abnormal noise level, and take corresponding emergency solutions according to the abnormal noise level.

[0098] The bidirectional long short-term memory network model is used to analyze the temporal dependency of the multidimensional feature vector, capture the long-term and short-term dependencies in the noise signal, and weight the output of the bidirectional long short-term memory network model through the attention mechanism to highlight the abnormal features at critical moments, improve the sensitivity and accuracy of abnormality recognition, and obtain the context vector. Based on the context vector and the preset warning threshold, the noise anomaly level is obtained according to the time series features after attention weighting, and the emergency solution corresponding to the noise anomaly level is called for response processing;

[0099] The calculation formula for weighting the output of the bidirectional long short-term memory network model through the attention mechanism is:

[0100]

[0101] Among them, α t is the attention weight at time t, h t is the hidden state of the bidirectional long short-term memory network model at time t, h k is the hidden state of the bidirectional long short-term memory network model at time k, W h is the trainable matrix in the attention mechanism, b h is the bias vector in the attention mechanism, v is the trainable weight vector in the attention mechanism, tanh(·) is the hyperbolic tangent activation function, exp(·) is the exponential function, and c is the context vector.

[0102] Specifically, this embodiment extracts features from classified noise signals based on convolutional neural networks, and can extract multi-dimensional feature vectors such as frequency features, intensity features, and spatial distribution characteristics from the time-frequency image of the noise signal. Through the structural design of multiple convolutional layers and pooling layers, the convolution operation can capture features of different scales and directions. The application of batch normalization and activation functions further enhances the expressiveness of features and the generalization performance of the model, ensuring the comprehensiveness and accuracy of feature extraction.

[0103] A fully connected layer is added to the last layer of the convolutional neural network to map the extracted multi-dimensional feature vector to a low-dimensional feature space, reducing the complexity of the data dimension while retaining key feature information.

[0104] By combining the long short-term memory network model with the time series data of classified noise signals, it is possible to capture the long-term and short-term dependencies in the noise signals, identify and extract abnormal noise data (including excessive noise and burst noise). Through the design of the bidirectional long short-term memory network model, the ability to capture the time series characteristics of the noise signal is further enhanced, ensuring the sensitivity and accuracy of abnormal noise detection.

[0105] By using the attention mechanism to weight the output of the bidirectional long short-term memory network model, the abnormal features at key moments can be highlighted, context vectors can be generated, and the recognition ability and accuracy of abnormal noise can be further improved. The weighted calculation of the attention mechanism can effectively focus on the most important time points in the noise signal and avoid the interference of irrelevant information.

[0106] Based on the context vector and the preset warning threshold, the noise abnormality level can be accurately obtained, and corresponding emergency solutions can be taken according to the noise abnormality level. This ensures the real-time monitoring and warning capabilities of abnormal noise, and can quickly respond to abnormal noise of different levels.

[0107] By combining the convolutional neural network and the bidirectional long short-term memory network model, and utilizing the synergy of multidimensional feature extraction and time series analysis, the generalization performance and robustness of the model in complex noise environments are significantly improved, and it can adapt to various types of noise signals and complex environmental changes.

[0108] The characteristic data of the separated noise signal, the classified noise signal data, the detection data of the abnormal noise event and the early warning record are stored in the cloud database, specifically including:

[0109] Construct a hierarchical data storage structure in the cloud database, index and store the characteristic data of separated noise signals and classified noise signal data according to timestamp and spatial location information, and classify and store the detection data and warning records of abnormal noise events according to event type and severity;

[0110] The hierarchical data storage structure includes an original data layer, a feature data layer and an event data layer;

[0111] The original data layer stores the original sampling data of the separated noise signal;

[0112] The feature data layer stores data such as frequency characteristics, intensity characteristics, and time characteristics of noise signals;

[0113] The event data layer stores information such as the type, occurrence time, duration, and processing status of abnormal noise events;

[0114] Each layer of data is associated through a unique identifier, supporting multi-dimensional data query and analysis;

[0115] Regularly back up and clean up data stored in the cloud database, compress and archive historical data that exceeds the preset storage period, and establish a data access permission control mechanism to ensure data security and traceability;

[0116] Automatically back up data according to the preset time period to generate incremental backup and full backup;

[0117] Compress data that exceeds the preset storage period and migrate the compressed data to the archive storage area;

[0118] Establish a multi-level data access permission system, including administrator permission, operator permission and view permission;

[0119] Record data access, modification and deletion operation logs to achieve full tracking of data operations.

[0120] Specifically, this embodiment constructs a hierarchical data storage structure in a cloud database, indexes and stores the characteristic data of separated noise signals and classified noise signal data according to timestamps and spatial location information, and classifies and stores the detection data and warning records of abnormal noise events according to event type and severity, thereby achieving efficient organization and rapid retrieval of data and improving the efficiency and response speed of data management.

[0121] By automatically backing up data according to the preset time period, generating incremental backups and full backups, the integrity and security of the data are ensured; data that exceeds the preset storage period is compressed and migrated to the archive storage area, optimizing the utilization of storage space while retaining the traceability of historical data.

[0122] A multi-level data access permission system is established, including administrator permissions, operator permissions, and viewing permissions, which can flexibly control the scope of data access according to user roles to ensure data security and privacy; at the same time, data access, modification, and deletion operation logs are recorded to achieve full tracking of data operations.

[0123] The storage data based on the cloud database generates visualization results, and the visualization results include a noise distribution heat map and a characteristic data generation time series change map, specifically including:

[0124] Retrieve noise monitoring data from the cloud database, obtain the spatial location information and intensity information of the noise signal based on the noise monitoring data to generate a noise distribution heat map, and generate a noise characteristic change trend map based on the time series data of the noise monitoring data to achieve multi-dimensional data visualization;

[0125] Construct a base map of the monitoring area based on the geographic information system;

[0126] Mapping the spatial location information of the noise signal onto the base map;

[0127] Set different color gradients according to noise intensity;

[0128] The interpolation algorithm is used to perform spatial interpolation on the discrete monitoring point data to generate a continuous noise distribution heat map;

[0129] Supports comparative display of noise distribution in multiple time periods;

[0130] Dynamically update and interactively display the visualization results. The filtering method of the visualization results includes supporting user-defined time range, spatial range and noise type filtering. The visualization results are used for data export and report generation.

[0131] Set an adjustable time window to support switching display of real-time data and historical data;

[0132] Provide multi-level spatial area selection, including data display of the overall region, sub-region and specific monitoring points;

[0133] Support data filtering by noise type (industrial noise, traffic noise, natural noise);

[0134] Provides data export interface and supports report generation in multiple formats, including daily, weekly and monthly reports;

[0135] Record users' data access and export operations to ensure the traceability of data usage.

[0136] Specifically, this embodiment generates a noise distribution heat map and a noise characteristic change trend map by retrieving noise monitoring data from a cloud database, which can intuitively display the spatial distribution and time-varying characteristics of noise signals. The noise distribution heat map provides a visual distribution of noise intensity at different spatial locations, helping users to quickly locate noise sources; the noise characteristic change trend map shows the dynamic changes of noise signals over time, making it easier for users to analyze long-term trends and abnormal fluctuations in noise.

[0137] The visualization results support dynamic updates and interactive displays. Users can customize the time range, spatial range, and noise type for filtering as needed, and flexibly adjust the dimension and granularity of data display.

[0138] By setting an adjustable time window, it supports switching display of real-time data and historical data, which can not only meet the needs of real-time monitoring, but also perform retrospective analysis on historical data, providing comprehensive data support for noise management and decision-making.

[0139] It provides a multi-level spatial area selection function for the overall area, sub-region and specific monitoring points, which can display noise data at different levels according to user needs. It can not only provide a macroscopic understanding of the overall noise distribution, but also microscopically analyze the noise characteristics of specific monitoring points, thus improving the level of refinement of data display.

[0140] The visualization results support data export and report generation functions. Users can export filtered data and charts for further analysis or report generation, facilitating data sharing and application.

[0141] See also Figure 2 The present invention also provides an environmental noise detection device, the device comprising:

[0142] A signal acquisition module is used to collect environmental noise signals through a sound pickup device, wherein the sound pickup device includes a microphone array, and pre-process the environmental noise signals to obtain audio to be measured;

[0143] A gain compensation module, used for sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and the second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal;

[0144] A signal separation module is used to separate the gain multi-source noise signal by using a microphone array to obtain a plurality of separated noise signals, extract key characteristics of the separated noise signals by using spectrum analysis and support vector machine classification algorithm, and classify the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals;

[0145] The abnormal warning module is used to extract features of classified noise signals based on convolutional neural networks, and to perform time series data analysis on classified noise signals in combination with long short-term memory network models to obtain abnormal noise data, and to identify the abnormal noise data through a preset early warning mechanism to obtain the abnormal noise level, and to adopt emergency solutions corresponding to the abnormal noise level;

[0146] A data storage module, used to store characteristic data of separated noise signals, classified noise signal data, detection data of abnormal noise events and early warning records in a cloud database;

[0147] The visual display module is used to generate visualization results based on the storage data of the cloud database, and the visualization results include a noise distribution heat map and a time series change map generated by characteristic data.

[0148] Specifically, an environmental noise detection device of the present embodiment realizes efficient acquisition of environmental noise signals, dynamic gain compensation, accurate separation and classification of multi-source noise signals, real-time detection and warning of abnormal noise, safe storage and management of long-term monitoring data, and intuitive visual display of noise distribution and characteristic changes through the coordinated work of a signal acquisition module, a gain compensation module, a signal separation module, an abnormal warning module, a data storage module and a visual display module, thereby ensuring the flexibility and scalability of the device, being able to adapt to the noise detection needs in complex environments, significantly improving the intelligence, accuracy and practicality of noise detection, and providing support for noise monitoring, management and decision-making.

[0149] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement an environmental noise detection method.

[0150] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of an environmental noise detection method described in an embodiment of the present invention. The storage medium includes: a U disk, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and other media that can store program codes.

[0151] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting environmental noise, characterized in that: The following steps are involved: Collecting an environmental noise signal through a sound pickup device, wherein the sound pickup device includes a microphone array, and preprocessing the environmental noise signal to obtain an audio to be measured; Sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, and calculating a third energy intensity based on the first energy intensity and a second energy intensity of the audio to be tested, adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal, wherein the second energy intensity is the overall energy intensity of the audio to be tested; Using a microphone array to separate the gain multi-source noise signal to obtain a plurality of separated noise signals, using spectrum analysis and support vector machine classification algorithm to extract key characteristics of the separated noise signals, and classifying the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals; Based on the convolutional neural network, the classified noise signal is feature extracted, and the time series data of the classified noise signal is analyzed in combination with the long short-term memory network model to obtain abnormal noise data, and the abnormal noise data is identified through the preset early warning mechanism to obtain the abnormal noise level, and the emergency solution corresponding to the abnormal noise level is adopted; The characteristic data of separated noise signals, classified noise signal data, detection data of abnormal noise events and early warning records are stored in a cloud database; The visualization results are generated based on the storage data of the cloud database, and the visualization results include a noise distribution heat map and a characteristic data generation time series change map.

2. The environmental noise detection method according to claim 1, characterized in that: The environmental noise signal is collected by a sound pickup device, the sound pickup device includes a microphone array, and the environmental noise signal is preprocessed to obtain the audio to be measured, specifically including: Collecting an environmental noise signal through a distributed sound pickup device, and transmitting the collected environmental noise signal to a processor, wherein the processor preliminarily screens the environmental noise signal based on a preset first intensity threshold to obtain a preliminarily screened signal; The processor performs spectrum analysis and time distribution analysis on the preliminary screening signal, identifies a human voice signal below the first intensity threshold as an invalid segment, and replaces the invalid segment with audio data of a preset frequency to obtain the audio to be tested.

3. The environmental noise detection method according to claim 1, characterized in that: The method of sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and a second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal, wherein the second energy intensity is the overall energy intensity of the audio to be tested, specifically includes: The audio to be tested is sent to a sound card device, the sound card device detects a first energy intensity of the audio to be tested at a preset frequency, and calculates a third energy intensity based on a ratio of the first energy intensity to the second energy intensity; According to the difference between the third energy intensity and the preset reference energy intensity, the gain compensation coefficient of the sound pickup device is dynamically adjusted, and the gain compensation coefficient is applied to the signal processing of the audio to be measured to obtain the gain multi-source noise signal.

4. The environmental noise detection method according to claim 1, characterized in that: The method of using a microphone array to separate the gain multi-source noise signal to obtain a plurality of separated noise signals, using spectrum analysis and support vector machine classification algorithm to extract key characteristics of the separated noise signals, and classifying the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals, specifically includes: Based on the spatial distribution characteristics of the microphone array, a spatial positioning model of the gain multi-source noise signal is constructed, and the time delay of the gain multi-source noise signal reaching different microphones is calculated using the time difference estimation algorithm. According to the time delay information and the geometric structure of the microphone array, the beamforming algorithm is used to spatially separate the gain multi-source noise signal, extract the signals of different noise sources, perform preliminary energy screening on the separated noise signals, remove low-intensity signals or invalid signals, and obtain multiple separated noise signals; Perform fast Fourier transform on the separated noise signal, extract spectrum characteristic data, calculate the energy integral value of the separated noise signal, extract the intensity characteristic of the separated noise signal, analyze the duration characteristic of the separated noise signal based on time series, input the extracted frequency, intensity and duration characteristics into the support vector machine classification model, classify the separated noise signal in combination with the preset noise feature library, and obtain different types of classified noise signals; The calculation formula of the classification decision function of the support vector machine classification algorithm is: Among them, f(x) is the output value of the classification decision function, sign(·) is the sign function, α i is the weight coefficient of the i-th support vector, y i is the category label of the i-th support vector, K(w·x i ,w·x) is the kernel function, w is the feature weight vector, x is the input feature, x i is the i-th support vector, N is the number of support vectors, and b is the input bias of the classification decision function.

5. The environmental noise detection method according to claim 1, characterized in that: The feature extraction of the classified noise signal based on the convolutional neural network and the time series data analysis of the classified noise signal in combination with the long short-term memory network model are performed to obtain abnormal noise data, and the abnormal noise data is identified through a preset early warning mechanism to obtain the abnormal noise level, and the emergency solution corresponding to the abnormal noise level is adopted, which specifically includes: Based on the convolutional neural network, the classified noise signal is subjected to feature extraction, and a multi-dimensional feature vector including frequency feature, intensity feature and spatial distribution characteristic is extracted; A convolutional neural network structure consisting of multiple convolutional layers and pooling layers is used to process the time-frequency images of classified noise signals. Features of different scales and directions are extracted through convolution operations. Multidimensional feature vectors are standardized and nonlinearly transformed using batch normalization and activation functions. A fully connected layer is added to the last layer of the convolutional neural network structure to map the extracted multidimensional feature vectors to a low-dimensional feature space, thus obtaining a multidimensional feature vector including frequency features, intensity features, and spatial distribution characteristics. Combined with the long short-term memory network model, the time series data of the classified noise signal is analyzed to identify and extract abnormal noise data, which includes excessive noise and sudden noise. The abnormal noise data is monitored and warned in real time through a preset warning mechanism to determine the abnormal noise level, and take corresponding emergency solutions according to the abnormal noise level. The bidirectional long short-term memory network model is used to perform temporal dependency analysis on the multidimensional feature vector to capture the long-term and short-term dependencies in the noise signal. The output of the bidirectional long short-term memory network model is weighted through the attention mechanism to obtain the context vector. Based on the context vector and the preset warning threshold, the noise anomaly level is obtained according to the attention-weighted temporal characteristics, and the emergency solution corresponding to the noise anomaly level is called for response processing. The calculation formula for weighting the output of the bidirectional long short-term memory network model through the attention mechanism is: Among them, α t is the attention weight at time t, h t is the hidden state of the bidirectional long short-term memory network model at time t, h k is the hidden state of the bidirectional long short-term memory network model at time k, W h is the trainable matrix in the attention mechanism, b h is the bias vector in the attention mechanism, v is the trainable weight vector in the attention mechanism, tanh(·) is the hyperbolic tangent activation function, exp(·) is the exponential function, and c is the context vector.

6. The environmental noise detection method according to claim 5, characterized in that: The characteristic data of the separated noise signal, the classified noise signal data, the detection data of the abnormal noise event and the early warning record are stored in the cloud database, specifically including: Construct a hierarchical data storage structure in the cloud database, index and store the characteristic data of separated noise signals and classified noise signal data according to timestamp and spatial location information, and classify and store the detection data and warning records of abnormal noise events according to event type and severity; Regularly back up and clean up data stored in the cloud database, compress and archive historical data that exceeds the preset storage period, and establish a data access permission control mechanism.

7. The environmental noise detection method according to claim 6, characterized in that: The storage data based on the cloud database generates visualization results, and the visualization results include a noise distribution heat map and a characteristic data generation time series change map, specifically including: Retrieve noise monitoring data from the cloud database, obtain the spatial location information and intensity information of the noise signal based on the noise monitoring data to generate a noise distribution heat map, and generate a noise characteristic change trend map based on the time series data of the noise monitoring data; The visualization results are dynamically updated and interactively displayed. The filtering method of the visualization results includes supporting user-defined time range, spatial range and noise type filtering. The visualization results are used for data export and report generation.

8. An environmental noise detection device, used to execute an environmental noise detection method according to claims 1-7, characterized in that: The device comprises: A signal acquisition module is used to collect environmental noise signals through a sound pickup device, wherein the sound pickup device includes a microphone array, and pre-process the environmental noise signals to obtain audio to be measured; A gain compensation module, used for sending the audio to be tested to a sound card device, detecting a first energy intensity of the audio to be tested at a preset frequency, calculating a third energy intensity based on the first energy intensity and the second energy intensity of the audio to be tested, and adjusting a gain compensation coefficient of a sound pickup device according to the third energy intensity to obtain a gain multi-source noise signal; A signal separation module is used to separate the gain multi-source noise signal by using a microphone array to obtain a plurality of separated noise signals, extract key characteristics of the separated noise signals by using spectrum analysis and support vector machine classification algorithm, and classify the plurality of separated noise signals based on the key characteristics to obtain different types of classified noise signals; The abnormal warning module is used to extract features of classified noise signals based on convolutional neural networks, and to perform time series data analysis on classified noise signals in combination with long short-term memory network models to obtain abnormal noise data, and to identify the abnormal noise data through a preset early warning mechanism to obtain the abnormal noise level, and to adopt emergency solutions corresponding to the abnormal noise level; A data storage module, used to store characteristic data of separated noise signals, classified noise signal data, detection data of abnormal noise events and early warning records in a cloud database; The visual display module is used to generate visualization results based on the storage data of the cloud database, and the visualization results include a noise distribution heat map and a time series change map generated by characteristic data.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other via the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images