Severe hemorrhagic shock recognition system and method based on multi-modal data fusion

Abstract

This invention discloses a system and method for identifying severe hemorrhagic shock based on multimodal data fusion, belonging to the field of severe hemorrhagic shock identification technology. The method includes the following steps: constructing a unified frequency distribution matrix based on the time synchronization results of acoustic and visual signals, and dynamically calibrating the main frequency peaks of the acoustic and visual signals; relying on the dynamic calibration results of the frequency distribution matrix, performing phase perturbation rearrangement processing on frequency bands with high correlation of frequency peaks, and breaking potential harmonic resonance synchronization relationships by introducing micro-amplitude random phase slip, so that the acoustic and visual signals maintain uncoupled change characteristics in the temporal dimension. This invention achieves independent evolution and frequency coordination of multimodal signals through phase perturbation rearrangement and energy hierarchical mapping, avoiding phase lock-in collapse; and maintains dynamic energy balance through time-varying weight balancing and time-slip fusion, improving identification stability and accuracy, and achieving efficient real-time monitoring of severe hemorrhagic shock.

Description

Technical Field

[0001] This invention relates to the field of severe hemorrhagic shock identification technology, specifically to a severe hemorrhagic shock identification system and method based on multimodal data fusion. Background Technology

[0002] Multimodal data fusion-based identification of severe hemorrhagic shock refers to the intelligent determination of whether a patient is in a state of severe hemorrhagic shock by integrating physiological and behavioral data from multiple sources. This method typically collects and fuses multimodal signals such as heart rate, blood pressure, blood oxygen saturation, respiratory rate, skin temperature, voice features, facial expressions, posture changes, and even imaging or hemodynamic parameters. Through feature extraction, temporal modeling, and deep fusion algorithms (such as attention mechanisms and hybrid structures of convolutional and recurrent networks), a unified representation space of the patient's multidimensional vital signs is established. Based on this, the system uses the fused feature vectors to identify early physiological signs such as decreased circulatory volume, insufficient perfusion, and tissue hypoxia, achieving automatic grading and dynamic early warning from mild blood loss to severe shock. Unlike single-indicator monitoring, this method can comprehensively verify multi-source information, significantly improving identification accuracy and response speed, providing highly sensitive early diagnostic criteria for clinical emergency care.

[0003] The existing technology has the following shortcomings: In the process of identifying severe hemorrhagic shock through multimodal data fusion, when acoustic and visual signals resonate harmonicly within a specific frequency range, the feature phase-locking mechanism within the fusion model may collapse, leading to an imbalance in the phase relationship between multimodal time series. At this point, the model's dynamic feature mapping layer cannot correctly distinguish the energy transition trends of high-frequency and low-frequency components in the physiological signal. The judgment algorithm may be misled into a locally stable region, thus misidentifying actual transient fluctuations before a rapid blood pressure collapse as a stable state. This problem causes the system to lose its ability to perceive hemodynamic changes in the critical early stages, resulting in delayed shock identification and failed early warnings, severely impacting the timeliness of treatment and the reliability of the system's dynamic response.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a critical hemorrhagic shock identification system and method based on multimodal data fusion to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying severe hemorrhagic shock based on multimodal data fusion, characterized by comprising the following steps: Step 1: Based on the time synchronization results of acoustic and visual signals, construct a unified frequency distribution matrix, dynamically calibrate the main frequency peaks of acoustic and visual signals, and obtain a time-frequency reference for subsequent harmonic identification. Step 2: Based on the dynamic calibration results of the frequency distribution matrix, phase perturbation rearrangement is performed on the frequency bands with high correlation of frequency peaks. By introducing micro-amplitude random phase slip, potential harmonic resonance synchronization relationship is broken, so that the acoustic signal and visual signal maintain uncoupled change characteristics in the time dimension. Step 3: Based on the multimodal signal sequence after phase perturbation rearrangement, establish an energy hierarchical mapping structure to map the temporal changes of acoustic and visual signals into multi-layer energy distribution, realize the spatial separation of high-frequency energy transitions and low-frequency energy slow changes, and obtain multi-layer energy mapping characteristics; Step 4: Based on the distribution results of the energy layer mapping structure, a time-varying weight balancing mechanism is introduced to continuously regulate the contribution ratio of energy distribution in each layer, weaken the energy dominance effect of a single mode, maintain the overall stability of multi-modal energy fusion, and obtain the energy characteristic results after weight balancing. Step 5: Using the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel is established to dynamically extend the energy layering information in a continuous time window, thereby achieving long-term suppression of phase-locked collapse caused by harmonic resonance, and thus ensuring the stability and accuracy of the multimodal fusion model in identifying severe hemorrhagic shock in complex scenarios.

[0007] Preferably, the step of constructing a uniform frequency distribution matrix includes the following sub-steps: After obtaining the time synchronization results of acoustic and visual signals, a time-scale reshaping operation is performed. The sampling time axis of the acoustic signal is used as a reference to perform time-series mapping on the visual signal frame sequence, and a transition frame is supplemented by linear interpolation, so that the two types of signals are divided into multiple continuous time segments under a unified time step. Based on time segments, frequency feature distributions are extracted from acoustic and visual signals respectively, the signals are decomposed into energy sets of different frequency components, and two sets of frequency energy data are established in the same frequency range. The obtained frequency energy distribution data is used to identify and dynamically calibrate the main frequency peaks, track the frequency change trajectory over time and match the frequency peak positions of different modes to achieve cross-modal frequency correspondence. The dynamically calibrated peak frequency data are filled into a matrix according to the time and frequency dimensions to form a unified frequency distribution matrix that includes the energy intensity of acoustic and visual signals, which serves as the time-frequency reference for subsequent harmonic identification.

[0008] Preferably, during the construction of the unified frequency distribution matrix, the energy values ​​within adjacent time segments are smoothed, and local energy gaps on the frequency axis are filled by interpolation to maintain the continuity between the time axis and the frequency axis. This ensures that the calibrated frequency peaks of the acoustic and visual signals form a stable correspondence in the matrix, thereby improving the integrity and resolvability of the time-frequency reference.

[0009] Preferably, the step of performing phase perturbation rearrangement processing on frequency bands with high correlation to frequency peaks includes the following sub-steps: Based on the dynamic calibration results of the frequency distribution matrix, correlation scanning is performed on the frequency distribution data of different time segments to identify highly correlated frequency bands where acoustic and visual signals are highly synchronized in energy distribution; Phase features are extracted for high-correlation frequency bands, phase change sequences corresponding to each frequency point are obtained, and the phase difference distribution of acoustic and visual signals in the same frequency band is determined. Based on the phase difference change trend, a small-amplitude random phase shift is introduced in the high correlation frequency band to shift and rearrange the local phase trajectory in order to break the harmonic resonance synchronization relationship between the acoustic signal and the visual signal. The perturbed phase data is adjusted in the time and frequency domains and remapped into the frequency distribution matrix to maintain the uncoupled change characteristics of the acoustic and visual signals in the time dimension.

[0010] Preferably, the amplitude of the micro-amplitude random phase shift is dynamically limited according to the energy density and time span within the high correlation frequency band. The phase shift direction is either advanced or delayed. After the disturbance, the phase adjustment of the acoustic signal and the visual signal is continuously transitioned through time domain smoothing and frequency domain normalization, thereby further stabilizing the overall temporal characteristics of the frequency distribution matrix.

[0011] Preferably, the step of establishing the energy hierarchical mapping structure includes the following sub-steps: After obtaining the acoustic and visual signals after phase perturbation rearrangement, the energy changes of the two on a unified time axis are expanded, the total energy of different frequency intervals in each time segment is calculated, and the corresponding energy distribution matrix is ​​formed. Based on the energy distribution matrix, the energy range of acoustic and visual signals is divided into layers. The frequency bands with high energy concentration and large change rate are divided into high frequency layers, the frequency bands with slow energy change are divided into low frequency layers, and the mid frequency layer is set as the energy transition interval. Based on the hierarchical division results, the high-frequency layer, mid-frequency layer and low-frequency layer are mapped to different height ranges of spatial coordinates, and energy isolation bands are set between the layers to prevent energy interference, forming a spatial hierarchical energy structure. By coordinating the energy mapping structure between layers and adjusting its temporal continuity, the energy changes in each layer are made to remain smooth and extended in the time dimension, thereby obtaining multi-layer energy mapping characteristics.

[0012] Preferably, during the process of interlayer coordination and time continuity adjustment of the energy mapping structure, the distribution curves of each energy layer are smoothed along the time direction, and the energy overlap region between the high-frequency layer and the low-frequency layer is compressed to maintain the synchronous extension of the continuous transition of the high-frequency layer and the slow change of the low-frequency layer in the time dimension, thereby ensuring the hierarchical independence of the multilayer energy mapping characteristics in spatial distribution and the smoothness of energy transition.

[0013] Preferably, the step of introducing a time-varying weight balancing mechanism includes the following sub-steps: After obtaining the energy layer mapping structure, the average energy density, energy fluctuation range and energy change rate of the high-frequency layer, mid-frequency layer and low-frequency layer are extracted. The energy distribution curve of each layer on the time axis is established, and the energy ratio of acoustic signal and visual signal in each layer is analyzed. Based on the energy distribution curve, the initial weight allocation is performed on the energy of each layer, the total energy of each time segment is normalized, and the dynamic weight value is determined according to the energy change, thus forming a time-varying weight control framework. The weight values ​​are adjusted to balance the time continuity and energy coordination, so that the weight change trend is consistent with the energy change trend, thereby forming a stable energy contribution pattern within the time window. The weights, adjusted by time-varying equilibrium, are remapped to the energy hierarchy structure, and the energy of each layer is weighted and superimposed to obtain the energy characteristic results after weight balance.

[0014] Preferably, the step of establishing a time-slip fusion channel includes the following sub-steps: After obtaining the energy characteristic results after time-varying weight balance, the energy characteristic curve is divided into time windows, and the time domain is divided into multiple continuous and partially overlapping time windows, so that each time window contains the balanced energy state of high frequency layer, mid frequency layer and low frequency layer. The energy stratification information within the time window is sequentially connected, the energy change trend of each layer in the overlapping area of ​​adjacent time windows is extracted, and boundary smoothing is performed on the high-frequency layer, mid-frequency layer and low-frequency layer to ensure that the energy transitions continuously at the window boundary. After completing the smooth energy transition, the energy information between consecutive time windows is dynamically extended. The end energy value of the previous time window is selected as the initial reference for the next time window, and an energy slow release mechanism is introduced to balance the time extension relationship between high-frequency and low-frequency energy. The dynamically extended multi-layer energy data are superimposed along the time axis to form the overall energy characteristic result after time-slip fusion, so as to achieve long-term suppression of phase-locked collapse caused by harmonic resonance.

[0015] The critical hemorrhagic shock identification system based on multimodal data fusion includes a frequency distribution calibration module, a phase perturbation rearrangement module, an energy hierarchical mapping module, a time-varying weight balancing module, and a time-sliding fusion module. Frequency distribution calibration module: Based on the time synchronization results of acoustic and visual signals, a unified frequency distribution matrix is ​​constructed to dynamically calibrate the main frequency peaks of acoustic and visual signals; Phase perturbation rearrangement module: Based on the dynamic calibration results of the frequency distribution matrix, phase perturbation rearrangement is performed on the frequency band with high correlation of frequency peaks. By introducing micro-amplitude random phase slip, the potential harmonic resonance synchronization relationship is broken, so that the acoustic signal and the visual signal maintain the uncoupled change characteristics in the time dimension. Energy layer mapping module: Based on the multimodal signal sequence after phase perturbation rearrangement, an energy layer mapping structure is established to map the temporal changes of acoustic and visual signals into multi-layer energy distribution and obtain multi-layer energy mapping features; Time-varying weight balancing module: Based on the distribution results of the energy hierarchical mapping structure, a time-varying weight balancing mechanism is introduced to continuously adjust the contribution ratio of energy distribution in each layer, weaken the energy dominance effect of a single mode, and obtain the energy characteristic results after weight balancing. Time-sliding fusion module: Utilizing the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel is established to dynamically extend the energy stratification information in a continuous time window, thereby suppressing phase-locked collapse caused by harmonic resonance in the long term.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention introduces phase perturbation rearrangement and energy hierarchical mapping mechanisms during multimodal data fusion, enabling acoustic and visual signals to maintain independent evolution in both time and frequency dimensions, effectively avoiding phase-locked collapse caused by harmonic resonance. Through a unified frequency distribution matrix and dynamic calibration reference, time synchronization and frequency coordination of multimodal signals are achieved, allowing the model to accurately capture energy change trends of different modes in the high and low frequency ranges. This maintains high sensitivity in the early stages of hemodynamic mutations, improving the stability of shock prodromal identification.

[0017] This invention utilizes the dynamic synergy of time-varying weight balancing and time-sliding fusion channels to continuously extend and adaptively adjust multi-layer energy characteristics within a time window, maintaining the continuity and overall balance of energy distribution. This approach weakens the dominant role of a single mode or frequency layer in the fusion result, enabling multi-modal energy to form a stable dynamic equilibrium structure during long-term monitoring. This effectively suppresses the cumulative interference of harmonic resonances, ensuring the system's identification accuracy and dynamic response consistency in complex scenarios, and providing reliable technical support for the real-time monitoring of severe hemorrhagic shock. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of the method for identifying severe hemorrhagic shock based on multimodal data fusion according to the present invention.

[0020] Figure 2 This is a schematic diagram of the modules of the critical hemorrhagic shock identification system based on multimodal data fusion of the present invention. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0022] This invention provides, for example Figure 1 The method for identifying severe hemorrhagic shock based on multimodal data fusion, as shown, includes the following steps: Step 1: Based on the time synchronization results of acoustic and visual signals, construct a unified frequency distribution matrix, dynamically calibrate the main frequency peaks of acoustic and visual signals, and obtain a time-frequency reference for subsequent harmonic identification. The specific implementation method for this step is as follows: After obtaining the time synchronization results of acoustic and visual signals, a timescale reshaping operation is performed to address the differences in sampling rate, temporal resolution, and acquisition duration between the two signals. Specifically, using the sampling time axis of the acoustic signal as a global reference, the sampling start point, end point, and sampling interval information of the acoustic signal are extracted to establish a complete time series. Then, the frame sequence of the acquired visual signal is time-mapped, mapping the acquisition time of each frame to the time axis of the acoustic signal. If the frame rate of the visual signal is lower than the sampling rate of the acoustic signal, transition time points are inserted between adjacent frames, and supplementary frames are generated using linear interpolation to precisely align their timestamps with the sampling points of the acoustic signal. Based on this, the acoustic and visual signals are divided into multiple consecutive time segments of equal length according to a unified time step, with each time segment containing the same number of sampling points or frames. Each time segment represents the patient's physiological activity state within a specific short period, ensuring a complete temporal correspondence between the acoustic and visual signals, laying a precise temporal foundation for subsequent frequency analysis.

[0023] After time-scale reshaping, frequency feature distributions are extracted for both acoustic and visual signals within each time segment. Specifically, the acoustic signal is divided into time intervals within each segment, and the resulting signal sequences are sequentially decomposed into frequency components to extract the energy amplitude values ​​corresponding to each frequency component. After processing, each time segment forms a dataset composed of different frequency components and their energies, reflecting the frequency energy changes of the acoustic signal within that segment. Simultaneously, brightness variation analysis is performed on the continuous frame sequence of the visual signal. The average brightness of the skin region, facial feature points, and blood vessel distribution areas in each frame is used as time series points. The brightness variation amplitude between adjacent frames is calculated, and the resulting brightness variation sequence is expanded into frequency components to obtain the brightness variation frequency distribution corresponding to the time segment of the acoustic signal. Throughout the frequency distribution extraction process, the same frequency interval division method is used, for example, dividing the frequency range from 0.1Hz to 20Hz into multiple equally spaced frequency bands, with each band corresponding to a frequency point, thus ensuring consistency between the acoustic and visual signals on the frequency axis. After this processing, each time segment corresponds to two sets of comparable frequency energy data, one from acoustic signals and the other from visual signals. The two sets of data are completely consistent on the frequency scale, providing a standardized comparison basis for frequency peak calibration.

[0024] After obtaining the frequency energy distribution data of acoustic and visual signals, the main frequency peaks of both are identified and dynamically calibrated. Specifically, for the acoustic signal energy distribution curve within each time segment, several dominant frequency peaks are identified according to energy intensity from high to low, and their corresponding frequency and energy values ​​are recorded. The same operation is performed on the frequency energy distribution of the visual signal to extract the main frequency peaks of brightness changes in the corresponding time segment. After identifying the main peaks of each time segment, the frequency peaks of adjacent time segments are continuously tracked, and the trajectory of peak frequency change over time is calculated to obtain a continuous sequence of frequency changes over time. Subsequently, the main frequency peaks of the acoustic and visual signals in the same time segment are matched, and a correspondence is established based on the similarity of peak position and energy amplitude to ensure that the frequency peaks of different modes at the same time point can be mutually referenced on the frequency axis. During the matching process, if the frequency peak deviation between the acoustic and visual signals exceeds a preset threshold, the peaks of the two signals are fine-tuned by adjusting the frequency interval position or recalculating the mean energy distribution to ensure that the main peaks of both are consistent in time and frequency. Through this dynamic calibration, the time correlation and frequency correspondence between the frequency peaks of different modes are established, forming a set of cross-modal frequency peak matching sequences.

[0025] After obtaining dynamically calibrated frequency peak data, a unified frequency distribution matrix is ​​constructed to form a time-frequency reference for subsequent harmonic identification. Specifically, the entire analysis period is divided into several consecutive time windows. With time as the horizontal axis and frequency as the vertical axis, the calibrated frequency peaks and their energy amplitudes of the acoustic and visual signals within each time segment are filled into the corresponding positions in the matrix. Each cell in the matrix represents the energy intensity of the corresponding mode at a specific time and frequency. When both the acoustic and visual signals have peaks at the same frequency point, their energies are superimposed according to time weights to form a multimodal energy point. As the time window progresses, the entire matrix is ​​gradually filled into a complete time-frequency distribution map. This distribution map not only reflects the energy change trajectory of the acoustic and visual signals in the frequency domain but also reveals their relative evolution trend over time. To enhance the continuity and resolvability of the matrix, the energy values ​​of adjacent time windows are smoothed during the filling process to make the energy changes on the time axis more stable, and local energy gaps on the frequency axis are interpolated to maintain the integrity of the matrix. The resulting unified frequency distribution matrix has a strict correspondence in both time and frequency dimensions, and can serve as a unified reference framework for subsequent harmonic identification, phase perturbation rearrangement, and energy layer mapping, providing a stable time-frequency benchmark for the fusion identification of multimodal signals in complex physiological scenarios.

[0026] Through the above steps, the entire process of acoustic and visual signal acquisition, from initial acquisition to frequency distribution matrix construction, achieves time synchronization, frequency consistency, and dynamic peak correspondence. This process effectively solves the problems of time offset and frequency drift caused by multimodal signals under different sampling rates and time-domain resolutions, enabling subsequent multimodal fusion analysis to be performed under a unified time-frequency reference condition. This ensures that the frequency characteristics of each modality signal can be accurately corresponded and comprehensively utilized during the identification of severe hemorrhagic shock.

[0027] Step 2: Based on the dynamic calibration results of the frequency distribution matrix, phase perturbation rearrangement is performed on the frequency bands with high correlation of frequency peaks. By introducing micro-amplitude random phase slip, potential harmonic resonance synchronization relationship is broken, so that the acoustic signal and visual signal maintain uncoupled change characteristics in the time dimension. The specific implementation method for this step is as follows: Based on the dynamic calibration results of the frequency distribution matrix, a correlation scan is performed on the frequency distribution data of different time segments in the matrix. This correlation scan uses time as the horizontal dimension and frequency as the vertical dimension to analyze the energy correspondence between acoustic and visual signals at each frequency point. Specifically, for each time segment, the energy difference between acoustic and visual signals at the same frequency point is calculated point by point. By comparing the energy change trends at consecutive frequency points, frequency bands with high synchronicity in energy distribution between the two types of signals are identified. These frequency bands often exhibit simultaneous energy peaks and consistent energy growth and decay trends, indicating a potential harmonic resonance relationship between acoustic and visual signals in this frequency band. Through this correlation scan, highly correlated frequency bands that may lead to harmonic resonance can be identified on the time axis, providing a precise range for phase perturbation processing. This step ensures that during subsequent phase adjustment, perturbations only act on frequency regions with genuine resonance risk, avoiding unnecessary interference to unrelated frequency bands.

[0028] After identifying the highly correlated frequency bands, phase features are extracted from the acoustic and visual signals in these bands. Specifically, each highly correlated frequency band in the frequency distribution matrix is ​​used as an analysis unit, and the phase change sequence corresponding to each frequency point within that band is extracted. The phase change of the acoustic signal can be derived by analyzing the temporal interval between adjacent energy peaks in the speech waveform, while the phase change of the visual signal can be determined by analyzing the temporal difference of the brightness change period in consecutive frames. In this way, the temporal displacement trend of the two modes within the same frequency band can be obtained. Next, the phase change sequences of the acoustic and visual signals are aligned in chronological order, and the changes in their phase difference over time are compared. If the phase difference remains constant or fluctuates periodically over a long period within a certain frequency band, it can be determined that there is a strong coupling relationship within that frequency band. At this point, it is necessary to prepare an initial reference point for phase slippage for subsequent phase perturbation operations. By extracting phase features and determining the distribution of the phase difference, a basis can be provided for subsequent phase perturbation, giving the perturbation direction and amplitude a clear temporal and phase positioning foundation.

[0029] After obtaining the phase characteristics of the highly correlated frequency bands, phase perturbation and rearrangement are performed on these bands. Specifically, based on the phase difference change trend determined in the previous step, several representative phase reference points are selected for each time segment of the highly correlated frequency band, and a small-amplitude random phase shift is introduced at these reference points. This phase shift is not a large-amplitude time offset, but a slight shift on the original phase trajectory, which breaks the strict synchronization between acoustic and visual signals at the same frequency by adjusting the local time position. The direction of the phase shift can be advanced or delayed, and the amplitude can be dynamically allocated according to the energy density and time span within the frequency range, so that the perturbation can effectively break the resonance lock without destroying the continuity of the original signal. After introducing the phase shift, the phase sequence of the entire frequency range is rearranged, so that the order of the perturbed phase points on the time axis is slightly changed, thus forming a nonlinear distribution in the time dimension. This rearrangement makes the phase changes of acoustic and visual signals in the same frequency range no longer show a synchronous growth or synchronous decay trend, but instead exhibit independent temporal evolution characteristics. In this way, the resonant coupling between acoustic and visual signals in the frequency space is successfully weakened, making their temporal characteristics more independent.

[0030] After phase perturbation and rearrangement, the processing results undergo consistency adjustment in both the time and frequency domains to ensure the perturbed signal remains smooth and continuous across the global time range. Specifically, the phase-perturbed acoustic and visual signals are remapped onto the time axis, and the phase spacing between adjacent frequency points in each time segment is smoothed to ensure continuous transition of the time series within a local range, avoiding abrupt changes caused by phase slippage. Subsequently, the phase sequence within the entire time window is re-normalized to keep the phase adjustment in different frequency intervals within a reasonable range, preventing phase drift from accumulating over time and forming a shift trend. After time adjustment, the processing results are refilled into the frequency distribution matrix to update the frequency characteristics of the corresponding time segments within the matrix, making the perturbed and rearranged phase data the new frequency reference data. At this point, the frequency distribution matrix no longer has large-scale phase-locked intervals, and the changing trends of the acoustic and visual signals at the same frequency point are no longer synchronized, but instead exhibit independent fluctuations and uncoupled change characteristics. This process not only eliminates the potential risk of harmonic resonance, but also provides a more stable time-series input for the subsequent energy layering mapping step, enabling the subsequent energy distribution analysis to accurately reflect the true changes in the multimodal signal.

[0031] Through the continuous implementation of the above steps, phase perturbation and rearrangement are achieved between acoustic and visual signals within the high correlation region of frequency peaks. This fundamentally breaks the phase-locked synchronization relationship caused by harmonic resonance, allowing the temporal variation processes of different modes to maintain relatively independent evolutionary trends within the same frequency range. This process ensures that the high-frequency and low-frequency characteristics of each mode will not be biased due to synchronous resonance during subsequent multimodal energy fusion, effectively improving the temporal stability and recognition reliability in the identification of severe hemorrhagic shock.

[0032] Step 3: Based on the multimodal signal sequence after phase perturbation rearrangement, establish an energy hierarchical mapping structure to map the temporal changes of acoustic and visual signals into multi-layer energy distribution, realize the spatial separation of high-frequency energy transitions and low-frequency energy slow changes, and obtain multi-layer energy mapping characteristics; The specific implementation method for this step is as follows: After obtaining the phase-perturbed and rearranged acoustic and visual signals, their energy changes along a unified time axis are comprehensively expanded to extract their energy distribution characteristics in both time and frequency dimensions. Specifically, the perturbed acoustic signal is sequentially divided into time segments, and the total energy of the signal in different frequency ranges within each time segment is calculated, forming an energy distribution matrix with time as the horizontal dimension and frequency as the vertical dimension. Simultaneously, the same processing is performed on the visual signal, converting the average brightness fluctuation of each frame in the brightness change sequence into energy values ​​at the corresponding frequencies, and arranging them chronologically to strictly correspond with the energy distribution of the acoustic signal on the time coordinate. This processing method yields two sets of energy distribution datasets that are aligned in time and comparable in frequency. At this point, each time segment contains the energy performance of both the acoustic and visual signals within different frequency ranges, providing a complete data foundation for subsequent energy level segmentation.

[0033] After obtaining complete energy distribution data, the energy ranges of acoustic and visual signals are hierarchically divided to distinguish the dynamic changes of high-frequency and low-frequency energy. Specifically, based on the energy distribution curves within each time segment, frequency regions with high energy concentration are identified. Frequency bands with high energy concentration and large change rates are classified as the high-frequency layer, while frequency bands with relatively stable energy distribution and small change amplitudes are classified as the low-frequency layer. To further refine the hierarchical structure, a mid-frequency layer can be set between the high-frequency and low-frequency layers to represent the energy transition range between them. During the division process, considering the differences in energy distribution between acoustic and visual signals, both signals are normalized to ensure that the energy values ​​of different modes can be directly compared within the same order of magnitude. Through this hierarchical division, multiple energy hierarchies can be established within the same time segment, with each layer corresponding to the energy distribution state within a specific frequency range. After the division is completed, the energy distributions of each layer are continuously arranged on the time axis to form a preliminary multi-layered energy framework.

[0034] After dividing the energy levels, a mapping structure is constructed for each energy layer, enabling the energy at different levels to be expressed independently in space without interference. Specifically, the high-frequency, mid-frequency, and low-frequency layers are mapped to different height ranges in spatial coordinates, placing the high-frequency layer in the upper region, the mid-frequency layer in the middle region, and the low-frequency layer in the lower region. The energy value of each layer unfolds in spatial coordinates as a continuous energy trajectory, with its horizontal distribution corresponding to the time axis and its vertical distribution corresponding to energy intensity, thus forming a layered spatial energy distribution structure. To enhance the distinction between layers, energy isolation bands are set between layers, ensuring that rapid fluctuations in high-frequency energy do not interfere with the stable distribution of low-frequency energy. At this point, the energy layers of acoustic and visual signals are mapped to different positions within the same spatial framework, achieving dual alignment of time and frequency through an intra-layer correspondence mechanism. Through this spatial mapping structure, it can be clearly seen that within the same time segment, the energy transitions of the high-frequency layer and the gradual changes in the low-frequency layer are independent yet maintain a temporal correspondence, thereby achieving a layered spatial expression of multimodal energy characteristics.

[0035] After obtaining the multi-layer energy mapping structure, the overall mapping structure undergoes inter-layer coordination and temporal continuity adjustments to ensure that energy changes between different layers maintain consistent extensibility in the temporal dimension. Specifically, the energy distribution curve of each layer is smoothed along the time direction, giving high-frequency energy fluctuations a continuous temporal extension characteristic and avoiding energy jumps caused by local abrupt changes. Simultaneously, energy transition adjustments are made to the mid-frequency and low-frequency layers to maintain a rhythmic correspondence with the high-frequency layers in time, enabling each layer to form a stable energy synchronization band as time progresses. Subsequently, the upper and lower boundaries of the energy distribution of each layer are adjusted in the spatial dimension, effectively compressing the energy overlap area between the high-frequency and low-frequency layers, thereby avoiding inter-layer energy interference. The adjusted energy mapping structure remains smooth and continuous on the time axis, maintains clear hierarchy on the spatial axis, and forms a natural transition in energy intensity. At this point, the temporal changes of acoustic and visual signals have been transformed into a hierarchical spatial energy distribution, forming multi-layer energy mapping characteristics. This energy mapping feature not only reflects the dynamic evolution of energy at each frequency over time, but also clearly distinguishes the transitions of high-frequency energy from the slow fluctuations of low-frequency energy in space, thus providing a precise input basis for subsequent energy weight balancing and time-slip fusion.

[0036] Through the above steps, a hierarchical energy mapping framework was successfully established based on the phase-perturbed rearranged multimodal signal sequence, enabling the temporal variation characteristics of acoustic and visual signals to be fully expressed in a spatially layered manner. This process achieves effective spatial separation of high-frequency and low-frequency energy, preserving both the sensitive response characteristics of high-frequency information in the multimodal signal and the stable variation trend of low-frequency information.

[0037] Step 4: Based on the distribution results of the energy layer mapping structure, a time-varying weight balancing mechanism is introduced to continuously regulate the contribution ratio of energy distribution in each layer, weaken the energy dominance effect of a single mode, maintain the overall stability of multi-modal energy fusion, and obtain the energy characteristic results after weight balancing. The specific implementation method for this step is as follows: After obtaining the energy layer mapping structure, a comprehensive analysis of the distribution characteristics of each energy layer is conducted to determine its energy contribution in both temporal and spatial dimensions. Specifically, indicators such as average energy density, energy fluctuation range, and energy change rate over the entire time window are extracted from the high-frequency, mid-frequency, and low-frequency layers. These indicators visually reflect the energy activity of each layer at different time intervals. For example, high-frequency layers typically exhibit rapid energy changes, high peak values, and large fluctuation amplitudes, while low-frequency layers show gradual energy changes and longer durations. After comprehensively statistically analyzing these energy characteristics, energy distribution curves for each layer on the time axis are established, allowing visualization of the energy trend of each layer over time. Simultaneously, the energy proportion relationship between acoustic and visual signals in each layer is analyzed, recording the relative contribution ratio of the two signals in each energy layer within the same time segment, providing a basic reference for subsequent weight balancing. Through this step, the energy distribution of each layer in the entire fusion structure and the energy dominance relationship among multiple modes can be clearly understood.

[0038] After clarifying the temporal distribution characteristics of each energy layer, a preliminary weight allocation is performed on the energy contributions of the high-frequency, mid-frequency, and low-frequency layers to form the initial state for time-varying weight control. Specifically, the total energy of each layer within the same time segment is normalized to maintain a constant sum of energy across the three layers. Different initial weights are assigned to different layers based on the dynamic changes in their energy distribution curves. For example, when the energy of the high-frequency layer experiences a short-term surge, its weight is reduced to prevent it from occupying an excessively high proportion in the overall energy fusion; conversely, when the energy of the low-frequency layer decays rapidly over a certain period, its weight is appropriately increased to maintain overall energy balance. Through this dynamic allocation method, a time-varying weight control framework is initially established, allowing the energy fusion process to be flexibly adjusted according to changes in time and frequency, rather than being fixed at a static ratio. At this point, each time segment corresponds to a set of dynamic weight values, which determine the contribution ratio of each energy layer to the overall energy output at that moment, providing initial conditions for subsequent continuous control.

[0039] After obtaining the initial weight allocation results, a dual balance adjustment is made to the weights of each layer, ensuring both temporal continuity and energy coordination. Specifically, in the time dimension, the weight values ​​of adjacent time segments are smoothly connected to ensure a continuous transition of weight changes along the time axis, avoiding abrupt changes. This smooth transition method allows the energy contributions of the high-frequency, mid-frequency, and low-frequency layers to form a natural connection in time, thus reflecting the actual energy evolution law of the multimodal signal. In the energy dimension, the weight values ​​are corrected according to the direction and rate of energy change of each layer within adjacent time segments, ensuring that the weight change trend is consistent with the energy change trend. For example, when the energy of the high-frequency layer gradually decreases within a certain time period, the weight should also be reduced accordingly to reduce its impact on the total energy; while when the energy of the low-frequency layer slowly increases, the weight value gradually increases to maintain the balance of the overall fused output. Through this dual regulation of time and energy, the energy of different layers can be dynamically coordinated in an adaptive manner, ultimately forming a stable energy contribution pattern throughout the entire time window. This balancing process not only achieves the time-varying nature of the weights, but also ensures the synergy between the multiple layers of energy, so that each layer can maintain its own change characteristics during the fusion process without disrupting the overall balance.

[0040] After adjusting the time-varying weight balance, the balanced energy distribution is remapped back into the energy hierarchy structure, forming the weighted energy characteristic result. Specifically, within each time segment, the energy of the high-frequency, mid-frequency, and low-frequency layers is weighted and superimposed according to the adjusted weight values ​​to obtain the comprehensive energy output value of that time segment. Subsequently, the weighted results of all time segments are arranged along the time axis to form a complete time-varying energy curve. This curve can simultaneously reflect the overall dynamic change characteristics of the multimodal signal under the energy balance state of each layer. Based on this, the joint energy distribution trend of acoustic and visual signals is further analyzed to confirm whether the energy contribution of the two signals has been balanced under the weighted fusion state. In the entire energy mapping space, the high-frequency layer energy after balancing no longer has a dominant influence on the overall structure due to short-term peaks, and the low-frequency layer energy is not marginalized due to slow changes. All energy layers maintain coordinated and stable evolution characteristics during the time progression, ensuring that the overall energy distribution of multimodal fusion is always in a dynamic equilibrium state. The final energy feature results not only retain the feature information of each of the multimodal signals, but also realize the coordinated distribution of energy at different levels in space and time, providing a stable and reliable input basis for subsequent time-slip fusion.

[0041] Through the continuous implementation of the above steps, a time-varying weight balancing mechanism capable of continuously adjusting the contribution relationship between energy layers is established based on the distribution results of the energy layer mapping structure. This mechanism effectively avoids the dominance effect of a single mode or a single energy layer within a specific time period, ensuring the coordination and stability of energy distribution during the fusion process of multimodal signals.

[0042] Step 5: Using the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel is established to dynamically extend the energy layering information in a continuous time window, thereby achieving long-term suppression of phase-locked collapse caused by harmonic resonance, thus ensuring the stability and accuracy of the multimodal fusion model in identifying severe hemorrhagic shock in complex scenarios. The specific implementation method for this step is as follows: After obtaining the energy characteristic results after time-varying weight balancing, time windows are divided to establish a sliding time frame that reflects the continuous evolution trend of energy characteristics. Specifically, based on the energy characteristic curve, the entire time domain is divided into multiple continuous and partially overlapping time windows of fixed length. Each time window contains a complete segment of multi-layer energy distribution information, including the equilibrium energy states of high-frequency, mid-frequency, and low-frequency layers. Adjacent time windows are connected through time overlap regions to achieve data transition, allowing energy changes to be continuously transmitted in time without discontinuities. Through this time sliding division method, a dynamically updated time-sliding sequence is formed, enabling energy characteristics to not only reflect instantaneous states but also possess time extension attributes, providing a time channel for subsequent energy layer extension. In this process, the overlap ratio of the time sliding windows can be dynamically set according to the rate of energy change. When energy changes rapidly, the time window is shortened to improve response speed; when energy changes slowly, the time window is extended to enhance smoothness. This time window structure ensures the continuity of energy characteristics over a long time axis, laying the foundation for subsequent dynamic extension.

[0043] After establishing time-sliding windows, the energy stratification information within each time window is sequentially linked to ensure a continuous transition of energy at the window boundaries. Specifically, for the energy distribution curves of the same energy layer in adjacent time windows, the energy change trend in the overlapping region is extracted, and the degree of difference between energy layers at the boundaries of adjacent time windows is calculated. Based on this degree of difference, the energy values ​​of the high-frequency, mid-frequency, and low-frequency layers are smoothed at the boundaries, allowing for a natural transition between energy layers at the connection points of different time windows. For example, if the energy of the high-frequency layer is on an upward trend at the end of the previous time window and on a downward trend at the beginning of the next time window, it is adjusted continuously in the transition region to maintain the consistency of energy change. At the same time, in the low-frequency layer, its energy variation range is extended to ensure the long-term continuity of low-frequency energy distribution in the time dimension. Through this combination of intra-layer smoothing and inter-layer linkage, the energy of multiple layers maintains overall continuity as time progresses, thus forming a coherent energy evolution structure in the time dimension. This step ensures that the energy characteristics do not abruptly or break during the time-slip channel, making the energy distribution traceable and extensible.

[0044] After smoothly connecting energy layers within each time window, the energy information between consecutive time windows is dynamically extended to establish a temporal continuity of energy distribution in the time-sliding channel. Specifically, the endpoint energy value of each energy layer in the previous time window is selected as the initial energy reference for the next time window, allowing the energy distribution to transition naturally along the time axis. Simultaneously, to prevent short-term fluctuations in high-frequency energy from affecting overall stability, an energy release mechanism is introduced in the time-sliding channel. This allows high-frequency layer energy to extend gradually over time, while low-frequency layer energy persists throughout the entire time period. Thus, during time-sliding, the high-frequency layer exhibits short-period rapid response characteristics, while the low-frequency layer displays long-period stable regulation characteristics. Together, they form a two-layer temporal dynamic structure of energy change in the time channel. Furthermore, between adjacent time windows, the energy ratio of each layer is maintained stable through energy balance mapping, ensuring that multiple layers of energy extend synchronously in the time-sliding channel without imbalance. Through this dynamic extension process, energy at different levels can form a complementary and balancing relationship in a continuous time series, thereby avoiding energy accumulation in a single mode or a single level, preventing harmonic resonance from reforming, and ensuring that energy distribution is stable in the time dimension for a long time.

[0045] After completing the dynamic extension of energy stratification, the multi-layer energy data obtained in the time-sliding channel are comprehensively mapped to form the overall energy characteristic result after time-sliding fusion. Specifically, the energy distribution results of all time windows are superimposed along the time axis, and the temporal change trajectory of the same layer of energy is combined into a continuous energy curve. Subsequently, the extension information of the high-frequency layer, mid-frequency layer, and low-frequency layer in the time-sliding channel is integrated according to their corresponding time positions, so that the overall energy distribution forms a smooth and continuous three-dimensional energy structure in the time dimension. This structure can reflect the changing trend of different frequencies of energy on the time axis and their inter-layer coordination relationship. Through this fusion method, the energy characteristics exhibit continuous extension in time and maintain layered independence in space, ensuring that the energy of each layer can maintain its individual dynamic characteristics and form overall coordination with other layers during the temporal evolution. Finally, the energy characteristic result output by the time-sliding fusion channel can continuously suppress the phase-locking effect caused by harmonic resonance over a long period of time, keeping the energy distribution in a balanced state in the time and spatial dimensions, thereby effectively avoiding the recurrence of phase-locked collapse. This result not only improves the stability and continuity of multimodal energy fusion over time, but also ensures the dynamic response capability in the identification process of severe hemorrhagic shock, enabling the identification model to accurately capture the trend of physiological parameter changes and maintain high-efficiency identification performance in complex scenarios.

[0046] Through the above steps, and utilizing the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel capable of dynamic extension within a continuous time window was successfully established. This process achieves natural connection and long-range balance of energy layering information in the time dimension, preventing phase relocking and energy aggregation effects in the time evolution of multimodal signals, and ensuring that the entire fusion recognition process maintains stable and reliable recognition capabilities even in complex dynamic environments.

[0047] After completing the construction of the time-slip fusion channel and obtaining the overall energy characteristic results after time-slip fusion, the discriminant feature parameters reflecting the dynamic changes of multimodal energy are further extracted based on the overall energy characteristic results. These include energy transition amplitude, interlayer energy distribution differences, time extension trend, and high- and low-frequency energy coupling degree. The discriminant feature parameters are then input into the pre-constructed severe hemorrhagic shock identification model or judgment rule for comprehensive analysis. The corresponding severe hemorrhagic shock identification results and their risk levels or warning signals are output, thereby realizing real-time identification and dynamic assessment of whether the target object is in a state of severe hemorrhagic shock.

[0048] This invention introduces phase perturbation rearrangement and energy hierarchical mapping mechanisms during multimodal data fusion, enabling acoustic and visual signals to maintain independent evolution in both time and frequency dimensions, effectively avoiding phase-locked collapse caused by harmonic resonance. Through a unified frequency distribution matrix and dynamic calibration reference, time synchronization and frequency coordination of multimodal signals are achieved, allowing the model to accurately capture energy change trends of different modes in the high and low frequency ranges. This maintains high sensitivity in the early stages of hemodynamic mutations, improving the stability of shock prodromal identification.

[0049] This invention utilizes the dynamic synergy of time-varying weight balancing and time-sliding fusion channels to continuously extend and adaptively adjust multi-layer energy characteristics within a time window, maintaining the continuity and overall balance of energy distribution. This approach weakens the dominant role of a single mode or frequency layer in the fusion result, enabling multi-modal energy to form a stable dynamic equilibrium structure during long-term monitoring. This effectively suppresses the cumulative interference of harmonic resonances, ensuring the system's identification accuracy and dynamic response consistency in complex scenarios, and providing reliable technical support for the real-time monitoring of severe hemorrhagic shock.

[0050] This invention provides, for example Figure 2 The critical hemorrhagic shock identification system based on multimodal data fusion shown includes a frequency distribution calibration module, a phase perturbation rearrangement module, an energy hierarchical mapping module, a time-varying weight balancing module, and a time-sliding fusion module; Frequency distribution calibration module: Based on the time synchronization results of acoustic and visual signals, a unified frequency distribution matrix is ​​constructed to dynamically calibrate the main frequency peaks of acoustic and visual signals; Phase perturbation rearrangement module: Based on the dynamic calibration results of the frequency distribution matrix, phase perturbation rearrangement is performed on the frequency band with high correlation of frequency peaks. By introducing micro-amplitude random phase slip, the potential harmonic resonance synchronization relationship is broken, so that the acoustic signal and the visual signal maintain the uncoupled change characteristics in the time dimension. Energy layer mapping module: Based on the multimodal signal sequence after phase perturbation rearrangement, an energy layer mapping structure is established to map the temporal changes of acoustic and visual signals into multi-layer energy distribution and obtain multi-layer energy mapping features; Time-varying weight balancing module: Based on the distribution results of the energy hierarchical mapping structure, a time-varying weight balancing mechanism is introduced to continuously adjust the contribution ratio of energy distribution in each layer, weaken the energy dominance effect of a single mode, and obtain the energy characteristic results after weight balancing. Time-sliding fusion module: Utilizing the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel is established to dynamically extend the energy stratification information in a continuous time window, thereby suppressing phase-locked collapse caused by harmonic resonance in the long term.

[0051] The critical hemorrhagic shock identification method based on multimodal data fusion provided in this invention is implemented through the aforementioned critical hemorrhagic shock identification system based on multimodal data fusion. For details of the specific methods and processes of the critical hemorrhagic shock identification system based on multimodal data fusion, please refer to the embodiments of the critical hemorrhagic shock identification method based on multimodal data fusion described above, which will not be repeated here.

[0052] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for identifying severe hemorrhagic shock based on multimodal data fusion, characterized in that, Includes the following steps: Step 1: Based on the time synchronization results of acoustic and visual signals, construct a unified frequency distribution matrix and dynamically calibrate the main frequency peaks of acoustic and visual signals; Step 2: Based on the dynamic calibration results of the frequency distribution matrix, perform phase perturbation rearrangement processing on the frequency band with high correlation of frequency peaks. By introducing micro-amplitude random phase slip, the potential harmonic resonance synchronization relationship is broken, so that the acoustic signal and the visual signal maintain the uncoupled change characteristics in the time dimension. Step 3: Based on the multimodal signal sequence after phase perturbation rearrangement, establish an energy hierarchical mapping structure to map the temporal changes of acoustic and visual signals into multi-layer energy distributions, thereby obtaining multi-layer energy mapping features; Step 4: Based on the distribution results of the energy layer mapping structure, a time-varying weight balancing mechanism is introduced to continuously adjust the contribution ratio of the energy distribution of each layer, weaken the energy dominance effect of a single mode, and obtain the energy characteristic results after weight balancing. Step 5: Using the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel is established to dynamically extend the energy stratification information in a continuous time window, thereby suppressing the phase-locked collapse caused by harmonic resonance in the long term.

2. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 1, characterized in that, The steps for constructing a unified frequency distribution matrix include the following sub-steps: After obtaining the time synchronization results of acoustic and visual signals, a time-scale reshaping operation is performed. The sampling time axis of the acoustic signal is used as a reference to perform time-series mapping on the visual signal frame sequence, and a transition frame is supplemented by linear interpolation, so that the two types of signals are divided into multiple continuous time segments under a unified time step. Based on time segments, frequency feature distributions are extracted from acoustic and visual signals respectively, the signals are decomposed into energy sets of different frequency components, and two sets of frequency energy data are established in the same frequency range. The obtained frequency energy distribution data is used to identify and dynamically calibrate the main frequency peaks, track the frequency change trajectory over time and match the frequency peak positions of different modes to achieve cross-modal frequency correspondence. The dynamically calibrated frequency peak data are filled into the matrix according to the time and frequency dimensions to form a unified frequency distribution matrix that includes the energy intensity of acoustic and visual signals.

3. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 2, characterized in that, In the process of constructing the unified frequency distribution matrix, the energy values ​​in adjacent time segments are smoothed and the local energy gaps on the frequency axis are filled by interpolation to maintain the continuity between the time axis and the frequency axis, so that the frequency peaks of the acoustic signal and the visual signal after calibration form a stable correspondence in the matrix.

4. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 2, characterized in that, The steps for performing phase perturbation rearrangement on frequency bands with high correlation to frequency peaks include the following sub-steps: Based on the dynamic calibration results of the frequency distribution matrix, correlation scanning is performed on the frequency distribution data of different time segments to identify highly correlated frequency bands where acoustic and visual signals are highly synchronized in energy distribution; Phase features are extracted for high-correlation frequency bands, phase change sequences corresponding to each frequency point are obtained, and the phase difference distribution of acoustic and visual signals in the same frequency band is determined. Based on the phase difference change trend, a small-amplitude random phase shift is introduced in the high correlation frequency band to shift and rearrange the local phase trajectory, thereby breaking the harmonic resonance synchronization relationship between the acoustic signal and the visual signal. The perturbed phase data is adjusted in the time and frequency domains and remapped into the frequency distribution matrix to maintain the uncoupled change characteristics of the acoustic and visual signals in the time dimension.

5. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 4, characterized in that, The amplitude of the micro-amplitude random phase shift is dynamically limited based on the energy density and time span within the high correlation frequency band. The phase shift direction is either advanced or delayed, and after perturbation, time-domain smoothing and frequency-domain normalization are performed to ensure a continuous transition in the phase adjustment of the acoustic and visual signals.

6. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 4, characterized in that, The steps to establish an energy hierarchical mapping structure include the following sub-steps: After obtaining the acoustic and visual signals after phase perturbation rearrangement, the energy changes of the two on a unified time axis are expanded, the total energy of different frequency intervals in each time segment is calculated, and the corresponding energy distribution matrix is ​​formed. Based on the energy distribution matrix, the energy range of acoustic and visual signals is divided into layers. The frequency bands with high energy concentration and large change rate are divided into high frequency layers, the frequency bands with slow energy change are divided into low frequency layers, and the mid frequency layer is set as the energy transition interval. Based on the hierarchical division results, the high-frequency layer, mid-frequency layer and low-frequency layer are mapped to different height ranges of spatial coordinates, and energy isolation bands are set between the layers to prevent energy interference, forming a spatially hierarchical energy structure. By coordinating the energy mapping structure between layers and adjusting its temporal continuity, the energy changes in each layer are made to remain smooth and extended in the time dimension, thus obtaining multi-layer energy mapping characteristics.

7. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 6, characterized in that, In the process of coordinating between layers and adjusting the temporal continuity of the energy mapping structure, the distribution curves of each energy layer are smoothed along the time direction, and the energy overlap region between the high-frequency layer and the low-frequency layer is compressed.

8. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 6, characterized in that, The steps for introducing a time-varying weight balancing mechanism include the following sub-steps: After obtaining the energy layer mapping structure, the energy indices of the high-frequency layer, mid-frequency layer and low-frequency layer are extracted, the energy distribution curves of each layer on the time axis are established, and the energy ratio of acoustic signal and visual signal in each layer is analyzed. Based on the energy distribution curve, the initial weight allocation is performed on the energy of each layer, the total energy of each time segment is normalized, and the dynamic weight value is determined according to the energy change, thus forming a time-varying weight control framework. The weight values ​​are adjusted to balance the continuity of time and the coordination of energy, so that the trend of weight change is consistent with the trend of energy change. The weights, adjusted by time-varying equilibrium, are remapped to the energy hierarchy structure, and the energy of each layer is weighted and superimposed to obtain the energy characteristic results after weight balance.

9. The method for identifying severe hemorrhagic shock based on multimodal data fusion according to claim 8, characterized in that, The steps to establish a time-slide fusion channel include the following sub-steps: After obtaining the energy characteristic results after time-varying weight balance, the energy characteristic curve is divided into time windows, and the time domain is divided into multiple continuous and partially overlapping time windows, so that each time window contains the balanced energy state of high frequency layer, mid frequency layer and low frequency layer. The energy stratification information within the time window is sequentially connected, the energy change trend of each layer in the overlapping area of ​​adjacent time windows is extracted, and boundary smoothing is performed on the high-frequency layer, mid-frequency layer and low-frequency layer to ensure that the energy transitions continuously at the window boundary. After completing the smooth energy transition, the energy information between consecutive time windows is dynamically extended. The end energy value of the previous time window is selected as the initial reference for the next time window, and an energy slow release mechanism is introduced to balance the time extension relationship between high-frequency and low-frequency energy. The dynamically extended multi-layer energy data are superimposed along the time axis to form the overall energy characteristic result after time-slip fusion, thereby achieving long-term suppression of phase-locked collapse caused by harmonic resonance.

10. A system for identifying severe hemorrhagic shock based on multimodal data fusion, used to implement the method for identifying severe hemorrhagic shock based on multimodal data fusion as described in any one of claims 1-9, characterized in that, It includes a frequency distribution calibration module, a phase perturbation rearrangement module, an energy layer mapping module, a time-varying weight balancing module, and a time-sliding fusion module; Frequency distribution calibration module: Based on the time synchronization results of acoustic and visual signals, a unified frequency distribution matrix is ​​constructed to dynamically calibrate the main frequency peaks of acoustic and visual signals; Phase perturbation rearrangement module: Based on the dynamic calibration results of the frequency distribution matrix, phase perturbation rearrangement is performed on the frequency band with high correlation of frequency peaks. By introducing micro-amplitude random phase slip, the potential harmonic resonance synchronization relationship is broken, so that the acoustic signal and the visual signal maintain the uncoupled change characteristics in the time dimension. Energy layer mapping module: Based on the multimodal signal sequence after phase perturbation rearrangement, an energy layer mapping structure is established to map the temporal changes of acoustic and visual signals into multi-layer energy distribution and obtain multi-layer energy mapping features; Time-varying weight balancing module: Based on the distribution results of the energy hierarchical mapping structure, a time-varying weight balancing mechanism is introduced to continuously adjust the contribution ratio of energy distribution in each layer, weaken the energy dominance effect of a single mode, and obtain the energy characteristic results after weight balancing. Time-sliding fusion module: Utilizing the energy characteristic results after time-varying weight balancing, a time-sliding fusion channel is established to dynamically extend the energy stratification information in a continuous time window, thereby suppressing phase-locked collapse caused by harmonic resonance in the long term.

Images