Signal noise reduction processing method and device, vehicle, computer equipment and storage medium
By using vectorization mapping and multi-scale decomposition techniques, combined with neighborhood correlation analysis and threshold criteria, the problem of noise interference in rail train monitoring data was solved, achieving high-quality signal processing and ensuring safe train operation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CRRC TANGSHAN CO LTD
- Filing Date
- 2026-03-26
- Publication Date
- 2026-05-19
AI Technical Summary
Train monitoring data in complex geographical environments is subject to significant noise interference. Conventional filtering techniques are prone to destroying high-frequency details of faults, resulting in a decline in the quality of train status monitoring data and affecting safe operation.
The original signal is converted into a vector input signal using vectorization mapping technology. A joint discriminant is constructed through multi-scale decomposition and neighborhood correlation analysis. Noise components are suppressed by combining a preset threshold criterion. Finally, inverse transformation processing is performed to obtain a high-quality denoised signal.
It significantly reduces noise interference, improves signal clarity and accuracy, provides reliable data support for train fault detection and condition assessment, and ensures safe operation.
Smart Images
Figure CN122064920A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of rail transit technology, specifically to a signal noise reduction processing method, apparatus, vehicle, computer equipment, and storage medium. Background Technology
[0002] Rail trains travel through complex geographical environments, such as tunnels, canyons, and steep slopes. These complex geographical environments can cause significant noise interference to the monitoring data of various key systems in the rail train.
[0003] For example, the electric arc radiation from the overhead contact line and the electromagnetic reflection from the tunnel wall structure create a strong interference field, causing a sharp drop in the signal-to-noise ratio of key monitoring data of the EMU train. This results in the high-frequency vibration characteristics of key components or systems being completely masked by irregular electromagnetic noise. Another example is the strong vibration impact generated by the braking of the train on a steep slope. This causes high-frequency mechanical noise to superimpose on the data collected by strain sensors and acceleration sensors, further obscuring the key dynamic characteristics of key components or systems. This type of noise is characterized by "high frequency randomness and interweaving with signal characteristics".
[0004] To address the above issues, conventional filtering techniques tend to destroy high-frequency details of faults during the noise reduction process, leading to a loss of crucial diagnostic information. Solving the data noise problem is urgently needed to provide high-quality data for train status monitoring and ensure safe train operation.
[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this application, and therefore may contain information that is not part of the prior art known to those skilled in the art. Summary of the Invention
[0006] This application provides a signal noise reduction processing method, device, vehicle, computer equipment, and storage medium, which can provide high-quality data for train status monitoring and improve train operation safety.
[0007] A first aspect of this application provides a signal noise reduction processing method, including: The original signal is vectorized and mapped, and a vector input signal is constructed based on the vectorization mapping result; The vector input signal is decomposed into low-frequency and high-frequency components at the corresponding scales. Based on the transform domain index parameter of each of the high-frequency components, multiple high-frequency components that have neighborhood correlation in spatial location are determined, and a joint discriminant quantity reflecting the statistical characteristics of multiple adjacent high-frequency components is determined. Based on the joint discriminant and the preset threshold criterion, the high-frequency components of the current decomposition layer are processed to suppress the noise components in the original signal and obtain noise-reduced sample points. Each of the noise-reduced sample points is subjected to an inverse transform to obtain the corresponding output signal; Based on the multiple output signals, the noise reduction signal corresponding to the original signal is determined.
[0008] In an optional embodiment of this application, the step of vectorizing the original signal and constructing a vector input signal based on the vectorization mapping result includes: The original signal is subjected to multi-wavelet preprocessing to obtain vector input sample points; The vector input sample points are time-domain shifted to generate multiple vector input signals that differ in phase from the vector input sample points. The time-domain translation process includes applying different time-domain translation amounts to the vector input sample points within a preset translation range. The data length of the time-domain translation amount is taken from [16, 128], and / or the data length of the time-domain translation amount is an integer power of 2.
[0009] In an optional embodiment of this application, the step of performing multi-scale decomposition on the vector input signal to obtain low-frequency and high-frequency components at the corresponding scale includes: The vector input signal is subjected to hierarchical decomposition based on a preset multi-component filtering structure. At each scale level, low-frequency components that characterize the structural features of the vector input signal and high-frequency components that characterize the details are obtained. Wherein, the hierarchical decomposition process satisfies:
[0010] in, Indicates the scale layer index. Indicates the first The low-frequency components of the layer, The time-domain shift is indicated as The The low-frequency components of the layer, The time-domain shift is indicated as of High-frequency components of the layer, This represents the filter weights used for the low-frequency components. This represents the filtering weights used for the high-frequency components. This indicates the position index of the vector input signal in the corresponding scale layer.
[0011] In an optional embodiment of this application, determining multiple high-frequency components with neighborhood correlation in spatial location based on the transform domain index parameter of each of the high-frequency components, and determining a joint discriminant reflecting the statistical characteristics of the multiple adjacent high-frequency components, includes: The high-frequency components that satisfy at least one of the following conditions—having the same decomposition scale, the same subband, and the same translation amount—are identified as high-frequency components with neighborhood correlation at the said spatial location. Based on multiple high-frequency components with the same neighborhood correlation, determine the covariance matrix corresponding to the high-frequency components of the current decomposition layer. The threshold discrimination variable is determined based on the high-frequency components of the current decomposition layer and their corresponding covariance matrix; The joint discriminant quantity is determined based on the threshold discriminant variable.
[0012] In an optional embodiment of this application, determining the covariance matrix corresponding to the high-frequency components of the current decomposition layer based on multiple high-frequency components having the neighborhood correlation includes: Robust scaling estimates for each of the multiple high-frequency components are calculated separately. These robust scaling estimates are obtained using the median absolute deviation method and are defined as follows: ; Based on the robust scaling estimator, each high-frequency component of the multiple high-frequency components is processed, and the covariance matrix corresponding to the current decomposition layer is constructed based on the processed high-frequency components. The expression for the covariance matrix is as follows:
[0013] in, , , , ,and and They respectively represent the high-frequency components The first and second lines.
[0014] In an optional embodiment of this application, the signal noise reduction processing method satisfies one or more of the following: The step of determining the threshold discriminant variable based on the high-frequency components of the current decomposition layer and their corresponding covariance matrix includes: multiplying the transpose of the high-frequency components by the inverse of the covariance matrix as the threshold discriminant variable; wherein, the threshold discriminant variable ; Determining the joint discriminant based on the threshold discriminant variable includes: determining the joint discriminant based on the threshold discriminant variable and the threshold discriminant variables adjacent to the threshold discriminant variable at that spatial location; wherein, the joint discriminant... .
[0015] In an optional embodiment of this application, the noise reduction sample points are obtained in the following manner:
[0016] in, Indicates the noise reduction sample points; Represents the high-frequency component; The joint discriminant quantity represents the following; This represents the threshold value corresponding to the preset threshold criterion, where, , For the first Layer detail signal points.
[0017] In an optional embodiment of this application, each of the noise-reduced sample points undergoes an inverse transform process to obtain the corresponding output signal, including: A translation-invariant multi-wavelet inverse transform is performed on each of the denoised sample points to reconstruct the denoised sample points and obtain a vector output signal; wherein, the vector output signal The This represents the inverse operation of the time-domain translation process; Indicates the ( The coefficients of the low-pass filter; Indicates the ( The coefficients of the high-pass filter; Represents the high-frequency component; the This refers to the low-frequency component; Each vector output signal is subjected to a cyclic inverse translation operation, and multiple wavelet post-processing is performed each time the cyclic inverse translation operation is performed, until a one-dimensional output signal is obtained.
[0018] A second aspect of this application provides a signal noise reduction processing apparatus, comprising: The first building unit is used to perform vectorization mapping on the original signal and construct a vector input signal based on the vectorization mapping result; The decomposition unit is used to perform multi-scale decomposition on the vector input signal to obtain the low-frequency component and high-frequency component at the corresponding scale. The second construction unit is used to determine multiple high-frequency components that have neighborhood correlation in spatial location based on the transform domain index parameter of each of the high-frequency components, and to determine a joint discriminant quantity that reflects the statistical characteristics of multiple adjacent high-frequency components. The noise reduction unit is used to process the high-frequency components of the current decomposition layer based on the joint discriminant and the preset threshold criterion, suppress the noise components in the original signal, and obtain noise-reduced sample points. The processing unit is configured to perform inverse transform processing on each of the noise-reduced sample points to obtain the corresponding output signal, and to determine the noise-reduced signal corresponding to the original signal based on the multiple output signals.
[0019] A third aspect of this application provides a vehicle, including a signal noise reduction processing device as described in any of the foregoing embodiments.
[0020] A fourth aspect of this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the signal noise reduction processing method described in any of the foregoing embodiments.
[0021] A fifth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the signal noise reduction processing method described in any of the foregoing embodiments.
[0022] A sixth aspect of the present application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the signal noise reduction processing method described in any of the foregoing embodiments.
[0023] In summary, this invention, by introducing vectorization mapping technology, can map the original signal into a vector input signal, achieving a multi-dimensional representation of the original signal. Based on the mapped vector input signal, multi-scale decomposition is performed to determine the low-frequency and high-frequency components of the vector input signal at different scales, thus laying the foundation for subsequent noise reduction processing. Through spatial neighborhood correlation analysis of multiple high-frequency components, a joint discriminant quantity reflecting the statistical characteristics of multiple adjacent high-frequency components is constructed. This joint discriminant quantity provides a key basis for subsequent processing and can accurately suppress noise components in the original signal. Furthermore, based on the joint discriminant quantity and high-frequency component information, and combined with a preset threshold criterion, the high-frequency components are processed, achieving effective noise suppression and obtaining noise-reduced sample points. Finally, after inverse transformation processing, an output signal corresponding to the original signal is obtained, and based on the results of multiple output signals, a high-quality noise-reduced signal is accurately obtained. This method can significantly reduce noise interference with train status monitoring data, improve signal clarity and accuracy, and provide more reliable data support for train fault detection, status assessment and predictive maintenance, thereby effectively ensuring the safe operation of trains, reducing safety hazards caused by inaccurate data or noise interference, and enhancing the overall stability and reliability of the train monitoring system. Attached Figure Description
[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of a signal noise reduction processing method provided in an embodiment of this application; Figure 2 This is a flowchart of determining a joint discriminant quantity provided in an embodiment of this application; Figure 3 This is a schematic diagram of a signal noise reduction processing method provided in an embodiment of this application; Figure 4 This is a waveform diagram showing the time-domain waveform and spectrum of a raw signal; Figure 5 This is a waveform diagram of a noise-reduced signal using translation-invariant multi-wavelet adjacent coefficient processing, provided in an embodiment of this application. Figure 6 This is a waveform diagram of a noise-reduced signal processed using multi-wavelet noise reduction. Figure 7 This is a schematic diagram showing the result of a signal noise reduction processing device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the hardware structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0025] In the process of developing this application, the inventors discovered that current conventional filtering techniques are ineffective and can destroy the authenticity of the original signal.
[0026] For example, in a wavelet transform-based data denoising method, the denoising result will alternately exhibit large peak values near singular points. These peak values are not inherent to the original signal, but are generated by artificial interference during the denoising process, and are called the Gibbs phenomenon.
[0027] Due to the localization characteristics of wavelet transform, such artificial oscillations of Gibbs phenomena are closely related to the arrangement of singular points, and the Gibbs phenomena near each singular point have the same characteristics, but there is a certain phase difference between them.
[0028] Traditional noise reduction methods set thresholds based on the different statistical characteristics of wavelet decomposition coefficients, while ignoring the correlation between wavelet coefficients. This makes it impossible to extract high-quality data for signals with rich structural features and extremely low signal-to-noise ratios.
[0029] Therefore, after studying signal denoising methods, the inventors discovered that the multi-wavelet data denoising method can overcome the shortcomings of traditional single wavelets, which cannot simultaneously possess multiple excellent properties such as orthogonality, symmetry, and short support.
[0030] Specifically, the core logic of multi-wavelet data denoising methods is to utilize the difference in coefficients between signal and noise in the multi-wavelet transform domain: the multi-wavelet coefficients of signals typically have strong correlation and large amplitude, and can correspond to key structural features or important signal characteristics such as image edges and textures; while the multi-wavelet coefficients of noise are mostly randomly distributed, with small amplitudes and no obvious correlation. By selectively processing the transformed coefficients, the dual goals of noise suppression and effective information preservation can be achieved.
[0031] Furthermore, the adjacent coefficient data denoising method is a type of denoising technique that focuses on the correlation between adjacent coefficients in the signal transform domain. Its core is to break through the limitations of traditional single-coefficient threshold processing and treat adjacent coefficients as a whole for analysis, thereby accurately distinguishing between signals and noise and better preserving the local features of the signal while denoising.
[0032] Based on the fundamental difference in the coefficient distribution of signal and noise in the transform domain, the transform coefficients of useful signals (such as wavelet coefficients and cosine coefficients) are not isolated. Their adjacent coefficients often exhibit strong correlations due to the continuous characteristics of the corresponding signals (such as mechanical fault impact signals and image textures), and their overall amplitude is relatively large. In contrast, noise coefficients are randomly distributed, with small amplitudes and no obvious correlation between adjacent noise coefficients. By treating adjacent coefficients as a whole for comprehensive calculation and threshold judgment, signal components can be effectively enhanced, random noise suppressed, and the problem of losing signal details in traditional single-coefficient processing can be solved.
[0033] Based on this, this application provides a signal denoising method. By introducing vectorization mapping technology, the original signal can be mapped into a vector input signal, realizing a multi-dimensional representation of the original signal. Based on the mapped vector input signal, multi-scale decomposition is performed to determine the low-frequency and high-frequency components of the vector input signal at different scales, thus laying the foundation for subsequent denoising processing. Through spatial neighborhood correlation analysis of multiple high-frequency components, a joint discriminant quantity reflecting the statistical characteristics of multiple adjacent high-frequency components is constructed. This joint discriminant quantity provides a key basis for subsequent processing and can accurately suppress noise components in the original signal. Furthermore, based on the joint discriminant quantity and high-frequency component information, and combined with a preset threshold criterion, the high-frequency components are processed to achieve effective noise suppression and obtain denoised sample points. Finally, after inverse transformation processing, an output signal corresponding to the original signal is obtained, and based on the results of multiple output signals, a high-quality denoised signal is accurately obtained. This method can significantly reduce noise interference with train status monitoring data, improve signal clarity and accuracy, and provide more reliable data support for train fault detection, status assessment and predictive maintenance, thereby effectively ensuring the safe operation of trains, reducing safety hazards caused by inaccurate data or noise interference, and enhancing the overall stability and reliability of the train monitoring system.
[0034] In other words, this application introduces the concept of multi-wavelet to propose a translation-invariant multi-wavelet adjacent coefficient denoising technique based on the traditional single-wavelet adjacent coefficient data denoising method. By relying on the multi-resolution analysis characteristics of multi-wavelets, multiple scaling functions and wavelet functions can be used to simultaneously capture the low-frequency trend and high-frequency details of the signal, solving the defect of detail loss in traditional single-wavelet denoising. At the same time, by strengthening the spatial correlation of signal coefficients, it can accurately distinguish between the "random discrete characteristics of noise" and the "periodic characteristics of data", achieving accurate separation of noise and effective signal.
[0035] The solutions in this application embodiment can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0036] To make the technical solutions and advantages of the embodiments of this application clearer, the exemplary embodiments of this application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.
[0037] See Figure 1 , Figure 1 This is a flowchart of a signal noise reduction processing method provided in an embodiment of this application, which can perform the following noise reduction processing steps: S101 performs vectorization mapping on the original signal and constructs a vector input signal based on the vectorization mapping result.
[0038] Specifically, the raw signal usually comes from multiple devices or sensors and contains factors such as noise and interference, making it difficult to use directly for subsequent processing. Therefore, the raw signal needs to be preprocessed.
[0039] To map the original signal into vector form, effective features need to be extracted. The method of feature extraction depends on the type of signal and the specific application requirements.
[0040] For example, wavelet transform is used to perform multi-scale analysis on signals that have instantaneous changes.
[0041] Feature extraction yields a set of feature values that characterize the original signal. Subsequently, a vectorization mapping method is used to convert these feature values into multidimensional vectors.
[0042] Once the feature vectors are obtained, vector input signals can be constructed according to requirements.
[0043] Through the above steps, the obtained vector input signal can effectively represent the characteristics of the original signal, thereby reducing the difficulty of subsequent signal denoising.
[0044] In one embodiment, step S101 may include: performing multi-wavelet pre-processing on the original signal to obtain vector input sample points; performing time-domain translation processing on the vector input sample points to generate multiple vector input signals that differ in phase from the vector input sample points; wherein, the time-domain translation processing includes applying different time-domain translation amounts to the vector input sample points within a preset translation range.
[0045] Specifically, the cyclic translation operator is introduced. And the nth vector input sample point after multi-wavelet preprocessing is defined as .
[0046] Next, define To The translation amount is The time-domain translation operation, where the translation range is Therefore, we can obtain One vector input signal A new signal with a phase difference.
[0047] That is to Based on this, multiple consecutive time-domain translation processes are performed to obtain multiple results related to this. Vector input signals with different phase differences .
[0048] In one embodiment, , .
[0049] Furthermore, the data length of the time-domain shift is taken from [16, 128], and / or the data length of the time-domain shift is an integer power of 2.
[0050] Specifically, in the translation-invariant multiwavelet algorithm, the cyclic translation amount This is a key parameter. Since the data length processed by the multi-wavelet transform in this scheme is generally a power of 2, and the multi-wavelet decomposition involves sampling every two data points, therefore... The setting is usually an integer power of 2.
[0051] Furthermore, if If the selection is too small, it will be difficult to effectively eliminate the Gibbs phenomenon in the multi-wavelet denoising process; if Choosing an excessively large value will not only affect the processing speed, but multiple averaging operations may also smooth out local fault features in the signal.
[0052] Therefore, in this application, the following is set .
[0053] S102 performs multi-scale decomposition on the vector input signal to obtain the low-frequency and high-frequency components at the corresponding scales.
[0054] In some embodiments, by performing multi-scale decomposition processing on the vector input signal in sequence form, the vector input signal is divided into low-frequency components reflecting overall structural features and high-frequency components reflecting local variation features at different scale levels. Through multi-scale decomposition, the signal can be analyzed from different resolution perspectives, thereby obtaining multi-level signal composition information.
[0055] Specifically, the vector input signal is decomposed layer by layer according to a preset number of decomposition layers based on the scale. Each scale corresponds to one decomposition layer, which is used to extract the signal features at that scale.
[0056] In practical implementation, wavelet transform, multi-resolution analysis, filter bank decomposition, or other equivalent multi-scale analysis methods can be used, and this invention does not limit these methods.
[0057] In one embodiment, step S102 may include: performing hierarchical decomposition processing on the vector input signal based on a preset multi-component filtering structure, and obtaining low-frequency components for characterizing the structural features of the vector input signal and high-frequency components for characterizing detail changes at each scale level.
[0058] Specifically, at a certain scale, by applying a low-pass filter to the vector input signal, the low-frequency component at that scale is obtained, which is used to characterize the overall trend or smoothness of the signal; at the same time, by applying a high-pass filter to the vector input signal, the high-frequency component at the corresponding scale is obtained, which is used to characterize the local changes, edge information or detail features of the signal. Thus, the low-frequency component and the high-frequency component can be obtained.
[0059] In this embodiment, the hierarchical decomposition process satisfies:
[0060] in, Indicates the scale layer index. Indicates the first The low-frequency components of the layer, The time-domain shift is indicated as The The low-frequency components of the layer, The time-domain shift is indicated as of High-frequency components of the layer, This represents the filter weights used for the low-frequency components. This represents the filtering weights used for the high-frequency components. This indicates the position index of the vector input signal in the corresponding scale layer; This indicates the sample index.
[0061] It should be noted that the low-frequency components are obtained by further downsampling or reconstructing the signal from the previous scale and are used as the input signal for the next scale decomposition; while the high-frequency components are preserved as detailed information of that scale.
[0062] Repeat the above decomposition process until all preset scales are decomposed, thereby obtaining sets of low-frequency components and high-frequency components at multiple scales. In this way, through multi-scale decomposition, vector input signals can be analyzed at different resolution levels, enabling the effective separation and representation of the global and local features of the vector input signals.
[0063] S103, based on the transform domain index parameter of each of the high-frequency components, determine multiple high-frequency components that have neighborhood correlation in spatial location, and determine a joint discriminant quantity that reflects the statistical characteristics of multiple adjacent high-frequency components.
[0064] The transform domain index parameters include at least one of the following: decomposition scale, subband, and translation amount.
[0065] In some embodiments, for each high-frequency component at each scale, the neighborhood correlation of multiple high-frequency components in spatial location can be determined based on their corresponding transform domain index parameters, and then a joint discriminant quantity reflecting the statistical characteristics of multiple adjacent high-frequency components can be determined based on the neighborhood correlation.
[0066] Specifically, based on the determined high-frequency components, a threshold is set by treating several adjacent wavelet coefficients as a whole. When there is a strong correlation between the wavelet coefficients, the adjacent coefficient denoising method, which treats the coefficients as a whole, is significantly better than the traditional thresholding method that compares each coefficient individually. It can more effectively preserve the local feature information of the signal while filtering out noise.
[0067] Due to the correlation between the vector coefficients of multiple wavelet decomposition, a vector coefficient processing method is chosen to construct a joint discriminant that reflects the statistical characteristics of multiple adjacent high-frequency components.
[0068] In one embodiment, Figure 2 This is a flowchart of determining a joint discriminant quantity provided in an embodiment of this application, such as... Figure 2 As shown, the following steps can be performed: S201, the high-frequency component that satisfies at least one of the following conditions—having the same decomposition scale, the same sub-band, and the same translation amount—is determined as the high-frequency component with neighborhood correlation at the said spatial location.
[0069] Specifically, in the high-frequency components obtained by wavelet transform, the spatial neighborhood correlation between them is determined by the consistency of scale, subband, or location. When one of the scale, subband, or location is the same, it indicates that the two high-frequency components have a spatial neighborhood correlation.
[0070] The decomposition scale refers to the levels of wavelet decomposition. For example, level 1, level 2, and level 3. The higher the level, the coarser the decomposition.
[0071] A subband refers to the multiple directional subbands obtained after wavelet decomposition. Translation refers to the spatial displacement of the wavelet basis, that is, the position of the coefficients in the image.
[0072] In other words, it means treating all adjacent high-frequency signals that meet the requirements as a whole.
[0073] S202, Based on multiple high-frequency components with the neighborhood correlation, determine the covariance matrix corresponding to the high-frequency components of the current decomposition layer.
[0074] Specifically, by leveraging the neighborhood correlation of high-frequency components in spatial location to reduce feature loss, several closely adjacent wavelet coefficients are treated as a whole to set a threshold. This processing method can filter out noise while preserving the local features of the signal to the greatest extent.
[0075] Specifically, step S202 may include: Robust scaling estimators for each of the multiple high-frequency components are calculated separately. These robust scaling estimators are obtained using the median absolute deviation method and are defined as follows: .
[0076] Based on the robust scaling estimator, each high-frequency component of the multiple high-frequency components is processed, and the covariance matrix corresponding to the current decomposition layer is constructed based on the processed high-frequency components.
[0077] The expression for the covariance matrix is:
[0078] in, , , , ,and and They respectively represent the high-frequency components The first and second lines.
[0079] In short, assuming translation-invariant multi-wavelet decomposition, the th Layer translation amount is The high-frequency components are Then define a new variable. , to be used as the basic variable for threshold denoising. Furthermore, in order to improve the accuracy of covariance matrix estimation, It is obtained by using robust coefficient estimation.
[0080] S203, determine the threshold discrimination variable based on the high-frequency components of the current decomposition layer and their corresponding covariance matrix.
[0081] In some embodiments, after determining the covariance matrix, a threshold discriminant variable is further determined based on the high-frequency components and the covariance matrix that have a corresponding relationship.
[0082] For example, the product of the transpose of the high-frequency component and the inverse of the covariance matrix is used as the threshold discrimination variable.
[0083] For example, threshold discriminant variables .
[0084] S204, Determine the joint discriminant based on the threshold discriminant variable.
[0085] In some embodiments, a threshold discriminant variable defines the criterion for judging abnormal signals. Based on this threshold discriminant variable, a joint discriminant quantity can be further determined to better improve the noise reduction effect.
[0086] For example, step S204 may include: determining a joint discriminant based on the threshold discriminant variable and the threshold discriminant variables that are adjacent to the threshold discriminant variable at the spatial location.
[0087] Among them, joint discriminant quantity .
[0088] S104. Based on the joint discriminant and the preset threshold criterion, the high-frequency components of the current decomposition layer are processed to suppress the noise components in the original signal and obtain noise-reduced sample points.
[0089] In some embodiments, adaptive wavelet coefficient processing is performed based on the constructed joint discriminant and information from high-frequency components. The purpose of this processing is to optimize the wavelet coefficients obtained from the multi-scale decomposition and suppress noise components.
[0090] Specifically, the joint discriminant can be used as the segmentation criterion, and the high-frequency components can be adjusted according to certain rules to obtain denoised sample points. This can effectively preserve the details of the signal while suppressing noise, ultimately yielding the denoised signal.
[0091] In one example, the noise reduction sample points are obtained in the following way:
[0092] in, Indicates the noise reduction sample points; Indicates high-frequency components; Indicates the joint discriminant; This represents the threshold value corresponding to the preset threshold criterion, where, , For the first Layer detail signal points.
[0093] Specifically, in the case of random noise, Satisfying the degree of freedom of Distribution. Considering the correlation between the multi-wavelet decomposition coefficients and their immediate neighbor coefficients, from Derivation of new variables This allows us to determine the noise reduction sample points.
[0094] S105 performs an inverse transform on each noise-reduced sample point to obtain the corresponding output signal.
[0095] In some embodiments, an inverse transform is performed on each denoised sample point to recover the time-domain signal. This step recombines the denoised components into a time-domain signal using an inverse wavelet transform (or inverse Fourier transform). The result of the inverse transform is a denoised signal with significantly reduced noise components and significantly improved signal quality compared to the original signal.
[0096] In one example, step S105 may include: Perform a translation-invariant multi-wavelet inverse transform on each denoised sample point to reconstruct the denoised sample points and obtain the vector output signal.
[0097] By introducing a translation invariance mechanism, the pseudo-oscillation problem caused by signal translation in traditional wavelet denoising is avoided.
[0098] For example, vector output signal The This represents the inverse operation of the time-domain translation process; Indicates the ( The coefficients of the low-pass filter; Indicates the ( The coefficients of the high-pass filter; Represents the high-frequency component; the This refers to the low-frequency component.
[0099] It should be noted that the vector output signal can be... As noise reduction sample points, they may differ in appearance, but they actually express the same meaning. That is, the expression in the vector output signal simply defines the relationship between different parameters.
[0100] For each vector output signal, a cyclic inverse translation operation is performed, and multiple wavelet post-processing is also performed each time the cyclic inverse translation operation is performed, until a one-dimensional output signal is obtained.
[0101] Specifically, after acquiring the vector output signal, a cyclic reverse translation operation is performed on the vector output signal to eliminate the influence of the aforementioned translation operation on the signal position and restore the signal to its original alignment state.
[0102] Subsequently, the signal after cyclic inverse translation is subjected to multi-wavelet post-processing. This multi-wavelet post-processing includes, but is not limited to, one or a combination of the following operations: synthesis of vector components of the vector signal; linear combination of multi-channel signals; and one-dimensional mapping of the multi-wavelet filtering results.
[0103] In this embodiment, a one-dimensional mapping of the multi-wavelet filtering results is used.
[0104] S106 determines the noise reduction signal corresponding to the original signal based on multiple output signals.
[0105] In one embodiment, an averaging method is used to determine the noise-reduced signal, i.e.:
[0106] in, Indicates the noise reduction signal; Indicates the number of output signals; This represents a one-dimensional output signal; This indicates the cyclic translation amount.
[0107] It should be noted that other appropriate methods can also be used to determine the denoised signal corresponding to the original signal. For example, weighted averaging, median fusion, etc.
[0108] See Figure 3 The diagram shown is a schematic representation of a signal noise reduction processing method provided in an embodiment of this application. Figure 3 As shown, it includes: The original signal is processed by multiple wavelet preprocessing to obtain vector input sample points. This corresponds to preprocessing.
[0109] Cyclic translation vector input sample points A vector input signal with a certain phase difference is obtained. This corresponds to loop processing.
[0110] For vector input signals Perform a translation-invariant multiwavelet transform on the vector input signal. Decompose to scale High-frequency components are obtained. This corresponds to multi-wavelet decomposition.
[0111] Using adjacent coefficient thresholds for high-frequency components Processing is performed to obtain noise-reduced sample points. This corresponds to noise reduction for adjacent coefficients.
[0112] For noise reduction sample points Perform translation-invariant multi-wavelet inverse transform to reconstruct the denoised vector output signal. This corresponds to multi-wavelet reconstruction.
[0113] Cyclic reverse translation The one-dimensional output signal is obtained by multi-wavelet post-processing. This corresponds to a cyclic reverse translation.
[0114] For one-dimensional output signal The final noise-reduced signal is obtained by averaging. This corresponds to the average one-dimensional output signal.
[0115] Thus, this invention integrates the multi-wavelet denoising method, the adjacent coefficient denoising method, and the translation-invariant denoising method to construct a translation-invariant multi-wavelet adjacent coefficient denoising method suitable for rail vehicles.
[0116] Specifically, the translation-invariant multi-wavelet adjacent coefficient denoising method uses time-domain translation operations to obtain a new signal with a certain phase difference from the input signal, thereby changing the position of singular points in the original data structure and reducing or eliminating the Gibbs phenomenon caused by the special position of singular points.
[0117] For a given signal, the anomalous amplitude can be minimized by selecting the optimal translation amount. However, for a signal containing multiple singularities, it is difficult to select the optimal translation amount for all singularities.
[0118] To address the issue of finding the optimal translation amount, a cyclic translation operation is performed on a certain range of translation amounts, followed by averaging the results. It can be seen that redundancy is introduced throughout the cyclic translation process. The averaging operation after cyclic translation better preserves the signal smoothness and exhibits superior noise reduction, closely approximating the true signal.
[0119] Furthermore, translation-invariant data denoising methods are an improved technique for addressing the "translation sensitivity" problem in traditional wavelet / multi-wavelet denoising. The core of these methods is to eliminate the inconsistency in denoising effects after signal translation through a logic of cyclic translation, multiple transformations, processing, inverse transformations, and result fusion. This allows for the more accurate preservation of local signal features (such as edges, impacts, and textures) while suppressing noise.
[0120] To better illustrate the noise reduction effect of the technical solution in this application, a specific example is provided.
[0121] See Figures 4 to 6 ,in, Figure 4 This is a waveform diagram showing the time-domain waveform and spectrum of an original signal. Figure 5 This is a waveform diagram of a noise-reduced signal processed using translation-invariant multi-wavelet adjacent coefficients, as provided in an embodiment of this application. Figure 6 This is a waveform diagram of a noise-reduced signal using multi-wavelet noise reduction processing.
[0122] Specifically, the test bench was used to simulate the actual operating conditions of a rail train, and the performance of the studied method was verified by measuring the vibration signal of the bogie bearing. The time-domain signal and spectrum of the measured dynamic signal were also measured.
[0123] like Figure 4 As shown, the test dynamic signals are highly complex, making it difficult to distinguish key features. Furthermore, high-frequency noise constitutes a large proportion of the spectrum. Directly analyzing the raw dynamic signals leads to difficulties in feature extraction and, in severe cases, can cause misjudgments in subsequent health status monitoring and diagnosis. Therefore, noise interference filtering is necessary before conducting health status monitoring of key train components or systems to improve the quality of the monitoring data.
[0124] Based on this, in the translation-invariant multi-wavelet cyclic translation... When the value is set to 32, the translation-invariant multi-wavelet adjacent coefficient denoising method is applied to the test dynamics respectively (see...). Figure 5 Multi-wavelet vector hard thresholding noise reduction (see...) Figure 6 Subgraph (a) in the figure), multi-wavelet vector soft thresholding denoising (see subgraph (a)). Figure 6 Subgraph (b) in the image) and adjacent coefficient denoising (see [link]). Figure 6 Subgraph (c) in the middle).
[0125] The noise reduction results from the different methods described above show that Figure 6 The noise reduction methods shown can eliminate most of the background noise. Although they can extract some impact features, they do not identify the weak impact features under strong noise interference.
[0126] And in Figure 5 The translation-invariant multi-wavelet adjacent coefficient denoising method shown can comprehensively and accurately extract all periodic impact features with a period of approximately 12.8 ms, or 78.125 Hz, which is consistent with the dynamic characteristic frequency tested in the experiment. This proves the effectiveness of the data denoising technology based on translation-invariant multi-wavelet adjacent coefficients.
[0127] It should be noted that, Figures 4 to 6The waveforms and processing results shown are for illustrative purposes only and should not be used to limit this application.
[0128] Therefore, the translation-invariant multi-wavelet adjacent coefficient data denoising method fully utilizes the advantages of each method, including: multi-wavelets provide a more accurate representation of the signal; the adjacent coefficient method utilizes the spatial correlation of the signal in the wavelet domain, which can better distinguish between signal and noise than independent coefficient processing, and better preserves the edge and texture of the signal; translation invariance eliminates pseudo-Gibbs oscillations, making the denoising results visually smoother, and is usually also better in terms of mean square error.
[0129] The translation-invariant multi-wavelet adjacent coefficient data denoising method shown in this invention can effectively eliminate noise interference in the complex environment of rail vehicles, and can comprehensively and accurately extract all data periodic features, greatly improving data quality, providing effective data support for train status monitoring and data analysis, and ensuring reliable train operation.
[0130] More specifically, this method generates a new signal with a phase difference from the original signal by using time-domain translation operations, thereby changing the position of singular points in the signal. Then, through cyclic translation and averaging operations, it significantly reduces or even eliminates the Gibbs phenomenon caused by singular points. At the same time, the averaging process it includes can efficiently reduce noise while maintaining the smoothness of the signal, making the processed signal closer to the real signal.
[0131] The denoising properties of translation-invariant multi-wavelets, combined with the synergistic effect of adjacent coefficients, give it extremely strong anti-interference capabilities. Practice shows that this method can accurately extract the characteristic frequencies of weak faults from strong noise, providing a reliable basis for early fault diagnosis of rail vehicle equipment. It solves the problem of difficult feature extraction of weak faults under harsh operating conditions, and can comprehensively and accurately extract all periodic impact features of the data, eliminate noise, and collect high-quality train operation data.
[0132] It should be understood that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order constraint on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the diagram may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.
[0133] The signal noise reduction processing method has been described in detail above through some embodiments. In order to enable those skilled in the art to better understand and implement it, the corresponding device is also described in detail below through some embodiments.
[0134] See Figure 7 The diagram shown is a schematic representation of the result of a signal noise reduction processing device provided in an embodiment of this application. Figure 7 As shown, the signal noise reduction processing device 700 may include: The first construction unit 710 is used to perform vectorization mapping on the original signal and construct a vector input signal based on the vectorization mapping result; Decomposition unit 720 is used to perform multi-scale decomposition on the vector input signal to obtain the low-frequency component and high-frequency component at the corresponding scale. The second construction unit 730 is used to determine multiple high-frequency components that have neighborhood correlation in spatial location based on the transform domain index parameter of each of the high-frequency components, and to determine a joint discriminant quantity that reflects the statistical characteristics of multiple adjacent high-frequency components. The noise reduction unit 740 is used to process the high-frequency components of the current decomposition layer based on the joint discriminant and the preset threshold criteria, suppress the noise components in the original signal, and obtain noise-reduced sample points. The processing unit 750 is used to perform inverse transform processing on each noise reduction sample point to obtain the corresponding output signal, and to determine the noise reduction signal corresponding to the original signal based on multiple output signals.
[0135] For further descriptions of signal noise reduction processing devices, please refer to the examples above.
[0136] It is understandable that the above division of units is only a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, the above units can be implemented by the processor calling software.
[0137] This application also provides a vehicle that may include the signal noise reduction processing device of any of the foregoing embodiments.
[0138] In one embodiment, a computer device is provided, the computer device including: a memory and a processor, the memory storing a computer program, the processor executing steps of a signal noise reduction processing method when running the computer program.
[0139] See Figure 8 , Figure 8 This is a schematic diagram of the hardware structure of a computer device provided in an embodiment of this application.
[0140] Figure 8The computer device shown includes a memory 81, a processor 82, and a transceiver 83. The processor 82 is coupled to the memory 81 and the transceiver 83. The memory 81 can be located inside or outside the terminal. The memory 81, processor 82, and transceiver 83 can be connected via a communication bus. The transceiver 83 is used to communicate with other devices or communication networks.
[0141] Optionally, transceiver 83 can be a transmitter. Memory 81 stores a computer program that can run on processor 82. When processor 82 runs the computer program, transceiver 83 executes the steps in the signal noise reduction processing method provided in the above embodiments.
[0142] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, can perform any of the steps in the signal noise reduction processing method described above.
[0143] In one embodiment, a computer program product is provided, including a computer program / instructions that, when executed by a processor, can perform any of the steps in the signal noise reduction processing method described above.
[0144] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0145] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0146] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0147] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0148] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0149] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A signal noise reduction processing method, characterized in that, include: The original signal is vectorized and mapped, and a vector input signal is constructed based on the vectorization mapping result; The vector input signal is decomposed into low-frequency and high-frequency components at the corresponding scales. Based on the transform domain index parameter of each of the high-frequency components, multiple high-frequency components that have neighborhood correlation in spatial location are determined, and a joint discriminant quantity reflecting the statistical characteristics of multiple adjacent high-frequency components is determined. Based on the joint discriminant and the preset threshold criterion, the high-frequency components of the current decomposition layer are processed to suppress the noise components in the original signal and obtain noise-reduced sample points. Each of the noise-reduced sample points is subjected to an inverse transform to obtain the corresponding output signal; Based on the multiple output signals, the noise reduction signal corresponding to the original signal is determined.
2. The signal noise reduction processing method according to claim 1, characterized in that, The process of vectorizing the original signal and constructing a vector input signal based on the vectorization result includes: The original signal is subjected to multi-wavelet preprocessing to obtain vector input sample points; The vector input sample points are time-domain shifted to generate multiple vector input signals that differ in phase from the vector input sample points. The time-domain translation process includes applying different time-domain translation amounts to the vector input sample points within a preset translation range. The data length of the time-domain translation amount is taken from [16, 128], and / or the data length of the time-domain translation amount is an integer power of 2.
3. The signal noise reduction processing method according to claim 2, characterized in that, The step of performing multi-scale decomposition on the vector input signal to obtain low-frequency and high-frequency components at the corresponding scales includes: The vector input signal is subjected to hierarchical decomposition based on a preset multi-component filtering structure. At each scale level, low-frequency components that characterize the structural features of the vector input signal and high-frequency components that characterize the details are obtained. Wherein, the hierarchical decomposition process satisfies: ； in, Indicates the scale layer index. Indicates the first The low-frequency components of the layer, The time-domain shift is indicated as The The low-frequency components of the layer, The time-domain shift is indicated as of High-frequency components of the layer, This represents the filter weights used for the low-frequency components. This represents the filtering weights used for the high-frequency components. This indicates the position index of the vector input signal in the corresponding scale layer.
4. The signal noise reduction processing method according to claim 1, characterized in that, The process of determining multiple high-frequency components with neighborhood correlation in spatial location based on the transform domain index parameter of each high-frequency component, and determining a joint discriminant reflecting the statistical characteristics of multiple adjacent high-frequency components, includes: The high-frequency components that satisfy at least one of the following conditions—having the same decomposition scale, the same subband, and the same translation amount—are identified as high-frequency components with neighborhood correlation at the said spatial location. Based on multiple high-frequency components with the same neighborhood correlation, determine the covariance matrix corresponding to the high-frequency components of the current decomposition layer. The threshold discrimination variable is determined based on the high-frequency components of the current decomposition layer and their corresponding covariance matrix; The joint discriminant quantity is determined based on the threshold discriminant variable.
5. The signal noise reduction processing method according to claim 4, characterized in that, The step of determining the covariance matrix corresponding to the high-frequency components of the current decomposition layer based on multiple high-frequency components with the neighborhood correlation includes: Robust scaling estimates are calculated for each of the multiple high-frequency components exhibiting neighborhood correlation. These robust scaling estimates are obtained using the median absolute deviation method and are defined as follows: ; Based on the robust scaling estimator, each high-frequency component of the multiple high-frequency components is processed, and the covariance matrix corresponding to the current decomposition layer is constructed based on the processed high-frequency components. The expression for the covariance matrix is as follows: ； in, , , , ,and and They respectively represent the high-frequency components The first and second lines.
6. The signal noise reduction processing method according to claim 4, characterized in that, Meet one or more of the following conditions: The step of determining the threshold discriminant variable based on the high-frequency components of the current decomposition layer and their corresponding covariance matrix includes: multiplying the transpose of the high-frequency components by the inverse of the covariance matrix as the threshold discriminant variable; wherein, the threshold discriminant variable ; Determining the joint discriminant based on the threshold discriminant variable includes: determining the joint discriminant based on the threshold discriminant variable and the threshold discriminant variables adjacent to the threshold discriminant variable at that spatial location; wherein, the joint discriminant... .
7. The signal noise reduction processing method according to claim 6, characterized in that, The noise reduction sample points are obtained using the following method: ； in, Indicates the noise reduction sample points; Represents the high-frequency component; The joint discriminant quantity is represented by the following: This represents the threshold value corresponding to the preset threshold criterion, where, , For the first Layer detail signal points.
8. The signal noise reduction processing method according to claim 2, characterized in that, Perform an inverse transform on each of the noise-reduced sample points to obtain the corresponding output signal, including: A translation-invariant multi-wavelet inverse transform is performed on each of the denoised sample points to reconstruct the denoised sample points and obtain a vector output signal; wherein, the vector output signal The This represents the inverse operation of the time-domain translation process; Indicates the ( The coefficients of the low-pass filter; Indicates the ( The coefficients of the high-pass filter; Represents the high-frequency component; the This refers to the low-frequency component; Each vector output signal is subjected to a cyclic inverse translation operation, and multiple wavelet post-processing is performed each time the cyclic inverse translation operation is performed, until a one-dimensional output signal is obtained.
9. A signal noise reduction processing device, characterized in that, include: The first building unit is used to perform vectorization mapping on the original signal and construct a vector input signal based on the vectorization mapping result; The decomposition unit is used to perform multi-scale decomposition on the vector input signal to obtain the low-frequency component and high-frequency component at the corresponding scale. The second construction unit is used to determine multiple high-frequency components that have neighborhood correlation in spatial location based on the transform domain index parameter of each of the high-frequency components, and to determine a joint discriminant quantity that reflects the statistical characteristics of multiple adjacent high-frequency components. The noise reduction unit is used to process the high-frequency components of the current decomposition layer based on the joint discriminant and the preset threshold criterion, suppress the noise components in the original signal, and obtain noise-reduced sample points. The processing unit is configured to perform inverse transform processing on each of the noise-reduced sample points to obtain the corresponding output signal, and to determine the noise-reduced signal corresponding to the original signal based on the multiple output signals.
10. A vehicle, characterized in that, include: The signal noise reduction processing device as described in claim 9.
11. A computer device, comprising: A memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the signal noise reduction processing method according to any one of claims 1 to 8.
12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the signal noise reduction processing method according to any one of claims 1 to 8; And / or, a computer program product comprising a computer program / instructions, characterized in that, when executed by a processor, the computer program / instructions implement the steps of the signal noise reduction processing method according to any one of claims 1 to 8.