BCG signal processing system based on spectral characteristics and computer program product

Abstract

The invention relates to the technical field of BCG signal processing, in particular to a BCG signal processing system based on spectral characteristics and a computer program, and the system comprises a preprocessing module which is used for preprocessing a received original signal; the spectrum feature extraction module is used for converting the preprocessed original signal into a time-frequency domain signal according to a short-time Fourier transform algorithm; performing feature extraction on the time-frequency domain signals according to a feature extraction algorithm to obtain multi-spectrum features; the signal quality classification module is used for inputting the multi-spectrum features into a preset signal quality classification model to obtain a corresponding quality classification result; and the signal optimization module is used for optimizing the multi-spectrum features according to the quality classification result and a preset signal optimization strategy to obtain an optimization signal for assisting health analysis. The method has the beneficial effects that the method is not easily interfered by environmental vibration, the real-time performance is high, and meanwhile, the robustness and the accuracy of the BCG signal are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of BCG signal processing, and in particular to a BCG signal processing system and computer program product based on spectrum characteristics. Background Art

[0002] The BCG (Ballistocardiography) signal is a physiological signal that reflects the tiny displacements of the human body caused by the contraction and relaxation of the heart. It offers the advantages of non-invasiveness, portability, and sustainable monitoring. In recent years, it has shown broad application prospects in areas such as home health monitoring, sleep quality assessment, and early screening for cardiovascular disease. Currently, BCG signal processing methods primarily include traditional time-domain analysis, frequency-domain analysis, and time-frequency analysis. Time-domain analysis primarily extracts BCG waveform features through filtering and threshold detection. Frequency-domain analysis uses methods such as Fourier transforms to study the frequency composition of the signal. Time-frequency analysis combines methods such as wavelet transforms to perform multi-scale analysis of the signal.

[0003] However, these traditional methods face challenges in practical applications, such as signal acquisition being susceptible to interference from environmental vibrations, significant individual differences leading to the failure of fixed thresholds, inadequate feature extraction, difficulty meeting real-time requirements, and insufficient noise immunity. Therefore, a new method for effectively processing BCG signals is urgently needed to improve their accuracy and reliability in real-world health monitoring. Summary of the Invention

[0004] (1) Technical issues to be resolved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a BCG signal processing system and computer program product based on spectral characteristics, which solve the technical problems that the existing BCG signal processing method is easily affected by environmental vibration interference, individual differences lead to the failure of fixed thresholds, feature extraction is insufficient, real-time requirements are difficult to meet, and anti-noise ability is insufficient.

[0006] (2) Technical solution

[0007] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, an embodiment of the present invention provides a BCG signal processing system based on spectrum characteristics, including:

[0009] A preprocessing module, used for preprocessing the received original signal;

[0010] The spectrum feature extraction module converts the pre-processed original signal into a time-frequency domain signal according to the preset short-time Fourier transform algorithm;

[0011] and extracting features from the time-frequency domain signal according to a preset feature extraction algorithm to obtain corresponding multi-spectral features;

[0012] The feature extraction algorithm includes short wavelet packet decomposition algorithm, fast Fourier transform algorithm and information entropy transform algorithm;

[0013] A signal quality classification module, configured to input the multi-spectral features into a preset signal quality classification model to obtain a corresponding quality classification result;

[0014] a signal optimization module, configured to optimize the multi-spectral features according to the quality classification results and a preset signal optimization strategy to obtain an optimized signal for assisting health analysis;

[0015] The signal quality classification module includes a classification model obtained by training a pre-built deep learning model through multi-spectral features of known quality classification results.

[0016] Optionally, the preprocessing includes: high-frequency denoising, adaptive segmentation and signal normalization;

[0017] The preprocessing module preprocesses the received original signal, including:

[0018] According to the pre-set wavelet threshold denoising algorithm, the received original signal is subjected to high-frequency denoising processing;

[0019] According to the preset signal statistical algorithm, the original signal after high-frequency denoising is adaptively segmented;

[0020] Perform signal normalization on the original signal after adaptive segmentation processing.

[0021] Optionally, the spectrum feature extraction module extracts features from the time-frequency domain signal according to a preset feature extraction algorithm to obtain corresponding multi-spectrum features, including:

[0022] Extracting energy distribution characteristics of the time-frequency domain signal in different frequency bands according to a preset short wavelet packet decomposition algorithm;

[0023] Obtaining corresponding power spectral density features according to the time-frequency domain signal and a preset fast Fourier transform algorithm;

[0024] According to the energy distribution characteristics and power spectrum density characteristics, as well as the pre-set information entropy transformation algorithm, the corresponding spectrum entropy characteristics are obtained;

[0025] The multi-spectral features include energy distribution features, power spectrum density features and spectrum entropy features.

[0026] Optionally, the signal quality classification model includes: a local feature extraction sub-model, a temporal feature extraction sub-model and a feature fusion classification sub-model;

[0027] The signal quality classification module inputs the multi-spectral features into a preset signal quality classification model to obtain a corresponding quality classification result, including:

[0028] Inputting the multi-spectral features into a pre-set local feature extraction sub-model and a temporal feature extraction sub-model respectively, and obtaining local features and temporal features corresponding to the multi-spectral features respectively; the local features and temporal features both include corresponding feature dimensions and sequence lengths;

[0029] Inputting the local features and temporal features into a pre-set feature fusion classification sub-model respectively to obtain corresponding quality classification results;

[0030] The local feature extraction sub-model includes a feature extraction model obtained by training a CNN network using multi-spectral features of known local features;

[0031] The time series feature extraction sub-model includes a feature extraction model obtained by training the LSTM network using multi-spectral features of known time series features;

[0032] The feature fusion classification sub-model includes a classification model obtained by training a transformer network using local features and temporal features corresponding to the same multi-spectral features with known quality classification results.

[0033] Optionally, the feature fusion classification sub-model is used to:

[0034] According to the local features, the temporal features and the preset attention mechanism, a local attention matrix corresponding to the local features and a temporal attention matrix corresponding to the temporal features are obtained respectively;

[0035] According to the local attention matrix, the temporal attention matrix, the local features and the temporal features, the fusion features corresponding to the multi-spectral features are obtained;

[0036] According to the fusion feature and the preset attention score weight matrix, and the preset formula 1, the attention weight corresponding to the fusion feature is obtained; the formula 1 is:

[0037] S=Softmax(CW score );

[0038] Among them, S is the attention weight corresponding to the fusion feature, C is the fusion feature, and W score is the pre-set attention score weight matrix, and Softmax is the normalization function;

[0039] According to the fusion feature and the attention weight corresponding to the fusion feature, and the preset formula 2, the fusion feature vector corresponding to the fusion feature is obtained; the formula 2 is:

[0040] c = SC;

[0041] Where c is the fusion feature vector;

[0042] The fused feature vector corresponding to the fused feature is input into the preset first fully connected layer, and the quality classification result corresponding to the multi-spectral feature is obtained according to the preset formula 3; the formula 3 is:

[0043] Y=W f ·dropout[tanh(W p c)];

[0044] Among them, Y is the quality classification result corresponding to the multi-spectral feature, W f and W p These are the preset weights of the first fully connected layer.

[0045] Optionally, the feature fusion classification sub-model obtains fusion features corresponding to the multi-spectral features according to the local attention matrix, the temporal attention matrix, the local features and the temporal features, including:

[0046] Inputting the local attention matrix and the temporal attention matrix into a preset second fully connected layer for feature transformation respectively;

[0047] Performing a residual connection between the local attention matrix after feature transformation and the local features to obtain a corresponding local feature residual connection result;

[0048] Performing a residual connection between the temporal attention matrix after feature transformation and the temporal features to obtain a corresponding temporal feature residual connection result;

[0049] The local feature residual connection results and the temporal feature residual connection results are concatenated to obtain the corresponding fusion features.

[0050] Optionally, the feature fusion classification sub-model performs feature concatenation on the residual connection results of the local features and the residual connection results of the temporal features to obtain corresponding fusion features, including:

[0051] According to the preset formula 4, the local feature residual connection result and the temporal feature residual connection result are feature spliced to obtain the corresponding fusion feature; the formula 4 is:

[0052] C = concat(L1, L2, dim = n);

[0053] Among them, C is the corresponding fusion feature, concat is the splicing operation, L1 is the local feature residual connection result, L2 is the temporal feature residual connection result, and dim=n is the pre-set feature dimension.

[0054] Optionally, the attention mechanism includes: a spatial attention mechanism and a self-attention mechanism;

[0055] Then, the feature fusion classification sub-model obtains the local attention matrix corresponding to the local features and the temporal attention matrix corresponding to the temporal features according to the local features, temporal features and the preset attention mechanism, including:

[0056] According to the local features and a preset self-attention mechanism, a local attention matrix corresponding to the local features is obtained;

[0057] According to the temporal features and the preset temporal attention mechanism, a temporal attention matrix corresponding to the temporal features is obtained.

[0058] Optionally, the signal optimization strategy includes a mapping relationship table between quality classification results and multi-spectral feature optimization processing algorithms;

[0059] The signal optimization module optimizes the multi-spectral features according to the quality classification results and the preset signal optimization strategy to obtain an optimized signal for assisting health analysis, including:

[0060] According to the quality classification result and the preset signal optimization strategy, a corresponding optimization processing algorithm is selected to optimize the multi-spectral features to obtain an optimized signal for assisting health analysis;

[0061] The optimization processing algorithm includes one or more of adaptive filtering, wavelet denoising, data interpolation, adversarial network, feature enhancement, time alignment, time synchronization, and resampling.

[0062] In a second aspect, an embodiment of the present invention provides a computer program product, including a computer program, which implements the above-mentioned BCG signal processing system based on spectrum characteristics when executed by a processor.

[0063] (3) Beneficial effects

[0064] The beneficial effects of the present invention are as follows: a BCG signal processing system based on spectral characteristics of the present invention suppresses the signal noise of the original signal by using preprocessing and feature extraction, analyzes the signal quality through a signal quality classification model, and optimizes the signal based on the classification results. Compared with the existing technology, it is not easily affected by environmental vibration interference, has strong real-time performance, and significantly improves the robustness and accuracy of the BCG signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A schematic diagram of the structure of a BCG signal processing system based on spectrum characteristics provided by an embodiment of the present invention;

[0066] Figure 2 A flowchart of a BCG signal processing method based on spectrum characteristics is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0068] A BCG signal processing system based on spectral characteristics proposed in an embodiment of the present invention suppresses the signal noise of the original signal through preprocessing and feature extraction, analyzes the signal quality through a signal quality classification model, and optimizes the signal based on the classification results. Compared with the existing technology, it is not easily affected by environmental vibration interference, has strong real-time performance, and significantly improves the robustness and accuracy of the BCG signal.

[0069] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0070] Example 1

[0071] This embodiment provides a BCG signal processing system based on spectrum characteristics, such as Figure 1 Shown, including:

[0072] A preprocessing module, used for preprocessing the received original signal;

[0073] The spectrum feature extraction module converts the pre-processed original signal into a time-frequency domain signal according to the preset short-time Fourier transform algorithm;

[0074] and extracting features from the time-frequency domain signal according to a preset feature extraction algorithm to obtain corresponding multi-spectral features;

[0075] The feature extraction algorithm includes short wavelet packet decomposition algorithm, fast Fourier transform algorithm and information entropy transform algorithm;

[0076] A signal quality classification module, configured to input the multi-spectral features into a preset signal quality classification model to obtain a corresponding quality classification result;

[0077] a signal optimization module, configured to optimize the multi-spectral features according to the quality classification results and a preset signal optimization strategy to obtain an optimized signal for assisting health analysis;

[0078] The signal quality classification module includes a classification model obtained by training a pre-built deep learning model through multi-spectral features of known quality classification results.

[0079] A BCG signal processing system based on spectral characteristics proposed in an embodiment of the present invention suppresses the signal noise of the original signal through preprocessing and feature extraction, analyzes the signal quality through a signal quality classification model, and optimizes the signal based on the classification results. Compared with the existing technology, it is not easily affected by environmental vibration interference, has strong real-time performance, and significantly improves the robustness and accuracy of the BCG signal.

[0080] Example 2

[0081] This embodiment provides a BCG signal processing method based on spectrum characteristics, such as Figure 2 Shown, including:

[0082] First, the collected BCG raw signal undergoes preprocessing, including signal noise reduction, segmentation, and normalization, to improve the reliability of subsequent processing. Specifically, wavelet threshold denoising is used to remove high-frequency noise, and adaptive segmentation is performed based on the statistical characteristics of the signal. The signal is normalized to the [0, 1] interval.

[0083] Secondly, multi-spectral feature extraction is performed. The original BCG signal is converted to the time-frequency domain through the short-time Fourier transform (STFT) to obtain the corresponding time-frequency domain signal. Combined with the wavelet packet decomposition algorithm, the energy distribution characteristics of the time-frequency domain signal in different frequency bands are extracted. At the same time, statistical features such as the frequency density (PSD) and spectral entropy of the signal are calculated to form multi-spectral features.

[0084] When performing wavelet packet decomposition, the appropriate wavelet basis is first selected based on the time-frequency domain signal. Different wavelet bases are suitable for different types of data. The signal is then decomposed into different layers. Each layer further divides the signal's frequency range. The wavelet packet decomposition of the time-frequency domain signal is performed using the selected wavelet basis and decomposition level.

[0085] Energy distribution feature extraction: At each wavelet packet node, the energy of the corresponding frequency band can be calculated. Typically, energy can be obtained by calculating the square sum of all coefficients within the frequency band. The energy distribution within each frequency band is statistically analyzed as one of the signal characteristics.

[0086] Power spectral density (PSD) can be calculated using the Fast Fourier Transform (FFT). The PSD shows how the signal power varies with frequency, revealing which frequency components dominate the signal. Common methods include the Welch method, which improves the stability of PSD estimation through piecewise averaging.

[0087] Spectral entropy calculation: Spectral entropy is a measure of spectral complexity, calculated based on the PSD or energy distribution. It reflects the uncertainty or complexity of the signal spectrum. To calculate it, the PSD or energy distribution must first be normalized to form a probability density function. The spectral entropy is then calculated using the information entropy formula.

[0088] The energy distribution, PSD value, spectrum entropy and other features of different frequency bands obtained in the above steps are combined to form a multi-spectral feature.

[0089] Subsequently, a deep learning network structure was designed, employing a hybrid architecture of convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) to automatically learn and classify the spectral characteristics of BCG signals. CNNs are used to extract local features (i.e., local spectral features), while LSTMs are used to capture the temporal characteristics of signals (i.e., temporal dependencies).

[0090] Finally, the introduction of an attention mechanism and residual connections enhances the model's ability to learn key features for classification and evaluation of signal quality. A cross-validation strategy is employed to optimize model parameters and improve the algorithm's generalization performance. An adaptive learning rate and early stopping strategy are employed to ensure the stability and efficiency of model training.

[0091] The multi-spectral features are optimized through the quality classification results and a pre-set signal optimization strategy to obtain an optimized signal for use in assisting health analysis.

[0092] This embodiment provides a BCG signal processing method based on spectral characteristics, which improves the comprehensiveness and effectiveness of feature extraction by using multi-spectral features; significantly enhances the anti-interference ability of the algorithm by leveraging the adaptive characteristics of the deep learning model; utilizes a hybrid network structure to achieve intelligent adaptation to individual differences; and ensures the stability and reliability of the model in practical applications through an attention mechanism and optimization strategy.

[0093] Subsequent experimental results show that the processing method of this embodiment has achieved significant improvements in signal feature extraction accuracy, anti-interference ability and processing precision compared with traditional methods, providing a new technical path for the intelligent processing of BCG signals.

[0094] Example 3

[0095] This embodiment provides a BCG signal processing system based on spectrum characteristics, including:

[0096] The pre-processing module is used to perform high-frequency denoising on the received original signal according to a preset wavelet threshold denoising algorithm;

[0097] According to the preset signal statistical algorithm, the original signal after high-frequency denoising is adaptively segmented;

[0098] Perform signal normalization on the original signal after adaptive segmentation processing.

[0099] The spectrum feature extraction module is used to convert the pre-processed original signal into a time-frequency domain signal according to a preset short-time Fourier transform algorithm;

[0100] Extracting energy distribution characteristics of the time-frequency domain signal in different frequency bands according to a preset short wavelet packet decomposition algorithm;

[0101] Obtaining corresponding power spectral density features according to the time-frequency domain signal and a preset fast Fourier transform algorithm;

[0102] According to the energy distribution characteristics and power spectrum density characteristics, as well as the pre-set information entropy transformation algorithm, the corresponding spectrum entropy characteristics are obtained;

[0103] The multi-spectral features include energy distribution features, power spectrum density features and spectrum entropy features.

[0104] A signal quality classification module is configured to input the multi-spectral features into a pre-set local feature extraction sub-model and a temporal feature extraction sub-model, respectively, to obtain local features and temporal features corresponding to the multi-spectral features; the local features and temporal features each include a corresponding feature dimension and sequence length;

[0105] Inputting the local features and temporal features into a pre-set feature fusion classification sub-model respectively to obtain corresponding quality classification results;

[0106] The local feature extraction sub-model includes a feature extraction model obtained by training a CNN network using multi-spectral features of known local features;

[0107] The time series feature extraction sub-model includes a feature extraction model obtained by training the LSTM network using multi-spectral features of known time series features;

[0108] The feature fusion classification sub-model includes a classification model obtained by training a transformer network using local features and temporal features corresponding to the same multi-spectral features with known quality classification results.

[0109] Furthermore, the feature fusion classification sub-model is used to:

[0110] According to the local features and a preset self-attention mechanism, a local attention matrix corresponding to the local features is obtained;

[0111] According to the temporal features and the preset temporal attention mechanism, a temporal attention matrix corresponding to the temporal features is obtained.

[0112] Temporal Attention Mechanism is mainly used to process data with obvious time dimension, with the aim of helping the model focus on the most important time points or time periods in the input sequence.

[0113] The self-attention mechanism is applied to various tasks of understanding the internal relationships of sequences and capturing the dependency between any two positions in the sequence.

[0114] In terms of computational methods, both the temporal attention mechanism and the self-attention mechanism are based on a query, key, and value computation framework. However, in their implementation, the temporal attention mechanism primarily focuses on weighting the importance of time, typically performing attention calculations in only one direction (for example, from front to back), making it suitable for sequence-sensitive tasks. The self-attention mechanism, on the other hand, simultaneously considers information at all positions in the sequence, allowing the model to focus on dependencies between any positions in the sequence.

[0115] Inputting the local attention matrix and the temporal attention matrix into a preset second fully connected layer for feature transformation respectively;

[0116] Performing a residual connection between the local attention matrix after feature transformation and the local features to obtain a corresponding local feature residual connection result;

[0117] Performing a residual connection between the temporal attention matrix after feature transformation and the temporal features to obtain a corresponding temporal feature residual connection result;

[0118] According to the preset formula 4, the local feature residual connection result and the temporal feature residual connection result are feature spliced to obtain the corresponding fusion feature; the formula 4 is:

[0119] C = concat(L1, L2, dim = n);

[0120] Among them, C is the corresponding fusion feature, concat is the splicing operation, L1 is the local feature residual connection result, L2 is the temporal feature residual connection result, and dim=n is the pre-set feature dimension.

[0121] According to the fusion feature and the preset attention score weight matrix, and the preset formula 1, the attention weight corresponding to the fusion feature is obtained; the formula 1 is:

[0122] S=Softmax(CW score );

[0123] Among them, S is the attention weight corresponding to the fusion feature, C is the fusion feature, and W score is the pre-set attention score weight matrix, and Softmax is the normalization function;

[0124] According to the fusion feature and the attention weight corresponding to the fusion feature, and the preset formula 2, the fusion feature vector corresponding to the fusion feature is obtained; the formula 2 is:

[0125] c = SC;

[0126] Where c is the fusion feature vector;

[0127] The fused feature vector corresponding to the fused feature is input into the preset first fully connected layer, and the quality classification result corresponding to the multi-spectral feature is obtained according to the preset formula 3; the formula 3 is:

[0128] Y=W f ·dropout[tanh(W p c)];

[0129] Among them, Y is the quality classification result corresponding to the multi-spectral feature, W f and W p These are the preset weights of the first fully connected layer.

[0130] a signal optimization module, which selects a corresponding optimization processing algorithm based on the quality classification result and a pre-set signal optimization strategy, optimizes the multi-spectral features, and obtains an optimized signal for assisting health analysis;

[0131] The optimization processing algorithm includes one or more of adaptive filtering, wavelet denoising, data interpolation, adversarial network, feature enhancement, time alignment, time synchronization, and resampling;

[0132] The signal optimization strategy includes a mapping relationship table between quality classification results and multi-spectral feature optimization processing algorithms.

[0133] This embodiment provides a BCG signal processing method based on spectral characteristics, which improves the comprehensiveness and effectiveness of feature extraction by using multi-spectral features; significantly enhances the anti-interference ability of the algorithm by leveraging the adaptive characteristics of the deep learning model; utilizes a hybrid network structure to achieve intelligent adaptation to individual differences; and ensures the stability and reliability of the model in practical applications through an attention mechanism and optimization strategy.

[0134] In practical applications, the introduction of spectrum analysis technology to preprocess BCG signals effectively addresses the poor interference resistance of existing methods. Experiments show that in ambient noise levels of 40-70 decibels, this method improves signal processing noise immunity by approximately 30% compared to traditional methods, while maintaining over 95% feature extraction stability.

[0135] Secondly, a quality classification model was used to analyze BCG signals, and the results were used to optimize the signals, significantly improving the measurement accuracy of physiological parameters. The average error in heart rate measurement was reduced from ±3 beats / minute (of the traditional method) to ±1 beat / minute, and the average error in respiratory rate measurement was reduced from ±2 beats / minute to ±0.5 beats / minute, providing more reliable data support for clinical diagnosis.

[0136] Furthermore, the signal processing pipeline enables efficient real-time monitoring. The processing delay is consistently controlled within 100ms, meeting the real-time requirements of continuous monitoring. Computing resources are reduced by 25% compared to traditional methods, making it more suitable for deployment on portable devices.

[0137] Finally, in tests of different age groups (20-70 years old) and different body positions (supine, side-lying), the recognition accuracy rate can be maintained stably (>92%), and it supports multiple models of BCG sensors, facilitating promotion and application in different scenarios.

[0138] Example 4

[0139] This embodiment provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the BCG signal processing system based on spectral characteristics described in claim embodiment 1 or embodiment 3, and the BCG signal processing method based on spectral characteristics described in embodiment 2.

[0140] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0141] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0142] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0143] In the description of this specification, the terms "one embodiment", "some embodiments", "embodiments", "examples", "specific examples" or "some examples" refer to the specific features, structures, materials or characteristics described in conjunction with the embodiment or example and included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0144] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A BCG signal processing system based on spectrum characteristics, characterized in that: include: A preprocessing module, used for preprocessing the received original signal; The spectrum feature extraction module converts the pre-processed original signal into a time-frequency domain signal according to the preset short-time Fourier transform algorithm; and extracting features from the time-frequency domain signal according to a preset feature extraction algorithm to obtain corresponding multi-spectral features; The feature extraction algorithm includes short wavelet packet decomposition algorithm, fast Fourier transform algorithm and information entropy transform algorithm; A signal quality classification module, configured to input the multi-spectral features into a preset signal quality classification model to obtain a corresponding quality classification result; a signal optimization module, configured to optimize the multi-spectral features according to the quality classification results and a preset signal optimization strategy to obtain an optimized signal for assisting health analysis; The signal quality classification module includes a classification model obtained by training a pre-built deep learning model through multi-spectral features of known quality classification results.

2. The BCG signal processing system based on spectrum characteristics according to claim 1, characterized in that: The preprocessing includes: high frequency denoising, adaptive segmentation and signal normalization; The preprocessing module preprocesses the received original signal, including: According to the pre-set wavelet threshold denoising algorithm, the received original signal is subjected to high-frequency denoising processing; According to the preset signal statistical algorithm, the original signal after high-frequency denoising is adaptively segmented; Perform signal normalization on the original signal after adaptive segmentation processing.

3. The BCG signal processing system based on spectrum characteristics according to claim 1, characterized in that: The spectrum feature extraction module extracts features from the time-frequency domain signal according to a preset feature extraction algorithm to obtain corresponding multi-spectral features, including: Extracting energy distribution characteristics of the time-frequency domain signal in different frequency bands according to a preset short wavelet packet decomposition algorithm; Obtaining corresponding power spectral density features according to the time-frequency domain signal and a preset fast Fourier transform algorithm; According to the energy distribution characteristics and power spectrum density characteristics, as well as the pre-set information entropy transformation algorithm, the corresponding spectrum entropy characteristics are obtained; The multi-spectral features include energy distribution features, power spectrum density features and spectrum entropy features.

4. The BCG signal processing system based on spectrum characteristics according to claim 1, characterized in that: The signal quality classification model includes: a local feature extraction sub-model, a temporal feature extraction sub-model and a feature fusion classification sub-model; The signal quality classification module inputs the multi-spectral features into a preset signal quality classification model to obtain a corresponding quality classification result, including: Inputting the multi-spectral features into a pre-set local feature extraction sub-model and a temporal feature extraction sub-model respectively, and obtaining local features and temporal features corresponding to the multi-spectral features respectively; the local features and temporal features both include corresponding feature dimensions and sequence lengths; Inputting the local features and temporal features into a pre-set feature fusion classification sub-model respectively to obtain corresponding quality classification results; The local feature extraction sub-model includes a feature extraction model obtained by training a CNN network using multi-spectral features of known local features; The time series feature extraction sub-model includes a feature extraction model obtained by training the LSTM network using multi-spectral features of known time series features; The feature fusion classification sub-model includes a classification model obtained by training a transformer network using local features and temporal features corresponding to the same multi-spectral features with known quality classification results.

5. The BCG signal processing system based on spectrum characteristics according to claim 4, characterized in that: The feature fusion classification sub-model is used to: According to the local features, the temporal features and the preset attention mechanism, a local attention matrix corresponding to the local features and a temporal attention matrix corresponding to the temporal features are obtained respectively; According to the local attention matrix, the temporal attention matrix, the local features and the temporal features, the fusion features corresponding to the multi-spectral features are obtained; According to the fusion feature and the preset attention score weight matrix, and the preset formula 1, the attention weight corresponding to the fusion feature is obtained; the formula 1 is: S=Softmax(CW score ); Among them, S is the attention weight corresponding to the fusion feature, C is the fusion feature, and W score is the pre-set attention score weight matrix, and Softmax is the normalization function; According to the fusion feature and the attention weight corresponding to the fusion feature, and the preset formula 2, the fusion feature vector corresponding to the fusion feature is obtained; the formula 2 is: c = SC; Where c is the fusion feature vector; The fused feature vector corresponding to the fused feature is input into the preset first fully connected layer, and the quality classification result corresponding to the multi-spectral feature is obtained according to the preset formula 3; the formula 3 is: Y=W f ·dropout[tanh(W p ·c)]; Among them, Y is the quality classification result corresponding to the multi-spectral feature, W f and W p These are the preset weights of the first fully connected layer.

6. The BCG signal processing system based on spectrum characteristics according to claim 5, characterized in that: The feature fusion classification sub-model obtains fusion features corresponding to multi-spectral features based on the local attention matrix, the temporal attention matrix, the local features and the temporal features, including: Inputting the local attention matrix and the temporal attention matrix into a preset second fully connected layer for feature transformation respectively; Performing a residual connection between the local attention matrix after feature transformation and the local features to obtain a corresponding local feature residual connection result; Performing a residual connection between the temporal attention matrix after feature transformation and the temporal features to obtain a corresponding temporal feature residual connection result; The local feature residual connection results and the temporal feature residual connection results are concatenated to obtain the corresponding fusion features.

7. The BCG signal processing system based on spectrum characteristics according to claim 6, characterized in that: The feature fusion classification sub-model concatenates the residual connection results of local features with the residual connection results of temporal features to obtain the corresponding fusion features, including: According to the preset formula 4, the local feature residual connection result and the temporal feature residual connection result are feature spliced to obtain the corresponding fusion feature; the formula 4 is: C = concat(L1, L2, dim = n); Among them, C is the corresponding fusion feature, concat is the splicing operation, L1 is the local feature residual connection result, L2 is the temporal feature residual connection result, and dim=n is the pre-set feature dimension.

8. The BCG signal processing system based on spectrum characteristics according to claim 5, characterized in that: The attention mechanism includes: spatial attention mechanism and self-attention mechanism; Then, the feature fusion classification sub-model obtains the local attention matrix corresponding to the local features and the temporal attention matrix corresponding to the temporal features according to the local features, temporal features and the preset attention mechanism, including: According to the local features and a preset self-attention mechanism, a local attention matrix corresponding to the local features is obtained; According to the temporal features and the preset temporal attention mechanism, a temporal attention matrix corresponding to the temporal features is obtained.

9. The BCG signal processing system based on spectrum characteristics according to claim 1, characterized in that: The signal optimization strategy includes a mapping relationship table between quality classification results and multi-spectral feature optimization processing algorithms; The signal optimization module optimizes the multi-spectral features according to the quality classification results and the preset signal optimization strategy to obtain an optimized signal for assisting health analysis, including: According to the quality classification result and the preset signal optimization strategy, a corresponding optimization processing algorithm is selected to optimize the multi-spectral features to obtain an optimized signal for assisting health analysis; The optimization processing algorithm includes one or more of adaptive filtering, wavelet denoising, data interpolation, adversarial network, feature enhancement, time alignment, time synchronization, and resampling.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the BCG signal processing system based on spectrum characteristics as claimed in any one of claims 1 to 9 is implemented.

Images