Method and apparatus for intelligent processing and feature selection of multi-sensor signals

Abstract

This application relates to the field of signal processing and mechanical equipment condition monitoring technology, and provides a method and apparatus for intelligent processing and feature selection of multi-sensor signals. The method first acquires multi-sensor signals, preprocesses them, and then extracts features to obtain a multi-dimensional feature set. Next, it calculates the comprehensive score of each feature in the multi-dimensional feature set and constructs multiple feature domains, dividing each feature in the multi-dimensional feature set into at least one feature domain. Within each feature domain, typical features are determined based on the correlation between each feature and other features and the comprehensive score of the feature itself. Finally, the feature identifiers of the typical features in each feature domain are combined, and feature data for each feature identifier is extracted from the multi-dimensional feature set to obtain the target feature subset. This method achieves the acquisition of a high-quality, low-redundancy optimal feature subset without manual intervention. The selected feature subset provides a more reliable data foundation for subsequent tasks such as condition recognition and fault diagnosis.

Description

Technical Field

[0001] This application relates to the field of signal processing and mechanical equipment condition monitoring technology, and in particular to a method and apparatus for intelligent processing and feature selection of multi-sensor signals. Background Technology

[0002] In the field of mechanical equipment condition monitoring and fault diagnosis, it is typically necessary to deploy various sensors to collect signals reflecting the operating status of the equipment. Common types include accelerometers, vibration sensors, acoustic emission sensors, force sensors, and temperature sensors. Processing the raw signals collected by these sensors and extracting effective signal features is one of the important prerequisites for subsequent condition monitoring and intelligent fault diagnosis.

[0003] Existing technologies have the following limitations in signal preprocessing, feature extraction, and feature selection:

[0004] (1) Fixed signal filtering parameters result in poor adaptability.

[0005] Traditional signal filtering methods such as median filtering and mean filtering use fixed window sizes and fixed thresholds, which cannot adaptively adjust to changes in the local statistical characteristics of the signal. When the signal contains non-stationary components or impulse noise, fixed-parameter filtering can easily lead to over-smoothing of the useful signal or incomplete noise removal, resulting in signal distortion.

[0006] (2) Feature extraction is incomplete and information utilization is low.

[0007] Most existing methods extract only a small number of time-domain statistical features while ignoring frequency-domain or time-frequency-domain features, making it difficult to fully characterize the essential properties of the signal. For multi-sensor data, the features of each channel are usually simply spliced ​​together, lacking in-depth exploration of the signal's intrinsic physical meaning.

[0008] (3) Feature selection is highly subjective and has high redundancy.

[0009] After constructing a high-dimensional feature set, selecting a feature dataset that is highly correlated with and stable to the target variable becomes a challenge. Existing methods often rely on manual experience or simple univariate methods such as correlation coefficient ranking, without considering the redundancy between features or the stability of features under different operating conditions. This makes the selected feature set prone to containing a large amount of redundant information, which not only increases the computational burden but may also reduce the accuracy, computational efficiency, and generalization ability of condition monitoring or fault diagnosis.

[0010] Therefore, there is an urgent need for a method that can adaptively process multi-sensor signals, comprehensively extract multi-domain features, and intelligently select representative features to solve the problems of fixed parameters, one-sided features, and subjective selection in existing technologies. Summary of the Invention

[0011] In view of this, embodiments of this application provide a method and apparatus for intelligent processing and feature selection of multi-sensor signals to solve the problem that in the prior art, the sensor signal processing methods are not comprehensive and rich enough to provide high-quality data for subsequent monitoring and diagnosis in mechanical equipment condition monitoring and fault diagnosis.

[0012] A first aspect of this application provides a method for intelligent processing and feature selection of multi-sensor signals, including:

[0013] Acquire signals from multiple sensors; the signals from multiple sensors include at least one of vibration signals, acoustic emission signals, force signals, and temperature signals;

[0014] The multi-sensor signals are preprocessed using an adaptive switching median filter to obtain the filtered signal.

[0015] The filtered signal is subjected to time-domain feature extraction and frequency-domain feature extraction, and the extracted time-domain features and frequency-domain features are merged to obtain a multi-dimensional feature set;

[0016] Each feature in the multidimensional feature set is comprehensively scored to obtain the overall score of each feature;

[0017] Construct N feature domains, dividing each feature in the multidimensional feature set into at least one feature domain; N is a positive integer greater than 1;

[0018] In each feature domain, typical features are selected based on the correlation between each feature and other features and the comprehensive score of the feature itself;

[0019] The feature identifiers of the typical features obtained from each feature domain are combined into a union. Based on the union, the corresponding feature data are extracted from the multidimensional feature set to form a subset of the target features.

[0020] A second aspect of this application provides a multi-sensor signal intelligent processing and feature selection device, comprising:

[0021] The acquisition module is configured to acquire signals from multiple sensors; the multiple sensor signals include at least one of vibration signals, acoustic emission signals, force signals, and temperature signals.

[0022] The preprocessing module is configured to preprocess the multi-sensor signals using an adaptive switched median filter to obtain the filtered signal.

[0023] The feature extraction module is configured to extract time-domain features and frequency-domain features from the filtered signal, and then merge the extracted time-domain features and frequency-domain features to obtain a multi-dimensional feature set.

[0024] The scoring module is configured to comprehensively score each feature in the multidimensional feature set to obtain a comprehensive score for each feature.

[0025] The extension module is configured to construct N feature domains, dividing each feature in the multidimensional feature set into at least one feature domain; N is a positive integer greater than 1.

[0026] The filtering module is configured to filter typical features in each feature domain based on the correlation between each feature and other features and the comprehensive score of the feature itself.

[0027] The merging module is configured to take the union of the feature identifiers of the typical features obtained from each feature domain, and extract the corresponding feature data from the multidimensional feature set based on the union to form the target feature subset.

[0028] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0029] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0030] The beneficial effects of this application embodiment compared with the prior art are as follows: This application embodiment first acquires multi-sensor signals, preprocesses them, and then extracts features to obtain a multi-dimensional feature set; then, it calculates the comprehensive score of each feature in the multi-dimensional feature set; next, it constructs multiple feature domains, divides each feature in the multi-dimensional feature set into at least one feature domain, and selects typical features in each feature domain based on the correlation between each feature and other features and the comprehensive score of the feature itself; finally, it takes the union of the feature identifiers of the typical features of each feature domain, and then extracts the feature data of each feature identifier from the multi-dimensional feature set to obtain the target feature subset. This achieves the goal of obtaining a high-quality, low-redundancy optimal feature subset without manual intervention. The selected feature subset can provide a more reliable data foundation for subsequent tasks such as state recognition and fault diagnosis. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating a multi-sensor signal intelligent processing and feature selection method provided in an embodiment of this application.

[0033] Figure 2 This is a flowchart illustrating the method for preprocessing multi-sensor signals using an adaptive switched median filter, as provided in an embodiment of this application.

[0034] Figure 3 This is a flowchart illustrating the method for comprehensively scoring each feature in a multidimensional feature set provided in this application embodiment.

[0035] Figure 4 This is a schematic diagram of the process for calculating the stability score of the target feature provided in the embodiments of this application.

[0036] Figure 5 This is a schematic diagram of a multi-sensor signal intelligent processing and feature selection device provided in an embodiment of this application.

[0037] Figure 6 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0038] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0039] The following will describe in detail, with reference to the accompanying drawings, a method and apparatus for intelligent processing and feature selection of multi-sensor signals according to embodiments of this application.

[0040] As mentioned above, in the field of mechanical equipment condition monitoring and fault diagnosis, there are limitations in the preprocessing, feature extraction and feature selection of sensor signals, such as poor adaptability of fixed signal filtering parameters, incomplete feature extraction, low information utilization, and strong subjectivity and redundancy in feature selection.

[0041] In view of this, the embodiments of this application provide a method for intelligent processing and feature selection of multi-sensor signals. By performing adaptive filtering, multi-domain feature extraction and intelligent feature screening on various sensor signals such as vibration, acoustic emission, force and temperature, a high-quality, low-redundancy representative feature subset is obtained, providing a reliable data foundation for subsequent state recognition and fault diagnosis.

[0042] Figure 1 This is a flowchart illustrating a multi-sensor signal intelligent processing and feature selection method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0043] In step S101, multi-sensor signals are acquired.

[0044] The multi-sensor signals include at least one of vibration signals, acoustic emission signals, force signals, and temperature signals.

[0045] In step S102, the multi-sensor signals are preprocessed using an adaptive switching median filter to obtain the filtered signals.

[0046] In step S103, time-domain features and frequency-domain features are extracted from the filtered signal, and the extracted time-domain features and frequency-domain features are merged to obtain a multi-dimensional feature set.

[0047] In step S104, each feature in the multidimensional feature set is comprehensively scored to obtain the comprehensive score of each feature.

[0048] In step S105, N feature domains are constructed, and each feature in the multidimensional feature set is divided into at least one feature domain.

[0049] N is a positive integer greater than 1.

[0050] In step S106, in each feature domain, typical features are selected based on the correlation between each feature and other features and the comprehensive score of the feature itself.

[0051] In step S107, the feature identifiers of the typical features obtained from each feature domain are combined into a union, and the corresponding feature data are extracted from the multidimensional feature set based on the union to form a target feature subset.

[0052] In some embodiments of this application, the method may be executed by a server or by a terminal device with certain processing capabilities.

[0053] In some embodiments of this application, the intelligent processing and feature selection method for multi-sensor signals can be used to process multiple sensor signals used for equipment monitoring and to filter out a target feature subset from the processed signals. This target feature subset can be used as input sample data for equipment condition monitoring or fault diagnosis algorithms. For example, if the equipment is an industrial cutting tool, and the equipment condition monitoring or fault diagnosis task requires prediction of tool wear, then tool wear can be used as the target variable to process and select features from the signals of multiple sensors associated with the tool to obtain the target feature subset required by the tool wear prediction algorithm.

[0054] In some embodiments of this application, multi-sensor signals can be acquired first, and then the multi-sensor signals can be preprocessed using an adaptive switching median filter to obtain filtered signals.

[0055] The multi-sensor signal data source can be files stored on a local disk, with file formats including CSV, Excel, Parquet, etc. In this embodiment, data can be loaded using a file reading function, supporting automatic recognition of multiple file formats. If the data contains multiple channels, each channel is considered an independent signal sequence, denoted as... , where i is the channel index. This is the sampling point number.

[0056] Next, time-domain and frequency-domain features are extracted from the filtered signal. The extracted time-domain and frequency-domain features are then combined to obtain a multi-dimensional feature set. Furthermore, each feature in the multi-dimensional feature set can be comprehensively scored to obtain a comprehensive score for each feature.

[0057] In some embodiments of this application, N feature domains may be constructed, dividing each feature in the multidimensional feature set into at least one feature domain. Then, in each feature domain, typical features are selected based on the correlation between each feature and other features and the comprehensive score of the feature itself.

[0058] Finally, the feature identifiers of the typical features obtained from each feature domain are combined into a union. Based on this union, the corresponding feature data is extracted from the multidimensional feature set to form a subset of the target features. The feature identifier can be a feature name or other identifier; there are no restrictions here.

[0059] According to the technical solution provided in the embodiments of this application, a multi-dimensional feature set is obtained by first acquiring multi-sensor signals, preprocessing them, and then extracting features. Then, the comprehensive score of each feature in the multi-dimensional feature set is calculated. Next, multiple feature domains are constructed, and each feature in the multi-dimensional feature set is divided into at least one feature domain. In each feature domain, typical features are selected based on the correlation between each feature and other features and the comprehensive score of the feature itself. Finally, the feature identifiers of the typical features of each feature domain are combined, and the feature data of each feature identifier is extracted from the multi-dimensional feature set to obtain the target feature subset. This achieves the goal of obtaining a high-quality, low-redundancy optimal feature subset without manual intervention. The selected feature subset can provide a more reliable data foundation for subsequent tasks such as state recognition and fault diagnosis.

[0060] Figure 2 This is a flowchart illustrating a method for preprocessing multi-sensor signals using an adaptive switched median filter, as provided in an embodiment of this application. Figure 2 As shown, the method includes:

[0061] In step S201, the validity of the multi-sensor signals is checked and missing value interpolation is performed.

[0062] In step S202, boundary filling is performed on the interpolated multi-sensor signals.

[0063] In step S203, a sliding window is constructed, and the sliding window is used to extract samples of the multi-sensor signals after boundary filling to obtain multiple signal sample sequences.

[0064] In step S204, the median and median absolute deviation of each signal sample sequence are calculated, and the dynamic threshold is determined based on the median absolute deviation of each signal sample sequence and the preset threshold factor.

[0065] In step S205, all samples in the multi-sensor signal after boundary filling are traversed through the signal sample sequence one by one, and samples whose absolute value of the difference between the original value of the sample and the median of the current signal sample sequence is greater than the dynamic threshold are identified as noise points.

[0066] In step S206, the values ​​of the noise points are replaced with the median of the signal sample sequence to obtain the filtered signal.

[0067] In some embodiments of this application, performing validity checks and missing value interpolation on multi-sensor signals can be done by processing the input multi-sensor signals. Perform an effectiveness check. If the multi-sensor signal length... Less than the preset minimum length If the signal is not a number, the original signal is returned directly; if the multi-sensor signal contains Not a Number (NaN) values, linear interpolation is used to fill them.

[0068] When filling, set the valid value index set as follows: ,Right now This represents the index position of all non-NaN values ​​in the signal, with a corresponding value of For a missing index m, its padding value is... ;in, This represents a linear interpolation function. When interpolating, for a missing index m, linear interpolation only uses... The two elements located before and after m and closest to m. and Perform interpolation calculations.

[0069] Next, to reduce boundary effects, boundary padding can be applied to the interpolated multi-sensor signals. This application supports multiple padding modes: reflect, nearest neighbor, constant (filled with zeros), mirror, and wrap. For aperiodic signals, reflect mode is recommended to avoid introducing spurious transitions. For cyclic signals such as integer-cycle sampling from rotating machinery, wrap mode can be used. Let the window radius be... Where W is the window size (odd number), and the length of the filled signal is... .

[0070] A sliding window can be constructed and adaptive switching median (MAD) filtering calculations can be performed. The window size is set to... (The default value in this application embodiment is 7; if the user inputs an even number, it will be automatically adjusted to an odd number). A sliding window matrix is ​​constructed through vectorization operations, and each window contains consecutive... One sample. For the center point index. ( Its window data is ;in, The signal after padding. Let be the window radius. The median of each window is calculated in parallel. and median absolute deviation .

[0071] For high-frequency sampling signals exceeding 10 kHz, the window can be appropriately increased. A value of 15 or higher yields more stable local statistics. For low-frequency or rapidly changing signals, a smaller window can be used to preserve signal details. The default value of 7 is suitable for most mechanical vibration signals.

[0072] It can also calculate a dynamic threshold based on a preset threshold factor k (default 3.0) and the MAD of each window. ;in, For a very small positive number (such as This is used to prevent the threshold from being zero when MAD is zero.

[0073] The parameter k controls the sensitivity of noise discrimination. A larger k value results in a more lenient discrimination, preserving more original points. A smaller k value filters out noise more effectively, but may smooth out useful signals. The default value can be set to 3.0, based on the 3x MAD criterion, suitable for Gaussian noise environments. For strong impulse noise, it can be reduced to 2.0~2.5.

[0074] Using each center point as a sampling point, iterate through each sampling point n and calculate its original value. With window value The absolute difference. If this difference is greater than the dynamic threshold of the corresponding window. ,Right now If the point is identified as a noise point, it will be replaced with [the point that is not specified in the original text]. Otherwise, retain the original value. Output filtered signal. .

[0075] By employing this method, and by performing on / off processing on each sampling point, median replacement is only performed when necessary, effectively preserving the edge and detail information of the signal while suppressing impulse noise.

[0076] In some embodiments of this application, when performing time-domain feature extraction on the filtered signal, the sampling rate is determined in the following ways: automatically detected from a user-provided file; or, if the data in the filtered signal contains a time series, the average sampling rate is calculated by time intervals; or, a preset global sampling rate is used.

[0077] The time-domain features include at least one of the following: mean, root mean square value, peak value, peak-to-peak value, variance, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin factor, zero-crossing rate, Wilson amplitude, mean absolute value, integral of squares, median absolute deviation, absolute standard deviation of difference, signal rate of change, root mean square amplitude, and mean square value.

[0078] In some embodiments of this application, frequency domain feature extraction of the filtered signal may include: performing a fast Fourier transform on the filtered signal to obtain the signal spectrum; determining the amplitude and corresponding frequency using the one-sided spectrum of the signal spectrum, and calculating the power spectrum; and determining frequency domain features based on at least one of the amplitude, corresponding frequency, and power spectrum.

[0079] The frequency domain features include at least one of the following: spectral centroid, spectral spread, spectral skewness, spectral kurtosis, spectral energy, spectral entropy, maximum spectral amplitude, maximum spectral frequency, median frequency, and frequency variance.

[0080] In other words, when performing time-domain feature extraction, the sampling rate can be determined first. This application's embodiments support multiple methods for determining the sampling rate: 1) automatically detected from user-provided file names; 2) if the data contains time columns, calculating the average sampling rate over time intervals. ;in, 3) Use the user-preset global sampling rate for the time interval.

[0081] Then, for each signal channel, all or some of the time-domain features listed above are calculated in the filtered signal.

[0082] When extracting frequency domain features, a Fast Fourier Transform (FFT) can be performed on the filtered signal to obtain its spectrum. The amplitude and corresponding frequency can be determined using the one-sided spectrum of the signal, and the power spectrum can then be calculated. Based on these frequency domain parameters, all or some of the time domain features listed above can be calculated.

[0083] When merging the extracted time-domain and frequency-domain features, the time-domain and frequency-domain features extracted from each channel can be combined to form a multi-dimensional feature vector. The feature naming convention can be "channel name_feature name" to distinguish different channels. After concatenating the features from all channels, the feature set of the entire data file can be obtained, denoted as a matrix. Where M is the number of samples (which can be a feature vector of a file or a time period), and D is the feature dimension.

[0084] The multidimensional feature set obtained in this way has relatively thin feature dimensions and contains limited information, which is also quite scattered. To enrich the feature dimensions and information contained in the constructed target feature subset, and to make the information more relevant to the target variable, the multidimensional feature set can be expanded in dimension and the features can be filtered. Here, the target variable is the output variable of the algorithm that needs to use the target feature subset.

[0085] In some implementations, each feature in the multidimensional feature set can be comprehensively scored first to calculate the comprehensive score of each feature.

[0086] Figure 3 This is a flowchart illustrating a method for comprehensively scoring each feature in a multi-dimensional feature set, as provided in an embodiment of this application. For example... Figure 3 As shown, the method includes the following steps:

[0087] In step S301, multiple correlation measures between the target features and the target variables are calculated.

[0088] The target feature is any feature in the multidimensional feature set; the multiple correlation measures include at least the Pearson correlation coefficient, Spearman rank correlation coefficient, Kendall rank correlation coefficient, distance correlation coefficient, and maximum information coefficient.

[0089] In step S302, the relevance measures are weighted and summed to obtain the relevance score of the target feature.

[0090] In step S303, the stability score of the target feature is calculated.

[0091] In step S304, the relevance score and stability score of the target feature are weighted and summed to obtain the comprehensive score of the target feature.

[0092] In other words, a relevance score can be calculated for each feature. For each feature... Calculate its relationship with the target variable Multiple correlation measures are used to measure the correlation between features. Multiple correlation values ​​are calculated using all or some of the methods listed above. Each correlation value is normalized to the range [0,1]. Weights are assigned to each method (equal weights by default). Then, the correlation scores of each feature are weighted and summed to obtain the correlation score of each feature.

[0093] Figure 4 This is a schematic diagram of the process for calculating the stability score of the target feature provided in the embodiments of this application. Figure 4 As shown, the method includes the following steps:

[0094] In step S401, the variability index of the target feature across M samples is calculated.

[0095] Among them, the variability index includes at least one of the coefficient of variation, kurtosis, skewness and interquartile range; a sample corresponds to the target feature value extracted from the self-filtered signal within a preset time window; M is a positive integer.

[0096] In step S402, the target feature value sequence is divided into multiple windows according to the time sequence or working condition sequence of the samples. The standard deviation of the target feature value in each window is calculated, and the average of the standard deviations of all windows is obtained to obtain the average window standard deviation.

[0097] In step S403, the ratio of the average window standard deviation to the overall mean of the target feature is determined as the window coefficient of variation, and the window stability score is determined based on the window coefficient of variation.

[0098] In step S404, the variability index and the window stability score are weighted and fused to obtain the stability score of the target feature.

[0099] In other words, we can first calculate the variability index of each feature across M samples. Here, a sample refers to the feature value of a signal segment within a time window. For example, if the time window is 1 second, a segment of cutting force signal can be captured every 1 second. If a time-domain feature and a statistical domain feature are extracted from any segment of the cutting force signal, then two samples can be obtained. In short, the samples here do not refer to sampling points in the original signal, but rather to features extracted from signal unit segments.

[0100] Calculate a feature When scoring window stability, a sample refers to the value of the feature across M different segments. For example, the mean square value feature calculated from multiple vibration signals captured with a recording period of 1 second represents multiple samples of the mean square value feature.

[0101] Then, the window stability index is calculated using a sliding window. The window size is 10% of the sample size (minimum 10), the sliding step is half the window size, and the mean value w for each window is calculated. and standard deviation Standard deviation of all windows Calculating the average yields the average window standard deviation. , Let be the number of windows. At this point, the window variation coefficient... It can be represented as , Features The overall mean over all M samples. Based on this, the window stability index can be calculated. .

[0102] Finally, the variability index and the window stability score are weighted and fused to obtain the stability score of the target feature. ;in, , , , and For the weighting coefficients, satisfying ; This is the coefficient of variation value. Skewness value, This is the kurtosis value. The normalized interquartile range is equal to the quotient of the interquartile range and the maximum value of the target feature in the sample.

[0103] Furthermore, the relevance score and stability score of the target feature are weighted and summed to obtain the comprehensive score of the target feature. The weights for the relevance score and stability score can be set according to actual needs, such as determining them based on the specific application scenario. In one example, the relevance score weight can be set to 0.7 and the stability score weight to 0.3. In another example, if the relationship between the target variable and the feature is clear and the data quality is high, the relevance weight can be increased to 0.8 or higher; if the signal is greatly affected by changes in operating conditions, the stability weight should be increased to 0.4 or higher to ensure the robustness of the selected feature.

[0104] In some embodiments of this application, the feature domain can also be expanded. Features can be categorized into different feature domains based on their physical computational origin, providing a domain-specific screening basis for subsequent selection of representative features within the domain, while also ensuring that the selected feature set has physical diversity in the time domain, frequency domain, and other dimensions.

[0105] In some implementations, N feature domains can be constructed, and the N feature domains include at least a time domain feature domain, a frequency domain feature domain, a time-frequency feature domain, a statistical feature domain, and a nonlinear feature domain.

[0106] Among them, the features in the time domain feature domain are calculated from the original time series signals of multi-sensor signals and are used to reflect the statistical characteristics of the signal on the time axis; the features in the frequency domain feature domain are calculated based on the spectrum or power spectrum obtained by time-frequency transformation and are used to reflect the distribution characteristics of the signal at different frequency components; the features in the time-frequency feature domain are extracted from the signal based on time-frequency analysis methods and are used to reflect the energy distribution or modal characteristics of the signal in the joint time and frequency domains; the time-frequency analysis methods include at least wavelet transform, empirical mode decomposition, and Hilbert-Huang transform; the features in the statistical feature domain are calculated based on higher-order statistics or probability distribution characteristics and are used to reflect the distribution pattern of the signal; the features in the nonlinear feature domain are calculated based on nonlinear dynamics or information wheel methods and are used to reflect the complexity, chaotic characteristics, or information content of the signal.

[0107] In other words, the definition and classification criteria of the feature domain can be given first. Each feature has been assigned a domain label during the extraction stage based on its mathematical calculation source. The embodiments of this application define the following feature domains and their calculation sources:

[0108] Time-domain characteristics: These are calculated directly from the original time-series signal and reflect the statistical properties of the signal on the time axis. They include mean, variance, root mean square, peak value, kurtosis, skewness, waveform factor, and impulse factor.

[0109] Frequency domain characteristics: Calculated based on the spectrum or power spectrum obtained from the Fourier transform, reflecting the distribution characteristics of the signal at different frequency components. These include centroid frequency, frequency variance, root mean square frequency, spectral entropy, and spectral kurtosis.

[0110] Time-frequency domain characteristics: Extracted from signals using time-frequency analysis methods such as wavelet transform, empirical mode decomposition, and Hilbert-Huang transform, reflecting the energy distribution or modal characteristics of the signal in the joint time and frequency domains. These include wavelet packet energy, the energy or statistics of each intrinsic mode function, and instantaneous frequency.

[0111] Statistical domain characteristics: These are calculated based on higher-order statistics or probability distribution characteristics, focusing on the distribution pattern of the signal. They include percentiles, interquartile ranges, histogram statistics, and higher-order moments.

[0112] Nonlinear domain characteristics: These are calculated based on nonlinear dynamics or information theory methods and reflect the complexity, chaotic characteristics, or information content of a signal. They include approximate entropy, sample entropy, fuzzy entropy, permutation entropy, mutual information, fractal dimension, and Lyapunov exponent.

[0113] The definitions of the aforementioned domains are related to the physical meaning of the features, and each feature is uniquely assigned to a feature domain category based on its computational algorithm. Furthermore, if a feature's computational algorithm involves the definitions of multiple domains, its classification is based on its core mathematical theory. For example, entropy-based features, because they are based on information theory or dynamics theory, are classified into the nonlinear domain.

[0114] Then, representative feature selection within the domain is carried out. In each feature domain, the correlation matrix between each feature is calculated; in response to the number of features in the target feature domain exceeding a preset threshold, the typical features of this feature domain are determined by the maximum correlation and minimum redundancy strategy or the diversity selection strategy; the target feature domain can be any feature domain.

[0115] The maximum relevance and minimum redundancy strategy includes: taking the feature with the highest comprehensive score as the first selected feature; among the remaining features, constructing an objective function with the aim of maximizing the relevance between the feature and the target variable and minimizing the redundancy between features; calculating the objective function value of each feature in descending order of comprehensive score; iteratively selecting the feature that maximizes the objective function to add to the selected feature set until the number of features in the selected feature set reaches the preset number of features or all features have been traversed.

[0116] The diversity selection strategy includes: sorting the comprehensive scores of each feature from high to low; sequentially determining whether the correlation between the current feature and the selected features is lower than a preset correlation threshold, and if so, selecting it as a typical feature. Here, the selected features refer to all selected features in the selected feature set.

[0117] In other words, within each feature domain, the correlation matrix between features can be used to determine the relationship between features. (element ) and overall score Q j’ Select representative features; among them and All are feature indices within the feature domain. The specific method is as follows: if the number of features in the domain does not exceed a preset maximum value, all features are retained; otherwise, the maximum relevance minimum redundancy (mRMR) strategy or a diversity selection strategy is adopted.

[0118] The mRMR strategy selects the feature with the highest overall score, and then iteratively selects features with low average relevance to the selected features and high overall scores. The objective function is: The mRMR strategy maximizes the relevance between features and the objective while minimizing feature redundancy. It selects the feature with the highest overall score as the first selected feature, and then iteratively selects features that maximize the relevance to the objective function. Maximizing features Represents the set of remaining features. Features The overall score, where S is the set of selected features. Features and absolute value of the correlation coefficient This is the balance coefficient (default value is 1).

[0119] The first term of the objective function This ensures that the selected features have a high relevance to the target; the second term of the objective function penalizes the average relevance with the selected features, reducing feature redundancy. Iterative optimization then yields a subset of features that are both highly relevant to the target and have low redundancy among themselves.

[0120] The diversity strategy sorts features by their overall score and selects features whose relevance to the selected features is below a threshold θ (default 0.8). If the number of features is insufficient, the highest-scoring unselected feature is added. The threshold θ is typically set between 0.8 and 0.9. A lower threshold requires higher independence between features and fewer features are selected. A higher threshold allows for greater redundancy.

[0121] The aforementioned selection of representative features within the domain helps avoid the loss of information in other physical dimensions due to human subjectivity or model-induced bias towards a particular type of feature, thus improving the interpretability and robustness of the feature set. Furthermore, the selection process, utilizing the comprehensive score of features and the correlation between features within the domain, is characterized by its simplicity and interpretability of physical meaning.

[0122] After completing the selection of typical features, the selected typical feature data for each feature domain can be saved as a file. A detailed selection report is also generated, recording the total number of original features, the number of features after selection, the distribution of each feature domain, the overall score and sub-scores for each selected feature, and metadata about the selection process (including parameters such as weights and thresholds used). The report is saved in text and JSON formats for easy reference later.

[0123] The technical solution provided in this application obtains a comprehensive correlation score by calculating and weighting the Pearson, Spearman, Kendall, distance correlation coefficients and maximum information coefficients between each feature and the target variable. This overcomes the limitations of a single correlation index in measuring nonlinear and non-monotonic relationships and comprehensively evaluates the degree of association between features and the target variable.

[0124] A feature stability assessment is proposed, which calculates variability indices such as coefficient of variation, kurtosis, skewness, and interquartile range of feature values, and combines them with the window stability index of the sliding window to obtain a stability score. Based on the score, features that remain stable under different working conditions and time periods can be effectively identified, thereby improving the robustness of the selected feature subset.

[0125] The system automatically identifies features and divides them into time domain, frequency domain, time-frequency domain, statistical domain, and nonlinear domain. Within each domain, redundancy removal is performed based on the correlation matrix and comprehensive score between features. This not only preserves the representative features of each domain and maintains the diversity of the feature set, but also reduces the redundancy between features.

[0126] The feature selection process is integrated into an automated workflow, from preprocessing and multi-domain feature extraction to intelligent selection, obtaining a high-quality, low-redundancy optimal feature subset without manual intervention. The selected feature subset provides a more reliable data foundation for subsequent tasks such as state recognition and fault diagnosis.

[0127] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0128] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0129] Figure 5 This is a schematic diagram of a multi-sensor signal intelligent processing and feature selection device provided in an embodiment of this application. Figure 5 As shown, the device includes:

[0130] The acquisition module 501 is configured to acquire multi-sensor signals; the multi-sensor signals include at least one of vibration signals, acoustic emission signals, force signals, and temperature signals.

[0131] The preprocessing module 502 is configured to preprocess the multi-sensor signals using an adaptive switching median filter to obtain the filtered signal.

[0132] The feature extraction module 503 is configured to extract time-domain features and frequency-domain features from the filtered signal, and merge the extracted time-domain features and frequency-domain features to obtain a multi-dimensional feature set.

[0133] The scoring module 504 is configured to perform a comprehensive score on each feature in the multidimensional feature set to obtain a comprehensive score for each feature.

[0134] Extension module 505 is configured to construct N feature domains, dividing each feature in the multidimensional feature set into at least one feature domain; N is a positive integer greater than 1.

[0135] The filtering module 506 is configured to filter typical features in each feature domain based on the correlation between each feature and other features and the comprehensive score of the feature itself.

[0136] The merging module 507 is configured to take the union of the feature identifiers of the typical features obtained from each feature domain, and extract the corresponding feature data from the multidimensional feature set based on the union to form a target feature subset.

[0137] According to the technical solution provided in the embodiments of this application, a multi-dimensional feature set is obtained by first acquiring multi-sensor signals, preprocessing them, and then extracting features. Then, the comprehensive score of each feature in the multi-dimensional feature set is calculated. Next, multiple feature domains are constructed, and each feature in the multi-dimensional feature set is divided into at least one feature domain. In each feature domain, typical features are selected based on the correlation between each feature and other features and the comprehensive score of the feature itself. Finally, the feature identifiers of the typical features of each feature domain are combined, and the feature data of each feature identifier is extracted from the multi-dimensional feature set to obtain the target feature subset. This achieves the goal of obtaining a high-quality, low-redundancy optimal feature subset without manual intervention. The selected feature subset can provide a more reliable data foundation for subsequent tasks such as state recognition and fault diagnosis.

[0138] In some implementations, the multi-sensor signals are preprocessed using an adaptive switching median filter, including: performing validity checks and missing value interpolation on the multi-sensor signals; filling the boundaries of the interpolated multi-sensor signals; constructing a sliding window and using the sliding window to extract samples one by one from the boundary-filled multi-sensor signals to obtain multiple signal sample sequences; calculating the median and median absolute deviation of each signal sample sequence, and determining a dynamic threshold based on the median absolute deviation of each signal sample sequence and a preset threshold factor; traversing all samples in the boundary-filled multi-sensor signals one by one, and identifying samples whose absolute difference between the original value and the median of the current signal sample sequence is greater than the dynamic threshold as noise points; replacing the values ​​of the noise points with the median of the current signal sample sequence to obtain the filtered signal.

[0139] In some implementations, when extracting time-domain features from the filtered signal, the sampling rate is determined in the following ways: automatically detected from a user-provided file; or, if the data in the filtered signal contains a time series, the average sampling rate is calculated by time intervals; or, a preset global sampling rate is used. The time-domain features include at least one of the following: mean, root mean square value, peak value, peak-to-peak value, variance, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin factor, zero-crossing rate, Wilson amplitude, mean absolute value, integral of squares, median absolute deviation, absolute standard deviation of difference, rate of change of signal, root square amplitude, and mean square value.

[0140] In some implementations, frequency domain feature extraction of the filtered signal includes: performing a fast Fourier transform on the filtered signal to obtain the signal spectrum; determining the amplitude and corresponding frequency using the one-sided spectrum of the signal spectrum, and calculating the power spectrum; determining frequency domain features based on at least one of the amplitude, corresponding frequency, and power spectrum; the frequency domain features include at least one of the following: spectral centroid, spectral spread, spectral skewness, spectral kurtosis, spectral energy, spectral entropy, maximum spectral amplitude, maximum spectral frequency, median frequency, and frequency variance.

[0141] In some implementations, a comprehensive score is given for each feature in the multidimensional feature set, including: calculating multiple correlation measures between the target feature and the target variable; wherein the target feature is any feature in the multidimensional feature set; the multiple correlation measures include at least Pearson correlation coefficient, Spearman rank correlation coefficient, Kendall rank correlation coefficient, distance correlation coefficient, and maximum information coefficient; weighted summation of each correlation measure to obtain the correlation score of the target feature; calculation of the stability score of the target feature; and weighted summation of the correlation score and stability score of the target feature to obtain the comprehensive score of the target feature.

[0142] In some implementations, calculating the stability score of the target feature includes: calculating the variability index of the target feature over M samples; the variability index includes at least one of coefficient of variation, kurtosis, skewness, and interquartile range; extracting the target feature value from the self-filtered signal within a preset time window corresponding to a sample; M is a positive integer; dividing the target feature value sequence into multiple windows using a sliding window according to the time order or working condition order of the samples, calculating the standard deviation of the target feature value within each window, and averaging the standard deviations of all windows to obtain the average window standard deviation; determining the ratio of the average window standard deviation to the overall mean of the target feature as the window coefficient of variation, and determining the window stability score based on the window coefficient of variation; and weightedly fusing the variability index and the window stability score to obtain the stability score of the target feature.

[0143] In some implementations, the N feature domains include at least a time-domain feature domain, a frequency-domain feature domain, a time-frequency feature domain, a statistical feature domain, and a nonlinear feature domain. Features in the time-domain feature domain are calculated from the original time-series signal of the multi-sensor signal and are used to reflect the statistical characteristics of the signal on the time axis. Features in the frequency-domain feature domain are calculated based on the spectrum or power spectrum obtained from the time-frequency transform and are used to reflect the distribution characteristics of the signal at different frequency components. Features in the time-frequency feature domain are extracted from the signal using time-frequency analysis methods and are used to reflect the energy distribution or modal characteristics of the signal in the joint time and frequency domains. Time-frequency analysis methods include at least wavelet transform, empirical mode decomposition, and Hilbert-Huang transform. Features in the statistical feature domain are calculated based on higher-order statistics or probability distribution characteristics and are used to reflect the distribution pattern of the signal. Features in the nonlinear feature domain are calculated based on nonlinear dynamics or information wheel methods and are used to reflect the complexity, chaotic characteristics, or information content of the signal.

[0144] In some implementations, typical features are selected within each feature domain based on the correlation between each feature and other features and the comprehensive score of the feature itself, including:

[0145] In each feature domain, calculate the correlation matrix between the features;

[0146] In response to the determination that the number of features in the target feature domain exceeds a preset threshold, a maximum relevance and minimum redundancy strategy or a diversity selection strategy is adopted to determine the typical features of this feature domain; the target feature domain can be any feature domain; wherein, the maximum relevance and minimum redundancy strategy includes: taking the feature with the highest comprehensive score as the first selected feature; among the remaining features, constructing an objective function with the aim of maximizing the relevance between the feature and the target variable and minimizing the redundancy between features, calculating the objective function value of each feature in descending order of comprehensive score, iteratively selecting the feature that maximizes the objective function to add to the selected feature set, until the number of features in the selected feature set reaches the preset number of features or all features have been traversed; the diversity selection strategy includes: sorting the comprehensive scores of each feature in descending order; sequentially judging whether the relevance between the current feature and the selected features is lower than a preset relevance threshold, and if so, selecting it as a typical feature.

[0147] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0148] Figure 6 This is a schematic diagram of the electronic device provided in an embodiment of this application. For example... Figure 6As shown, the electronic device 6 of this embodiment includes a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601. When the processor 601 executes the computer program 603, it implements the steps in the various method embodiments described above. Alternatively, when the processor 601 executes the computer program 603, it implements the functions of each module / unit in the various device embodiments described above.

[0149] Electronic device 6 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 6 may include, but is not limited to, processor 601 and memory 602. Those skilled in the art will understand that... Figure 6 This is merely an example of electronic device 6 and does not constitute a limitation on electronic device 6. It may include more or fewer components than shown, or different components.

[0150] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0151] The memory 602 can be an internal storage unit of the electronic device 6, such as a hard disk or RAM of the electronic device 6. The memory 602 can also be an external storage device of the electronic device 6, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 6. The memory 602 can also include both internal and external storage units of the electronic device 6. The memory 602 is used to store computer programs and other programs and data required by the electronic device.

[0152] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0153] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0154] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for intelligent processing and feature selection of multi-sensor signals, characterized in that, include: Acquire signals from multiple sensors; the multiple sensor signals include at least one of vibration signals, acoustic emission signals, force signals, and temperature signals; The multi-sensor signals are preprocessed using an adaptive switching median filter to obtain the filtered signal. The filtered signal is subjected to time-domain feature extraction and frequency-domain feature extraction, and the extracted time-domain features and frequency-domain features are merged to obtain a multi-dimensional feature set; Each feature in the multidimensional feature set is comprehensively scored to obtain a comprehensive score for each feature; Construct N feature domains, and divide each feature in the multidimensional feature set into at least one feature domain; N is a positive integer greater than 1; In each feature domain, typical features are selected based on the correlation between each feature and other features and the comprehensive score of the feature itself; The feature identifiers of the typical features obtained from each feature domain are taken as a union, and the corresponding feature data are extracted from the multidimensional feature set according to the union to form a target feature subset.

2. The multi-sensor signal intelligent processing and feature selection method according to claim 1, characterized in that, The multi-sensor signals are preprocessed using an adaptive switched median filter, including: The multi-sensor signals are subjected to validity checks and missing value interpolation. Boundary filling is performed on the interpolated multi-sensor signals; Construct a sliding window and use it to extract sample-by-sample signals from the multi-sensor signals after boundary filling, resulting in multiple signal sample sequences. Calculate the median and median absolute deviation for each signal sample sequence, and determine the dynamic threshold based on the median absolute deviation of each signal sample sequence and the preset threshold factor; The sample sequence is traversed one by one through all samples in the multi-sensor signal after boundary filling, and samples whose absolute value of the difference between the original value of the sample and the median of the current signal sample sequence is greater than the dynamic threshold are identified as noise points. The filtered signal is obtained by replacing the values ​​of the noise points with the median of the sample sequence of this signal.

3. The multi-sensor signal intelligent processing and feature selection method according to claim 1, characterized in that, When performing time-domain feature extraction on the filtered signal, the sampling rate is determined as follows: Automatically detected from user-provided files; Alternatively, if the data in the filtered signal contains a time series, the average sampling rate can be calculated using the time intervals. Alternatively, use the preset global sampling rate; The time-domain features include at least one of the following: mean, root mean square value, peak value, peak-to-peak value, variance, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin factor, zero-crossing rate, Wilson amplitude, mean absolute value, integral of squares, median absolute deviation, absolute standard deviation of difference, signal rate of change, root mean square amplitude, and mean square value.

4. The multi-sensor signal intelligent processing and feature selection method according to claim 1, characterized in that, Frequency domain feature extraction of the filtered signal includes: Perform a Fast Fourier Transform on the filtered signal to obtain the signal spectrum; The amplitude and corresponding frequency are determined using the one-sided spectrum of the signal spectrum, and the power spectrum is calculated. The frequency domain characteristics are determined based on at least one of the amplitude, corresponding frequency, and power spectrum. The frequency domain features include at least one of the following: spectral centroid, spectral spread, spectral skewness, spectral kurtosis, spectral energy, spectral entropy, maximum spectral amplitude, maximum spectral frequency, median frequency, and frequency variance.

5. The multi-sensor signal intelligent processing and feature selection method according to claim 1, characterized in that, A comprehensive score is given for each feature in the multidimensional feature set, including: Calculate multiple correlation measures between target features and target variables; where the target feature is any feature in a multidimensional feature set; the multiple correlation measures include at least Pearson correlation coefficient, Spearman rank correlation coefficient, Kendall rank correlation coefficient, distance correlation coefficient, and maximum information coefficient. The relevance scores of the target features are obtained by weighted summation of the various relevance measures. Calculate the stability score of the target features; The relevance score and stability score of the target feature are weighted and summed to obtain the comprehensive score of the target feature.

6. The intelligent processing and feature selection method for multi-sensor signals according to claim 5, characterized in that, Calculate the stability score of the target features, including: Calculate the variability index of the target feature across M samples; the variability index includes at least one of coefficient of variation, kurtosis, skewness, and interquartile range; a sample corresponds to the target feature value extracted from the self-filtered signal within a preset time window; M is a positive integer; According to the time sequence or working condition sequence of the samples, the target feature value sequence is divided into multiple windows using a sliding window. The standard deviation of the target feature value in each window is calculated, and the average of the standard deviations of all windows is obtained to get the average window standard deviation. The ratio of the average window standard deviation to the overall mean of the target feature is determined as the window coefficient of variation, and the window stability score is determined based on the window coefficient of variation. The stability score of the target feature is obtained by weighted fusion of the variability index and the window stability score.

7. The multi-sensor signal intelligent processing and feature selection method according to claim 1, characterized in that, The N feature domains include at least time-domain feature domains, frequency-domain feature domains, time-frequency feature domains, statistical feature domains, and nonlinear feature domains; The features in the time-domain feature domain are calculated from the original time-series signals of the multi-sensor signals and are used to reflect the statistical characteristics of the signals on the time axis. The features in the frequency domain feature domain are calculated based on the spectrum or power spectrum obtained by time-frequency transformation, and are used to reflect the distribution characteristics of the signal at different frequency components. The features in the time-frequency feature domain are extracted from the signal based on time-frequency analysis methods and are used to reflect the energy distribution or modal characteristics of the signal in the joint time and frequency domains; the time-frequency analysis methods include at least wavelet transform, empirical mode decomposition, and Hilbert-Huang transform. The features of the statistical feature domain are calculated based on higher-order statistics or probability distribution characteristics, and are used to reflect the distribution pattern of the signal. The features of the nonlinear feature domain are calculated based on nonlinear dynamics or information wheel methods, and are used to reflect the complexity, chaotic characteristics, or information content of the signal.

8. The intelligent processing and feature selection method for multi-sensor signals according to claim 1, characterized in that, Within each feature domain, typical features are selected based on the correlation between each feature and other features, as well as the overall score of the feature itself. These include: In each feature domain, calculate the correlation matrix between the features; In response to the number of features in the target feature domain exceeding a preset threshold, the typical features of this feature domain are determined using either the maximum relevance and minimum redundancy strategy or the diversity selection strategy; the target feature domain can be any feature domain. The maximum correlation and minimum redundancy strategy includes: The feature with the highest overall score will be selected as the first feature. Among the remaining features, an objective function is constructed with the aim of maximizing the correlation between the feature and the target variable and minimizing the redundancy between features. The objective function values ​​of each feature are calculated in descending order of the comprehensive score. The feature that maximizes the objective function is iteratively selected and added to the selected feature set until the number of features in the selected feature set reaches the preset number of features or all features have been traversed. The diversity selection strategy includes: Sort the scores of each feature from highest to lowest. The system sequentially determines whether the correlation between the current feature and the selected features is lower than a preset correlation threshold. If so, it selects the current feature as a typical feature.

9. A multi-sensor signal intelligent processing and feature selection device, characterized in that, include: The acquisition module is configured to acquire multi-sensor signals; the multi-sensor signals include at least one of vibration signals, acoustic emission signals, force signals, and temperature signals. The preprocessing module is configured to preprocess the multi-sensor signals using an adaptive switched median filter to obtain a filtered signal. The feature extraction module is configured to perform time-domain feature extraction and frequency-domain feature extraction on the filtered signal, and merge the extracted time-domain features and frequency-domain features to obtain a multi-dimensional feature set; The scoring module is configured to perform a comprehensive score on each feature in the multidimensional feature set to obtain a comprehensive score for each feature. The extension module is configured to construct N feature domains, dividing each feature in the multidimensional feature set into at least one feature domain; N is a positive integer greater than 1. The filtering module is configured to filter typical features in each feature domain based on the correlation between each feature and other features and the comprehensive score of the feature itself. The merging module is configured to take the union of the feature identifiers of the typical features obtained from each feature domain screening, and extract the corresponding feature data from the multidimensional feature set according to the union to form a target feature subset.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-sensor signal intelligent processing and feature selection method as described in any one of claims 1 to 8.

Images