Brain-controlled massage chair system based on wavelet transform, phase space reconstruction and deep learning

Abstract

This invention discloses a brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning, belonging to the field of EEG signal processing technology. It includes: a single-channel EEG signal acquisition module for acquiring the user's EEG signals; a signal processing module, including a preprocessing denoising unit, a decomposition and reconstruction unit, a deep learning classification unit, and a user database storage unit, for processing EEG signals and classifying user intentions and thoughts; a control module for controlling the drive circuit; and a massage module for executing massage. This invention proposes a massage chair system that acquires the user's EEG signals in real time and adjusts the massage position according to real-time needs. It uses a single-channel EEG signal acquisition device to reduce signal redundancy and computational complexity, enabling high-speed execution of control commands. Through TQWT and phase space reconstruction, the system performs full-band analysis of the signal and extracts and classifies features using a deep learning classification unit, extracting effective features in a data-driven manner and accurately responding to immediate needs.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal processing technology, and more specifically, to a brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning. Background Technology

[0002] Currently, people's demand for a healthy lifestyle is constantly increasing. Massage chairs, as a way to relieve physical fatigue and reduce stress, are becoming increasingly popular, and their market demand is growing. At present, massage chairs in my country are mainly limited to manual control and voice input control. Traditional manual control requires users to operate via buttons or remote controls, which can be cumbersome and disruptive for users already in a relaxed state. While voice input control offers some convenience, it is still limited by the accuracy of voice recognition and interference from environmental noise. Neither of these control methods can meet the personalized needs and immediate adjustment requirements of users.

[0003] In the current environment, patent application CN201910004126.9 collects EEG data from 11 electrodes and analyzes the EEG data under massage conditions through a signal processing module. It then compares this data with a personal database to determine the comfort level at that massage intensity, thereby controlling the massage chair. Patent application CN202410093155.8 analyzes the frequency and power of alpha and theta waves in the collected EEG signals to determine the user's fatigue level and selectively applies massage and plays audio based on the fatigue level.

[0004] While the existing control methods described above basically meet some of the user's personalized needs and requirements for real-time adjustments, they still have the following drawbacks: multi-channel EEG acquisition devices cause inconvenience in wearing and increase the redundancy and complexity of the acquired EEG signals. Processing only certain bands of the EEG signals reduces classification accuracy. Since EEG signals are inherently highly random, nonlinear, non-stationary, and non-Gaussian signals, existing EEG signal processing methods for massage chairs controlled by EEG signals do not fully consider the temporal and spectral characteristics of EEG signals. Furthermore, because EEG signals are often corrupted by various eye and muscle artifacts, traditional methods based on manual feature extraction are computationally cumbersome and extract incomplete features, resulting in poor classification performance. Therefore, this invention proposes a brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning. Summary of the Invention

[0005] To overcome the inconvenience of wearing multi-channel EEG acquisition devices and the increased redundancy and complexity of signal processing in existing technologies, this invention provides a brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning. This system addresses the shortcomings of existing EEG signal processing methods for massage chairs, which do not fully consider the temporal and spectral characteristics of EEG signals, and whose traditional manual feature extraction methods are computationally cumbersome and extract incomplete features, resulting in poor classification performance.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] The brain-controlled massage chair system based on wavelet transform, phase space reconstruction and deep learning includes: a single-channel EEG signal acquisition module, a signal processing module, a control module and a massage module;

[0008] The single-channel EEG signal acquisition module acquires the user's EEG signals;

[0009] The signal processing module receives and processes the EEG signal. The signal processing module includes a preprocessing denoising unit, a decomposition and reconstruction unit, a deep learning classification unit, and a user database storage unit. The preprocessing denoising unit preprocesses and denoises the EEG signal. The decomposition and reconstruction unit performs multi-level wavelet decomposition on the preprocessed and denoised EEG signal using adjustable Q-factor wavelet transform and reconstructs the phase space matrix from the decomposed signal. The deep learning classification unit includes an attention-based multi-scale hybrid convolutional neural network model. The attention-based multi-scale hybrid convolutional neural network model takes the phase space matrix as input and the classification result of the user's feedback intent as output.

[0010] The user database storage unit stores the classification results and related EEG signal data, which are used to adjust and optimize the parameters of the attention-based multi-scale hybrid convolutional neural network model.

[0011] The control module is used to control the drive circuit, and the massage module is used to perform massage. The control module receives the classification results from the signal processing module and activates the corresponding drive circuit, and dynamically adjusts the mode and massage position of the massage module in real time.

[0012] As a preferred embodiment, the single-channel EEG signal acquisition module uses a portable single-channel EEG signal acquisition device, which includes a sensor, an electrode for acquiring EEG signals, a reference electrode REF, and a ground electrode GND.

[0013] As a preferred embodiment, the preprocessing denoising includes: using a notch filter to remove interference from power lines on the EEG signal, and performing bandpass filtering on the EEG signal to remove out-of-band noise, thereby obtaining a preprocessed denoised EEG signal matrix. Where K is the total number of signal data points.

[0014] As a preferred embodiment, the frequency response of the filter bank composed of low-pass and high-pass filters used in the process of performing multi-layer wavelet decomposition on the preprocessed and denoised EEG signal through adjustable Q-factor wavelet transform is as follows:

[0015]

[0016] Where α is the low-pass scaling scale LPS of the frequency response of the low-pass filter G0, β is the high-pass scaling scale HPS of the frequency response of the high-pass filter G1, and j represents the number of decomposition layers.

[0017] After preprocessing, the EEG signal undergoes tunable Q-factor wavelet transform to obtain J+1 signal matrices with different time and frequency resolutions:

[0018]

[0019] Where K is the total number of signal data points.

[0020] As a preferred embodiment, the signal matrix X obtained after wavelet transform with adjustable Q factor TQWT The expression is a time series vector:

[0021] v = [v1…v K ] T

[0022] The reconstructed phase space is then represented as:

[0023] {Y i =[v i v i+τ …v i+(d-1)τ ] T for i = 1, 2, ..., K - (d - 1)τ}

[0024] Where d is the embedding dimension and τ is the time delay;

[0025] The signal matrix X TQWT Phase space reconstruction yields the phase space matrix:

[0026]

[0027] As a preferred embodiment, the attention-based multi-scale hybrid convolutional neural network model uses the phase space matrix as input, wherein the phase space matrix is ​​first augmented in dimension before being used as input to obtain the input signal matrix:

[0028]

[0029] The input signal matrix serves as the input feature matrix to the attention-based multi-scale hybrid convolutional neural network model.

[0030] As a preferred embodiment, the attention-based multi-scale hybrid convolutional neural network model includes a deep convolutional module, a feature attention module, and a classification module.

[0031] As a preferred embodiment, the deep convolution module includes two parallel sub-networks: a first sub-network and a second sub-network;

[0032] The first sub-network includes a first convolutional computation unit, a second convolutional computation unit, a first feature fusion unit, and a second feature fusion unit connected in sequence;

[0033] The first convolutional computation unit includes a first convolutional layer, a first batch of normalized layers, and a first ELU activation layer connected in sequence. The kernel size of the first convolutional layer is (4, 3, 1), where 4 is the number of kernels, and 3 and 1 are the sizes of the kernels in the spatial and temporal directions, respectively. The second convolutional computation unit includes a second convolutional layer, a second batch of normalized layers, and a second ELU activation layer connected in sequence. The kernel size of the second convolutional layer is (8, 1, 8), where 8 is the number of kernels, and 1 and 8 are the sizes of the kernels in the spatial and temporal directions, respectively.

[0034] The first feature fusion unit includes a third convolutional layer, a third batch normalization layer, a third ELU activation layer, and a first two-dimensional average pooling layer connected in sequence. The kernel size of the third convolutional layer is (8, 1, 4), and the pooling window size of the first two-dimensional average pooling layer is (1, 8), where 1 is the height of the pooling window and 8 is the width of the pooling window. The second feature fusion unit includes a fourth convolutional layer, a fourth batch normalization layer, a fourth ELU activation layer, and a second two-dimensional average pooling layer connected in sequence. The kernel size of the fourth convolutional layer is (8, J+1, 1), and the size of the second two-dimensional average pooling layer is (1, 8), where J+1 represents the number of decomposition layers of the adjustable Q-factor wavelet transform (TQWT).

[0035] The second sub-network includes a third convolutional computation unit, a fourth convolutional computation unit, a third feature fusion unit, and a fourth feature fusion unit connected in sequence;

[0036] The third convolutional computation unit includes a fifth convolutional layer, a fifth batch normalization layer, and a fifth ELU activation layer connected in sequence. The kernel size of the fifth convolutional layer is (4, 3, 1). The fourth convolutional computation unit includes a sixth convolutional layer, a sixth batch normalization layer, and a sixth ELU activation layer connected in sequence. The kernel size of the sixth convolutional layer is (8, 1, 4).

[0037] The third feature fusion unit includes a seventh convolutional layer, a seventh batch normalization layer, a seventh ELU activation layer, and a third two-dimensional average pooling layer connected in sequence. The kernel size of the seventh convolutional layer is (8, 1, 4), and the pooling window size of the third two-dimensional average pooling layer is (1, 8). The fourth feature fusion unit includes an eighth convolutional layer, an eighth batch normalization layer, an eighth ELU activation layer, and a fourth two-dimensional average pooling layer connected in sequence. The kernel size of the eighth convolutional layer is (8, J+1, 1), and the size of the fourth two-dimensional average pooling layer is (1, 8).

[0038] The outputs of the first sub-network and the second sub-network are reduced in dimensionality in the spatial dimension and merged in the temporal dimension to generate the output feature matrix of the deep convolutional module:

[0039]

[0040] The expressions for the ELU activation functions of all ELU activation layers are as follows:

[0041]

[0042] The expressions for all two-dimensional average pooling layers are:

[0043]

[0044] Where I(i+m,j+n) is the value of the input feature matrix at position (i+m,j+n), O(i,j) is the value of the output feature matrix at position (i,j), and k h and k w These represent the height and width of the pooling window, respectively.

[0045] As a preferred embodiment, the feature attention module includes a channel attention unit and a transformer connected in sequence;

[0046] The channel attention unit comprises a first one-dimensional global average pooling layer, a ninth convolutional layer, and a first sigmoid activation layer connected in sequence, and the output feature matrix X Conv As input to the feature attention module, the formula for one-dimensional global average pooling is:

[0047]

[0048] Where x Conv (c,t) represents the input matrix X. Conv The value at position (c,t), y pool (c,1) represents the value of the output vector at position (c,1);

[0049] The formula for the sigmoid activation function is:

[0050]

[0051] The output y of the first one-dimensional global average pooling layer pool The transpose is then used as input to the ninth convolutional layer, and the channel attention weight vector is then interposed with X. Conv Multiplying yields the channel attention feature matrix. Transposing the channel attention feature matrix by its channel dimension and time dimension gives...

[0052] Will The input to the Transformer is used to learn the global correlation in the feature matrix;

[0053] The Transformer includes a position embedding unit, a multi-head attention network, a layer normalization unit, a feedforward network, and a first Softmax activation layer. The position embedding formula in the position embedding unit is as follows:

[0054] PE (pos,2i) =sin(pos / 10000) 2i / d )

[0055] PE (pos,2i+1) =cos(pos / 10000) 2i / d )

[0056] The multi-head attention network will input the feature matrix X. CA Divided into h parts, each part is The multi-head attention network includes multiple self-attention units, each of which receives input... Linear transformation to query Q l Key K l Sum V l Q l and K l The dot product can calculate the global positional dependencies. The result is passed through the first Softmax activation layer to obtain the weight matrix, i.e., the attention score. Then, the attention score is p-valued by the dot product. l Weighting these global dependencies and adding them to the features can be represented as follows:

[0057] MHA(Q,K,V)=[head0;...;head h-1 ],

[0058]

[0059] Where d represents matrix K l To maintain the integrity of the input information, mitigate the vanishing gradient problem, and accelerate convergence, the Transformer uses residual connections at the locations of the multi-head attention network and the feedforward network, based on the channel dimension length. Finally, it generates the output matrix of the feature attention module.

[0060]

[0061] As a preferred embodiment, the classification module includes a first linear layer and a second Softmax activation layer. The number of neurons in the first linear layer is equal to the number of classification categories. Based on the neuron index corresponding to the maximum output value of the second Softmax activation layer, the user's feedback intent corresponding to each EEG signal collected by the user is obtained and output as the classification result.

[0062] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0063] This invention proposes a massage chair system that dynamically adjusts the massage position based on the user's immediate needs by acquiring the user's electroencephalogram (EEG) signals in real time, thereby achieving highly personalized and intelligent health services. It utilizes a more portable single-channel EEG signal acquisition device to reduce signal redundancy and computational complexity, enabling the massage chair to execute user control commands at high speed. Simultaneously, it performs full-band analysis of the acquired EEG signals through adjustable Q-factor wavelet transform (TQWT) and phase space reconstruction, comprehensively considering their time-frequency and nonlinear information. Furthermore, it automatically extracts and classifies the signals using a deep learning method based on attention-based hybrid convolutional neural networks, achieving effective feature extraction and precise response to the user's immediate needs in a data-driven manner. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the structure and process of the brain-controlled massage chair system based on wavelet transform, phase space reconstruction and deep learning in this application;

[0065] Figure 2 This is a schematic diagram of the structure of the attention-based hybrid convolutional neural network deep learning model of this application;

[0066] Figure 3 This is a schematic diagram of the transformer structure of the feature attention module in the deep learning model based on attention-based hybrid convolutional neural networks of this application. Detailed Implementation

[0067] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0068] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0069] Example 1

[0070] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure and process of the brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to this application. Embodiment 1 provides a brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning, including:

[0071] Single-channel EEG signal acquisition module:

[0072] In this embodiment 1, the portable single-channel EEG signal acquisition device used in the single-channel EEG signal acquisition module employs a NeuroSky TGAM1 chip as the sensor to acquire single-channel EEG signals. The single-channel EEG signal acquisition device has three metal contact points, i.e., electrodes, including an electrode located at TP10 in the right ear for acquiring EEG signals, a reference electrode (REF), and a ground electrode (GND). Its sampling rate is 512Hz, and the acquired EEG signals are transmitted to the signal processing module.

[0073] Signal processing module:

[0074] The signal processing module includes a preprocessing denoising unit, a decomposition and reconstruction unit, a deep learning classification unit, and a user database storage unit.

[0075] 1. Pre-processing noise reduction unit:

[0076] This includes a notch filter and a bandpass filter. In Example 1, the preprocessing denoising unit uses a 50Hz notch filter to remove interference from the power line and a bandpass filter to remove out-of-band noise from the signal within the 0-150Hz range, resulting in a preprocessed and denoised EEG signal matrix. Where K is the total number of signal data points.

[0077] 2. Decomposition and Reconstruction Units:

[0078] This includes tunable Q-factor wavelet transform (TQWT) and phase space reconstruction.

[0079] (1) Tunable Q-factor Wavelet Transform (TQWT): Tunable Q-factor wavelet transform (TQWT) performs multi-level wavelet decomposition on the preprocessed EEG signal. By selecting appropriate parameters such as Q, r, and J, TQWT can effectively analyze complex oscillatory signals. TQWT performs tree-structure decomposition using a filter bank consisting of a high-pass filter and a low-pass filter. The decomposition of the next layer is based on the low-pass signal of the previous layer, and this tree structure is repeated until there are J+1 subbands. The frequency response of the dual-channel filter bank consisting of the high-pass filter and the low-pass filter used in the TQWT decomposition process is shown below:

[0080]

[0081] Where α is the low-pass scaling factor (LPS) of the frequency response G0 of the low-pass filter, β is the high-pass scaling factor (HPS) of the frequency response G1 of the high-pass filter, and j represents the decomposition layer number. After TQWT, the preprocessed EEG signal will yield J+1 sub-bands with different time and frequency resolutions, i.e., the signal matrix. Where K is the total number of signal data points.

[0082] (2) Phase space reconstruction: Based on the pre-processed and denoised EEG signal, its one-dimensional signal vector expression can be known as the time series vector v = [v1…v2]. K ] T The phase space used for reconstruction is then represented as:

[0083] {Y i =[v i v i+τ …v i+(d-1)τ ] T for i = 1, 2, ..., K - (d - 1)τ}

[0084] Where d is the embedding dimension and τ is the time delay. The decomposition and reconstruction unit in the signal processing module transforms the signal matrix X obtained after the adjustable Q-factor wavelet transform. TQWT Constructing the phase space matrix through phase space reconstruction It contains rich nonlinear dynamic characteristics of EEG signals. The phase space matrix needs to be augmented before being used as input to deep learning classification units to obtain the input signal matrix:

[0085]

[0086] 3. Deep Learning Classification Unit: Please refer to [link / reference] Figure 2 , Figure 2This is a schematic diagram of the structure of the attention-based hybrid convolutional neural network deep learning model of this application. Embodiment 1 proposes a deep learning model for classifying EEG signals as a deep learning classification unit. This model is based on an attention-based multi-scale hybrid convolutional neural network (AMHCNet). The model takes the input signal matrix as input and outputs the feedback intent ψ corresponding to each sample for each user.

[0087] ψ∈{head, back, waist, legs, balance}

[0088] The deep learning unit, namely the attention-based multi-scale hybrid convolutional neural network (AMHCNet), includes a deep convolution module, a feature attention module, and a classification module. The deep convolution module utilizes the spatiotemporal and frequency features extracted from the four convolutions, capturing potential nonlinear dynamic characteristics such as chaotic behavior and nonlinear oscillations. The feature attention module first extracts channel attention from the feature matrix output by the deep convolution module, obtaining attention weights, which are then multiplied by the feature matrix to obtain the channel attention feature matrix. A transformer then captures the global correlations within the channel attention feature matrix, making the model focus more on the most informative parts of the input data, thereby improving the ability to identify key features and the overall model performance. The classification module classifies the aforementioned weighted feature matrix, outputting the classification result for the feedback intent ψ corresponding to each sample for each user. The deep learning classification unit includes the following modules:

[0089] (1) Deep Convolution Module: The deep convolution module includes two parallel sub-networks. The first sub-network includes a first convolutional computation unit, a second convolutional computation unit, a first feature fusion unit, and a second feature fusion unit connected in sequence. The first convolutional computation unit includes a first convolutional layer, a first batch of normalization layers, and a first ELU activation layer connected in sequence. The kernel size of the first convolutional layer is (4, 3, 1), where 4 is the number of kernels, and 3 and 1 are the sizes of the kernels in the spatial and temporal directions, respectively. The formula for the ELU activation function is:

[0090]

[0091] The second convolutional computation unit comprises a second convolutional layer, a second batch normalization layer, and a second ELU activation layer connected in sequence. The kernel size of the second convolutional layer is (8, 1, 8). The first feature fusion unit comprises a third convolutional layer, a third batch normalization layer, a third ELU activation layer, and a first two-dimensional average pooling layer connected in sequence. The kernel size of the third convolutional layer is (8, 1, 4), and the pooling window size of the first two-dimensional average pooling layer is (1, 8), where 1 is the height of the window and 8 is the width of the window. The formula for the two-dimensional average pooling layer is:

[0092]

[0093] Where I(i+m,j+n) is the value of the input feature matrix at position (i+m,j+n), O(i,j) is the value of the output feature matrix at position (i,j), and k h and k w These represent the height and width of the pooling window, respectively.

[0094] The second feature fusion unit comprises a fourth convolutional layer, a fourth batch normalization layer, a fourth ELU activation layer, and a second two-dimensional average pooling layer connected in sequence. The kernel size of the fourth convolutional layer is (8, J+1, 1), and the size of the second two-dimensional average pooling layer is (1, 8), where J+1 represents the number of decomposition layers of TQWT.

[0095] The second sub-network comprises a third convolutional computation unit, a fourth convolutional computation unit, a third feature fusion unit, and a fourth feature fusion unit connected in sequence. The third convolutional computation unit comprises a fifth convolutional layer, a fifth batch normalization layer, and a fifth ELU activation layer connected in sequence, with the kernel size of the fifth convolutional layer being (4, 3, 1). The fourth convolutional computation unit comprises a sixth convolutional layer, a sixth batch normalization layer, and a sixth ELU activation layer connected in sequence, with the kernel size of the sixth convolutional layer being (8, 1, 4). The third feature fusion unit comprises a seventh convolutional layer, a seventh batch normalization layer, a seventh ELU activation layer, and a third two-dimensional average pooling layer connected in sequence. The kernel size of the seventh convolutional layer is (8, 1, 4), and the pooling window size of the third two-dimensional average pooling layer is (1, 8). The fourth feature fusion unit comprises an eighth convolutional layer, an eighth batch normalization layer, an eighth ELU activation layer, and a fourth two-dimensional average pooling layer connected in sequence. The kernel size of the eighth convolutional layer is (8, J+1, 1), and the size of the fourth two-dimensional average pooling layer is (1, 8). The outputs of the two parallel sub-networks are reduced in the spatial dimension (second dimension) and merged in the temporal dimension (third dimension) to generate the output feature matrix of the deep convolutional module.

[0096] (2) Feature Attention Module: This module includes channel attention units and a transformer connected in sequence. The channel attention unit comprises a first one-dimensional global average pooling layer, a ninth convolutional layer, and a first sigmoid activation layer connected in sequence. The output feature matrix X is then processed. Conv As input to the feature attention module, the formula for one-dimensional global average pooling is:

[0097]

[0098] Where x Conv (c,t) represents the input matrix X. Conv The value at position (c,t), y pool(c,1) represents the value of the output vector at position (c,1). The output y of the first one-dimensional global average pooling layer... pool The transpose is used as input to the ninth convolutional layer, with a kernel size of (1, 1, 3). The sigmoid activation function is defined as follows:

[0099]

[0100] The channel attention weight vector y is obtained by transposing the output of the first sigmoid activation layer. weight The channel attention weight vector is then compared with X. Conv Multiplying these matrices yields the channel attention feature matrix. Then, transposing this feature matrix by its channel and time dimensions gives... It is used as input to the Transformer to learn the global correlation in the feature matrix.

[0101] Please see Figure 3 , Figure 3 This is a schematic diagram of the transformer structure in the feature attention module of the attention-based hybrid convolutional neural network deep learning model of this application. The transformer includes a position embedding unit, a multi-head attention network, a layer normalization unit, and a feedforward network. The position embedding unit allows input features to have relative positional relationships while preserving data relevance. The formula for position embedding is as follows:

[0102] PE (pos,2i) =sin(pos / 10000) 2i / d )

[0103] PE (pos,2i+1) =cos(pos / 10000) 2i / d )

[0104] The multi-head attention network takes the input feature matrix X as input. CA Divided into h parts, each part is Multi-head attention networks consist of multiple self-attention units, each of which receives input... Linear transformation to query (Q) l ), key (K) l ) and value (V) l ), Q l and K l The dot product can calculate the global positional dependencies. The result is passed through the first Softmax activation layer to obtain the weight matrix, i.e., the attention score. Then, the attention score is multiplied by the dot product on V. l We then weight these global dependencies and add them to the features. This process can be described as follows:

[0105] MHA(Q,K,V)=[head0;...;head h-1 ],

[0106]

[0107] Where d represents matrix K l The length of the channel dimension. The Transformer uses residual connections in the multi-head attention network and feedforward network to maintain the integrity of the input information, alleviate the gradient vanishing problem, and accelerate convergence. Finally, it generates the output matrix of the feature attention module.

[0108] (3) Classification Module: This module consists of a first linear layer and a second Softmax activation layer. The first linear layer has 5 neurons, representing the number of classification categories. The formula for the Softmax function is:

[0109]

[0110] Take the neuron index corresponding to the maximum output value of the second Softmax activation layer to obtain the feedback intention ψ corresponding to each EEG signal sample collected by the user.

[0111] 4. User Database Storage Unit: The user database storage unit stores the classification results and related EEG signal data output by the deep learning classification unit. The classification results and related EEG signal data are used to adjust and optimize the parameters of the attention-based multi-scale hybrid convolutional neural network model.

[0112] Control module:

[0113] The control module includes a control unit comprised of head, back, waist, and leg massagers, as well as an overall drive circuit. It inputs the classification results obtained from the signal processing module into the massage chair's control module, thereby activating the corresponding massage program. The control module automatically adjusts various parts of the massage module, such as the head, back, waist, and legs, selecting the appropriate massage mode. Based on the user's feedback and intention commands, it provides targeted massage to the areas of interest. The control unit converts the processed feedback and intention commands into specific massage actions, achieving precise control of different parts of the massage chair by controlling various drive circuits, high-precision motors, pressure sensors, and position sensors.

[0114] Massage module:

[0115] It includes various massage units of the massage chair, which are used to perform massage actions according to the massage modes and key massage points set by the control module.

[0116] Example 2

[0117] This embodiment 2 is based on the intelligent brain-controlled massage chair system of embodiment 1, and is used to illustrate the usage method of the intelligent brain-controlled massage chair system of embodiment 1, including the following steps:

[0118] S1: In this Example 2, the single-channel EEG signal acquisition device is an EEG signal acquisition headband. The user sits on the massage chair and puts on the EEG signal acquisition headband, imagining the desired massage location and pattern. The single-channel EEG signal acquisition module begins acquiring the user's current EEG signal and inputs the acquired EEG signal into the signal processing module.

[0119] S2: The signal processing module preprocesses the EEG signal through a preprocessing and denoising unit to remove power line interference and out-of-band noise. The decomposition and reconstruction unit uses Tiltable Q-factor wavelet transform (TQWT) to perform multi-level wavelet decomposition on the preprocessed and denoised EEG signal, decomposing the EEG signal into multiple sub-bands with different time and frequency resolutions to extract comprehensive time-frequency information. The signal after TQWT decomposition is reconstructed in phase space to fully capture the dynamic behavior of the EEG signal and its nonlinear dynamic characteristics in high-dimensional mode. The resulting phase space matrix is ​​then augmented and used as input to the deep learning classification unit.

[0120] S3: The deep learning classification unit takes the phase space matrix with increased dimensions as input and outputs the user's feedback intent through an attention-based multi-scale hybrid convolutional neural network (AMHCNet), thereby classifying the user's different needs.

[0121] S4: The user database storage unit sets up a personalized database for the brain-controlled massage chair user. This database stores EEG signal-related data and classification results from the deep learning classification unit to train the system and adapt to the user's feedback commands regarding massage position and mode. Steps S1-S3 continuously optimize and adjust the parameters of the deep learning classification unit, using this as the standard for outputting classification results to the control module for controlling the massage chair.

[0122] S5: The control module activates the corresponding massage program based on the classification results output by the deep learning classification unit after parameter optimization. The control module automatically adjusts various parts of the massage module, such as the head, back, waist, and legs, selects the corresponding massage mode, and controls the massage units of the corresponding massage module to perform targeted massage on the areas of interest according to the user's feedback and intention commands.

[0123] Example 3

[0124] This embodiment 3 is based on embodiment 1, and further explains how other implementation methods can be used to achieve the purpose of the present invention to a certain extent.

[0125] 1. Without using TQWT and phase space reconstruction methods, the decomposition and reconstruction unit of the system in Example 1 is deleted. Instead, the collected EEG signals are preprocessed and denoised before being input into the attention-based hybrid convolutional neural network of the deep learning classification unit in Example 1 for classification.

[0126] 2. Instead of using deep learning models, we extract the entropy features of different frequency bands of EEG signals and use algorithms such as support vector machines or random forests to classify the input signal matrix.

[0127] In summary, compared to existing technologies, the embodiments of the present invention dynamically adjust the massage position and mode according to the user's immediate needs, achieving a highly personalized and intelligent health service. Furthermore, by strictly categorizing user commands into five areas—head, back, waist, legs, and balance—the corresponding EEG signals generated during signal acquisition become more easily identifiable. Compared to existing multi-channel EEG signal acquisition units, the embodiments of the present invention use a single-channel EEG signal acquisition unit, improving device convenience and user comfort, and significantly reducing the complexity and redundancy of signal processing, thereby increasing the device's response speed to user ideas. Compared to techniques that only analyze and process certain frequency bands of EEG signals, the embodiments of the present invention employ a signal processing method using tunable Q-factor wavelet transform (TQWT) and phase space reconstruction, which can more comprehensively preserve and distinguish different categories of EEG signals. This invention fully utilizes the different time-frequency resolutions and nonlinear dynamic characteristics of EEG signals to improve the accuracy of model classification. Instead of manual feature extraction and complex feature engineering, this invention automatically learns and extracts information from EEG signals through a deep learning model, simplifying the process of recognizing user intent commands. Furthermore, compared to existing technologies that extract specific and limited features, the attention-based hybrid convolutional neural network proposed in this invention can comprehensively understand and capture the spatiotemporal information and nonlinear characteristics of EEG signals. Through attention mechanism-based feature fusion, the model focuses more on the most informative parts of the input data, thereby improving the ability to recognize key features and the overall model performance.

[0128] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. 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 claims of the present invention.

Claims

1. A brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning, characterized in that, The system includes: a single-channel EEG signal acquisition module, a signal processing module, a control module, and a massage module; The single-channel EEG signal acquisition module acquires the user's EEG signals; The signal processing module receives and processes the EEG signal. The signal processing module includes a preprocessing denoising unit, a decomposition and reconstruction unit, a deep learning classification unit, and a user database storage unit. The preprocessing denoising unit preprocesses and denoises the EEG signal. The decomposition and reconstruction unit performs multi-level wavelet decomposition on the preprocessed and denoised EEG signal using adjustable Q-factor wavelet transform and reconstructs the phase space matrix from the decomposed signal. The deep learning classification unit includes an attention-based multi-scale hybrid convolutional neural network model. The attention-based multi-scale hybrid convolutional neural network model takes the phase space matrix as input and the classification result of the user's feedback intent as output. The user database storage unit stores the classification results and related EEG signal data, which are used to adjust and optimize the parameters of the attention-based multi-scale hybrid convolutional neural network model. The control module is used to control the drive circuit, and the massage module is used to perform massage. The control module receives the classification results from the signal processing module and activates the corresponding massage program and drive circuit, and dynamically adjusts the mode and massage position of the massage module in real time. The attention-based multi-scale hybrid convolutional neural network model includes a deep convolutional module, a feature attention module, and a classification module. The deep convolutional module includes two parallel sub-networks: a first sub-network and a second sub-network. The first sub-network includes a first convolutional computation unit, a second convolutional computation unit, a first feature fusion unit, and a second feature fusion unit connected in sequence. The second sub-network includes a third convolutional computation unit, a fourth convolutional computation unit, a third feature fusion unit, and a fourth feature fusion unit connected in sequence. The feature attention module includes a channel attention unit and a Transformer connected in sequence. The classification module includes a first linear layer and a second Softmax activation layer.

2. The brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 1, characterized in that, The single-channel EEG signal acquisition module uses a portable single-channel EEG signal acquisition device, which includes a sensor, electrodes for acquiring EEG signals, a reference electrode REF, and a ground electrode GND.

3. The brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 1, characterized in that, The preprocessing denoising includes: using a notch filter to remove interference from the power line on the EEG signal, and performing bandpass filtering on the EEG signal to remove out-of-band noise, thereby obtaining a preprocessed and denoised EEG signal and a signal matrix. ,in It represents the total number of signal data points.

4. A brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 3, characterized in that, The frequency response of the filter bank consisting of low-pass and high-pass filters used in the process of performing multi-layer wavelet decomposition on the preprocessed and denoised EEG signal through adjustable Q-factor wavelet transform is as follows: for for in It is the frequency response of the low-pass filter. Low-pass scaling scale LPS, It is the frequency response of the high-pass filter. The high-pass scaling scale HPS, where j represents the number of decomposition layers; After preprocessing, the EEG signal undergoes tunable Q-factor wavelet transform to obtain J+1 signal matrices with different time and frequency resolutions: in, It represents the total number of signal data points.

5. A brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 4, characterized in that, The expression for the one-dimensional signal vector of the preprocessed and denoised EEG signal is a time series vector: The reconstructed phase space is then represented as: in It is the embedded dimension. It's a time delay; signal matrix Phase space reconstruction yields the phase space matrix: 。 6. A brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 5, characterized in that, The attention-based multi-scale hybrid convolutional neural network model uses the phase space matrix as input. When used as input, the phase space matrix is ​​first augmented to obtain the input signal matrix. The input signal matrix is ​​then used as the input feature matrix of the attention-based multi-scale hybrid convolutional neural network model.

7. The brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 1, characterized in that, The first convolutional computation unit includes a first convolutional layer, a first batch of normalized layers, and a first ELU activation layer connected in sequence. The kernel size of the first convolutional layer is (4, 3, 1), where 4 is the number of kernels, and 3 and 1 are the sizes of the kernels in the spatial and temporal directions, respectively. The second convolutional computation unit includes a second convolutional layer, a second batch of normalized layers, and a second ELU activation layer connected in sequence. The kernel size of the second convolutional layer is (8, 1, 8), where 8 is the number of kernels, and 1 and 8 are the sizes of the kernels in the spatial and temporal directions, respectively. The first feature fusion unit includes a third convolutional layer, a third batch normalization layer, a third ELU activation layer, and a first two-dimensional average pooling layer connected in sequence. The kernel size of the third convolutional layer is (8, 1, 4), and the pooling window size of the first two-dimensional average pooling layer is (1, 8), where 1 is the height of the pooling window and 8 is the width of the pooling window. The second feature fusion unit includes a fourth convolutional layer, a fourth batch normalization layer, a fourth ELU activation layer, and a second two-dimensional average pooling layer connected in sequence. The kernel size of the fourth convolutional layer is (8, J+1, 1), and the size of the second two-dimensional average pooling layer is (1, 8), where J+1 represents the number of decomposition layers of the adjustable Q-factor wavelet transform (TQWT). The third convolutional computation unit includes a fifth convolutional layer, a fifth batch normalization layer, and a fifth ELU activation layer connected in sequence. The kernel size of the fifth convolutional layer is (4, 3, 1). The fourth convolutional computation unit includes a sixth convolutional layer, a sixth batch normalization layer, and a sixth ELU activation layer connected in sequence. The kernel size of the sixth convolutional layer is (8, 1, 4). The third feature fusion unit comprises a seventh convolutional layer, a seventh batch normalization layer, a seventh ELU activation layer, and a third two-dimensional average pooling layer connected in sequence. The kernel size of the seventh convolutional layer is (8, 1, 4), and the pooling window size of the third two-dimensional average pooling layer is (1, 8). The fourth feature fusion unit comprises an eighth convolutional layer, an eighth batch normalization layer, an eighth ELU activation layer, and a fourth two-dimensional average pooling layer connected in sequence. The kernel size of the eighth convolutional layer is (8, J+1, 1), and the size of the fourth two-dimensional average pooling layer is (1, 8). The outputs of the first sub-network and the second sub-network are reduced in dimensionality in the spatial dimension and merged in the temporal dimension to generate the output feature matrix of the deep convolutional module: The expressions for the ELU activation functions of all ELU activation layers are as follows: The expressions for all two-dimensional average pooling layers are: in, For the input feature matrix at position The value, To output the feature matrix at position The value, and These represent the height and width of the pooling window, respectively.

8. A brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 7, characterized in that, The channel attention unit includes a first one-dimensional global average pooling layer, a ninth convolutional layer, and a first sigmoid activation layer connected in sequence. Output feature matrix As input to the feature attention module, the formula for one-dimensional global average pooling is: in Represents the input matrix exist The value of the position, Indicates the output vector in The value of the position; The formula for the sigmoid activation function is: The output of the first one-dimensional global average pooling layer Transpose the vector and use it as input to the ninth convolutional layer, then combine the channel attention weight vector with... Multiplying yields the channel attention feature matrix. Transposing the channel attention feature matrix by its channel dimension and time dimension results in... ; Will The input to the Transformer is used to learn the global correlation in the feature matrix; The Transformer includes a position embedding unit, a multi-head attention network, a layer normalization unit, a feedforward network, and a first Softmax activation layer. The position embedding formula in the position embedding unit is as follows: The multi-head attention network will input the feature matrix. Divided into h parts, each part is The multi-head attention network includes multiple self-attention units, each of which receives input... Linear transformation to query ,key Sum , and The dot product can calculate the global positional dependencies. The result is passed through the first Softmax activation layer to obtain the weight matrix, i.e., the attention score. Then, the attention score is padded with the dot product. Weighting these global dependencies and adding them to the features can be represented as follows: in Representation matrix The length of the channel dimension is determined by the Transformer's use of residual connections at the locations of the multi-head attention network and the feedforward network to maintain the integrity of the input information, mitigate the gradient vanishing problem, and accelerate convergence. Finally, the output matrix of the feature attention module is generated. 。 9. A brain-controlled massage chair system based on wavelet transform, phase space reconstruction, and deep learning according to claim 8, characterized in that, The number of neurons in the first linear layer is equal to the number of classification categories. Based on the neuron index corresponding to the maximum output value of the second Softmax activation layer, the user's feedback intent corresponding to each collected EEG signal is obtained and output as the classification result.

Images