68 results about "Biological signal processing" patented technology

The invention provides an electroencephalogram signal artifact removing method and device, equipment and a medium, and relates to the technical field of biological signal processing, collected electroencephalogram signal data is processed to obtain an optimal modal number and an optimal bandwidth parameter, and the optimal modal number and the optimal bandwidth parameter are used for conducting self-adaptive variational mode decomposition on the electroencephalogram signal data to obtain a plurality of modal components; then carrying out multi-dimensional feature analysis to obtain a plurality of feature indexes for judging an artifact suspicion mode and an effective mode component; performing short-time Fourier transform on the artifact suspicion mode, and constructing a time-frequency confidence map to guide weighted time-frequency independent component separation on the artifact suspicion mode to obtain an artifact component; suppressing the artifact component to obtain a suppressed independent component, and performing weighted reconstruction and multi-component fusion on the suppressed independent component and the effective modal component based on the time-frequency confidence map to obtain an artifact-removed electroencephalogram signal; and the fidelity, the robustness and the real-time performance of the electroencephalogram signal are improved.

The invention is suitable for the technical field of intelligent wearable devices and biological signal processing, and provides a heart rate monitoring method and system of an intelligent wearable device, and the method comprises the following steps: detecting the wearing state and the connection state of the device; waking up the sensor module to collect multi-modal data according to the wearing state and the connection state of the equipment; the multi-modal data comprises an ECG signal, a PPG signal and motion data; the multi-modal data is preprocessed, then the ECG signals and the PPG signals are corrected according to the motion data, motion interference is eliminated, and the corrected ECG signals and PPG signals are obtained; selecting a heart rate calculation mode according to the motion data and the corrected ECG signal and PPG signal, and performing real-time calculation to obtain heart rate data; abnormal recognition is carried out on the real-time heart rate data, and multi-level early warning response is carried out. According to the invention, the power consumption defect in the prior art can be solved, the interference of a motion scene is reduced, the health monitoring accuracy and reliability are improved, and the user experience and health management efficiency can be optimized.

The invention belongs to the technical field of biological signal processing and monitoring, and particularly relates to a multi-mode-based non-contact heart rate intelligent monitoring method which comprises the steps that a ballistocardiogram signal of a user is collected through a piezoelectric film sensor, meanwhile, a photoelectric volume pulse wave signal of the user is collected through a photoelectric sensor, and a multi-mode physiological signal is formed; performing space-time alignment processing on the multi-modal physiological signal, eliminating low-frequency breathing interference and motion noise by adopting an adaptive filtering algorithm, and outputting a pre-processed signal; inputting the preprocessed signal into a space-time coupling feature analysis engine, extracting time sequence features through a bidirectional long-short-term memory network, and extracting spatial features through a full convolutional network; adopting a multi-head self-attention mechanism to carry out dynamic weight fusion on the time sequence features and the spatial features to obtain fusion features; and calculating a real-time heart rate value based on the fused features, and judging a heart rate abnormal state. The problem that the heart rate monitoring difficulty is large is well solved.

The application belongs to the technical field of smart home, wearable device, biological signal processing and artificial intelligence control, and specifically discloses a self-adaptive audio adjustment sleep-aiding method based on sleep state recognition, which comprises the following steps: a non-invasive touch interaction mode is introduced to simplify the starting process of the sleep-aiding process; key physiological indexes such as the heart rate and body movement of a user are continuously monitored, and subtle changes in the data are analyzed to determine the physiological state of the user and the reaction of the user to environmental changes, so that the change in the acoustic environment is ensured to be always within the comfortable range that can be accepted by the user; the amplitude of the volume adjustment is controlled to be below the perception threshold of the user, and a special disturbance observation window is set to evaluate the physiological influence of each fine adjustment operation, so that inappropriate adjustment can be found and cancelled in time before substantial interference is caused. The application significantly improves the reliability of the user experience and sleep-aiding effect.

The invention discloses a sign language recognition method based on double-arm electromyographic signals, and belongs to the technical field of human-computer interaction and biological signal processing. The method comprises the steps that multi-channel electromyographic signals generated when sign language gestures are executed are synchronously collected through electromyographic arm rings worn on the left forearm and the right forearm of a user; performing preprocessing and feature extraction on the signal to obtain a time sequence feature sequence of left and right arms; the time sequence feature sequence is input into a pre-trained two-arm collaborative recognition model, and the model outputs a sign language gesture recognition result by fusing the spatial-temporal features of the left arm and the right arm; and finally, the recognized text information is converted into voice to be output. According to the method, the cooperation and time sequence relation of the double-arm electromyographic signals is creatively utilized, the problems that a traditional visual recognition method is greatly interfered by the environment, privacy is invaded, and double-hand linkage complex gestures cannot be effectively analyzed through single-arm electromyographic recognition are solved, and natural, accurate and real-time recognition and translation of the double-hand sign language gestures are achieved.

The application provides a contralateral movement direction prediction method for heterolateral lower limb myoelectric signals and a related device, and belongs to the technical field of human biological signal processing and pattern recognition. The application uses the combined features of the surface myoelectric signals of the left and right lower limbs, the left foot plantar pressure and the right foot three-dimensional acceleration to construct a training set and a test set; a three-class gait prediction model is constructed, the three-class gait prediction model is trained based on the constructed training set, the three-class gait prediction model training result is tested based on the constructed test set, and a trained three-class gait prediction model is obtained; the combined features of the surface myoelectric signals of the left and right lower limbs, the left foot plantar pressure and the right foot three-dimensional acceleration are input into the trained three-class gait prediction model for right foot movement direction prediction, and a right foot movement direction prediction result is obtained. The application solves the problem of low accuracy of contralateral movement direction prediction for heterolateral lower limb myoelectric signals.

The invention belongs to the technical field of smart home, wearable equipment, biological signal processing and artificial intelligence control, and particularly discloses a sleep state recognition-based adaptive audio adjustment sleep-aiding method, which comprises the following steps of: simplifying the starting process of a sleep-aiding process by introducing a non-intrusive touch interaction mode; by continuously monitoring key physiological indexes such as heart rate and body movement of the user and analyzing subtle changes of the data, the physiological state of the user and the response to environmental changes are judged, and it is ensured that the changes of the acoustic environment are always in a comfortable area which can be accepted by the physiology of the user; the volume adjustment amplitude is controlled below the perception threshold value of the user, and the special disturbance observation window is set up to evaluate the physiological influence of each fine adjustment operation, so that improper adjustment can be found and revoked in time before substantive interference is caused. The user experience and the reliability of the sleep aiding effect are remarkably improved.

This invention discloses a method for accurate heart rate calculation based on multi-source multimodal fusion, relating to the fields of biosignal processing and health monitoring technology. The method includes: simultaneously acquiring multi-source array data such as biosignals, environmental information, and motion status, and constructing a time-aligned dataset; preprocessing and extracting features from the data to obtain multi-dimensional features such as pulse peak interval, R-wave interval, environmental interference level, and exercise intensity level; identifying the current dynamic scene based on features and adaptively assigning heart rate calculation weights to photoelectric and electrocardiogram signals; calculating two preliminary heart rate values ​​in parallel and performing weighted fusion; combining individual feature calibration and a time-series trend model to perform anomaly judgment and correction on the preliminary heart rate values, and outputting the final heart rate value and confidence score. This method, through multi-source fusion and scene-adaptive weighting, achieves multi-scene adaptive, anti-interference, and personalized accurate heart rate monitoring.

The invention relates to the technical field of computer vision and biological signal processing, and discloses a multi-task learning-oriented facial heart rate and signal confidence joint prediction method and system. The method aims at solving the problems that in the prior art, only heart rate prediction precision is concerned, confidence coefficient evaluation is neglected, a multi-task cooperation mechanism is missing, loss function design is unbalanced, and an output result is unreliable in a dynamic environment. The method comprises the following steps: collecting and preprocessing face video data, extracting time sequence features, synchronously executing heart rate regression prediction and signal confidence estimation through a shared feature encoder and an independent decoder, balancing each task weight by adopting a self-adaptive weighted multi-task loss function, and post-processing a prediction result. The system comprises a face video data acquisition module, a time sequence feature extraction module, a multi-task joint learning module, a task collaborative optimization module and a post-processing and output module. According to the technical scheme, synchronous output of heart rate prediction and signal confidence estimation can be achieved, and the reliability and practicability of a detection result are improved.

This invention relates to the field of non-contact biosignal processing technology, providing a method, device, electronic device, and storage medium for determining psychological state assessment parameters. The method includes: inputting millimeter-wave radar signals into a psychological state assessment model to obtain psychological state assessment parameters; the psychological state assessment model includes a physiological feature extraction module, a micromotor feature extraction module, a multimodal feature fusion module, and a psychological state inference module. This invention reconstructs electrocardiogram waveforms through the physiological feature extraction module to obtain detailed physiological features, extracts micromotor features through the micromotor feature extraction module, and integrates features reflecting internal physiological changes and external behavioral manifestations through the multimodal feature fusion module. This forms a more comprehensive and multidimensional fused feature set for comprehensive analysis by the psychological state inference module, greatly improving the comprehensiveness and accuracy of psychological state assessment in unconstrained daily environments.

This invention provides the principle and method for influencing cellular functional excitatory expression through misaligned distribution of acoustic wave frequencies, relating to the interdisciplinary field of biological signal processing. The principle and method for influencing cellular functional excitatory expression through misaligned distribution of acoustic wave frequencies includes the following steps: Step 1: Target cell pretreatment, selecting cells to be regulated, separating, purifying, and culturing to the logarithmic growth phase, adjusting the cell density to... 1×104‑1×106cells / cm2 The cells are placed in a biocompatible culture medium. The method of this invention significantly improves the specificity of regulation. Through a frequency-function specific matching mechanism, frequency groups targeting specific cellular functions are screened. Combined with a multidimensional orbital misalignment design to avoid frequency interference, each frequency precisely activates the corresponding signaling pathway. The single-function excitation expression level of cells is more than twice that of the traditional single-frequency method, and the non-specific functional excitation rate is reduced by more than 80%, while the efficiency of action is greatly improved.

The invention discloses a non-contact blood pressure measurement method and system based on transfer learning, and belongs to the technical field of biological signal processing. In order to solve the problems of scarcity of labeled samples and difficulty in model generalization in non-contact blood pressure measurement, knowledge migration from source domain wearable equipment signals to target domain radar signals is realized by constructing a multi-modal feature encoder and a cross-modal attention module and combining a migration learning strategy of source domain pre-training and target domain staged fine tuning. According to the method, systolic pressure and diastolic pressure can be accurately predicted on a small-scale target domain data set, personalized calibration requirements are eliminated, and the precision and applicability of non-contact blood pressure measurement are improved.

The application relates to the technical field of data processing, and discloses a multi-modal neural biological signal processing method and device, a server and a storage medium. The method comprises the following steps: acquiring a multi-modal neural biological signal to be processed, and preprocessing the multi-modal neural biological signal; inputting the preprocessed multi-modal neural biological signal into a deep learning model, extracting deep features of each kind of modal neural biological signal in the multi-modal neural biological signal based on the deep learning model; inputting the deep features into a feature fusion layer for feature fusion to obtain target fusion features; inputting the target fusion features into a regression layer through a full connection layer, performing vital sign prediction on the target fusion features by using the regression layer, and generating a biological vital sign prediction result. Through implementation of the technical solution, the deep features can be effectively captured, the features do not need to be artificially designed or selected, the effective prediction of the biological representation of the neural signal is realized, and the prediction accuracy of the biological representation is improved.

The present application relates to the field of biological signal processing, in particular to a frequency estimation method based on rootMUSIC and signal dimension conversion, a storage medium and equipment. The method comprises the following steps: chest displacement signals extracted from radar echoes are modeled as the sum of sine waves with integer multiple frequencies of basic respiration and heart rate, and corresponding single-channel analytic signals are generated through Hilbert transform; the single-channel analytic signals are converted into multi-channel data through a signal rearrangement method, the chest wall signal model is equivalent to a uniform linear array multi-shot receiving signal model, so that the vital sign frequency and the signal source direction of arrival establish a one-to-one correspondence; the received data of the multi-shot receiving signal model is processed through an alternating direction multiplier method framework, and a denoised signal matrix is solved; the rootMUSIC algorithm is used for the denoised signal matrix to solve the frequency estimation. The present application converts the signal model, makes the frequency estimation range equivalent to the angle estimation range, and improves the frequency estimation accuracy.

The application discloses a body state monitoring method and a wristband type emergency monitoring ring, relates to the technical field of biological signal processing, and the body state monitoring method comprises the following steps: emitting red light of a preset first wavelength to the wrist part of a human body, collecting reflected light signals, obtaining a waveform physiological signal, determining a monitoring index of the human body, inputting the monitoring index into a warning model, obtaining a stress state probability of the human body, judging whether a warning is triggered or not based on the stress state probability and a preset probability threshold, and if the warning is triggered, issuing an alarm and sending corresponding body state data to a preset medical platform. The wristband type emergency monitoring ring can be worn on the wrist of a player to realize convenient detection, and through real-time monitoring of physiological data of the human body and a warning model, stress state detection and prediction are realized, so that the body state of the player can be monitored conveniently, in real time and accurately, and the effect of body monitoring of the player is improved.

The invention relates to a biological signal processing technology, in particular to a noninvasive continuous blood pressure estimation system based on a subject adaptive feature modulation mechanism, which is characterized in that an individual prior embedding vector is generated through an individual prior embedding vector generation module; the subject adaptive feature extraction module performs real-time multi-level dynamic modulation on the preprocessed time sequence physiological signal according to an individual prior embedded vector, and outputs a high-dimensional time sequence feature map; the sequence processing module carries out time sequence dependence modeling and context refining on the high-dimensional time sequence characteristic pattern, and outputs time sequence characteristics; the cross-modal attention fusion module splices the individual prior embedded vectors and the time sequence features; the spliced feature vectors are input into a multi-head attention fusion module, dynamic cross-modal feature weighted fusion is achieved, and high-dimensional feature vectors are obtained; and the blood pressure regression output module is used for mapping the high-dimensional feature vector through a full connection layer and carrying out regression to obtain estimated values of systolic pressure and diastolic pressure. The problem that an existing static mapping normal form cannot deal with individual heterogeneity is solved.

The application relates to the technical field of biological signal processing, and discloses a multi-modal physiological signal coupling analysis method, a system, a terminal and a storage medium. The method comprises the following steps: synchronously collecting multi-modal physiological signals of a subject when the subject performs a specific experimental paradigm; constructing a multi-modal dynamic causal model comprising a neuron dynamics model and an observation model corresponding to each signal; adopting a staged Bayesian inversion strategy to perform parameter estimation, fixing a first type of signal parameter to invert a second type of signal related parameter, fixing the inverted parameter to invert the first type of signal parameter, and obtaining a joint posterior distribution; and finally generating a biomarker of the subject based on the model parameters obtained through inversion. Through the staged inversion strategy, the technical problems of high computational complexity and unstable parameter estimation caused by the large time scale difference of multi-modal data and the large number of model parameters are effectively solved, and the biomarker can be generated and used for clinical motor function evaluation.

The invention relates to the technical field of biological signal processing and intelligent health equipment, in particular to a dream monitoring and visualization method and system based on multi-modal physiological signal fusion. The intelligent monitoring terminal is integrated in the portable sleep eyeshade and comprises a flexible dry electrode array (used for EEG acquisition), an integrated photoelectric sensor (used for PPG signal acquisition), a miniature millimeter wave radar chip (used for non-contact EOG acquisition), a signal processing module, a wireless transmission module, a semiconductor temperature control module and a double-beat audio module; and the user terminal APP is used for receiving, storing and displaying sleep data (a sleep staging chart and REM period information), a dream intensity index (DII) and a dream cloud chart, and generating a sleep health report and trend analysis. The method has the advantages that high-precision recognition is achieved, the problems that a single sensor is prone to interference and high in misjudgment rate are solved through EEG, EOG and PPG three-mode signal fusion judgment, and the REM period recognition accuracy is improved by 40% or above compared with single EEG monitoring.

This invention discloses a method for identifying the physiological state of adult product users based on multi-sensor data acquisition, relating to the fields of intelligent health monitoring and biosignal processing technology. This invention uses non-negative matrix factorization (NMF) technology to decompose multi-channel non-steady-state electromyographic signals collected from the pelvic floor and core muscle groups into multiple muscle coercive elements and their temporal activation coefficients. Each coercive element represents a fixed pattern of muscle cooperating under neural drive, while the activation coefficient reflects the change of this pattern over time. Sexual arousal, as a specific neurophysiological process, induces the activation of coercive elements with specific spatiotemporal patterns. By tracking the activation and evolution of specific coercive patterns, the neural control fingerprint reflecting sexual arousal is extracted from the original signal contaminated by motion noise. This effectively distinguishes between ordinary muscle contractions caused by device use and physical activity and autonomous neuromuscular activities related to sexual responses, overcoming the defects of susceptibility to interference and insufficient specificity.

This invention discloses a dual-domain self-supervised EEG signal representation learning method based on decoupled fusion, belonging to the fields of artificial intelligence and biosignal processing technology. The method includes: acquiring and preprocessing EEG samples, segmenting them into patches; simultaneously extracting and fusing temporal and frequency domain features for each patch; decoupling the features into a temporal-centered view and a spatial-centered view through learnable gating; adding positional encoding and then performing a curriculum-based structured masking strategy; inputting the masked feature stream into a DeFuse encoder, extracting deep spatiotemporal representations through stacked decoupled fusion layers, and predicting the mask content; training the network with a multi-objective loss function including temporal reconstruction loss, spectral fidelity loss, and spatial covariance loss as the optimization objective. This invention can learn structurally complete and generalizable universal representations from unlabeled EEG data, and its performance on downstream tasks such as emotion recognition, motor imagery classification, and anomaly detection is significantly superior to existing methods.

This invention relates to the fields of medical device technology, biosignal processing, and artificial intelligence-assisted diagnostic technology. Specifically, it relates to a non-invasive dynamic monitoring system and method for gastric function based on quantum sensing and multimodal artificial intelligence. The system includes a quantum dot biosensor array, a multimodal signal fusion processor, an artificial intelligence dynamic evaluation algorithm, and a cloud-based collaborative diagnostic platform. The quantum sensor uses graphene quantum dot material and achieves 128-channel parallel acquisition through capacitive coupling. In this invention, by setting up the quantum dot biosensor, the sensitivity reaches the single-photon level, the resolution is 0.1μV, the frequency response range is 0.01-100Hz, and the anti-interference capability is 90dB common-mode rejection ratio, thus improving signal acquisition accuracy and anti-interference capability.

The present application relates to a kind of variable gain chopper amplifier for biological signal sensing system, belong to the field of analog integrated circuit design, including two chopping switches S1 and S2, bias circuit, gain control module, cascode module and filter.First, power supply and bias circuit provide stable power supply voltage and static working point for the whole circuit, then modulate useful signal to chopping frequency through chopping switch S1, signal is converted into a certain amount of current after voltage through gain control module, and then current is converted into amplified ac voltage signal through cascode module, signal is modulated back to baseband through chopping switch S2, finally, high-frequency part is filtered out through low-pass filter, and the amplified low-frequency useful signal is obtained.The present application not only effectively reduces the low-frequency noise of operational amplifier by chopping principle, suppresses the imbalance of amplifier, and realizes the function of variable gain, is suitable for wide input dynamic range low-frequency biological signal processing.

A biological signal measurement system (1) is provided with a biological signal measurement device (10) and a data analysis device (20). A biological signal measurement device (10) is provided with a wireless communication unit (11) and a sensor (14) that measures a biological signal based on a response of a subject to a stimulus. A data analysis device (20) is provided with a data processing unit (24) that performs predetermined data processing on a biological signal, and a wireless communication unit (23) that performs data communication with a wireless communication unit (11). A data analysis device (20) is provided with a stimulation signal generation unit (21) that generates a stimulation signal for applying stimulation. A biological signal measurement device (10) is provided with: a trigger signal acquisition unit (13) that acquires a trigger signal corresponding to a stimulation signal; and a synchronous sampling unit (15) that synchronously samples the trigger signal and the biological signal. The wireless communication unit (23) transmits a stimulation signal to the wireless communication unit (11). The wireless communication unit (11) receives the stimulation signal from the wireless communication unit (23). A wireless communication unit (11) transmits a trigger signal and a biological signal obtained by synchronous sampling to a wireless communication unit (23). The wireless communication unit (23) receives the trigger signal and the biological signal from the wireless communication unit (11). The data processing unit (24) performs predetermined data processing on the basis of the received biosignal and trigger signal.

This invention discloses a signal processing system and method based on dual chopper modulation and demodulation. The invention employs a redundant design of the dual chopper modulation and demodulation network, and through a switching mechanism between the primary and backup networks, significantly improves the reliability and stability of the system in complex application environments such as aerospace, effectively solving the problem of single chopper network failure. The redundancy switching controller can automatically select the optimal signal path based on the real-time detection results of the status monitor, ensuring seamless switching between the primary and backup networks, improving the system's adaptability and response speed. The dual chopper modulation and demodulation network uses a periodic modulation strategy, which can effectively suppress low-frequency noise while maintaining high signal accuracy and stability, overcoming the limitations of traditional chopper technology in low-frequency noise processing. This invention has strong engineering practicality, can meet the application requirements of wide input dynamic range and high precision, and is suitable for various complex biological signal processing scenarios.

The application provides an electroencephalogram deartifact method, device, equipment and medium, relates to the technical field of biological signal processing, and processes collected electroencephalogram data to obtain optimal modal number and optimal bandwidth parameters, which are used for adaptive variational modal decomposition of the electroencephalogram data, obtaining a plurality of modal components, then performing multidimensional feature analysis to obtain a plurality of feature indexes, which are used for determining artifact suspected modal and effective modal components; performing short-time Fourier transform on the artifact suspected modal to construct a time-frequency confidence chart to guide weighted time-frequency independent component separation on the artifact suspected modal to obtain artifact components; suppressing the artifact components to obtain suppressed independent components, and performing weighted reconstruction and multi-component fusion on the suppressed independent components and the effective modal components based on the time-frequency confidence chart to obtain deartifact electroencephalogram; and the fidelity, robustness and real-time performance of the electroencephalogram are improved.

The present application provides a biological health system, a physiological state regulation method, a wearable device, and a storage medium. The biological health system comprises: a detection assembly configured to collect a biological signal of a user in real time; a processing assembly configured to calculate, on the basis of a phase of the biological signal, a physiological state regulation parameter and generate, on the basis of the physiological state regulation parameter, a stimulation signal; and an output assembly configured to output the stimulation signal, so that an coupling effect of the biological signal and the stimulation signal regulates a physiological state of the user. In the present application, the biological signal of the user is acquired in real time, and the personalized physiological state regulation parameter of the user is calculated on the basis of the phase of the biological signal so as to generate the stimulation signal in real time, thereby implementing physiological state regulation of the user by means of the coupling effect of the stimulation signal and the biological signal.

The present application relates to biological signal processing technology, and is a non-invasive continuous blood pressure estimation system based on a subject adaptive feature modulation mechanism, an individual prior embedding vector is generated by an individual prior embedding vector generation module; a subject adaptive feature extraction module performs real-time multi-level dynamic modulation on the preprocessed time series physiological signal according to the individual prior embedding vector, and outputs a high-dimensional time series feature map; a sequence processing module performs time series dependence modeling and context refining on the high-dimensional time series feature map, and outputs a time series feature; a cross-modal attention fusion module splices the individual prior embedding vector and the time series feature; the spliced feature vector is input into a multi-head attention fusion module to realize dynamic cross-modal feature weighted fusion and obtain a high-dimensional feature vector; and a blood pressure regression output module maps the high-dimensional feature vector through a full connection layer to regress to obtain an estimated value of systolic pressure and diastolic pressure. The present application solves the problem that the existing static mapping paradigm cannot cope with individual heterogeneity.

The invention provides a lamb behavior recognition system and method based on motion monitoring, and relates to the technical field of flexible sensing and biological signal processing, and the method comprises the steps: obtaining a lamb motion signal and a behavior video, carrying out the feature extraction of the motion signal and the behavior video, constructing a lamb motion-visual fusion block, constructing a behavior feature input matrix based on the lamb motion-visual fusion block, and carrying out the recognition of the lamb behavior. Constructing a lamb behavior recognition model based on multi-modal data fusion and multi-layer perception, training and verifying the lamb behavior recognition model based on the behavior feature input matrix, adjusting related parameters according to a verification result, and performing model evaluation to obtain a final lamb behavior recognition model, and obtaining an unknown lamb behavior feature input matrix. And inputting the lamb behavior recognition result into a final lamb behavior recognition model, and outputting a corresponding behavior recognition result to complete lamb behavior recognition. The behavior recognition can be realized based on the motion signal and the behavior video.

The invention relates to a Riemannian manifold motor imagery electroencephalogram decoding method combined with a U network, and relates to the field of biological signal processing and brain-computer interfaces, and the method comprises the following steps: a multi-scale space-time convolution module extracts a multi-scale feature map from motor imagery electroencephalogram signals by using a space convolution layer and a time convolution layer; the Riemann encoder structure module encodes the multi-scale feature map into features of different levels; the Riemann decoder structure module performs reverse mapping on the low-level features and the high-level features in the Riemann encoder based on a jump connection method of the Riemann center of gravity, and reconstructs to obtain decoding features; a reconstruction error is calculated by comparing decoding features with an original feature map, and then a U network formed by a Riemann encoder structure module and a Riemann decoder structure module is optimized. According to the method, neglect of non-Euclidean features in the conventional decoding process is remarkably relieved, the sensitivity of the network to the spatial-temporal features of the electroencephalogram signals is improved, and therefore a feasible technical solution is provided for development of brain-computer interfaces.

The application provides a breathing signal processing method and system, and relates to the technical field of biological signal processing and deep learning. The method comprises the following steps: analyzing a first breathing signal by using a sliding window, and calculating the moving average standard deviation of the breathing signal in the sliding window; if the moving average standard deviation is less than the maximum standard deviation of apnea, the count is increased by 1, otherwise the count is cleared; if the count reaches the count threshold, an apnea event occurs; the peak value of the first breathing signal is detected by using the sliding window, if the peak value is greater than the minimum wave peak threshold and the interval between adjacent wave peaks is greater than the preset time length, a sigh event occurs; the breathing signal of the apnea and sigh event is removed to obtain a second breathing signal; the time domain and space domain features of the second breathing signal are extracted; the time domain and space domain features are subjected to multi-modal feature fusion, and the second breathing signal is subjected to breathing event classification based on the multi-dimensional feature representation obtained by fusion. The recognition accuracy and efficiency of key breathing events such as apnea and sigh can be significantly improved.