A method for skin impedance affective data acquisition and modeling

Abstract

The application discloses a kind of skin impedance emotion data acquisition and modeling method, the method comprises the following steps: step 1: using hardware equipment to measure the skin impedance information of human impedance model, and using double frequency method to carry out data post-processing;Step 2: design experiment, stimulate emotional fluctuation to generate physiological characteristic signal and predict and process signal;Step 3: combined with the skin impedance information of human impedance model, respiratory rate and heartbeat frequency carry out emotion prediction modeling.The application establishes human impedance emotion database, and corresponding respiratory signal and heartbeat frequency when emotional fluctuation, it is more comprehensive that the human physiological information acquisition related to emotion, improves the credibility of emotional prediction.

Description

Technical Field

[0001] This invention discloses a method for collecting and modeling skin impedance emotional data, belonging to the field of information technology. Background Technology

[0002] Human bodily functions are regulated by the autonomic nervous system, which is in turn influenced by emotions. Therefore, by measuring physiological signals, one can infer a person's emotional state. Designing a method to collect various human-derived physiological parameters, such as skin impedance information, and modeling and mapping them to predict physiological states such as emotions has profound significance and impact on incorporating biofeedback and emotional responsiveness into personalized medicine.

[0003] Because human body impedance is a complex coupling of resistors and capacitors of varying magnitudes, the human body can be represented as a complex circuit. Emotions are closely related to various derived physiological parameters of the human body. Based on the distribution of impedance in the skin, various information related to the human body can be obtained, such as respiratory rate, heart rate, aortic blood flow, skin temperature, blood volume, and pulse. Changes in these physiological parameters are often caused by emotional fluctuations. For example, human skin impedance often changes with physical activity and emotional reactions; when a person is tense or excited, sweat glands secrete sweat, increasing the amount of sweat under the skin and the amount of tissue fluid in the body. Since tissue fluid contains a large amount of electrolytes, this leads to changes in the body's resistance.

[0004] Current biological circuit models are typically constructed using constant impedance elements, which limits their applicability and fails to meet the accuracy requirements for describing changes in human emotions. Therefore, a method for collecting and modeling skin impedance emotional data is needed, along with the establishment of a biological impedance model with high response accuracy, and the collection of more human-derived physiological parameters, to achieve real-time and accurate description of human emotional changes under frequently changing and sudden situations.

[0005] Prior to this invention, the disadvantages of traditional methods were that it was difficult to accurately detect the fluctuations and trends of surface and deep impedance in the skin, the methods for measuring skin impedance were not convenient enough and were too complicated, and the collected human physiological information was often not comprehensive enough, resulting in insufficient reliability and confidence in emotion prediction. Summary of the Invention

[0006] This invention relates to the measurement of electrical resistance in human body parts such as skin. More specifically, this invention relates to a data acquisition and human emotion prediction modeling method that uses hardware to measure the impedance in human skin, thereby obtaining human-derived vital signs parameters such as respiratory rate and heart rate to predict human emotions; applied in the field of novel smart wearable technology, namely, real-time monitoring of human information and emotional state, recording health status, and then targeted treatment.

[0007] This invention specifically proposes a method for collecting and modeling emotional data using skin impedance. It extracts multi-level skin impedance information, further utilizes this information to extract respiratory rate and heart rate, and finally combines these three physiological signals to describe and predict emotions. The method includes the following steps:

[0008] Step 1: Use hardware devices to measure the skin impedance information of the human body impedance model and use the dual-frequency method for data post-processing;

[0009] Step 2: Design an experiment to stimulate emotional fluctuations to generate physiological characteristic signals and predict and process these signals;

[0010] Step 3: Combine skin impedance information, respiratory rate and heart rate from the human body impedance model to perform emotion prediction modeling.

[0011] Furthermore, step 1 includes the following steps:

[0012] Step 1.1: Skin impedance information measurement; Connect the hardware and electrodes, and generate preset sinusoidal frequency signals sequentially through the DDS signal generator. First, input a sinusoidal signal with frequency F1 for measurement, and record the data after the serial port output stabilizes; then change the sinusoidal signal to frequency F2, record the data after the serial port output stabilizes, and obtain two sets of frequency data results.

[0013] Step 1.2: Skin impedance calculation; Organize the data, combine the discrete Fourier transform method, use the dual-frequency method, substitute into relevant formulas to calculate, and obtain the linear resistance R of the simulated deep skin tissue in the human body impedance model. s Simulates the linear resistance R of the stratum corneum of the skin epidermis p And the linear capacitance C of the simulated skin epidermal stratum corneum p The value of the three parts;

[0014] Step 1.3: Phase and amplitude correction; Since the internal circuitry of the hardware has different amplification effects on signals generated at different frequencies, the amplitude and phase of the resulting impedance will have certain deviations. For circuits of different frequencies, the circuit components with small fixed input values ​​are matched and measured to correct the amplitude and phase of the impedance measurement results.

[0015] Step 1.4: Chip internal resistance correction; The internal resistance generated by the internal circuit simulation is on the same order of magnitude as the skin impedance. After measurement and experimentation, the method for establishing the internal resistance is estimated, and the impedance measurement results are recorrected to make the results more accurate.

[0016] Step 1.5: Final result calculation; subtract the error obtained after two corrections to obtain the accurate values ​​of the three parts of the human body impedance model;

[0017] Furthermore, step 2 includes the following steps:

[0018] Step 2.1: Classify human emotions into two types of emotion prediction: positive and negative. These correspond to elevated emotions and depressed emotions. Elevated emotions include high-intensity emotions such as excitement, joy, and fear, while depressed emotions include low-intensity emotions such as sadness, disappointment, and relaxation.

[0019] Step 2.2: Select 20 test participants, aged 20 to 30, who are students, including 10 women and 10 men;

[0020] Step 2.3: Using pre-set videos or music that produce obvious emotional tendencies and fluctuations as a set of emotional stimuli, guide test subjects to watch videos or listen to music to stimulate emotional fluctuations, while simultaneously conducting hardware wearing and experimental measurements;

[0021] Step 2.4: Obtain the human skin resistance emotional experiment data files corresponding to various emotions. The file name of the experimental data file includes the file number, emotion category, tester gender, acquisition method, tester number, and data number identifier information of the stimulus emotion dataset.

[0022] Furthermore, step 3 includes the following steps:

[0023] Step 3.1: After the above steps, the values ​​of each part of the human body impedance model are obtained through collection, measurement and processing. The values ​​of the three parts of the human body impedance model are used as the data source to build a model and predict the respiratory rate and heart rate data.

[0024] Step 3.2: Establish two emotion judgment models. The first model takes only multi-level skin impedance information as input and outputs an emotion type prediction. The second model takes multi-level skin impedance information, respiratory rate and heart rate as input and outputs an emotion type prediction.

[0025] Step 3.3: The model is built using machine learning or neural network models, and all models used are end-to-end physiological signal-emotion prediction models.

[0026] The positive effects and advantages of this invention are as follows:

[0027] 1. A human impedance emotion database was established, which includes human impedance information of 20 experimental subjects (10 males and 10 females), as well as respiratory signals and heart rate during corresponding emotional fluctuations. The collection of human physiological information related to emotions is more comprehensive, which improves the reliability of emotion prediction.

[0028] 2. A method for measuring multi-level skin impedance using dual-frequency measurement, namely the dual-frequency method, is proposed. Unlike the previous single-frequency method, it uses multiple different frequency signals to excite the circuit and uses DFT to calculate and transform the solution. Compared with the previous method, the results are more accurate, real-time and robust.

[0029] 3. More human-derived physiological signals, such as respiratory rate and heart rate, are extracted from existing human impedance information. A model for emotion prediction based on these various human-derived physiological signals is designed, built, and trained. This model describes more types of physiological information about emotions and provides more accurate prediction results. Attached Figure Description

[0030] Figure 1 Patent overall overview flowchart;

[0031] Figure 2 Flowchart for extracting human skin impedance information using hardware devices and dual-frequency method;

[0032] Figure 3 The human skin impedance model used in this patent;

[0033] Figure 4 shows the amplitude shift (a) and phase shift (b) caused by the circuit response at different frequencies;

[0034] Figure 5 Flowchart of the standard procedure for collecting human emotional impedance data;

[0035] Figure 6 Schematic diagram of modeling derived vital sign parameters;

[0036] Figure 7 A schematic diagram of emotion recognition and classification modeling. Detailed Implementation

[0037] The overall flowchart of this invention is as follows: Figure 1 As shown in the attached diagrams, the specific details of each part are illustrated below. Figures 2-7 The technical solution described in this invention will be further elaborated below.

[0038] like Figure 2 The diagram shows a flowchart of extracting human skin impedance information using hardware devices and a dual-frequency method. It is mainly divided into two parts: the dual-frequency extraction method and the secondary correction method for the acquisition results, including offset and internal resistance.

[0039] I. The extraction method using the dual-frequency method, the steps of which are as follows:

[0040] Step 1: Use hardware devices to measure human body characteristic information such as skin impedance, and use the dual-frequency method for data post-processing;

[0041] Step 1.1: Connect the hardware and electrodes, and generate preset sinusoidal frequency signals sequentially using the DDS signal generator. First, input a sinusoidal signal with frequency F1 for measurement, and record the data after the serial port output stabilizes. Then, change the sinusoidal signal to frequency F2, record the data after the serial port output stabilizes, and obtain two sets of data results.

[0042] Step 1.2: Organize the data, combine the Discrete Fourier Transform method with the dual-frequency method, substitute into relevant formulas to calculate, and obtain the R of the human body impedance model. s R p and C p The value of the three parts;

[0043] Step 1.3: Since the hardware chip has different amplification effects for signals generated at different frequencies, the amplitude and phase of the resulting impedance will have certain deviations. Matching measurements are performed on the input resistance, capacitance and inductance for different frequency circuits to correct the amplitude and phase of the impedance measurement results.

[0044] Step 1.4: The internal resistance generated by the hardware chip simulation is approximately on the same order of magnitude as the impedance of human skin. After multiple measurements and experiments, the method for establishing the internal resistance is estimated, and the impedance measurement results are recorrected to make the results more accurate.

[0045] Step 1.5: Subtract the error obtained after two corrections to obtain the accurate values ​​of the three parts of the human body impedance model;

[0046] First, the human skin impedance model used in this invention is introduced, such as... Figure 3 As shown, the linear resistance R s Used to simulate the impedance of deep skin tissue, linear capacitance C p and linear resistance R p The parallel combination is used to simulate the impedance of the stratum corneum of the skin epidermis, where U1=sinω1t and U2=sinω2t represent the sinusoidal voltages of two frequencies (ω1 and ω2) in the dual-frequency method, respectively.

[0047] From the series-parallel impedance formula, the relationship between the stratum corneum impedance Z(ω) and the angular frequency of the external signal can be expressed as: (where ω represents the frequency of the sinusoidal signal)

[0048]

[0049] After sorting, we can obtain:

[0050]

[0051] Then, the hardware and electrodes were connected, and a preset sinusoidal frequency signal was generated sequentially through the DDS signal generator. First, a sinusoidal signal with an angular frequency of ω1 was input for measurement. After a period of time, the serial port output stabilized and the data was recorded. Then, the frequency was changed to input a sinusoidal signal with an angular frequency of ω2. After the serial port output stabilized, the data was recorded, and two sets of data results were obtained. The measured skin impedances were Z1=X1+iY1 and Z2=X2+iY2 (where X and Y represent the real part and imaginary part of the measured total skin impedance Z, respectively). Then, the results were derived using the Discrete Fourier Transform (DFT).

[0052] Based on the results, the equation is simplified and solved:

[0053]

[0054]

[0055]

[0056]

[0057] Substituting formulas (5) and (6) into (3) and (4) respectively, we can eliminate the denominators.

[0058]

[0059]

[0060] Subtracting formula (7) from formula (8) yields:

[0061]

[0062] Substituting (9) into (7) or (8) yields the linear resistance R used to simulate the impedance of deep skin tissue. s :

[0063]

[0064] From formulas (5) and (6):

[0065]

[0066]

[0067] The two equations can be divided to eliminate each other. have to:

[0068]

[0069] Substituting formula (9) into formula (13):

[0070]

[0071] The linear resistance R used to simulate the stratum corneum of the skin's epidermis was obtained. p :

[0072]

[0073] From (9) and (15), we obtain the linear capacitance C used to simulate the stratum corneum of the skin epidermis. p :

[0074]

[0075] Specifically, in the above formula, ω1 and ω2 correspond to the two frequencies in the dual-frequency method, and X1,X2 and Y1,Y2 are the real and imaginary parts of the two skin impedance measurement results Z1 and Z2 in the dual-frequency method, respectively.

[0076] II. The principles and methods of correction are as follows:

[0077] As described in step 1 above, the linear resistance R of the deep skin tissue s The linear capacitance C of the skin epidermal stratum corneum impedance p And the linear resistance R of the simulated skin epidermal stratum corneum p All results were obtained using a DDS signal generator combined with the dual-frequency method. However, due to the DDS's transformation for different frequencies and the DDS's internal resistance, the measurement results have some deviation. We performed two corrections: one for the circuit variations at different frequencies and the other for the chip's internal resistance.

[0078] First, the DDS signal generator has different circuit settings for signals of different frequencies, which results in different amplification effects for signals of different frequencies. This causes a certain deviation in the amplitude and phase of the obtained impedance results, and this bias needs to be corrected.

[0079] Experimental simulations were conducted using MATLAB, and the results are shown in Figure 4 (Figure 4(a) shows the amplitude shift of different parts of the human body impedance under different frequency signals, and Figure 4(b) shows the phase shift of different parts of the human body impedance under different frequency signals). A sinusoidal signal generated by a DDS signal generator and its circuit model were constructed to simulate its response to sinusoidal signals of different angular frequencies. First, a 1K ohm resistor was connected at the connection point of the human body impedance electrodes, and the frequency was changed from 40kHz to 80kHz. Experimental simulations were performed and plotted, allowing the calculation of the amplitude and phase offsets at different frequencies caused by circuit changes. This enabled the correction of impedance measurement results, making the results more accurate.

[0080] The second correction was performed to address the internal resistance of the DDS signal generator. Since the distance between the two electrodes at this measurement end was only 5cm, the human body resistance measurement results at this distance were affected by emotional fluctuations causing changes in the sympathetic nervous system, sweating on the skin surface, and the upward flow of subcutaneous fluid containing electrolytes, resulting in a change in human body impedance on the order of approximately 500-1000 ohms. This order of magnitude is the same as the internal resistance of the DDS signal generator. Therefore, it was necessary to design an experiment to reasonably estimate the internal resistance.

[0081] The experimental design is as follows: a 1μF capacitor is connected, and a 40kHz-80kHz sinusoidal signal is generated by a signal generator for the experiment, recording the response waveform data; then a 1μH inductor is connected, and the above operation is repeated, recording the data; the two are combined for calculation, and finally the internal resistance r0 of the DDS signal generator is estimated to be approximately 170 ohms. After two corrections, the scientific validity, reliability, and interpretability of the measurement results are further improved. The correction formula is shown below, where... R0 is the phase angle of the measured impedance Z, and r0 is the internal resistance calculated after correction.

[0082]

[0083]

[0084] III. The methods for creating, collecting, and establishing the resistance-based emotional database are as follows:

[0085] The design and preparation of the experiment include the following steps:

[0086] Step 2: Design experiments to stimulate emotional fluctuations and predict and process physiological characteristic signals;

[0087] Step 2.1: Classify human emotions into two types of emotion prediction: positive and negative. These correspond to elevated emotions and depressed emotions. Elevated emotions include high-intensity emotions such as excitement, joy, and fear, while depressed emotions include low-intensity emotions such as sadness, disappointment, and relaxation.

[0088] Step 2.2: Select 20 test participants, aged 20 to 30, who are students, including 10 women and 10 men;

[0089] Step 2.3: Using pre-set videos or music that can produce obvious emotional tendencies and fluctuations as a set of emotional stimuli, guide testers to watch videos or listen to music to stimulate emotional fluctuations, while simultaneously conducting hardware wearing and experimental measurements;

[0090] Step 2.4: Obtain the human skin resistance emotional experiment data files corresponding to various emotions. The file name of the experimental data file includes the file number, emotion category, tester gender, acquisition method, tester number, and data number identifier information of the stimulus emotion dataset.

[0091] Every step in the entire process of collecting and creating the resistivity emotion database should follow specific standards, including test subject standards, emotion trigger script design standards, testing standards, result file naming standards, and experimental record standards. The specific production standards and their meanings are shown in Table 1.

[0092] Table 1

[0093]

[0094] Tester specifications: Select 20 testers, aged 20 to 30, whose occupation is students, including 10 men and 10 women;

[0095] The selection of emotional trigger scripts for experimental testing, i.e., videos / music, mainly considers the following three aspects: 1) They should be able to evoke a certain emotional tendency; 2) The selected videos or music should be sufficiently diverse in type, and the number of trigger scripts corresponding to each type of emotion should be the same; 3) The selection of emotional trigger scripts should be moderate, and large-scale emotional triggers should not be selected to protect the psychological and emotional health of the testers and avoid causing discomfort to the testers.

[0096] Based on the above principles, we selected the video and music shown in the supplementary materials as the test script.

[0097] To facilitate the organization and retrieval of emotional impedance data and to provide an intuitive description of the key characteristics of individual emotions, we designed a naming convention for emotional impedance data files. Both emotional impedance files and derived human information feature files must be named according to this convention, as shown in Table 2 below. File names consist of 8 characters and are divided into 6 parts.

[0098] Table 2

[0099] character length 1 1 1 2 2 1 character range P,N F,M V,M 01～10 01-20 1-9

[0100] 1) Emotion Category: The emotion category identifier occupies 1 character. The emotion information determination in this invention is divided into positive and negative categories. Positive is an excited, happy or angry state of high emotion, and the corresponding negative is a depressed state of emotion such as loss, sadness and despair. It is composed of the first lowercase letter of the emotion type fluctuation level, P: emotion is positive; N: emotion is negative.

[0101] 2) Tester's gender: The amplitude and phase of the human body impedance signal differ depending on the speaker's gender. Gender is indicated by one character: F: female; M: male.

[0102] 3) Acquisition methods: Different acquisition methods for emotional impedance information result in more sufficient data and more diverse samples, avoiding randomness. There are two acquisition methods: V: video; M: music.

[0103] 4) Tester ID: We record and number the testers' information. Based on the current number of testers, the tester ID occupies two numbers, ranging from 01 to 10.

[0104] 5) Test Case Number: We record the emotional trigger scripts and number them. Based on the scale of the existing emotional trigger scripts, the script identifier occupies 2 numbers, ranging from 01 to 20.

[0105] 6) Number of tests: This indicates the number of times an emotion-triggered script has been used. It occupies one number and ranges from 1 to 9.

[0106] Data acquisition and creation process as follows Figure 5 As shown. In the emotional state judgment of this invention, we define a calm state as the basic emotional state of a person, and elevated and depressed emotions as two emotional fluctuation states to be predicted. First, the test subject is asked to conduct a 4-minute basic emotional state test, during which some normal routine formulas and personal facts are provided for the test subject to copy and repeat, and the emotion at this time is taken as the calm state of emotional state; then the test subject is asked to perform a 5-minute distraction task (such as choosing from a Sudoku game, a maze game, or copying geometric shapes), and then to perform the first random emotional task, that is, to randomly choose a video or music from the pre-selected list for experimental measurement; this process is repeated until 4 emotional stimuli and measurements are completed.

[0107] Each emotional stimulus and measurement session was planned to last 5 minutes, or 300 seconds. This means each measurement session consisted of 300 seconds of data. To ensure the rigor and scientific validity of the experiment, the first and last 30 seconds of the measurement results were removed, assuming the test subject had not yet entered the corresponding emotional state; only the middle 240 seconds of measurement data were collected. This data was then divided into 10-second units, resulting in 24 sets of 10-second data points from each 240-second measurement session. In summary, one test subject spent approximately 45 minutes conducting the measurement and recording, obtaining a total of 96 sets of experimental data. After obtaining the measurement results, the data were recorded and processed strictly according to the aforementioned dataset preparation and naming conventions.

[0108] IV. The modeling of human skin impedance and other physiological signals into emotion prediction is as follows:

[0109] Step 3: An emotion prediction modeling method combining human impedance information, respiratory signals, and aortic blood flow;

[0110] Step 3.1: After the above steps, the values ​​of each part of the human body impedance model are obtained through collection, measurement and processing. The values ​​of the three parts of the human body impedance model are used as the data source to build a model and predict the respiratory rate and heart rate data.

[0111] Step 3.2: Establish two emotion judgment models. The first model takes only multi-level skin impedance information as input and outputs an emotion type prediction. The second model takes multi-level skin impedance information, respiratory rate and heart rate as input and outputs an emotion type prediction.

[0112] Step 3.3: The model is built using machine learning and neural network models such as LSTM and TCN. All models used are end-to-end physiological signal-emotion prediction models.

[0113] By following the steps described above, the impedance R of the simulated deep skin tissue in the human body impedance model can be obtained. s Simulates the linear capacitance C of the stratum corneum impedance of the skin epidermis. p and linear resistance R p To more accurately predict changes in human emotions, we can use the above results to model and obtain more other human-derived physiological parameters, and then use the data to build a prediction from data to emotion through machine learning or neural network models.

[0114] This invention proposes to use a Temporal Convolutional Network (TCN) to extract two additional derived vital signs parameters: respiratory rate and heart rate. The TCN model is an improvement on the CNN model, employing causal convolutions, making it more suitable for sequence models; and it utilizes dilated convolutions and residual modules to address the memory history problem. The network structure for extracting the aforementioned two derived vital signs parameters is as follows: Figure 6 As shown.

[0115] TCNs offer significant advantages over CNNs. Firstly, they possess parallelism. Unlike RNNs, where predictions for subsequent time steps must wait for their predecessors to complete, convolutions can be performed in parallel because each layer uses the same filters. Therefore, long input sequences can be processed as a whole during training and evaluation in a TCN, rather than sequentially as in an RNN. Secondly, their receptive field is more flexible. TCNs can change their receptive field size in various ways, such as stacking more causal convolutional layers, using larger dilation factors, or increasing filter sizes. This allows for better control over the model's memory footprint and makes it easier to adapt to different machine learning domains. Thirdly, their gradients are more stable. Unlike recurrent architectures, the backpropagation path in a TCN is not aligned with the temporal direction of the sequence. This avoids the gradient explosion problem, further reducing memory requirements during training.

[0116] After extracting the derived vital signs parameters, we established two models for emotion prediction. The first model takes human skin impedance information as input and outputs a predicted human emotion. The second model takes human skin impedance information, respiratory rate, and heart rate extracted through the aforementioned TCN as input and outputs a predicted human emotion. The models will employ neural network models, combining CNN and attention mechanisms, as well as an Encoder-Decoder architecture, referencing classic models such as LSTM, to establish a prediction system from the database to emotion. The basic models used in this invention are CNN and Bidirectional LSTM, with certain improvements made to the basic models. The final overall network model structure is as follows: Figure 7 As shown.

[0117] Bidirectional LSTM is a bidirectional Long Short-Term Memory (LSTM) network that allows neural networks to utilize both past and future sequence information simultaneously, improving data utilization and prediction accuracy. Bidirectional LSTM is also an RNN layer, making it easy to wrap around other RNN layers, facilitating model modification and addition. Bidirectional LSTM can be used to handle sequence tasks that require simultaneous consideration of past and future information, commonly used in natural language processing such as sentiment recognition and sentiment analysis, which is the task proposed in this invention. It can capture bidirectional dependencies in sequences, thereby improving the model's predictive performance and generalization ability. Bidirectional LSTM has significant advantages over ordinary LSTM. Ordinary LSTM outputs information from only one direction, while bidirectional LSTM adds a reverse LSTM layer, allowing the model to receive both forward and reverse sequence information simultaneously, concatenating the information from both directions as the output.

[0118] In summary, the task proposed in this invention is to use the proposed data acquisition method to collect a series of extracted human-derived feature information, and to use the trained model to predict the emotional state at that time. That is, to use the human's electrical resistance information (as well as breathing frequency and heart rate) within 10 seconds as input to predict the human's emotional state within those 10 seconds.

[0119] The training process of the model used is as follows:

[0120] 1. Determine the training batch size and batch, run a model combining CNN and bidirectional LSTM, perform forward propagation, and use the forward propagation of CNN to learn and extract representative features from human-derived feature information;

[0121] 2. Run the bidirectional LSTM forward and backward propagation, and use the bidirectional LSTM to classify the features extracted by the CNN in step 1, so as to obtain the output of all labels at any position;

[0122] 3. Perform backpropagation from output to input, calculate the gradient of the loss function with respect to each weight and bias, and update the weights and biases using the gradients calculated through backpropagation to reduce the value of the loss function;

[0123] 4. Repeat the above training and model parameter update process until the number of iterations reaches 100 epochs.

[0124] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for collecting and modeling skin impedance emotional data, characterized in that, The method includes the following steps: Step 1: Use hardware devices to measure the skin impedance information of the human body impedance model and use the dual-frequency method for data post-processing; Step 2: Design an experiment to stimulate emotional fluctuations to generate physiological characteristic signals and predict and process these signals; Step 3: Combine skin impedance information, respiratory rate, and heart rate from the human body impedance model to perform emotion prediction modeling; The hardware was set up to acquire multi-layer skin impedance data, and the proposed dual-frequency method and correction method were used for data post-processing. Step 1 includes the following steps: Step 1.1: Skin impedance information measurement; Connect the hardware and electrodes, and generate preset sinusoidal frequency signals sequentially through the DDS signal generator. First, input a sinusoidal signal with frequency F1 for measurement, and record the data after the serial port output stabilizes; then change the sinusoidal signal to frequency F2, record the data after the serial port output stabilizes, and obtain two sets of frequency data results. Step 1.2: Skin impedance calculation; Organize the data, combine the discrete Fourier transform method, use the dual-frequency method, substitute into relevant formulas to calculate, and obtain the linear resistance R of the simulated deep skin tissue in the human body impedance model. s Simulates the linear resistance R of the stratum corneum of the skin epidermis p And the linear capacitance C of the simulated skin epidermal stratum corneum p The value of the three parts; Step 1.3: Phase and amplitude correction; Since the internal circuitry of the hardware has different amplification effects on signals generated at different frequencies, the amplitude and phase of the resulting impedance will have certain deviations. For circuits of different frequencies, the circuit components with small fixed input values ​​are matched and measured to correct the amplitude and phase of the impedance measurement results. Step 1.4: Chip internal resistance correction; The internal resistance generated by the internal circuit simulation is on the same order of magnitude as the skin impedance. After measurement and experimentation, the method for establishing the internal resistance is estimated, and the impedance measurement results are recorrected to make the results more accurate. Step 1.5: Calculate the final result; subtract the error obtained after two corrections to obtain the accurate values ​​of the three parts of the human body impedance model.

2. The method according to claim 1, characterized in that, Step 2 of the experiment, which designs an emotion classification standard and stimulates changes in the emotions of participants, includes the following steps: Step 2.1: Classify human emotions into two types of emotion prediction: positive and negative. These correspond to elevated emotions and depressed emotions. Elevated emotions include high-intensity emotions such as excitement, joy, and fear, while depressed emotions include low-intensity emotions such as sadness, disappointment, and relaxation. Step 2.2: Select 20 test participants, aged 20 to 30, who are students, including 10 women and 10 men; Step 2.3: Using pre-set videos or music that produce obvious emotional tendencies and fluctuations as a set of emotional stimuli, guide test subjects to watch videos or listen to music to stimulate emotional fluctuations, while simultaneously conducting hardware wearing and experimental measurements; Step 2.4: Obtain the human skin resistance emotional experiment data files corresponding to various emotions. The file name of the experimental data file includes the file number, emotion category, tester gender, acquisition method, tester number, and data number identifier information of the stimulus emotion dataset.

3. The method according to claim 1, characterized in that, Step 3, which combines neural networks to establish prediction and classification of physiological signals into emotions, includes the following steps: Step 3.1: After the above steps, the values ​​of each part of the human body impedance model are obtained through collection, measurement and processing. The values ​​of the three parts of the human body impedance model are used as the data source to build a model and predict the respiratory rate and heart rate data. Step 3.2: Establish two emotion judgment models. The first model takes only multi-level skin impedance information as input and outputs an emotion type prediction. The second model takes multi-level skin impedance information, respiratory rate and heart rate as input and outputs an emotion type prediction. Step 3.3: The model is built using machine learning or neural network models, and all models used are end-to-end physiological signal-emotion prediction models.

Images