Sound barrier multi-risk fusion identification method and device based on deep learning

Abstract

The invention provides a sound barrier multi-risk fusion identification method and device based on deep learning, and relates to the field of data processing. The method comprises the following steps: acquiring detection data for target sound barrier risk detection; determining risk assessment data corresponding to the risk factors according to the detection data; carrying out data fusion on the risk assessment data by adopting a deep learning model to obtain risk category probability distribution; and determining whether the target sound barrier has a risk based on the risk category probability distribution. By implementing the technical scheme provided by the invention, the accuracy of sound barrier risk identification can be improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a method and device for multi-risk fusion recognition of sound barriers based on deep learning. Background Art

[0002] With the acceleration of the urbanization process, the problem of road traffic noise pollution has become increasingly serious and has become one of the important factors affecting the quality of residents' lives and public health. To reduce the impact of traffic noise on the surrounding environment, sound barriers are generally installed along urban roads, highways, and railways.

[0003] Currently, the risk detection of traditional sound barriers mainly relies on manual regular inspections. Due to the complex and changeable environment where sound barriers are located, relying solely on single-point measurements or empirical judgments easily leads to insufficient monitoring accuracy, restricting the accuracy of sound barrier risk identification.

[0004] Therefore, there is an urgent need for a method and device for multi-risk fusion recognition of sound barriers based on deep learning. Summary of the Invention

[0005] This application provides a method and device for multi-risk fusion recognition of sound barriers based on deep learning, which is convenient for improving the accuracy of sound barrier risk recognition.

[0006] In the first aspect of this application, a method for multi-risk fusion recognition of sound barriers based on deep learning is provided. The method includes: obtaining detection data for target sound barrier risk detection; determining risk assessment data corresponding to risk factors according to the detection data; using a deep learning model to perform data fusion on the risk assessment data to obtain a risk category probability distribution; and determining whether there is a risk for the target sound barrier based on the risk category probability distribution.

[0007] By adopting the above technical solutions, the detection data is automatically acquired, manual intervention is reduced, and the efficiency and coverage of data collection are improved. Continuous and long-term monitoring can be achieved, avoiding the problem of missing potential risks due to periodic limitations in traditional inspection methods. By calculating the risk assessment data corresponding to risk factors from the detection data, various risk sources such as structural damage (cracks, deformations), abnormal noise attenuation (aging of sound-absorbing materials), and environmental impacts (wind load, humidity) can be covered, forming a multi-dimensional assessment system. Combining different risk factor data helps to identify potential systemic problems rather than single-factor analysis, improving the accuracy of overall safety assessment. Using a deep learning model to perform data fusion on the risk assessment data can fully explore the complex correlations between various risk factors, rather than relying on simple threshold judgments or traditional statistical methods. By learning the patterns in historical data through a neural network, the adaptability to new risks can be enhanced, and the ability to identify unknown or complex risk categories can be improved. By outputting the probability distribution of risk categories through the deep learning model, the decision-making becomes more flexible. Graded early warnings can be made according to the probabilities of different risk categories, such as minor risks, medium risks, and severe risks, thus supporting refined maintenance management strategies. Based on the probability distribution of risk categories, it is possible to quickly determine whether there are risks in the sound barrier, avoiding misjudgment or delay caused by subjective factors during manual inspection. In the event of an emergency (such as structural damage caused by a wind disaster), the system can immediately give an early warning, reducing potential safety hazards and improving road safety. Therefore, it is convenient to improve the accuracy of sound barrier risk identification.

[0008] Optionally, the detection data includes vibration data. The determination of the risk assessment data corresponding to the risk factors according to the detection data specifically includes: determining a vibration signal according to the vibration data; performing continuous wavelet transform on the vibration signal to obtain time-frequency coefficients; performing time-series modeling on the time-frequency coefficients to generate time-series features; and obtaining vibration features based on the time-frequency coefficients and the time-series features, where the risk assessment data includes the vibration features.

[0009] By adopting the above technical solutions, the vibration data can directly reflect the structural health status of the sound barrier, including potential hazards such as cracks, looseness, and deformation. Vibration anomalies are usually early signals of structural damage, which can detect potential problems earlier than visual inspections and improve the early warning ability. Combining with the noise attenuation data, the health status of the sound barrier can be comprehensively analyzed to avoid the one-sidedness brought by a single data source. Continuous wavelet transform can map the vibration signal to the time-frequency domain and effectively capture the time variation of different frequency components. The traditional Fourier transform can only analyze frequency information and loses time information, while wavelet transform can provide both time and frequency resolutions and is suitable for the analysis of non-stationary signals. This method can identify sudden vibration anomalies, such as short-time shocks and local vibration enhancement caused by material fatigue, which helps to detect local structural defects. Since the vibration signal has time correlation, directly analyzing the data at a single moment may lose key information. Therefore, time series modeling is used to learn the change pattern in time. Through time series feature extraction, complex dynamic patterns such as long-term trends, short-term fluctuations, and sudden anomalies can be identified, improving the understanding of the development of structural damage. The extraction of vibration features not only depends on time-frequency coefficients but also combines time series features, avoiding the problem of information loss caused by a single feature. By fusing time-frequency features and time series features, a more comprehensive vibration feature expression is obtained, improving the perception ability of complex structural states. This method can adapt to different types of sound barrier materials and structural types, improving generality.

[0010] Optionally, when performing continuous wavelet transform on the vibration signal to obtain time-frequency coefficients, the following formula is specifically used for calculation: ; where W(a, b) is the wavelet transform coefficient, that is, the time-frequency coefficient, which reflects the local characteristics of the signal at scale a and translation b, W(t) is the vibration time series signal, is the complex conjugate of the wavelet basis function, a is used to control the scale, and b is used to control the translation.

[0011] By adopting the above technical solutions, vibration signals are usually non-stationary signals, that is, their frequency components change with time. Continuous wavelet transform can analyze time and frequency information simultaneously, provide different time resolutions at different scales, and is suitable for capturing sudden events (such as impact vibrations caused by crack propagation and loosening of connectors). By controlling the scale parameter, the global trend can be observed at large scales, and local details can be captured at small scales, improving the understanding of vibration signals. The core feature of wavelet transform is local analysis, that is, it can amplify or compress the local changes of the signal to capture features at different scales. This is especially important for detecting features such as short-term shocks and minute structural changes in vibration signals, helping to discover subtle but critical damage signals and revealing the change patterns of vibration signals at different time scales. These coefficients can be used to train a deep learning model, enabling the model to learn the time-frequency features under different damage patterns (such as crack propagation, material aging, and loosening faults), and improving the accuracy of fault classification and recognition. The translation parameter controls the time position of signal analysis, ensures the integrity of the signal, enables dynamic sliding analysis, and adapts to data in different time periods. This flexible parameter control makes wavelet transform applicable to various types of sound barrier structures and environmental conditions, improving the generalization ability. By combining time-frequency features and time-series features, the model can learn the laws of structural health evolution, thereby enhancing the effects of risk prediction and classification.

[0012] Optionally, the detection data further includes noise data. The determining of the risk assessment data corresponding to the risk factors according to the detection data specifically further includes: calculating a noise level based on the noise data; converting the noise level into a noise signal; extracting local features of the noise signal through a convolutional neural network; calculating attention weights through a self-attention mechanism, and performing weighted summation on each of the local features to obtain noise features, and the risk assessment data includes the noise features.

[0013] By adopting the above technical solution, the noise level is usually a scalar value and is difficult to be directly used for complex pattern analysis. This method converts the noise level into a noise signal (time series data), retains its dynamic characteristics that change over time, and makes the noise feature extraction more accurate. Convolutional neural networks are particularly suitable for processing the local features of noise signals and can automatically extract key features without the need for manual design of feature parameters. The local receptive field characteristic of CNN can effectively capture the noise patterns under different time windows and the anomalies of different frequency components. Traditional noise analysis methods are usually based on Fourier transform, while CNN can directly perform end-to-end learning in the time domain or frequency domain, reducing the dependence on pre-processing feature engineering and improving the flexibility of analysis. The influencing factors of noise signals are complex, and the noise characteristics in different parts and different time periods may have different importance. The self-attention mechanism can automatically learn which noise features are more important, assign different weights to different features, and achieve more accurate feature fusion. It improves the detection ability for progressive damage (such as material aging) and is more sensitive than the method based solely on vibration analysis. It adapts to different environments and structure types, improves the generalization ability, and makes it applicable to different types of sound barrier monitoring tasks.

[0014] Optionally, the noise level is calculated according to the noise data, and the specific calculation formula is as follows: ; where is the noise level at a distance d and service time t, L p0 is the noise level at the reference distance d0, d is the actually measured distance, d0 is the reference distance, α0 is the initial sound absorption coefficient, and λ is the material aging rate, which is used to describe the rate of attenuation of sound absorption performance over time.

[0015] By adopting the above technical solution, this formula dynamically calculates the noise level of the sound barrier at different measurement positions and service times by introducing two major factors of distance attenuation and material aging attenuation. Compared with the traditional fixed noise level calculation method, this formula not only considers the spatial attenuation of noise during propagation, but also combines the degradation effect of material sound absorption performance over time and the aging rate, which is more in line with the change law of sound barrier performance in the real environment. Through this formula, the prediction of the long-term noise attenuation trend can be realized, which can assist in the maintenance and replacement decision-making of the sound barrier and improve the scientificity and accuracy of urban traffic noise control.

[0016] Optionally, the detection data further includes wind load data. The determination of the risk assessment data corresponding to the risk factors according to the detection data specifically further includes: determining the wind pressure per unit area according to the wind load data; processing the wind pressure per unit area by using a multi-layer perceptron to obtain wind load features, and the risk assessment data includes the wind load features; The wind pressure per unit area is specifically calculated by the following formula: ; Among them, P w is the wind pressure per unit area, P air is the air density, v is the wind speed, and C d is the drag coefficient.

[0017] By adopting the above technical solution, the method quantifies the influence of wind load on the sound barrier by calculating the wind pressure per unit area, and introduces a multi-layer perceptron for feature extraction to accurately evaluate the wind load risk. The formula takes into account the air density, wind speed, and drag coefficient, and can dynamically adapt to the wind pressure changes under different meteorological conditions to ensure the physical rationality of the calculation. Compared with the traditional wind load analysis, this method combines the learning ability of neural networks, can automatically mine the wind load characteristics, and improve the evaluation accuracy of the sound barrier stability under complex wind environments (such as sudden strong winds and vortex effects). This method not only improves the prediction ability of the influence of wind load on the structure, but also provides a scientific basis for wind-induced vibration analysis and structural optimization, further enhancing the safety and durability of the sound barrier.

[0018] Optionally, the method uses a deep learning model to perform data fusion on the risk assessment data to obtain a risk category probability distribution, specifically including: inputting the vibration characteristics, the noise characteristics, and the wind load characteristics into the deep learning model; controlling the deep learning model to perform weighted fusion on the vibration characteristics, the noise characteristics, and the wind load characteristics by using a gating mechanism to obtain a fused feature; using the Transformer algorithm to perform global modeling on the fused feature to obtain a target feature; using the fully connected layer and softmax activation of the deep learning model to perform risk level classification on the target feature to generate the risk category probability distribution.

[0019] By adopting the above technical solution, the method performs fusion analysis on the vibration, noise, and wind load characteristics through a deep learning model, uses a gating mechanism to adaptively allocate feature weights, ensures that the contribution degrees of different risk factors are dynamically adjusted, and avoids information redundancy or distortion. Subsequently, the Transformer algorithm performs global modeling to capture long-distance dependence relationships, improves the expression ability of risk patterns, and enables the model to accurately identify the interaction effects among multiple complex risk factors. Finally, risk level classification is realized through the fully connected layer and Softmax activation to generate a risk category probability distribution, providing a transparent and quantifiable risk assessment result. This method not only improves the intelligent level of sound barrier risk prediction, but also has the advantages of cross-modal feature fusion, high-precision risk discrimination, and strong adaptability, can significantly enhance the early warning ability of structural safety hazards, and provide a scientific decision-making basis for sound barrier maintenance and optimization.

[0020] In a second aspect of the present application, a multi-risk fusion recognition device for a sound barrier based on deep learning is provided, characterized in that the device includes an acquisition module and a processing module, wherein the acquisition module is used to acquire detection data for detecting target sound barrier risks; the processing module is used to determine risk assessment data corresponding to risk factors according to the detection data; the processing module is further used to perform data fusion on the risk assessment data by using a deep learning model to obtain a risk category probability distribution; the processing module is further used to determine whether there is a risk for the target sound barrier based on the risk category probability distribution.

[0021] In a third aspect of the present application, an electronic device is provided. The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. Both the user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory so that the electronic device executes the method described above.

[0022] In a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores instructions, and when the instructions are executed, the method described above is executed.

[0023] In summary, one or more technical solutions provided in the present application have at least the following technical effects or advantages: By automatically acquiring detection data, manual intervention is reduced, and the efficiency and coverage of data collection are improved. It can achieve continuous and long-term monitoring, avoiding the problem of missing potential risks due to periodic limitations in traditional inspection methods. By calculating risk assessment data corresponding to risk factors from the detection data, various risk sources such as structural damage (cracks, deformation), abnormal noise attenuation (aging of sound-absorbing materials), and environmental impacts (wind load, humidity) can be covered, forming a multi-dimensional assessment system. Combining data of different risk factors helps to identify potential systemic problems rather than single-factor analysis, enhancing the accuracy of overall safety assessment. Using a deep learning model to perform data fusion on the risk assessment data can fully explore the complex correlations between various risk factors, rather than relying on simple threshold judgments or traditional statistical methods. By learning patterns in historical data through a neural network, the adaptability to new risks can be enhanced, and the ability to identify unknown or complex risk categories can be improved. By outputting the probability distribution of risk categories through the deep learning model, decision-making becomes more flexible. Graded warnings can be issued according to the probabilities of different risk categories, such as minor risk, medium risk, and severe risk, thus supporting refined maintenance management strategies. Based on the probability distribution of risk categories, it is possible to quickly determine whether there is a risk with the sound barrier, avoiding misjudgment or delay caused by subjective factors during manual inspection. In the event of an emergency (such as structural damage caused by a wind disaster), the system can immediately give an early warning, reducing potential safety hazards and enhancing road safety. Therefore, it is convenient to improve the accuracy of sound barrier risk identification. Description of the Drawings

[0024] Figure 1 It is a schematic flowchart of a method for multi-risk fusion identification of a sound barrier based on deep learning provided by an embodiment of the present application; Figure 2 It is another schematic flowchart of a method for multi-risk fusion identification of a sound barrier based on deep learning provided by an embodiment of the present application; Figure 3 It is a schematic block diagram of a device for multi-risk fusion identification of a sound barrier based on deep learning provided by an embodiment of the present application; Figure 4 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0025] Description of the reference numerals: 31, acquisition module; 32, processing module; 41, processor; 42, communication bus; 43, user interface; 44, network interface; 45, memory. Detailed Embodiments

[0026] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments.

[0027] In the description of the embodiments of this application, words such as "for example" or "for instance" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "for example" or "for instance" is intended to present relevant concepts in a specific manner.

[0028] In the description of the embodiments of this application, the meaning of the term "a plurality of" refers to two or more. For example, a plurality of systems refers to two or more systems, and a plurality of screen terminals refers to two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the technical features indicated. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "include", "comprise", "have" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0029] With the acceleration of the urbanization process, the problem of road traffic noise pollution has become increasingly serious, becoming one of the key factors affecting the quality of residents' lives and public health. In order to reduce the negative impact of traffic noise on the surrounding environment, sound barriers are generally set along urban roads, highways and railways.

[0030] However, the current risk detection of traditional sound barriers relies on manual regular inspections. Due to the complex and ever-changing environmental conditions where sound barriers are located, single-point measurement or relying on experience judgment is likely to lead to low monitoring accuracy, thus affecting the accuracy of sound barrier risk identification.

[0031] To solve the above technical problems, this application provides a method for multi-risk fusion identification of sound barriers based on deep learning. Refer to Figure 1 , Figure 1 is a schematic flow chart of a method for multi-risk fusion identification of sound barriers based on deep learning provided by an embodiment of this application. This method is applied to a server and includes steps S110 to S140. The above steps are as follows: S110, Obtain detection data for target sound barrier risk detection.

[0032] Specifically, in the modern Internet of Things environment, the server is the core of data storage, processing, and management. It is responsible for receiving, storing, processing, and analyzing data from various sensors and detection devices. Therefore, in the context of sound barrier risk detection, the server typically takes charge of centrally storing real-time or historical data collected from various types of sensors (such as vibration sensors, noise sensors, wind speed sensors, etc.). The target sound barrier refers to a specific sound barrier that needs to be monitored and risk-evaluated. In practical applications, these sound barriers may be facilities located along urban roads, highways, or railways, aiming to reduce the impact of traffic noise on the surrounding environment. The detection data refers to the multi-dimensional data related to the sound barrier collected by various devices and sensors. These data may include, but are not limited to: Vibration data: reflecting the vibration of the sound barrier caused by external forces (such as wind load, traffic vibration, etc.). Noise data: including the noise absorption effect of the sound barrier, the real-time change of the noise level, and any abnormal noise patterns. Wind load data: recording data such as wind speed and wind pressure, especially when the wind force is relatively large, which may have a negative impact on the sound barrier. Structural health monitoring data: such as physical damage data like cracks and deformations, indicating the structural integrity of the sound barrier. This process includes data acquisition, transmission, and storage. The server receives data from different types of sensors or measurement devices and usually transmits the data to the server through wireless communication protocols (such as Wi-Fi, LoRa, Zigbee, etc.). The purpose of obtaining data is to provide the necessary input for subsequent analysis and evaluation, ensuring the timeliness and accuracy of the data.

[0033] For example: Suppose a variety of sensors, such as vibration sensors, noise sensors, and wind speed sensors, are installed on a sound barrier beside a highway. The role of the server is to receive the data from these sensors and store it for subsequent processing. The specific process may be as follows: Vibration sensors: Installed at different positions of the sound barrier to monitor vibrations caused by factors such as traffic, wind, or weather changes. These sensors continuously record vibration signals and send them to the server. The server extracts valuable vibration characteristics from these data to assist in analyzing the structural health of the sound barrier. Noise sensors: These sensors record the noise isolation effect of the sound barrier. For example, the noise sensors measure the noise level differences at different positions in front of and behind the sound barrier. After the server obtains these data, it can analyze whether the sound absorption ability of the sound barrier has decreased, such as due to material aging or damage, resulting in weakened sound absorption performance. Wind speed sensors: The wind speed sensors monitor the changes in wind force, especially the impact on the sound barrier during strong winds or storms. The server combines these data with vibration data, noise data, etc. to help evaluate the potential risks of wind load on the sound barrier.

[0034] S120. Determine the risk assessment data corresponding to the risk factors according to the detection data.

[0035] Specifically, the server is responsible for collecting data from various sensors (such as vibration, noise, wind load, etc.) and analyzing and processing this data. The data itself is just raw information and cannot be directly used for decision-making or evaluation. Therefore, the server uses this raw detection data to generate risk assessment data that is more meaningful for decision-making, so as to provide a basis for subsequent risk judgment and decision-making. In the monitoring of sound barriers, there are various possible risk factors, such as structural damage (cracks, deformations), noise attenuation, wind load, etc. These risk factors affect the function and safety of the sound barrier. The server extracts indicators that can represent these risk factors from the analyzed detection data to form "risk assessment data". These data provide the "health status" of the current state of the sound barrier and help evaluate whether there are potential risks.

[0036] The server first processes the raw data, such as cleaning, denoising, standardization, etc., to ensure the accuracy of the data. Key features are extracted from the processed data to represent the current state of the sound barrier. For example, features such as vibration amplitude and frequency are extracted from the vibration signal; the trend of noise level change is extracted from the noise data. According to the preset standards or models, the extracted features are compared with historical data and known risk patterns to obtain risk assessment data. These data can be used to judge whether there is a risk of damage or performance degradation of the sound barrier.

[0037] In a possible implementation manner, the detection data includes vibration data. According to the detection data, the risk assessment data corresponding to the risk factors is determined, specifically including: determining the vibration signal according to the vibration data; performing continuous wavelet transform on the vibration signal to obtain time-frequency coefficients; performing time-series modeling on the time-frequency coefficients to generate time-series features; based on the time-frequency coefficients and time-series features, obtaining vibration features, and the risk assessment data includes the vibration features.

[0038] Specifically, vibration data refers to the vibration information generated by a sound barrier under different environmental conditions due to external forces (such as traffic, wind, or earthquakes). The vibration data collected by sensors (such as accelerometers, displacement sensors, etc.) can reflect the dynamic response of the sound barrier. These vibration data are very important for evaluating the health status of the sound barrier because excessive vibration of the structure may lead to damage or functional decline. Vibration data is essentially time series data, that is, the variation of vibration over a certain period of time. Based on these original vibration data, it is first necessary to determine the vibration signals therein, which may include extracting features such as the amplitude and frequency of the vibration. The vibration signal represents the movement or vibration condition of the sound barrier during a specific period. Wavelet transform is a mathematical tool used to decompose a signal into different frequency components while retaining time information. Different from the traditional Fourier transform, wavelet transform can not only capture the frequency domain characteristics of the signal but also retain the time domain information. Through continuous wavelet transform, the vibration signal can be transformed into time-frequency coefficients, that is, representing the frequency characteristics of the signal at different time scales. In this way, the frequency characteristics of the vibration signal are no longer static but dynamic characteristics closely combined with time. The obtained time-frequency coefficients contain multi-level information of the vibration signal, including the frequency distribution of the signal and the dynamic characteristics changing with time. Through time series modeling (usually using recurrent neural networks such as LSTM, etc.), these time-frequency coefficients can be modeled to capture the law of change of the vibration signal over time. The goal of this process is to extract time series features, that is, to describe the evolution pattern of the vibration signal in the time series, which helps to identify possible abnormal fluctuations or risk factors.

[0039] Combining the time-frequency coefficients and the time series features obtained from time series modeling, vibration features are finally extracted. These features include important information of the vibration signal in terms of time and frequency, such as the amplitude of the vibration, the frequency pattern, the fluctuation trend, etc. Through these vibration features, the system can determine whether there are structural problems (such as cracks, deformations, etc.) in the sound barrier or whether there are potential risks due to excessive vibration. These vibration features, as part of the risk assessment data, reflect the current health status of the sound barrier. If the vibration features are abnormal (such as an increase in amplitude, a frequency shift, etc.), it may mean that the sound barrier is facing a load beyond the designed tolerance, with a risk of structural damage or functional decline.

[0040] In a possible implementation, the vibration signal is subjected to continuous wavelet transform to obtain time-frequency coefficients, which are specifically calculated using the following formula: ; where W(a, b) is the wavelet transform coefficient, that is, the time-frequency coefficient, reflecting the local characteristics of the signal at scale a and translation b, W(t) is the vibration time series signal, is the complex conjugate of the wavelet basis function, a is used to control the scale, and b is used to control the translation.

[0041] Specifically, vibration signals are usually represented as time - series data, that is, the intensity of vibration changes over time. This signal reflects the dynamic response of the sound barrier and may contain high - frequency components (such as short - time impacts) and low - frequency components (such as long - term continuous vibrations). It may be difficult to obtain the changes of vibration signals in different frequency ranges by directly analyzing the time - series data. Therefore, wavelet transform is needed to obtain more comprehensive information. Wavelet transform coefficients are also known as time - frequency coefficients. Wavelet transform converts the vibration signal into a new representation, revealing the characteristics of the signal at different time scales (a) and time positions (b). The time - frequency coefficients reflect the local properties of the signal under scale (a) and time shift (b). In other words, wavelet transform can not only describe the frequency components of the signal (similar to Fourier transform), but also reveal how the frequency components change over time. Therefore, the time - frequency coefficients can provide both time information and frequency information. W(t) is a part of the vibration time - series signal, which usually refers to the complex conjugate of the signal (i.e., the complex - conjugate form of the signal). Complex conjugate is a common operation when dealing with complex signals, which can help obtain more accurate amplitude and phase information during signal analysis. In the context of wavelet transform, complex conjugate is used to calculate the inner product between the wavelet function and the signal to determine the characteristics of the signal at a specific scale and time position.

[0042] The scale controls the width of the wavelet basis function. The larger the scale, the wider the time window of the basis function, which can capture lower - frequency signal components; the smaller the scale, the smaller the time window of the basis function, which can capture higher - frequency signal components. Therefore, the scale (a) determines the frequency range of wavelet transform analysis, with large scales corresponding to low frequencies and small scales corresponding to high frequencies. Translation controls the position of the wavelet basis function on the time axis. By adjusting the translation parameter b, the wavelet function can analyze the signal at different time points, thus revealing the change characteristics of the signal in different time periods. Wavelet transform performs a "convolution" operation on the signal through the wavelet basis function to extract information of different frequency components. Complex conjugate is used here to ensure the accuracy of the transform, especially when dealing with complex signals, it helps maintain the consistency of frequency information and phase information.

[0043] In a possible implementation, the detection data also includes noise data. According to the detection data, risk assessment data corresponding to risk factors is determined, which specifically further includes: calculating the noise level based on the noise data; converting the noise level into a noise signal; extracting local features of the noise signal through a convolutional neural network; calculating attention weights through a self - attention mechanism, and performing weighted summation on each local feature to obtain a noise feature. The risk assessment data includes the noise feature.

[0044] Specifically, first, the server collects noise data related to the sound barrier. This data may come from noise sensors around the sound barrier and reflects the noise intensity in the area where the sound barrier is located. The noise level is a standardized measure of noise intensity, usually expressed in decibels. Converting the noise level to a noise signal means converting the calculation result of the noise intensity into a signal that can be used for further analysis. Here, the "noise signal" is a time series signal representing the noise changes at different time points. The noise signal can include various frequency components (low frequency, broadband noise, etc.). The noise signal is high-dimensional and may contain many complex fluctuations and patterns. To extract useful information from it, a convolutional neural network can be used. A convolutional neural network is a deep learning model commonly used for feature extraction of images and time series data. The CNN extracts local features from the noise signal by applying convolutional operations in a sliding window manner, such as the instantaneous changes of the noise signal, the fluctuations of frequency components, and the local periodic features in the signal. Local features usually represent the short-term change patterns of the signal, which are crucial for capturing abnormal noise behaviors.

[0045] The self-attention mechanism is a mechanism that can automatically assign weights to different features and is commonly found in the Transformer model in deep learning. Its role is to dynamically adjust the influence of each feature according to the importance of different features. In the processing of the noise signal, using the self-attention mechanism can calculate the importance of each local feature, assign higher weights to important features, and ignore or reduce less important features. The self-attention mechanism can help the model focus on the noise features that are most important for risk assessment. After the attention weights calculated by the self-attention mechanism are combined with the local features, a weighted sum is performed. This means that each feature will be assigned a weight according to its importance, and finally a comprehensive noise feature is obtained. The noise feature is a high-level representation processed by a deep learning model and can reflect the overall characteristics of the noise signal and the potential risks it may bring. The obtained noise feature will be used as the input data for subsequent risk assessment to help determine the health status and potential risks of the sound barrier. Finally, the noise features processed through the above steps will become part of the risk assessment data. These features will be input into a further risk assessment model to help determine whether there are potential risks in the sound barrier.

[0046] In a possible implementation, according to the noise data, the noise level is calculated, and the specific calculation formula is as follows: ; where is the noise level at a distance d and a service time of t, L p0The noise level at the reference distance d0, d is the actually measured distance, d0 is the reference distance, α0 is the initial sound absorption coefficient, and λ is the material aging rate, which is used to describe the rate of attenuation of the sound absorption performance over time.

[0047] Specifically, the noise level at a distance d and a service time of t describes the noise intensity of the sound barrier or noise source and takes into account the effects of distance and time on the noise level. This is usually a standard value obtained through experiments or measurements, representing the noise level under standard conditions. d is the distance at which the noise is actually measured. Usually, the noise level decreases with increasing distance, so the farther the distance, the lower the noise level generally. The reference distance refers to the standard reference distance used in the measurement. Usually, this reference distance is measured in a relatively fixed environment, such as 1 meter away from the sound source. The initial sound absorption coefficient describes the ability of the sound barrier material or the environment to absorb sound waves. The larger the sound absorption coefficient, the easier the noise is absorbed and the relatively lower the noise intensity. This coefficient is usually obtained through the experimental characteristics of the material. The material aging rate is used to describe the sound absorption performance of the material that decays over time. When the time t increases, the sound absorption ability of the material may decrease, resulting in more obvious noise propagation.

[0048] In a possible implementation, the detection data further includes wind load data. Based on the detection data, risk assessment data corresponding to risk factors is determined, which specifically further includes: determining the wind pressure per unit area according to the wind load data; processing the wind pressure per unit area using a multi-layer perceptron to obtain wind load characteristics, and the risk assessment data includes the wind load characteristics; The wind pressure per unit area is specifically calculated using the following formula: ; where P w is the wind pressure per unit area, P air is the air density, v is the wind speed, and C d is the drag coefficient.

[0049] Specifically, the wind pressure per unit area represents the pressure exerted by the wind on a unit area, and the unit is usually Pascal. The greater the wind pressure, the greater the impact of the wind force on the object. The air density, the unit is usually kilograms per cubic meter (kg / m³). The air density varies with temperature, humidity, and air pressure, and usually takes the standard air density at normal temperature, about 1.225 kg / m³. The wind speed, the unit is meters per second (m / s). The wind speed is a key factor affecting the wind pressure. The greater the wind speed, the greater the wind pressure. The wind speed is directly obtained through meteorological measurement tools or can be estimated through historical data. The drag coefficient represents the degree of obstruction of the object to the wind. This coefficient is determined according to factors such as the shape of the object, the smoothness of the surface, and the air flow conditions. For smooth objects, the drag coefficient is small; for rough or obstructed objects, the drag coefficient is large.

[0050] According to the above formula, a value is obtained for the calculation of the wind pressure per unit area, and this value reflects the impact of the wind force on the object. And for further risk assessment, this data needs to be processed and analyzed to extract relevant wind load characteristics. It is mentioned in this passage that a multi-layer perceptron is used to process the wind pressure per unit area. The multi-layer perceptron is a common neural network architecture with multiple hidden layers that can automatically learn features from the input data. Through training, the MLP can capture the complex non-linear relationships in the wind load data, thus providing more accurate wind load characteristics for risk assessment. After being processed by the MLP, these characteristics will be used as risk assessment data for subsequent risk classification or analysis. The wind load characteristics may include the change trend of the wind pressure, the fluctuation characteristics of the wind speed, etc. These characteristics can help the model judge whether there are abnormal wind force impacts, and then predict the potential risks of the wind force on the sound barrier or other structures.

[0051] S130. Use a deep learning model to perform data fusion on the risk assessment data to obtain the probability distribution of risk categories.

[0052] Specifically, in this step, the deep learning model is a tool for processing input data and performing feature learning. Deep learning, especially neural networks, can capture complex non-linear relationships between input data through multiple levels of structure. In this scenario, the deep learning model will be used to "fuse" risk assessment data from different sources to generate a comprehensive risk category probability distribution. "Data fusion" means combining different types of data (such as vibration characteristics, noise characteristics, wind load characteristics, etc.) to generate a comprehensive feature set to help the model more accurately judge and predict the risks of the target. These data may come from different sensors or measurement methods, and each type of data reflects different types of risk factors. Vibration characteristics come from the analysis of vibration signals, such as vibration frequency, amplitude and other characteristics, which can reveal whether physical damage has occurred to an object or structure. Noise characteristics come from the analysis of noise data, such as noise level, noise signal pattern and other characteristics, which can reveal whether there are abnormalities in the environment or structure. Characteristics such as wind speed and wind pressure in wind load characteristics indicate whether the wind load in the environment may affect the structure. The "risk category probability distribution" is the result output by the deep learning model, which represents the system's prediction of different risk levels. Usually, the output is a vector, where each element corresponds to the probability of a specific risk category. For example, assume there are three risk categories: no risk (0), low risk (1), high risk (2), then the output probability distribution may be [0.7, 0.2, 0.1], which means the system predicts the probability of the object having no risk is 70%, the probability of low risk is 20%, and the probability of high risk is 10%.

[0053] For example, assume that a sound barrier in a city is being monitored, and the goal is to determine whether there are potential structural risks by monitoring vibration, noise and wind load. Vibration data is the data collected by vibration sensors, analyzing the vibration mode of the sound barrier to determine whether there are structural cracks or deformations. Noise data is the sound intensity and frequency data collected by noise sensors, evaluating whether the noise attenuation effect is normal and whether there is aging of the sound absorption material. Wind load data is the wind speed and wind pressure data collected by meteorological equipment, evaluating whether the wind load may affect the structural stability. Vibration data obtains time-frequency coefficients through continuous wavelet transform and generates vibration characteristics by combining time series modeling. Noise data is transformed according to the noise level, and local characteristics of the noise signal are extracted by combining a convolutional neural network to further obtain noise characteristics. Wind load data calculates the wind pressure per unit area and uses a deep learning model to extract wind load characteristics.

[0054] All risk assessment data (vibration characteristics, noise characteristics, wind load characteristics) are input into a deep learning model. This model could be a multi-layer neural network or other types of deep learning architectures such as the Transformer model, and the specific structure depends on the requirements of the task and the characteristics of the data. In the deep learning model, the system automatically identifies which features are most important for risk judgment in different situations by learning the complex relationships between these input features. For example, the model may find that in the case of strong winds, the wind load characteristics are more predictive than the vibration characteristics, and in the case of material aging, the noise characteristics may be more crucial. After multiple levels of learning and calculation, the model outputs a probability distribution of risk categories. Suppose we have three risk categories: no risk, low risk, and high risk. The output of the model may be a vector containing the probabilities of these categories, for example: [0.1, 0.3, 0.6]. This means that the probability that the system judges the risk of the target sound barrier to be high risk is 60%, the probability of low risk is 30%, and the probability of no risk is 10%.

[0055] Therefore, the deep learning model effectively fuses data from different types of sensors, thus providing a more accurate judgment for the risk assessment of the target sound barrier. Data fusion enables the model to process different types of data at multiple levels, taking into account the complex interaction relationships of different risk factors, and finally providing a comprehensive risk assessment result in the form of a probability distribution of risk categories. Such an approach can reduce the subjectivity and limitations of manual inspections and improve the efficiency and accuracy of sound barrier risk monitoring.

[0056] In a possible implementation, referring to Figure 2 , Figure 2 is another process schematic diagram of a method for multi-risk fusion recognition of a sound barrier based on deep learning provided by an embodiment of this application. Specifically, it includes steps S210 to S240, and the above steps are as follows: S210, input the vibration characteristics, noise characteristics, and wind load characteristics into the deep learning model; S220, control the deep learning model to perform weighted fusion on the vibration characteristics, noise characteristics, and wind load characteristics using a gating mechanism to obtain fused features; S230, use the Transformer algorithm to perform global modeling on the fused features to obtain target features; S240, use the fully connected layer and softmax activation of the deep learning model to classify the risk levels of the target features and generate a probability distribution of risk categories.

[0057] Specifically, the gating mechanism is a mechanism that controls how the model combines different input features. Similar to the gating mechanism in "gated neural networks", its role is to weightedly fuse different features according to the importance of the input features. Specifically, the gating mechanism assigns a weight to each feature, determining its influence degree in the final fused feature. Different features (vibration, noise, wind load) may have different impacts on risk assessment, and the gating mechanism allows the model to adaptively adjust the weight of each feature. Suppose at a certain moment, the wind load has a greater impact on the structure of the sound barrier, while the change in the vibration feature is smaller. The gating mechanism will automatically reduce the influence of the vibration feature and increase the influence of the wind load feature, making the final fused feature better reflect the current risk state. The Transformer algorithm is a deep learning model based on the self-attention mechanism, widely used in natural language processing tasks. In this application, the Transformer is used to perform "global modeling" on the fused features, that is, to consider the relationships between all features, rather than just local feature interactions. The Transformer can capture the relationships of long-distance dependencies in the data, so it is suitable for tasks that require extracting global information from multiple features.

[0058] For example, after fusing the vibration, noise, and wind load features, the Transformer model will consider the relationships between these features. For example, there may be a certain connection between the wind load and noise features (such as when the wind speed is large, the noise attenuation ability will decrease), and the Transformer can understand these relationships through the self-attention mechanism. The "fully connected layer" is a common structure in deep neural networks, with connections between all input nodes and the nodes in the next layer. Through the fully connected layer, the model can synthesize the previously extracted features to form a high-dimensional feature representation. The "Softmax activation" function is used to convert these high-dimensional features into a probability distribution, enabling the output result to be used for classification tasks. Through the Softmax function, the model will output the probabilities of each possible risk category. Risk level classification usually includes multiple categories, such as: no risk, low risk, medium risk, high risk.

[0059] For example, through a fully connected layer and Softmax activation, the model may output the following probability distribution for risk categories: no risk: 0.05, low risk: 0.15, medium risk: 0.45, high risk: 0.35. This means that the system believes the probability that the sound barrier is in a medium-risk state is 45%, the probability of high risk is 35%, and the probabilities of low risk and no risk are relatively small. Through the above steps, the deep learning model finally outputs a probability distribution, reflecting the system's prediction of the risk of the target object (sound barrier). This is the final output after the model's comprehensive analysis of multiple risk factors, which can give the probability values for each risk category and help decision-makers determine whether the sound barrier needs repair or maintenance. Suppose the probability distribution of the risk categories output by the system is [0.1, 0.3, 0.5, 0.1], corresponding to four risk levels: no risk, low risk, medium risk, and high risk. This distribution means that the system believes the probability that the sound barrier is in the "medium-risk" state is 50%, the low risk is 30%, and the high risk and no risk are each 10%.

[0060] By using a deep learning model (especially adopting a gating mechanism and Transformer algorithm) to fuse and model features such as vibration, noise, and wind load. Through a fully connected layer and Softmax activation, the model can generate a risk assessment result for the sound barrier, that is, the probability distribution for each risk category. Compared with traditional manual judgment methods, this method can integrate multiple risk factors, provide a more accurate and comprehensive risk assessment, and has higher efficiency and better accuracy.

[0061] S140. Determine whether there is a risk for the target sound barrier based on the probability distribution of risk categories.

[0062] Specifically, first, the server will receive a "probability distribution of risk categories" generated by the deep learning model. This means that the model has calculated the probabilities of the sound barrier at different risk levels based on data from multiple sensors (such as vibration, noise, wind load). For example, the model may generate a probability distribution containing four risk levels (no risk, low risk, medium risk, high risk), reflecting the prediction confidence for each risk level.

[0063] Example: Suppose the probability distribution of the risk categories given by the model is as follows: no risk: 0.1, low risk: 0.2, medium risk: 0.5, high risk: 0.2. These probabilities reflect the confidence level of the model in different risk states of the target sound barrier. The server needs to determine whether the sound barrier is at risk based on the results of the probability distribution. This decision usually involves setting a threshold and making a judgment based on the probability values of the risk categories. If the probability value of a certain risk level exceeds the set threshold, the server will determine that the target sound barrier is in that risk state. Specifically, a risk judgment criterion is usually set. For example, if the probability values of medium risk or high risk exceed a certain threshold (such as 0.4), the server will determine that the sound barrier is at risk and further measures need to be taken (such as maintenance, enhanced monitoring, etc.). If the probability values of low risk or no risk dominate, the server will determine that the risk of the sound barrier is low and continuous monitoring can be maintained.

[0064] For example, according to the above probability distribution of risk categories (no risk 0.1, low risk 0.2, medium risk 0.5, high risk 0.2), assume the set judgment criterion is that if the probability values of medium risk and high risk exceed 0.4, it is considered at risk. Then, in this example, the probability of medium risk (0.5) exceeds the set threshold (0.4). Therefore, the server will determine that the sound barrier is at risk. After determining whether the sound barrier is at risk, the server can perform corresponding actions based on this decision. For example: at risk: the server will trigger an alarm to remind relevant personnel to conduct on-site inspections, increase monitoring efforts, or initiate repair procedures. No risk or low risk: the server will continue to monitor the sound barrier but will not trigger an emergency response for the time being. Suppose the server determines that a certain section of the sound barrier is "medium risk", then the server can automatically notify the maintenance personnel to conduct a detailed inspection of this section of the sound barrier to check for cracks, aging, or other structural problems, or further optimize the maintenance plan based on historical data analysis.

[0065] Therefore, the server analyzes the generated probability distribution of risk categories and determines whether further inspection or repair of the sound barrier is required based on the set threshold. This process can achieve automated risk monitoring and management, avoiding reliance on manual judgment, thereby improving the accuracy of risk assessment and the efficiency of response.

[0066] This application also provides a multi-risk fusion recognition device for sound barriers based on deep learning. Refer to Figure 3 , Figure 3This is a schematic diagram of the modules of a sound barrier multi-risk fusion recognition device based on deep learning provided by an embodiment of the present application. The device is a server, and the server includes an acquisition module 31 and a processing module 32. Among them, the acquisition module 31 acquires detection data for target sound barrier risk detection; the processing module 32 determines risk assessment data corresponding to risk factors according to the detection data; the processing module 32 performs data fusion on the risk assessment data using a deep learning model to obtain a risk category probability distribution; the processing module 32 determines whether there is a risk for the target sound barrier based on the risk category probability distribution.

[0067] In a possible implementation manner, the detection data includes vibration data. The processing module 32 determines risk assessment data corresponding to risk factors according to the detection data, specifically including: the processing module 32 determines a vibration signal according to the vibration data; the processing module 32 performs continuous wavelet transform on the vibration signal to obtain time-frequency coefficients; the processing module 32 performs time series modeling on the time-frequency coefficients to generate time series features; the processing module 32 obtains vibration features based on the time-frequency coefficients and the time series features, and the risk assessment data includes the vibration features.

[0068] In a possible implementation manner, when the processing module 32 performs continuous wavelet transform on the vibration signal to obtain time-frequency coefficients, the following formula is specifically used for calculation: ; where W(a, b) is the wavelet transform coefficient, that is, the time-frequency coefficient, which reflects the local characteristics of the signal at scale a and translation b, W(t) is the vibration time series signal, is the complex conjugate of the wavelet basis function, a is used to control the scale, and b is used to control the translation.

[0069] In a possible implementation manner, the detection data further includes noise data. According to the detection data, determining risk assessment data corresponding to risk factors specifically further includes: the processing module 32 calculates a noise level according to the noise data; the processing module 32 converts the noise level into a noise signal; the processing module 32 extracts local features of the noise signal through a convolutional neural network; the processing module 32 calculates attention weights through a self-attention mechanism and performs weighted summation on each local feature to obtain noise features, and the risk assessment data includes the noise features.

[0070] In a possible implementation manner, when the processing module 32 calculates the noise level according to the noise data, the following formula is specifically used for calculation: ; where is the noise level at a distance d and a service time of t, L p0The noise level at the reference distance d0, d is the actually measured distance, d0 is the reference distance, α0 is the initial sound absorption coefficient, and λ is the material aging rate, which is used to describe the rate of attenuation of the sound absorption performance over time.

[0071] In a possible implementation, the detection data further includes wind load data. The processing module 32 determines the risk assessment data corresponding to the risk factors according to the detection data, specifically including: the processing module 32 determines the wind pressure per unit area according to the wind load data; the processing module 32 processes the wind pressure per unit area using a multi-layer perceptron to obtain wind load characteristics, and the risk assessment data includes wind load characteristics; The wind pressure per unit area is specifically calculated using the following formula: ; where, P w is the wind pressure per unit area, P air is the air density, v is the wind speed, and C d is the drag coefficient.

[0072] In a possible implementation, the processing module 32 performs data fusion on the risk assessment data using a deep learning model to obtain the risk category probability distribution, specifically including: the processing module 32 inputs the vibration characteristics, noise characteristics, and wind load characteristics into the deep learning model; the processing module 32 controls the deep learning model to perform weighted fusion on the vibration characteristics, noise characteristics, and wind load characteristics using a gating mechanism to obtain the fused characteristics; the processing module 32 performs global modeling on the fused characteristics using the Transformer algorithm to obtain the target characteristics; the processing module 32 uses the fully connected layer and softmax activation of the deep learning model to classify the risk level of the target characteristics and generate the risk category probability distribution.

[0073] It should be noted that: when the device provided in the above embodiment realizes its functions, only the above-mentioned division of each functional module is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be elaborated here.

[0074] This application also provides an electronic device. Refer to Figure 4 , Figure 4 which is a schematic structural diagram of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 41, at least one network interface 44, a user interface 43, a memory 45, and at least one communication bus 42.

[0075] Among them, the communication bus 42 is used to realize the connection and communication between these components.

[0076] Among them, the user interface 43 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 43 may further include a standard wired interface and a wireless interface.

[0077] Among them, the network interface 44 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface).

[0078] Among them, the processor 41 may include one or more processing cores. The processor 41 connects various parts within the entire server through various interfaces and lines, and executes various functions of the server and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 45, and by calling data stored in the memory 45. Optionally, the processor 41 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 41 may integrate a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor 41 and may be implemented separately by a single chip.

[0079] Among them, the memory 45 may include a Random Access Memory (RAM), or may include a Read-Only Memory. Optionally, the memory 45 includes a non-transitory computer-readable storage medium. The memory 45 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 45 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above method embodiments, etc.; the data storage area may store the data involved in the above method embodiments. Optionally, the memory 45 may also be at least one storage device located far from the aforementioned processor 41. As Figure 4 shown, in the memory 45 as a computer storage medium, there may be included an operating system, a network communication module, a user interface module, and an application program of a method for multi-risk fusion recognition of a sound barrier based on deep learning.

[0080] In Figure 4 the electronic device shown, the user interface 43 is mainly used to provide an input interface for the user to obtain the data input by the user; and the processor 41 can be used to call the application program of a method for multi-risk fusion recognition of a sound barrier based on deep learning stored in the memory 45. When executed by one or more processors, the electronic device is caused to execute the method as described in one or more of the above embodiments.

[0081] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be adopted in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0082] This application also provides a computer-readable storage medium, and the computer-readable storage medium stores instructions. When executed by one or more processors, the electronic device is caused to execute the method as described in one or more of the above embodiments.

[0083] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0084] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0085] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0086] In addition, each functional unit in various embodiments of the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0087] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present application. And the aforementioned memory includes: various media such as USB flash drives, mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0088] The above are only exemplary embodiments of the present disclosure and should not be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. Those skilled in the art will easily think of other implementation schemes of the present disclosure after considering the specification and the disclosure of the practical truth. The present application aims to cover any variations, uses, or adaptive changes of the present disclosure. These variations, uses, or adaptive changes follow the general principles of the present disclosure and include common general knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for multi-risk fusion recognition of sound barriers based on deep learning, characterized in that, The method includes: Obtaining detection data for target sound barrier risk detection; Determining risk assessment data corresponding to risk factors according to the detection data; Using a deep learning model to perform data fusion on the risk assessment data to obtain a risk category probability distribution; Based on the risk category probability distribution, determining whether there is a risk for the target sound barrier.

2. The method for identifying multiple risks of a sound barrier based on deep learning according to claim 1, wherein The detection data includes vibration data. The determining of the risk assessment data corresponding to the risk factors according to the detection data specifically includes: Determining a vibration signal according to the vibration data; Performing continuous wavelet transform on the vibration signal to obtain time-frequency coefficients; Performing time series modeling on the time-frequency coefficients to generate time series features; Based on the time-frequency coefficients and the time series features, obtaining vibration features, and the risk assessment data includes the vibration features.

3. The method for identifying multiple risks of a sound barrier based on deep learning according to claim 2, wherein The performing of continuous wavelet transform on the vibration signal to obtain time-frequency coefficients is specifically calculated using the following formula: ； Among them, W(a, b) is the wavelet transform coefficient, that is, the time-frequency coefficient, which reflects the local characteristics of the signal at scale a and translation b. W(t) is the vibration time series signal. is the complex conjugate of the wavelet basis function. a is used to control the scale, and b is used to control the translation.

4. The method for identifying multiple risks of a sound barrier based on deep learning according to claim 2, characterized in that, The detection data further includes noise data. The determining of the risk assessment data corresponding to the risk factors according to the detection data specifically further includes: Calculating a noise level according to the noise data; Converting the noise level into a noise signal; Extracting local features of the noise signal through a convolutional neural network; Calculating attention weights through a self-attention mechanism, and performing weighted summation on each of the local features to obtain noise features, and the risk assessment data includes the noise features.

5. The method for multi-risk fusion recognition of a sound barrier based on deep learning according to claim 4, wherein The calculating of the noise level according to the noise data is specifically calculated using the following formula: ； Among them, is the noise level at a distance d and a service time of t, L p0 is the noise level at a reference distance d0, d is the actually measured distance, d0 is the reference distance, α0 is the initial sound absorption coefficient, and λ is the material aging rate, which is used to describe the rate of attenuation of sound absorption performance over time.

6. The method for identifying multiple risks of a sound barrier based on deep learning according to claim 4, wherein The detection data further includes wind load data. The determining of the risk assessment data corresponding to the risk factors according to the detection data specifically further includes: Determining the wind pressure per unit area according to the wind load data; Processing the wind pressure per unit area using a multi-layer perceptron to obtain wind load features, and the risk assessment data includes the wind load features; The wind pressure per unit area is specifically calculated using the following formula: ； Wherein, Pw is the wind pressure per unit area, Pair is the air density, v is the wind speed, and Cd is the drag coefficient.

7. The method for identifying multi-risk fusion of a sound barrier based on deep learning according to claim 6, wherein The using of a deep learning model to perform data fusion on the risk assessment data to obtain a risk category probability distribution specifically includes: Inputting the vibration features, the noise features, and the wind load features into the deep learning model; Controlling the deep learning model to perform weighted fusion on the vibration features, the noise features, and the wind load features using a gating mechanism to obtain fusion features; Performing global modeling on the fusion features using a Transformer algorithm to obtain target features; Using the fully connected layer and softmax activation of the deep learning model to perform risk level classification on the target features to generate the risk category probability distribution.

8. An acoustic barrier multi-risk fusion recognition device based on deep learning, characterized in that, The device includes an acquisition module (31) and a processing module (32), wherein, The acquisition module (31) is used to obtain detection data for target sound barrier risk detection; The processing module (32) is used to determine risk assessment data corresponding to risk factors according to the detection data; The processing module (32) is further configured to perform data fusion on the risk assessment data by using a deep learning model to obtain a risk category probability distribution; The processing module (32) is further configured to determine whether there is a risk in the target sound barrier based on the risk category probability distribution.

9. An electronic device, characterized in that, The electronic device includes a processor (41), a memory (45), a user interface (43), and a network interface (44). The memory (45) is used to store instructions. Both the user interface (43) and the network interface (44) are used to communicate with other devices. The processor (41) is configured to execute the instructions stored in the memory (45) so that the electronic device executes 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 instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images