Adult product user physiological state recognition method based on multi-sensor data collection

Abstract

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.

Description

Technical Field

[0001] This invention relates to the field of intelligent health monitoring and biosignal processing technology, specifically a method for identifying the physiological state of adult product users based on multi-sensor data acquisition. Background Technology

[0002] With the development of intelligent and personalized health technologies, the adult product industry is gradually evolving from simple mechanical drives to intelligent devices with physiological sensing and interactive capabilities. Integrating biosensors to monitor users' physiological responses in real time, and accurately and reliably identifying complex physiological states during use, especially the activity of neuromuscular and autonomic nervous systems related to sexual response, and dynamically adjusting the device's operating mode accordingly to enhance safety and personalization, has become an important development trend.

[0003] In existing technologies, there are significant limitations to inferring user status by integrating a single sensor: In dynamic usage scenarios of adult products, strong active movements and device vibrations generate a large number of motion artifacts, which seriously contaminate physiological signals. Traditional filtering methods are difficult to effectively remove these artifacts, resulting in a low signal-to-noise ratio.

[0004] Existing methods mostly rely on simple threshold judgments of indirect physiological parameters such as heart rate and skin conductance. These signals lack specificity and are easily affected by various factors such as emotions, environment, and physical condition, making it impossible to accurately distinguish and arouse related specific neuromuscular patterns from general physical activities or states of tension.

[0005] In privacy-sensitive scenarios, there is a lack of systematic solutions for achieving accurate status recognition and personalized services without disclosing raw biometric data.

[0006] Therefore, there is an urgent need in this field for a technical solution that can adaptively and privacy-preservingly identify the core physiological state of users in complex usage environments with high dynamics and strong interference, so as to support reliable interaction and personalized experience of smart adult products. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for recognizing the physiological state of adult product users based on multi-sensor data acquisition. This method integrates a multi-channel surface electromyography (EMG) sensor array, an inertial measurement unit, and a pressure sensor. It employs dynamic analysis technology based on muscle coordination patterns to enhance the specificity and anti-interference capability of state recognition. Through non-negative matrix factorization, it decomposes multi-channel non-steady-state EMG signals collected from the pelvic floor and core muscle groups into multiple muscle coordinators and their temporal activation coefficients. Each coordinator represents a fixed pattern of muscle coordination 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 coordinators with specific spatiotemporal patterns. By tracking the activation and evolution of specific coordinator 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 shortcomings of existing technologies that rely on indirect, single signals, are susceptible to interference, and lack specificity.

[0008] To solve the above-mentioned technical problems, this invention provides the following technical solution: a method for identifying the physiological state of adult product users based on multi-sensor data acquisition, the specific steps of which are as follows: S100: Collects multi-channel surface electromyography (EMG) signals of the pelvic floor and core muscle groups through a multi-channel surface EMG sensor array integrated into the body of the adult product; simultaneously, it collects motion data through an inertial measurement unit and contact pressure distribution data through a pressure sensor array. S200. Adaptive preprocessing is performed on the acquired multi-channel surface electromyography signals, including motion artifact elimination based on the motion data of the inertial measurement unit, and signal quality evaluation and weighting based on the contact pressure distribution data. S300: The preprocessed multi-channel surface electromyography signal is decomposed using non-negative matrix factorization technology, and muscle coercisors that characterize neural control patterns and their activation coefficients over time are extracted and tracked to construct a personalized muscle coercisor library. S400. Extract physiological state features based on the muscle coercion and its activation coefficient, including co-activation features, neural control features, and multimodal fusion features fused with the motion data and the contact pressure distribution data; S500: Input the extracted physiological state features into the physiological state classification model to identify multiple physiological states, including sexual arousal-specific states, and generate feedback adjustment instructions for controlling the working mode of the adult product based on the identification results.

[0009] Furthermore, the multi-channel surface electromyography sensor array is used to collect electromyographic activity signals of the pelvic floor muscles and related muscle groups; The inertial measurement unit is used to collect the motion posture, frequency, and amplitude signals of the adult product body; The pressure sensor array is used to collect the spatial distribution and dynamic waveform signals of contact pressure.

[0010] Furthermore, the adaptive preprocessing in S200 specifically includes: Based on the impedance signal output by the contact quality detection electrode, the input impedance and gain of the surface electromyography signal amplifier of the corresponding channel are adjusted in real time, and the data of the channel with poor contact are marked. The acceleration and angular velocity data collected by the inertial measurement unit are used to construct a motion artifact reference signal, and a normalized least mean square adaptive filter is used to filter out the interference components related to the motion artifact reference signal from the original surface electromyography signal. Based on the data from the pressure sensor array, the contact stability index of the sensor contact area of ​​each channel is calculated, and higher weights are assigned to the channel data with higher contact stability index values.

[0011] Furthermore, in step S300, the multi-channel surface electromyography (EMG) signals within each time window are treated as an observation matrix. This observation matrix is ​​then decomposed using non-negative matrix factorization (NMF) into a muscle synergy element matrix representing a fixed muscle synergy pattern and an activation coefficient matrix representing the activation intensity of each pattern over time. Specific steps include: The preprocessed multichannel surface electromyography signals were divided according to overlapping time windows; For the surface electromyography (EMG) signal matrix within each time window, its nonnegative matrix factorization is solved. The objective function is to minimize the reconstruction error, with a sparsity constraint on the activation coefficient matrix. The objective function is: The sparsity constraint is: ,in, For the current time window containing One channel, Surface electromyography signal matrix at each time point This is the muscle co-operation element matrix, where each column represents a muscle co-operation element, characterizing the participation weight of different muscles in this co-operation mode. This is an activation coefficient matrix, where each row represents the activation intensity sequence of the corresponding muscle coercive element within a time window. These are regularization coefficients used to control the activation coefficient matrix. The sparsity, Denotes the Frobenius norm. express Norm; The number of muscle coerci within each time window was determined using the minimum description length criterion. The value; The muscle co-element matrix obtained from the decomposition Clustering and updating are performed to maintain a dynamic and personalized muscle collaboration meta-library; In real-time analysis, for new surface electromyography (EMG) signal data points, the known muscle coordination element matrix is ​​solved. The activation coefficient vector is used to track the real-time activation intensity of each muscle coercor.

[0012] Furthermore, the construction and updating process of the personalized muscle coordination meta-library is as follows: During the initial calibration phase, users are guided to perform preset action patterns, and a personalized muscle synergy library is initialized based on the decomposition results. During real-time use, the residual energy between the surface electromyography (EMG) signal matrix and the current muscle co-resource library reconstruction signal is calculated. When the residual energy continuously exceeds the preset first threshold, the muscle coordination library is updated, and the newly emerging muscle coordination mode is added to the personalized muscle coordination library.

[0013] Furthermore, in S400: The co-activation features include: dominant muscle cofactors and their activation weights, the start time and duration of muscle cofactor activation, and the phase synchronization index between different muscle cofactor activation sequences. The neural control features include: the differential entropy of the activation coefficient sequence, the frequency of switching of dominant muscle coercive elements per unit time, and the sparsity of the activation coefficient vector. The multimodal fusion features include: the correlation between specific muscle co-activation modes and inertial motion data, and the correspondence between muscle co-activation modes and local pressure change modes.

[0014] Furthermore, in S500, the physiological state classification model is a hierarchical hybrid classification model built on machine learning and deep learning, and is generated by training on historical labeled data. The historical labeled data contains multiple sets of time-series physiological state feature vectors and their corresponding physiological state labels. The physiological state classification model includes a first layer of machine learning-based abnormal state filter, a second layer of deep learning-based time-series pattern analyzer, and a third layer of machine learning-based multi-state classifier.

[0015] Furthermore, the working principle of the physiological state classification model in recognizing physiological states is as follows: The first layer of filtering involves inputting the real-time extracted physiological state feature vectors into the abnormal state filter. The abnormal state filter is a single-class support vector machine used to identify and filter out data points that deviate from the normal physiological pattern and mark them as invalid signals. The second layer of temporal analysis involves inputting the filtered temporal physiological state feature vector sequence into the temporal pattern analyzer. The temporal pattern analyzer is a one-dimensional convolutional neural network used to extract higher-order patterns from the temporal changes of the feature vectors and output a state evolution pattern encoding vector that represents the current state evolution pattern. The third layer of state determination involves fusing the state evolution pattern encoding vector with the current physiological state feature vector and inputting it into the multi-state classifier. The multi-state classifier outputs the current physiological state category label, the confidence score corresponding to different categories, and the state evolution trend prediction.

[0016] Furthermore, the physiological state category labels can be configured to include: low arousal baseline state, high arousal excitation state, excessive tension inhibition state, and muscle coordination disorder state. The recognition results output by the physiological state classification model include: the physiological state category label determined at the current moment, and the confidence score corresponding to the physiological state category label.

[0017] Furthermore, the feedback adjustment command is used to control the stimulation parameters of the adult product, and the specific control logic includes: When the physiological state classification model identifies a physiological state category label belonging to the positive feedback state category and its confidence score is higher than the first preset threshold, a first type of feedback regulation instruction is generated to maintain the current stimulation mode. The positive feedback state category includes the high arousal state. When the physiological state classifier model identifies a physiological state category label belonging to the negative regulation state category and its confidence score is higher than the second preset threshold, a second type of feedback regulation instruction is generated to reduce the stimulus intensity and switch modes. The negative regulation state category includes excessive tension inhibition state and muscle coordination disorder state.

[0018] Compared with existing technologies, this method for recognizing the physiological state of adult product users based on multi-sensor data acquisition has the following advantages: I. This invention integrates a multi-channel surface electromyography (EMG) sensor array, an inertial measurement unit, and a pressure sensor. It employs a muscle synergy pattern dynamic analysis technique to enhance the specificity and anti-interference capability of state recognition. Through non-negative matrix factorization, it decomposes the multi-channel non-steady-state EMG signals collected from the pelvic floor and core muscle groups into multiple muscle synergy elements and their temporal activation coefficients. Each synergy element represents a fixed pattern of muscle synergy 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 synergy elements with specific spatiotemporal patterns. By tracking the activation and evolution of specific synergy 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 response, overcoming the shortcomings of existing technologies that rely on indirect, single signals, are susceptible to interference, and lack specificity.

[0019] Second, this invention achieves adaptive learning and privacy security by constructing a personalized muscle coordination meta-library and a hierarchical hybrid classification model. During initialization or use, the muscle coordination meta-library is dynamically constructed and updated by analyzing user data, making the model base adapt to the user's unique physiological characteristics. It adopts a three-layer classification architecture of anomaly filtering, time series analysis, and state determination to deeply understand the personalized patterns of user state evolution, thereby improving classification accuracy. It continuously learns and adapts to individual user patterns, ensuring that sensitive raw physiological data is not separated from the user's device, thus resolving the contradiction between personalized needs and privacy protection.

[0020] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0022] Figure 1 A flowchart illustrating the steps of a method for recognizing the physiological state of adult product users based on multi-sensor data acquisition. Figure 2 This is a flowchart illustrating the construction and updating process of the personalized muscle synergy library in an embodiment of the present invention; Figure 3This is a flowchart of a method for recognizing the physiological state of adult product users based on multi-sensor data acquisition. Detailed Implementation

[0023] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] To address the shortcomings of existing smart adult products, such as susceptibility to motion artifacts in dynamic usage scenarios, insufficient specificity in state recognition, and the difficulty in balancing personalized needs and privacy protection, this invention provides a method for recognizing the physiological state of adult product users based on multi-sensor data acquisition. This method aims to robustly and specifically identify the core physiological states related to sexual response from highly interfering and non-steady-state physiological signals by fusing data from multi-channel surface electromyography (EMG), inertial measurement units (IMU), and pressure sensors, and by introducing dynamic analysis and personalized modeling techniques for muscle coordination patterns. Based on this, the core physiological states of the user related to sexual response can be identified robustly and specifically, enabling personalized adaptive adjustment of the device's operating mode. This invention is primarily applied to smart adult products with physiological sensing and interaction capabilities. During the use of such devices, the user's active body movements, the device's vibrations, and complex physiological responses are coupled together, which traditional methods based on heart rate or simple EMG amplitude cannot identify. This invention constructs a complete technical system from physiological signal acquisition, specific feature extraction, to accurate state recognition and closed-loop feedback by using multi-sensor synchronous acquisition, adaptive signal preprocessing based on motion and contact data, non-negative matrix factorization-driven muscle cofactor extraction and tracking, multimodal feature fusion, and hierarchical hybrid classification model.

[0025] Specifically, such as Figure 1 As shown, the detailed implementation process of the method for recognizing the physiological state of adult product users based on multi-sensor data acquisition is as follows: The specific steps of this method are: S100: Collects multi-channel surface electromyography (EMG) signals of the pelvic floor and core muscle groups through a multi-channel surface EMG sensor array integrated into the body of the adult product; simultaneously, it collects motion data through an inertial measurement unit and contact pressure distribution data through a pressure sensor array. S200. Adaptive preprocessing is performed on the acquired multi-channel surface electromyography signals, including motion artifact elimination based on the motion data of the inertial measurement unit, and signal quality evaluation and weighting based on the contact pressure distribution data. S300: The preprocessed multi-channel surface electromyography signal is decomposed using non-negative matrix factorization technology, and muscle coercisors that characterize neural control patterns and their activation coefficients over time are extracted and tracked to construct a personalized muscle coercisor library. S400. Extract physiological state features based on the muscle coercion and its activation coefficient, including co-activation features, neural control features, and multimodal fusion features fused with the motion data and the contact pressure distribution data; S500: Input the extracted physiological state features into the physiological state classification model to identify multiple physiological states, including sexual arousal-specific states, and generate feedback adjustment instructions for controlling the working mode of the adult product based on the identification results.

[0026] In practice, after the adult product is powered on and enters its working mode, the multi-channel sEMG sensor array, IMU, and pressure sensor array begin to synchronously and continuously acquire data. The multi-channel sEMG sensor array, in the form of flexible circuits or fabric electrodes, is deployed at the points of contact with the user's pelvic floor muscles and related core muscles to collect high-density electromyographic activity signals from these muscle groups. The IMU, containing a triaxial accelerometer and gyroscope, is fixed to the main internal structure of the device to accurately acquire the motion posture, frequency, and amplitude signals of the adult product relative to the user's body. The pressure sensor array, in the form of a distributed matrix or thin film, is integrated at the main contact interface between the device and the user to collect the spatial distribution of contact pressure and its dynamic waveform changes in real time. All sensors are triggered and acquired synchronously to ensure data time alignment, laying the foundation for subsequent multimodal fusion analysis.

[0027] High-quality neuromuscular activity information is recovered from raw sEMG signals contaminated by motion interference and contact noise through adaptive preprocessing. The adaptive preprocessing specifically includes: Contact quality-based signal adjustment and labeling: Through the contact quality detection electrode, a signal reflecting the impedance of the electrode-skin interface is output. This impedance signal is monitored in real time. When an abnormal increase in the impedance of a certain channel is detected, the input impedance and gain of the preamplifier of that channel are automatically adjusted to optimize signal pickup. The data frames acquired by that channel are marked as low quality or suspicious, and their weight is reduced or they are excluded in subsequent analysis.

[0028] IMU-based motion artifact elimination: Severe active motion or equipment vibration can introduce motion-related artifacts into sEMG signals. This invention utilizes acceleration and angular velocity data acquired by an IMU to construct a motion artifact reference signal. A normalized least mean square adaptive filter is used, with the original sEMG signal as the main input and the motion artifact reference signal as the reference input. Through iterative learning, the filter estimates the transmission characteristics of motion artifacts in the sEMG signal and dynamically subtracts the estimated motion artifact components from the original signal, thereby outputting a clean sEMG signal that eliminates most motion interference.

[0029] Contact stability assessment based on pressure data: A contact stability index is calculated for the contact area corresponding to each electromyography (EMG) signal acquisition channel based on real-time data from the pressure sensor array. This index is then used to assign weights for subsequent analysis of the preprocessed surface EMG signals from each channel. The contact stability index... Regarding the first Each electromyography (EMG) signal acquisition channel, through its corresponding pressure sensor, acquires signals within the most recent time window. Internal pressure reading sequence The calculation shows that, ,in. Indicates time window Internal pressure reading The average value is used to assess the average contact pressure level. Indicates time window Internal pressure reading The standard deviation is used to assess the degree of fluctuation in contact pressure. Indicates time window The mean absolute value of the first derivative of the internal pressure reading is used to assess the drasticness of pressure changes. This is the maximum range of the pressure sensor, used to normalize the average pressure. These are weighting coefficients, and This is used to adjust the contribution weights of the average value, stability, and rate of change in the final index, resulting in the calculated contact stability index. The value ranges from 0 to 1. The closer the value is to 1, the more stable and reliable the electrode-skin contact of the channel is. In the subsequent dynamic analysis of muscle synergy patterns, the surface electromyography signal data of this channel will be given higher weight.

[0030] After S200 preprocessing, a multi-channel sEMG signal sequence with motion artifact suppression and stability assessment labels was obtained, along with time-synchronized IMU motion and stress distribution data. The underlying neural strategies controlling muscle activity, i.e., muscle synergy patterns, were extracted from the multi-channel sEMG signals. The preprocessed multi-channel sEMG signals were divided into overlapping time windows. For each time window, the sEMG signal data from m channels and n time points were arranged into an observation matrix. .

[0031] For each time window, the EMG matrix is ​​decomposed into a nonnegative matrix factorization (NMF) that is the product of two nonnegative matrices: ,in, This is called the muscle co-operational matrix, where each column represents a muscle co-operational element. Each element in this vector represents the participation weight of the corresponding muscle channel in that co-operational mode. This is called the activation coefficient matrix, where each row represents the activation intensity sequence of the corresponding muscle cofactor over time within that time window. k is the number of muscle cofactors decomposed. The objective function with sparsity constraints is used to solve this problem. ,in, Denotes the Frobenius norm. Describing the L1 norm, is the regularization coefficient, used to encourage the sparsity of the activation coefficient matrix H.

[0032] For each time window, the minimum description length (MDL) criterion is used to determine the optimal number of muscle synergists k. When the user uses the device for the first time or during recalibration, the user is guided to perform a series of preset action patterns. The sEMG data under these actions is collected, and NMF decomposition is performed. The muscle synergists W that appear stably in multiple repetitions are collected to form the user's initial personalized muscle synergist library.

[0033] During real-time use, the current user's muscle co-element library is maintained. For new data points or time windows, given the known (or last updated) co-element matrix W, the activation coefficient vector h, i.e., a column of H, is solved using optimization methods such as non-negative least squares. This enables real-time tracking of the activation intensity of each co-element. Specifically, for example... Figure 2 As shown, the construction and updating process of the personalized muscle synergy meta-library is as follows: During the initial calibration phase, users are guided to perform preset action patterns, and a personalized muscle synergy library is initialized based on the decomposition results. During real-time use, the residual energy between the surface electromyography (EMG) signal matrix and the current muscle co-resource library reconstruction signal is calculated. When the residual energy continuously exceeds the preset first threshold, the muscle coordination library is updated, and the newly emerging muscle coordination mode is added to the personalized muscle coordination library.

[0034] Based on the muscle coerci and their activation coefficient sequences obtained from S300, and combined with IMU motion data and pressure data, multi-dimensional features characterizing the user's physiological state are extracted, including co-activation features, neural control features, and multimodal fusion features fused with the motion data and the contact pressure distribution data, wherein: Co-activation features: extracted from the activation coefficient matrix H, including: dominant co-element index and its activation weight (the co-element with the highest activation intensity in the current time window); activation temporal features (start time, duration, and rising / falling slope of co-element activation); and coupling relationship between co-elements (phase synchronization index and coherence between activation sequences of different co-elements).

[0035] Neural control characteristics: extracted from the dynamic properties of activation coefficient sequences, reflecting the complexity and patterns of neural drive, including: differential entropy of activation sequences, characterizing the complexity or uncertainty of neural control; dominant coordinator switching frequency, the number of times the dominant coordinator changes per unit time, reflecting the flexibility of neural strategies; sparsity of activation coefficient vectors, reflecting the degree of concentration of muscle activity in coordinating patterns.

[0036] Multimodal fusion features: associating muscle synergy patterns with motion and stress information, including: the correlation between the activation intensity of specific coercors and the motion amplitude / frequency measured by the IMU; the temporal correspondence or mutual information between the pressure change patterns of specific pressure distribution areas and the activation patterns of specific muscle coercors, used to distinguish whether it is specific neuromuscular activity related to sexual arousal or muscle contraction caused by body movement or tension.

[0037] The time-series physiological state feature vectors extracted from the S400 are input into a pre-trained hierarchical hybrid classification model, which ultimately outputs the physiological state category and confidence level. The physiological state classification model employs a three-layer architecture. The first layer (abnormal state filter) uses a single-class support vector machine learning model, trained with feature vectors from normal usage states. This filter removes obviously abnormal or unreliable feature data caused by strenuous exercise, transient sensor malfunctions, etc., marking them as invalid and excluding them from subsequent analysis. The second layer (time-series pattern analyzer) uses a one-dimensional convolutional neural network (1D-CNN) or a long short-term memory network (LSTM). The input is the time-series feature vector sequence filtered by the first layer. This network automatically learns high-order, state transition-related abstract representations from the patterns of feature evolution over time, outputting a state evolution pattern encoding vector that condenses the current state evolution information. The third layer (multi-state classifier): employs a support vector machine or random forest. The input is a fusion vector of the state evolution pattern encoding vector output from the second layer and the original feature vector at the current moment. The output is the final physiological state category label and confidence scores for each category. The training of the model relies on historical labeled data, including sensor data collected synchronously by multiple users during device use, extracted feature vector sequences, and physiological state labels. The physiological state category labels can be configured to include: low arousal basal state, high arousal excitation state, excessive tension inhibition state, and muscle coordination disorder state.

[0038] During real-time operation, S100-S400 are continuously executed to generate a real-time feature vector stream. This real-time feature vector stream first passes through the first layer filter to remove outliers. The most recent effective feature sequence is then sent to the second layer time series analyzer to obtain the current state evolution code. This code is then fused with the features from the latest time and input into the third layer classifier to obtain the physiological state category label and its confidence score at the current time.

[0039] The recognition results of the hierarchical hybrid classification model are converted into control commands for the actuator of adult products. In this embodiment, the control logic is as follows: When a highly aroused state is detected and the confidence level is higher than the high threshold, a first-type feedback regulation instruction is generated to control the device to maintain the current effective stimulation mode or to make minor optimizations to prolong or deepen the positive state.

[0040] When an excessively tense or disordered state of muscle coordination is detected and the confidence level is higher than the medium threshold, a second type of feedback regulation instruction is generated. The control device immediately reduces the stimulation intensity, changes the stimulation frequency, or switches to a gentler mode, aiming to relieve the user's tension or discomfort and guide the physiological state to recover in a more coordinated direction.

[0041] For low arousal baseline states, different stimulation modes are explored based on preset personalized plans.

[0042] The entire data processing and model operation process is designed to be completed locally on the device. The personalized muscle coordination meta-library and classification model parameters are stored locally on the device, eliminating the need to upload raw physiological data to the cloud. Model updates are also based on incremental learning using local data, fully protecting user privacy.

[0043] like Figure 3 As shown, this embodiment also provides a method for identifying the physiological state of users of adult products based on multi-sensor data acquisition. The specific steps for identifying the physiological state of users of adult products are as follows: (1) Synchronous acquisition of data from multiple sensors A multi-channel surface electromyography sensor array, an inertial measurement unit, and a pressure sensor array are integrated into the body of the adult product.

[0044] Start the device and simultaneously collect surface electromyography signals, motion data, and pressure distribution data.

[0045] (2) Adaptive preprocessing of surface electromyography signals Based on the impedance signal output by the contact quality detection electrode, the input impedance and gain of the surface electromyography signal amplifier are adjusted in real time.

[0046] Data from channels with poor contact are marked to reduce their weight in subsequent analysis.

[0047] Motion artifact reference signals are constructed using acceleration and angular velocity data collected by the inertial measurement unit.

[0048] A normalized least mean square adaptive filter is used to filter out motion artifact interference from the original surface electromyography signal.

[0049] (3) Muscle synergist extraction and personalized library construction The preprocessed multichannel surface electromyography signals were divided according to overlapping time windows.

[0050] Nonnegative matrix decomposition was performed on the surface electromyography signal matrix within each time window.

[0051] The decomposition yields fixed muscle synergy patterns (muscle synergy element matrix) and the activation intensity of each pattern over time (activation coefficient matrix).

[0052] The number of muscle coerci within each time window is automatically determined using the minimum description length criterion.

[0053] Guide users to perform preset action patterns and initialize a personalized muscle coordination meta-library based on the decomposition results.

[0054] Calculate the residual energy between the current signal and the reconstructed signal from the co-resource library.

[0055] If the residual energy continues to exceed the preset threshold, the library will be updated, and the new co-element will be added to the library.

[0056] For new surface electromyography data points, the activation coefficients are solved under the known coercive matrix to achieve real-time tracking of the activation intensity of each coercive element.

[0057] (4) Extraction of features of multimodal physiological states Extracting co-activation features, neural control features, and multimodal fusion features. Construct a hierarchical hybrid classification model: anomaly state filter, time series pattern analyzer, and multi-state classifier.

[0058] The model is trained using historical labeled data.

[0059] The physiological state feature vectors extracted in real time are input into the model.

[0060] First layer: Filtering outlier data points.

[0061] The second layer analyzes the temporal characteristics and outputs the state evolution pattern encoding vector.

[0062] The third layer: fuses the encoded vector with the current features, and outputs the physiological state category label and confidence score.

[0063] (5) Feedback control command generation If the state is identified as highly aroused and has a high confidence level, an instruction to maintain the current stimulation pattern is generated.

[0064] If the condition is identified as an over-tense inhibition state or a state of muscle coordination disorder with a high confidence level, an instruction to reduce the stimulus intensity and switch modes is generated.

[0065] If the low-awakening base state is identified, generate a mode exploration command.

[0066] Feedback adjustment of adult products is performed based on adjustment instructions.

[0067] In summary, this invention achieves highly robust and accurate identification of users' specific physiological states in complex usage environments through a complete system from hardware acquisition, signal processing, feature extraction to intelligent recognition. Based on this, it forms a personalized and adaptive closed-loop interaction, significantly improving the user experience and safety of intelligent adult products.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for identifying the physiological state of adult product users based on multi-sensor data acquisition, characterized in that, The specific steps of this method are as follows: S100: Collects multi-channel surface electromyography (EMG) signals of the pelvic floor and core muscle groups through a multi-channel surface EMG sensor array integrated into the body of the adult product; simultaneously, it collects motion data through an inertial measurement unit and contact pressure distribution data through a pressure sensor array. S200. Adaptive preprocessing is performed on the acquired multi-channel surface electromyography signals, including motion artifact elimination based on the motion data of the inertial measurement unit, and signal quality evaluation and weighting based on the contact pressure distribution data. S300: The preprocessed multi-channel surface electromyography signal is decomposed using non-negative matrix factorization technology, and muscle coercisors that characterize neural control patterns and their activation coefficients over time are extracted and tracked to construct a personalized muscle coercisor library. S400. Extract physiological state features based on the muscle coercion and its activation coefficient, including co-activation features, neural control features, and multimodal fusion features fused with the motion data and the contact pressure distribution data; S500: Input the extracted physiological state features into the physiological state classification model to identify multiple physiological states, including sexual arousal-specific states, and generate feedback adjustment instructions for controlling the working mode of the adult product based on the identification results.

2. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 1, characterized in that, The multi-channel surface electromyography sensor array is used to collect electromyographic activity signals of the pelvic floor muscles and related muscle groups; The inertial measurement unit is used to collect the motion posture, frequency, and amplitude signals of the adult product body; The pressure sensor array is used to collect the spatial distribution and dynamic waveform signals of contact pressure.

3. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 1, characterized in that, The adaptive preprocessing in S200 specifically includes: Based on the impedance signal output by the contact quality detection electrode, the input impedance and gain of the surface electromyography signal amplifier of the corresponding channel are adjusted in real time, and the data of the channel with poor contact are marked. The acceleration and angular velocity data collected by the inertial measurement unit are used to construct a motion artifact reference signal, and a normalized least mean square adaptive filter is used to filter out the interference components related to the motion artifact reference signal from the original surface electromyography signal. The contact stability index of each channel sensor contact area is calculated based on the data from the pressure sensor array.

4. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 1, characterized in that, In step S300, the multi-channel surface electromyography (EMG) signals within each time window are used as an observation matrix. The observation matrix is ​​then decomposed using non-negative matrix factorization (NMF) into a muscle synergy element matrix representing a fixed muscle synergy pattern and an activation coefficient matrix representing the activation intensity of each pattern over time. Specific steps include: The preprocessed multichannel surface electromyography signals were divided according to overlapping time windows; For the surface electromyography (EMG) signal matrix within each time window, its nonnegative matrix factorization is solved. The objective function is to minimize the reconstruction error, with a sparsity constraint on the activation coefficient matrix. The objective function is: The sparsity constraint is: ,in, This is a matrix of surface electromyography (EMG) signals containing m channels and n time points within the current time window. This is the muscle co-operation element matrix, where each column represents a muscle co-operation element, characterizing the participation weight of different muscles in this co-operation mode. This is an activation coefficient matrix, where each row represents the activation intensity sequence of the corresponding muscle coercive element within a time window. These are regularization coefficients used to control the activation coefficient matrix. The sparsity, Denotes the Frobenius norm. express Norm; The value of the number of muscle coerci k within each time window is determined by the minimum description length criterion. The muscle co-element matrix obtained from the decomposition Clustering and updating are performed to maintain a dynamic and personalized muscle collaboration meta-library; In real-time analysis, for new surface electromyography (EMG) signal data points, the known muscle coordination element matrix is ​​solved. The activation coefficient vector is used to track the real-time activation intensity of each muscle coercor.

5. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 4, characterized in that, The construction and updating process of the personalized muscle coordination meta-library is as follows: During the initial calibration phase, users are guided to perform preset action patterns, and a personalized muscle synergy library is initialized based on the decomposition results. During real-time use, the residual energy between the surface electromyography (EMG) signal matrix and the current muscle co-resource library reconstruction signal is calculated. When the residual energy continuously exceeds the preset first threshold, the muscle coordination library is updated, and the newly emerging muscle coordination mode is added to the personalized muscle coordination library.

6. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 1, characterized in that, In S400: The co-activation features include: dominant muscle cofactors and their activation weights, the start time and duration of muscle cofactor activation, and the phase synchronization index between different muscle cofactor activation sequences. The neural control features include: the differential entropy of the activation coefficient sequence, the frequency of switching of dominant muscle coercive elements per unit time, and the sparsity of the activation coefficient vector. The multimodal fusion features include: the correlation between specific muscle co-activation modes and inertial motion data, and the correspondence between muscle co-activation modes and local pressure change modes.

7. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 1, characterized in that, In S500, the physiological state classification model is a hierarchical hybrid classification model based on machine learning and deep learning, and is generated by training on historical labeled data. The historical labeled data contains multiple sets of time-series physiological state feature vectors and their corresponding physiological state labels. The physiological state classification model includes a first layer of machine learning-based abnormal state filter, a second layer of deep learning-based time-series pattern analyzer, and a third layer of machine learning-based multi-state classifier.

8. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 7, characterized in that, The physiological state classification model works by recognizing physiological states as follows: The first layer of filtering involves inputting the real-time extracted physiological state feature vectors into the abnormal state filter. The abnormal state filter is a single-class support vector machine used to identify and filter out data points that deviate from the normal physiological pattern and mark them as invalid signals. The second layer of temporal analysis involves inputting the filtered temporal physiological state feature vector sequence into the temporal pattern analyzer. The temporal pattern analyzer is a one-dimensional convolutional neural network used to extract higher-order patterns from the temporal changes of the feature vectors and output a state evolution pattern encoding vector that represents the current state evolution pattern. The third layer of state determination involves fusing the state evolution pattern encoding vector with the current physiological state feature vector and inputting it into the multi-state classifier. The multi-state classifier outputs the current physiological state category label, the confidence score corresponding to different categories, and the state evolution trend prediction.

9. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 8, characterized in that, The physiological state category labels can be configured to include: low arousal baseline state, high arousal excitation state, excessive tension inhibition state, and muscle coordination disorder state. The recognition results output by the physiological state classification model include: the physiological state category label determined at the current moment, and the confidence score corresponding to the physiological state category label.

10. The method for identifying the physiological state of adult product users based on multi-sensor data acquisition according to claim 1, characterized in that, The feedback adjustment command is used to control the stimulation parameters of adult products, and the specific control logic includes: When the physiological state classification model identifies a physiological state category label belonging to the positive feedback state category and its confidence score is higher than the first preset threshold, a first type of feedback regulation instruction is generated to maintain the current stimulation mode. The positive feedback state category includes the high arousal state. When the physiological state classifier model identifies a physiological state category label belonging to the negative regulation state category and its confidence score is higher than the second preset threshold, a second type of feedback regulation instruction is generated to reduce the stimulus intensity and switch modes. The negative regulation state category includes excessive tension inhibition state and muscle coordination disorder state.

Images