UNet-BiLSTM detection method and system for cardiac interval of ballistocardiogram signal
By processing cardiac impact signals using a distributed sensor array and a UNet-BiLSTM fusion model, the problems of weak anti-interference capability and insufficient temporal modeling in existing technologies are solved, achieving high-precision cardiac interval detection and heart rate calculation, adapting to complex environments, and meeting medical and industrial requirements.
Patent Information
- Application Number
- CN202511715551.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-01-06
AI Technical Summary
In non-contact BCG signal monitoring, existing technologies have limitations: single LSTM models have weak anti-interference capabilities and large detection errors in low signal-to-noise ratio scenarios; UNet models lack time-series modeling capabilities and cannot cope with dynamic changes in heart rate; traditional methods have high false negative rates when signal quality is poor, and existing fusion models are not optimized for the main lobe spectrum of BCG signals, resulting in insufficient extraction of core features.
A distributed sensor array was used to collect cardiac impact signals. The UNet-BiLSTM fusion model was used for filtering, baseline correction and normalization. The UNet module was combined with the BiLSTM module to extract spatial features and capture bidirectional temporal dependence. The J peak of the cardiac impact signal was located and the cardiac interval and heart rate were calculated.
It improves the accuracy and robustness of cardiac interval detection, reduces errors, meets medical-grade accuracy requirements, lowers the false negative rate, enhances the system's reliability in complex environments, adapts to sudden changes in heart rate and electromagnetic interference, and improves user comfort.
Smart Images
Figure CN121265028A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of biomedical engineering and intelligent monitoring technology, specifically a UNet-BiLSTM detection method and system for cardiac impulse signal intercardia. Background Technology
[0002] Non-contact BCG signal monitoring has become an important means of long-term cardiac activity monitoring due to its non-invasive and convenient advantages. However, existing technologies have obvious limitations: single LSTM models have weak anti-interference capabilities and large detection errors in low signal-to-noise ratio scenarios; UNet models lack time-series modeling capabilities and cannot cope with dynamic changes in heart rate; traditional methods rely on fixed rules, and the false negative rate is >15% when the signal quality is poor or the heart rate is abnormal. Furthermore, existing fusion models are not optimized for the main lobe spectrum (1-15Hz) of BCG signals, resulting in insufficient extraction of core features.
[0003] Therefore, improvements are needed. Summary of the Invention
[0004] To address the above technical issues, this application provides a UNet-BiLSTM detection method and system for cardiac impulse signal intercardia.
[0005] The first objective of this invention is achieved through the following technical solution:
[0006] A UNet-BiLSTM method for detecting the cardiac interval of cardiac impulse signals includes:
[0007] Raw cardiac impact signals are acquired non-contactly using a distributed sensor array;
[0008] The original cardiac impact signal is filtered, baseline corrected, and standardized to generate a standardized signal segment;
[0009] The standardized signal segment is input into the UNet-BiLSTM fusion model, where the UNet module extracts spatial features and outputs a probability map, and the BiLSTM module captures bidirectional temporal dependencies and corrects the probability map to locate the J peak of the centroid impact signal.
[0010] The cardiac interval and heart rate are calculated based on the J-peak location results, and at least one of the following operations is performed: display, alert, and data transmission.
[0011] In a preferred embodiment, the step of non-contact acquisition of raw cardiac impact signals via a distributed sensor array includes:
[0012] The raw cardiac impact signal was acquired using flexible piezoelectric thin film sensors distributed in an array.
[0013] The original cardiac impact signal is digitized based on a sampling rate with a preset frequency.
[0014] At least one environmental parameter among ambient temperature, humidity, and electromagnetic interference intensity is collected simultaneously.
[0015] In a preferred embodiment, the step of filtering, baseline correction, and normalization of the original cardiac impact signal to generate a normalized signal segment includes:
[0016] The original cardiac impact signal is filtered by a bandpass filter, and the passband range covers the main lobe spectrum of the cardiac impact signal;
[0017] Baseline drift correction was performed using the moving average method.
[0018] The signal amplitude is standardized and mapped to a preset range.
[0019] In a preferred embodiment, the step of inputting the standardized signal segment into the UNet-BiLSTM fusion model, wherein the UNet module extracts spatial features and outputs a probability map, and the BiLSTM module captures bidirectional temporal dependencies and corrects the probability map to locate the J-peak of the centroid impact signal, includes:
[0020] The UNet module adopts a symmetrical structure of a preset layer encoder and a preset layer decoder;
[0021] The encoder extracts multi-scale spatial features through convolution and downsampling operations, and outputs a high-dimensional feature sequence.
[0022] The decoder recovers the signal resolution through upsampling and cross-layer connections, and outputs an initial J-peak probability map.
[0023] The BiLSTM module receives the high-dimensional feature sequence, captures bidirectional temporal context information through forward and backward long short-term memory networks, and outputs a temporal correction vector.
[0024] The initial J-peak probability map is fused with the time-series correction vector to generate a corrected J-peak probability map, and the precise coordinates of the J-peak are determined by a post-processing algorithm.
[0025] In a preferred embodiment, the step of calculating the cardiac interval and heart rate based on the J-peak location result and performing at least one of the following operational steps: display, warning, and data transmission, includes:
[0026] Based on the precise coordinates of the J peak, the time interval between adjacent J peaks is calculated as the cardiac interval, and the instantaneous heart rate value is derived.
[0027] The cardiac interval, heart rate value, and corresponding cardiac impact signal waveform are visualized in real time on the display terminal;
[0028] When an abnormal heart rate or an intercardiac interval error exceeding a preset threshold is detected, a multi-level early warning mechanism is triggered, which includes at least one of sound, light, and electrical signals.
[0029] The cardiac interval, the instantaneous heart rate value, and the triggered multi-level early warning mechanism are used to generate a feedback report, which is then sent to the user terminal.
[0030] In a preferred embodiment, the step of calculating the time interval between adjacent J peaks as the cardiac interval based on the precise coordinates of the J peaks, and deriving the instantaneous heart rate value, includes:
[0031] Based on the precise coordinate sequence of the J peaks, the time difference between the (i+1)th J peak and the ith J peak is calculated and used as the ith cardiac interval R_i.
[0032] According to the formula: HR=60 / R_i, the cardiac interval R_i is converted into the instantaneous heart rate value HR;
[0033] A moving average filter is applied to the multiple instantaneous heart rate values (HR) calculated continuously to smooth short-term fluctuations.
[0034] The second objective of this invention is achieved through the following technical solution:
[0035] A UNet-BiLSTM detection system for cardiac impulse signal intervals includes:
[0036] Module 1: Non-contact acquisition of raw cardiac impact signals via a distributed sensor array;
[0037] The second module: filters, corrects baselines, and standardizes the original cardiac impact signal to generate a standardized signal segment;
[0038] The third module: The standardized signal segment is input into the UNet-BiLSTM fusion model, where the UNet module extracts spatial features and outputs a probability map, and the BiLSTM module captures bidirectional temporal dependencies and corrects the probability map to locate the J peak of the centroid impact signal.
[0039] Module 4: Calculates cardiac interval and heart rate based on J-peak location results, and performs at least one of the following operations: display, alert, and data transmission.
[0040] The above-mentioned objective three of this application is achieved through the following technical solution:
[0041] A computer device includes 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 UNet-BiLSTM detection method for cardiac impulse signal interval described above.
[0042] The fourth objective of this application is achieved through the following technical solution:
[0043] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the UNet-BiLSTM method for detecting the cardiac interval of a cardiac impulse signal.
[0044] In summary, this application includes at least one of the following beneficial technical effects:
[0045] 1. In terms of detection accuracy, the mean absolute error (Eabs) of cardiac intervals is controlled at 9.31ms through the fusion architecture of UNet-BiLSTM, and the accuracy is improved to 98.04%. This represents a 28% reduction in error compared to the traditional UNet model and an 8.3% improvement in accuracy compared to the single LSTM model. It breaks through the limitation of existing technologies with errors >20ms in low signal-to-noise ratio (SNR=5dB) scenarios and meets the requirements of medical-grade accuracy.
[0046] 2. In terms of dynamic adaptability, the BiLSTM module captures bidirectional temporal dependencies and uses the correlation between preceding and following signals to correct positioning errors, reducing the false negative rate to below 1.5% when the heart rate changes abruptly (>120 bpm), thus solving the problem of false negatives and false positives when the heart rate is abnormal under traditional fixed rules; the number of UNet encoding-decoding layers is optimized to 5 layers, accurately matching the main lobe spectrum of the BCG signal (1-15Hz), strengthening the extraction of key features, and overcoming the shortcomings of existing fusion models in capturing core frequency band features.
[0047] 3. In terms of environmental robustness, the dynamic anti-interference mechanism can adjust network parameters according to the real-time signal-to-noise ratio. The signal distortion rate is reduced from 25% to 8% in electromagnetic interference scenarios, and the data integrity is maintained at 95% in a low temperature environment of -40℃, which is 57% better than the existing technology. It meets the industrial-grade reliability requirements of complex scenarios such as rail transit (compliant with GB / T 24338.4-2018 standard). Attached Figure Description
[0048] Figure 1 This is a flowchart of an embodiment of this application;
[0049] Figure 2 This is a flowchart of an embodiment of step S20 of this application;
[0050] Figure 3 This is a schematic block diagram of the computer device described in this application. Detailed Implementation
[0051] The following is in conjunction with the appendix Figure 1-3 This application will be described in further detail.
[0052] In one embodiment, such as Figure 1 As shown, this application discloses a UNet-BiLSTM method for detecting the cardiac interval of cardiac impulse signals, which specifically includes the following steps:
[0053] S10: Non-contact acquisition of raw cardiac impact signals via a distributed sensor array;
[0054] S20: The original cardiac impact signal is filtered, baseline corrected, and standardized to generate a standardized signal segment;
[0055] S30: Input the standardized signal segment into the UNet-BiLSTM fusion model, where the UNet module extracts spatial features and outputs a probability map, and the BiLSTM module captures bidirectional temporal dependencies and corrects the probability map to locate the J peak of the centroid impact signal.
[0056] S40: Calculate the cardiac interval and heart rate based on the J-peak location results, and perform at least one of the following operations: display, warning, and data transmission.
[0057] In this embodiment, a complete "acquisition-processing-analysis-output" technology chain is constructed to achieve accurate detection of non-contact cardiac impact signals. The scheme first uses a distributed sensor array to acquire high-quality raw signals, then performs multi-level preprocessing to eliminate noise interference and standardize data specifications. Next, a UNet-BiLSTM fusion model is used to collaboratively extract spatial features and temporal dependencies, effectively improving the accuracy and robustness of J-peak localization in complex environments. Finally, based on the localization results, cardiac interval calculation is performed and multimodal output is achieved. The overall system significantly improves user comfort and system usability while ensuring medical-grade detection accuracy.
[0058] Step S10 includes:
[0059] S101: The original cardiac impact signal is acquired using flexible piezoelectric thin film sensors distributed in an array.
[0060] S102: Digitize the original cardiac impact signal based on a sampling rate with a preset frequency;
[0061] S103: Simultaneously collect at least one environmental parameter among ambient temperature, humidity, and electromagnetic interference intensity.
[0062] In this embodiment, the distributed array effectively suppresses interference such as local body motion through spatial diversity effect, the flexible PVDF sensor accurately captures micron-level vibrations with its excellent force-to-electricity conversion characteristics, the high sampling rate ensures the complete preservation of the J-peak time domain characteristics, and the synchronous monitoring of multiple environmental parameters provides a basis for subsequent signal compensation and quality assessment, thereby achieving high-quality raw signal acquisition in complex usage scenarios.
[0063] like Figure 2 Step S20 includes:
[0064] S201: The original cardiac impact signal is filtered by a bandpass filter, and the passband range covers the main lobe spectrum of the cardiac impact signal;
[0065] S202: Baseline drift correction is performed using the moving average method;
[0066] S203: Standardize the amplitude of the signal and map it to a preset range.
[0067] In this embodiment, the 1-10Hz bandpass filter accurately matches the main lobe spectral characteristics of the BCG signal, effectively separating breathing interference from high-frequency noise; the adaptive window moving average method dynamically tracks the baseline drift trend, avoiding insufficient correction of traditional fixed windows during body movement; amplitude standardization eliminates the influence of individual physiological differences, enabling the model to focus on waveform morphology features, providing standardized input with uniform quality and prominent features for subsequent deep learning models.
[0068] Step S30 includes:
[0069] S301: The UNet module adopts a symmetrical structure of a preset layer encoder and a preset layer decoder;
[0070] S302: The encoder extracts multi-scale spatial features through convolution and downsampling operations and outputs a high-dimensional feature sequence;
[0071] S303: The decoder recovers the signal resolution through upsampling and cross-layer connections, and outputs an initial J-peak probability map;
[0072] S304: The BiLSTM module receives the high-dimensional feature sequence, captures bidirectional temporal context information through forward and backward long short-term memory networks, and outputs a temporal correction vector;
[0073] S305: The initial J-peak probability map is fused with the time-series correction vector to generate a corrected J-peak probability map, and the precise coordinates of the J-peak are determined by a post-processing algorithm.
[0074] In this embodiment, UNet's 5-layer encoder-decoder structure captures multi-scale features from local J-peak morphology to the complete cardiac cycle by progressively expanding the receptive field, and the cross-layer connection mechanism preserves key spatial details; BiLSTM uses the high-dimensional feature sequence output by the UNet encoder to learn the temporal pattern between the J-peaks before and after, and performs context-aware error correction on the initial probability map through the temporal correction vector. This fusion architecture effectively solves the localization ambiguity problem of a single model in low signal-to-noise ratio and heart rate variability scenarios.
[0075] Step S40 includes:
[0076] S401: Based on the precise coordinates of the J peak, calculate the time interval between adjacent J peaks as the cardiac interval, and derive the instantaneous heart rate value;
[0077] S402: Visualize the cardiac interval, heart rate value, and corresponding cardiac impact signal waveform in real time on the display terminal;
[0078] S403: When an abnormal heart rate or an intercardiac interval error is detected that exceeds a preset threshold, a multi-level early warning mechanism containing at least one of sound, light, and electrical signals is triggered.
[0079] S404: Generate a feedback report based on the cardiac interval, the instantaneous heart rate value, and the triggered multi-level early warning mechanism, and send it to the user terminal.
[0080] In this embodiment, the cardiac interval calculated based on the precise J-peak coordinates accurately reflects the heart rhythm, the multi-parameter visualization provides intuitive physiological state monitoring, the multi-level early warning mechanism enables timely intervention in abnormal situations through combined sound, light, and electricity warnings, and the structured report generation and remote transmission establish a closed loop from data collection to health management, significantly enhancing the system's practical value in medical monitoring and daily health monitoring.
[0081] Step S401 includes:
[0082] SE1: Based on the precise coordinate sequence of the J peaks, calculate the time difference between the (i+1)th J peak and the ith J peak, and use it as the ith cardiac interval R_i;
[0083] SE2: Based on the formula: HR=60 / R_i, convert the cardiac interval R_i into the instantaneous heart rate value HR;
[0084] SE3: Performs a moving average filter on multiple continuously calculated instantaneous heart rate values (HR) to smooth short-term fluctuations.
[0085] In this embodiment, differential calculation based on the J-peak sequence accurately captures the time interval between adjacent heartbeats. The standard formula HR=60 / R_i is used to ensure the physiological accuracy of heart rate conversion. Moving average filtering effectively smooths instantaneous fluctuations caused by breathing or slight body movement, and finally outputs a stable and reliable heart rate trend, providing reliable rhythm parameters for clinical diagnosis and health assessment.
[0086] 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.
[0087] In one embodiment, a UNet-BiLSTM detection system for the cardiac interval of a cardiac impulse signal is provided. This UNet-BiLSTM detection system for the cardiac interval of a cardiac impulse signal corresponds one-to-one with the UNet-BiLSTM detection method for the cardiac interval of a cardiac impulse signal described in the previous embodiment. The UNet-BiLSTM detection system for the cardiac interval of a cardiac impulse signal includes:
[0088] Module 1: Non-contact acquisition of raw cardiac impact signals via a distributed sensor array;
[0089] The second module: filters, corrects baselines, and standardizes the original cardiac impact signal to generate a standardized signal segment;
[0090] The third module: The standardized signal segment is input into the UNet-BiLSTM fusion model, where the UNet module extracts spatial features and outputs a probability map, and the BiLSTM module captures bidirectional temporal dependencies and corrects the probability map to locate the J peak of the centroid impact signal.
[0091] Module 4: Calculates cardiac interval and heart rate based on J-peak location results, and performs at least one of the following operations: display, alert, and data transmission.
[0092] Optional, also includes:
[0093] The fifth module: uses flexible piezoelectric thin-film sensors distributed in an array to acquire the original cardiac impact signal;
[0094] Module 6: Digitize the original cardiac impact signal based on a preset sampling rate;
[0095] Module 7: Simultaneously collect at least one environmental parameter from ambient temperature, humidity, and electromagnetic interference intensity.
[0096] Optional, also includes:
[0097] Module 8: The original cardiac impact signal is filtered using a bandpass filter, with the passband range covering the main lobe spectrum of the cardiac impact signal;
[0098] Module 9: Baseline drift correction using the moving average method;
[0099] Module 10: Standardize the amplitude of the signal and map it to a preset range.
[0100] Optional, also includes:
[0101] Module 11: The UNet module adopts a symmetrical structure of a preset layer encoder and a preset layer decoder;
[0102] Twelve modules: The encoder extracts multi-scale spatial features through convolution and downsampling operations, and outputs a high-dimensional feature sequence;
[0103] Module 13: The decoder recovers the signal resolution through upsampling and cross-layer connections, and outputs the initial J-peak probability map;
[0104] Fourteenth module: The BiLSTM module receives the high-dimensional feature sequence, captures bidirectional temporal context information through forward and backward long short-term memory networks, and outputs a temporal correction vector;
[0105] Module 15: The initial J-peak probability map is fused with the time-series correction vector to generate a corrected J-peak probability map, and the precise coordinates of the J-peak are determined through a post-processing algorithm.
[0106] Optional, also includes:
[0107] Module 16: Based on the precise coordinates of the J-peak, calculate the time interval between adjacent J-peaks as the cardiac interval, and derive the instantaneous heart rate value;
[0108] Module 17: Visualizes the cardiac interval, heart rate value, and corresponding cardiac impact signal waveform in real time on the display terminal;
[0109] Eighteen modules: When an abnormal heart rate or an intercardiac interval error exceeds a preset threshold, a multi-level early warning mechanism is triggered, which includes at least one of the sound, light, and electrical signals;
[0110] Module 19: Generates a feedback report based on the cardiac interval, the instantaneous heart rate value, and the triggered multi-level early warning mechanism, and sends it to the user terminal.
[0111] Optional, also includes:
[0112] Module 20: Based on the precise coordinate sequence of the J peaks, calculate the time difference between the (i+1)th J peak and the ith J peak, and use it as the ith cardiac interval R_i;
[0113] Module 21: Based on the formula: HR=60 / R_i, convert the cardiac interval R_i into the instantaneous heart rate value HR;
[0114] Module 22: Performs a moving average filter on multiple continuously calculated instantaneous heart rate values (HR) to smooth short-term fluctuations.
[0115] Specific limitations of the UNet-BiLSTM detection system for cardiac impulse signal intervals can be found in the above description of the UNet-BiLSTM detection method for cardiac impulse signal intervals, and will not be repeated here. Each module in the aforementioned UNet-BiLSTM detection system for cardiac impulse signal intervals can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.
[0116] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements... [method / mechanism].
[0117] In one embodiment, a computer device is provided, 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 perform the following steps:
[0118] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:
[0119] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0120] 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 used as 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.
[0121] The above-described 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 UNet-BiLSTM method for detecting heart impulse signal heart interval, characterized in that, The method comprises the following steps: non-contact acquisition of original cardiac impact signals by a distributed sensor array; filtering, baseline correction and standardization of the original cardiac impact signals to generate a standardized signal segment; inputting the standardized signal segment into a UNet-BiLSTM fusion model, wherein the UNet module extracts spatial features and outputs a probability map, the BiLSTM module captures bidirectional time sequence dependence and corrects the probability map, and the J peak of the cardiac impact signal is located; based on the J peak positioning result, the heart interval and heart rate are calculated, and at least one of the display, warning and data transmission operations is performed.
2. The UNet-BiLSTM detection method of heart impulse signal heart interval according to claim 1, characterized in that, The step of non-contact acquisition of original cardiac impact signals by a distributed sensor array comprises: using a flexible piezoelectric film sensor distributed in an array form to collect the original cardiac impact signals; digitizing the original cardiac impact signals based on a preset frequency sampling rate; synchronously collecting at least one environmental parameter such as ambient temperature, humidity and electromagnetic interference intensity. 3.The UNet-BiLSTM method for detecting heart impact signal heart interval according to claim 1, characterized in that, The step of filtering, baseline correction and standardization of the original cardiac impact signals to generate a standardized signal segment comprises: filtering the original cardiac impact signals by a band-pass filter, and the passband range covers the main lobe spectrum of the cardiac impact signals; baseline drift correction is performed by using a moving average method; amplitude standardization is performed on the signals, and the signals are mapped to a preset interval.
4. The UNet-BiLSTM detection method of heart impulse signal heart interval according to claim 1, characterized in that, The step of inputting the standardized signal segment into a UNet-BiLSTM fusion model, wherein the UNet module extracts spatial features and outputs a probability map, the BiLSTM module captures bidirectional time sequence dependence and corrects the probability map, and the J peak of the cardiac impact signal is located, comprises: The UNet module adopts a symmetric structure of a preset layer encoder and a preset layer decoder; The encoder extracts multi-scale spatial features by convolution and downsampling operations, and outputs a high-dimensional feature sequence; The decoder restores the signal resolution by upsampling and cross-layer connection, and outputs an initial J peak probability map; The BiLSTM module receives the high-dimensional feature sequence, captures bidirectional time sequence context information by forward and backward long short-term memory networks, and outputs a time sequence correction vector; fuse the initial J peak probability map and the time sequence correction vector to generate a corrected J peak probability map, and determine the accurate coordinates of the J peak by a post-processing algorithm.
5. The UNet-BiLSTM detection method of heart impulse signal heart interval according to claim 1, characterized in that, The step of calculating the heart interval and heart rate based on the J peak positioning result, and performing at least one of the display, warning and data transmission operations, comprises: According to the accurate coordinates of the J peak, the time interval of adjacent J peaks is calculated as the heart interval, and the instantaneous heart rate value is derived; the heart interval, the heart rate value, and the corresponding cardiac impact signal waveform are visualized in real time on a display terminal; when the heart rate is abnormal or the heart interval error exceeds a preset threshold, a multi-level warning mechanism including at least one of sound, light and electrical signals is triggered; the heart interval, the instantaneous heart rate value, and the triggered multi-level warning mechanism generate a feedback report and send it to a user terminal.
6. The UNet-BiLSTM detection method of heart impulse signal heart interval according to claim 1, characterized in that, The step of calculating the heart interval and heart rate based on the J peak positioning result, and performing at least one of the display, warning and data transmission operations, comprises: Based on the accurate coordinate sequence of the J-peak, a time difference between an i+1th J-peak and an ith J-peak is calculated as an ith cardiac interval R_i; The cardiac interval R_i is converted into an instantaneous heart rate value HR according to a formula: HR=60 / R_i; A plurality of the instantaneous heart rate values HR calculated continuously are subjected to a sliding average filtering to smooth short-term fluctuations. 7.A UNet-BiLSTM detection system for heart impulse signal heart interval, characterized in that, The method comprises: A first module: non-contact acquisition of a raw ballistocardiogram by a distributed sensor array; A second module: filtering, baseline correction and standardization processing of the raw ballistocardiogram to generate a standardized signal segment; A third module: input of the standardized signal segment into a UNet-BiLSTM fusion model, wherein a UNet module extracts spatial features and outputs a probability map, and a BiLSTM module captures bidirectional time sequence dependence and corrects the probability map to locate a J-peak of the ballistocardiogram; A fourth module: calculation of a cardiac interval and a heart rate based on the J-peak location result, and at least one of display, early warning and data transmission.
8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the UNet-BiLSTM detection method for a ballistocardiogram cardiac interval according to any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the steps of the UNet-BiLSTM detection method for a ballistocardiogram cardiac interval according to any one of claims 1 to 6.