A spinal posture detection method, device and equipment based on sEMG and SpO2
By combining sEMG and SpO2 signals, using BiLSTM neural network model, non-invasive and portable spinal posture detection is achieved, solving the problem of inability to evaluate spinal posture in the prior art, and improving the real-time and efficiency of spinal rehabilitation assessment .
Patent Information
- Application Number
- CN202211331276.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-28
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-10-28
AI Technical Summary
The prior art cannot effectively detect spinal posture in a state of motion, and traditional medical imaging technology has insufficient radiation, high costs, and the need to rest, so it is impossible to evaluate the recovery of spinal posture.
Using a combination of surface electromyography signal (sEMG) and blood oxygen saturation (SpO2), we use the method of collecting and analyzing signals, using the BiLSTM neural network model to extract features, detect spinal posture in real time, and provide spinal rehabilitation assessment in motor states.
Non-invasive and portable spinal posture detection is realized, which improves the real-time and efficiency of scoliosis detection, facilitates spinal rehabilitation assessment, and solves the problem of low EMG signal in the prior art.
Smart Images

Figure CN115886790B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical big data processing, and in particular to a spinal posture detection method, device and equipment based on sEMG and SpO2. Background Art
[0002] Spinal disease is a common disease in modern people. Medical imaging technologies such as computed tomography (CT) and magnetic resonance imaging (MRI) are usually used as auxiliary means for doctors to diagnose spinal diseases, thereby realizing the detection of spinal posture. However, CT has a large radiation dose and is not suitable for multiple irradiation; MRI detection is costly and requires subjective judgment by doctors. At the same time, medical imaging technologies require patients to remain still and cannot evaluate spinal posture in motion. Therefore, it is of great significance to find a non-invasive and portable method for spinal posture detection for the rehabilitation evaluation of spinal posture in motion of patients undergoing non-surgical treatment.
[0003] Electromyographic signal (EMG) is the superposition of action potentials of motor units in many muscle fibers in time and space. Surface electromyographic signal (sEMG) is the comprehensive effect of superficial muscle EMG and electrical activity on nerve trunks on the skin surface, which can reflect neuromuscular activity to a certain extent. SEMG has the advantages of non-invasiveness, non-traumaticity and simple operation in measurement. Therefore, sEMG has important practical value in clinical medicine, ergonomics, rehabilitation medicine and sports science.
[0004] Blood oxygen saturation (SpO2) refers to the percentage of oxygenated hemoglobin in the blood that is bound to oxygen to the total hemoglobin capacity that can be bound, and is an important indicator of the normal functioning of the human body. Studies have shown that muscle blood oxygen content, like myoelectric signals, can reflect paraspinal muscle activity during exercise.
[0005] Therefore, it is very meaningful to use a combination of sEMG and SpO2 to detect spinal posture and explore a quantitative method for spinal posture rehabilitation assessment during motion. Summary of the invention
[0006] In order to address the deficiencies in the prior art, the present application proposes a spinal posture detection method, device and equipment based on sEMG and SpO2. After collecting surface electromyography signals (sEMG) and blood oxygen saturation (SpO2) and performing data analysis, the analysis results of different spinal postures are output according to the features extracted from sEMG and SpO2.
[0007] The technical solution adopted by the present invention is as follows:
[0008] A spinal posture detection method based on sEMG and SpO2 comprises the following steps:
[0009] Step S110. Simultaneously collect the surface electromyography signal (sEMG) and the blood oxygen dual-wavelength signal of the object to be detected;
[0010] Step S120. The surface electromyography signal sEMG and the blood oxygen dual-wavelength signal are passed through a modulation circuit for amplification, filtering and noise reduction to obtain discrete values of the sEMG and blood oxygen dual-wavelength signals respectively;
[0011] Step S130. respectively extracting the time domain features and frequency domain features of sEMG, including the integral value IEMG, the root mean square value RMS, the intermediate frequency MF and the mean power frequency MPF, and extracting the time domain characteristic signal of the blood oxygen saturation SpO2;
[0012] Step S140. Based on the time domain features and frequency domain features of sEMG and the time domain features of SpO2 obtained in S130, draw a waveform diagram corresponding to each feature;
[0013] Step S150. Build and train a BiLSTM neural network model, use the LSTM neural network model to train and classify the IEMG, RMS, MF, MPF and SpO2 feature values extracted in step S130, and output the detection results.
[0014] Furthermore, the method of collecting sEMG of the subject to be detected is as follows: using patch electrodes to collect spinal electromyographic signals of the subject.
[0015] Furthermore, a method for collecting blood oxygen signals of a subject to be detected is as follows: a blood oxygen probe is used to collect a dual-wavelength blood oxygen signal of the subject's spine.
[0016] Furthermore, the wavelet transform method is used to denoise the blood oxygen dual-wavelength signal.
[0017] Furthermore, the IEMG, RMS, MF, MPF and SpO2 sequence signals within 20 seconds were extracted.
[0018] Furthermore, the BiLSTM neural network model is composed of a forward LSTM and a backward LSTM. The hidden layer vectors obtained by the forward LSTM and the backward LSTM are concatenated through an attention layer, and then a Softmax transfer function is used as an activation function.
[0019] Furthermore, the memory unit structure (forget gate, memory gate and output gate) of the LSTM neural network model is expressed as:
[0020] f t =σ(W f ·[h t-1 ,x t ]+b f )
[0021] it =σ(W i ·[h t-1 ,x t ]+b i )
[0022]
[0023] O t =σ(W o ·[h t-1 ,x t ]+b o )
[0024] h t =O t tanh(C t )
[0025] Among them, f t Represents the function of the forget gate, which determines the memory unit C at the previous moment t-1 The retention value of the state, h t-1 represents the output of the memory unit at the previous moment, x t Represents the input of the original signal, W f , W i , W c , W o Respectively represent the weight matrices corresponding to the forget gate, memory gate and output gate, b f 、b i 、b c 、b o is the bias vector corresponding to the forget gate, memory gate and output gate, i t The function representing the memory gate determines the current moment The retention value of C t Indicates the current state of the memory unit, C t-1 is the state of the memory unit at the previous moment, is the update function of the memory unit state at the current moment, tanh is the activation function, and σ(·) represents the Sigmoid function. t Represents the output gate signal, h t is the output of the memory unit at the current moment.
[0026] A spinal posture detection device based on sEMG and SpO2, comprising:
[0027] A data acquisition module, wherein the data acquisition module collects sEMG signals and blood oxygen dual-wavelength signals through an electromyography sensor and a near-infrared sensor respectively;
[0028] A hardware processing module, wherein the hardware processing module obtains discrete values of the sEMG signal and the blood oxygen dual-wavelength signal through a modulation circuit that amplifies, filters and reduces noise;
[0029] An information transmission module, which transmits the obtained sEMG signal and blood oxygen dual-wavelength signal to a PC for analysis;
[0030] A feature analysis module, which extracts feature values from sEMG signals, processes and analyzes blood oxygen dual-wavelength signals, and draws corresponding analysis result graphs;
[0031] A posture analysis module, which classifies sEMG signals and blood oxygen dual-wavelength signals through a BiLSTM neural network model, and detects and identifies different spinal postures in real time;
[0032] The result display module displays the analysis results on the corresponding interface of the PC.
[0033] Furthermore, the information transmission module uses Bluetooth to realize information transmission.
[0034] Furthermore, an electronic device includes a processor and a display, wherein the processor has built-in the above-mentioned spinal posture detection method based on sEMG and SpO2.
[0035] Beneficial effects of the present invention:
[0036] The present invention proposes a spinal posture detection method and device based on sEMG and SpO2. On the one hand, the sEMG signal collected by the electromyography sensor and the blood oxygen dual-wavelength signal collected by the near-infrared sensor are received, and the sEMG signal and the blood oxygen dual-wavelength signal are amplified, filtered and denoised by a modulation circuit to obtain discrete values of the sEMG and blood oxygen dual-wavelength signals, and then the sEMG time domain and frequency domain features (including IEMG, RMS, MF and MPF) are extracted, and the SpO2 signal is processed and analyzed at the same time, and the corresponding analysis result graph is drawn for intuitive observation; on the other hand, the IEMG, RMS, MF, MPF and SpO2 multi-dimensional data are trained and classified through the LSTM neural network model, and the model parameters of the LSTM are further optimized for real-time detection of corresponding different spinal postures, and the results are viewed on the PC side, which facilitates the real-time analysis and review of the spinal posture, and effectively solves the problem in the prior art that the patient's spinal posture cannot be observed in real time and intuitively due to the low timeliness of the electromyography signal. The detection results of the present invention can be used as a reference for scoliosis detection and spinal rehabilitation evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1The flowchart schematically shows a method for detecting spinal posture based on sEMG and SpO2 according to an exemplary embodiment of the present disclosure.
[0038] Figure 2 An example block diagram of an electromyography and blood oxygen signal acquisition and conditioning circuit according to an example embodiment of the present disclosure is schematically shown.
[0039] Figure 3 A block diagram schematically shows a spinal posture detection device based on sEMG and SpO2 according to an exemplary embodiment of the present disclosure.
[0040] Figure 4 A diagram schematically shows a spinal posture classification result according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0042] In this example implementation, a spinal posture detection method based on sEMG and SpO2 is first provided. The method is based on the analysis of sEMG and SpO2 and can be used for scoliosis detection and spinal rehabilitation assessment. Figure 1 As shown, the detection method may include the following steps:
[0043] Step S110: Receive the sEMG signal collected by the electromyography sensor and the blood oxygen dual-wavelength signal collected by the near-infrared sensor.
[0044] Step S120: The sEMG signal and the blood oxygen dual-wavelength signal are passed through a modulation circuit for amplification, filtering and noise reduction to obtain discrete values of the sEMG signal and the blood oxygen dual-wavelength signal respectively.
[0045] Step S130: extract the time domain features and frequency domain features (including IEMG, RMS, MF and MPF) of sEMG respectively, and extract the time domain features of SpO2.
[0046] Step S140: Draw waveforms of IEMG, RMS, MF, MPF and SpO2 for observation. Calculate each characteristic value using existing formulas and draw waveforms.
[0047] Step S150. Build a BiLSTM neural network model, use the BiLSTM neural network model to train and classify the IEMG, RMS, MF, MPF and SpO2 multi-dimensional data extracted in step S130, and output the detection results; further optimize the model parameters of BiLSTM for real-time detection of corresponding different spinal postures.
[0048] In the above-mentioned spinal posture detection method, on the one hand, the sEMG signal collected by the electromyography sensor and the blood oxygen dual-wavelength signal collected by the blood oxygen sensor are received, the sEMG signal and the blood oxygen dual-wavelength signal are amplified and noise-reduced, the sEMG signal is subjected to feature value extraction and the blood oxygen dual-wavelength signal is processed and analyzed; based on the above-mentioned feature value of the extracted sEMG signal and the information after the processing of the blood oxygen dual-wavelength signal, the corresponding analysis result graph is drawn for intuitive observation;
[0049] On the other hand, the sEMG signals are classified through the BiLSTM neural network model and combined with the blood oxygen dual-wavelength signal to detect and identify different spinal postures in real time. The results can be viewed on the PC to facilitate real-time analysis and review of spinal postures. This effectively solves the problem of the low timeliness of sEMG signals in existing technologies, which makes it impossible to observe the patient's spinal posture in real time. This improves the efficiency of scoliosis detection and portable spinal rehabilitation assessment.
[0050] The following will explain and illustrate the various steps involved in the above-mentioned medical data processing method in this example embodiment in detail with reference to the accompanying drawings.
[0051] In step S110, a sEMG signal collected by an electromyography sensor and a blood oxygen dual-wavelength signal collected by a near-infrared sensor are received.
[0052] In this example embodiment, before the experiment, the subject exposed his waist, maintained a standing posture, and wiped his skin with 75% alcohol. The arms were in a relaxed state. The electromyographic signal detection electrodes 1 and 2 were attached to the left and right sides of the lower spine midline of the scoliosis arc, 2 cm from the side. The blood oxygen probes 3 and 4 were attached to the left and right sides of the lower spine midline of the scoliosis arc, 2 cm from the side, and 1 cm from the electromyographic signal electrode. The subject first maintained a standing posture for 20 seconds, then bent 45 degrees, bending once every 5 seconds for a total of 4 times, 20 seconds. After each trial, the subject rested for 30 seconds to avoid fatigue caused by long-term muscle activity affecting the experimental results. The experiment was divided into four groups: upright, left bend, right bend, and forward bend, each group containing 80 trials. The subject's electromyographic signal, blood oxygen signal, and corresponding posture formed a data set, in which the posture was used as a label corresponding to the data.
[0053] In step S120, the sEMG signal and the blood oxygen dual-wavelength signal are respectively amplified and filtered by a modulation circuit for noise reduction to obtain discrete values of the sEMG signal and the blood oxygen dual-wavelength signal.
[0054] In this example embodiment, sEMG has a narrow distribution and a wide spectrum distribution, which is easily submerged by noise interference. Due to the instability of sEMG, general traditional analysis methods may lose non-fixed parts. Therefore, sEMG is processed together with wavelet decomposition to achieve signal denoising. Specifically, this paper selects sys4 as the mother wavelet, decomposes 5 layers of sEMG signals, and deploys the maximum-minimum threshold for threshold denoising. According to the sampling theorem, the maximum frequency of the signal obtained by the sampling frequency of 1600Hz is 500Hz, then the fifth-order approximate frequency band of the signal is less than 15.6Hz. In this way, low-frequency noise interference can be filtered out by subtracting the reconstructed signal of the fifth-order approximate frequency band from the original signal reconstructed by the wavelet. The commonly used feature extraction methods in the sEMG signal processing mode are mainly time domain analysis, frequency domain analysis, time-frequency domain analysis, etc. The time domain analysis method can clearly observe the shape of the signal changing over time. The frequency domain analysis method is mainly implemented by Fourier transform. The wavelet transform method is also used to denoise the collected blood oxygen dual-wavelength signal. Select appropriate wavelet basis and wavelet scale, then perform wavelet denoising on the signal to obtain better SpO2.
[0055] In step S130, the time domain and frequency domain features of the sEMG signal (including IEMG, RMS, MF and MPF) are extracted and the SpO2 signal is processed and analyzed at the same time. All sEMG signals and blood oxygen dual-wavelength signal data use the data within 1s as the translation window, 0.5s as the overlapping window, and the corresponding feature values (including IEMG, RMS, MF, MPF and SpO2) are obtained by translation calculation for 20s.
[0056] In this exemplary embodiment, sEMG time domain feature signal extraction and sEMG frequency domain feature signal extraction are performed, and SpO2 time and frequency domain feature signal extraction is performed. The details are as follows:
[0057] For sEMG signals, we first select sys4 as the mother wavelet, and then perform wavelet transform denoising on the signal. After obtaining a good denoising effect, we search and intercept relatively stable signal sections in the time domain, and then perform FFT transform on it, process the signal in the frequency domain, and obtain the result.
[0058] The formula for wavelet transform is as follows:
[0059] The wavelet transform expression calculation formula for processing continuous signals is shown in formula (1).
[0060]
[0061] Where x(t) is the continuous sampling signal, t is the time variable, τ is the translation factor, and a is the scale factor. is the wavelet basis function, is a conjugate function.
[0062] In order to eliminate the redundancy of continuous wavelet, the parameters a and b in the wavelet function are discretized and binary discretization is adopted. The wavelet transform expression calculation formula for processing discrete signals is shown in formula (2).
[0063]
[0064] Where ψ*(t) is the conjugate function of the wavelet basis, x(nT s ) is a discrete sampling signal, T s is the signal sampling period, n is the number of sampling points of the signal, j is the frequency resolution, and k is the translation. According to the wavelet multi-scale analysis, the original signal is divided into the approximate component CA and the high-frequency component CD. The denoising method is a combination of wavelet hard threshold and soft threshold. The high-scale wavelet coefficients are denoised by hard threshold, and the low-scale wavelet coefficients are denoised by soft threshold. Finally, the wavelet coefficients are reconstructed. The reconstruction formula is as follows (3).
[0065] S=cA n +cD1+cD2+…+cD n (3)
[0066] Where cA n is the approximate coefficient part of the decomposition, cD n For the detail coefficient part of the decomposition, the decomposition scale n is determined by the signal sampling rate.
[0067] For EMG signals, time domain feature extraction includes IEMG and RMS. These features can clearly show the average variation range and effective discharge value of the operations performed by the tester in the time domain.
[0068] The root mean square value (RMS) can well reflect the average power characteristics of a segment of sEMG, that is, it can be used to measure the load degree of muscles in different positions during a set of movements. Its calculation formula is shown in formula (4).
[0069]
[0070] The integrated sEMG (IEMG) value is the integrated value of the sEMG signal amplitude, which represents the firing rate of the motor nerve unit per unit time. Its calculation method is shown in formula (5).
[0071]
[0072] Here, EMG(t) is the sEMG signal, t is the time, and T is the period.
[0073] In this experiment, we selected RMS and IEMG as the time domain feature parameters of sEMG signals.
[0074] For EMG signals, frequency domain feature extraction includes mean power frequency (MPF) and intermediate frequency (MF). Frequency domain signals can show the change of power signals with frequency, which is helpful for observing frequency domain characteristics and avoiding instability caused by signal fluctuations over time.
[0075] Mean power frequency (MPF) refers to the average power spectrum obtained by fast Fourier transform of sEMG signal, which represents muscle fatigue. Its calculation formula is shown in formula (6).
[0076]
[0077] In the above formula, f is the frequency and p(f) is the power spectral density of the sEMG signal.
[0078] The intermediate frequency (MF) refers to half of the power spectrum obtained by fast Fourier transform of the sEMG signal. Its magnitude is related to the fast and slow muscle fibers in the muscle tissue involved in the activity. Its calculation formula is shown in formula (7).
[0079]
[0080] Here, f is the frequency and p(f) is the power spectral density of the sEMG signal.
[0081] For the blood oxygen signal, wavelet denoising using db5 wavelet is first used.
[0082] Based on the difference in HbO2 and HbR absorption spectra, a blood oxygen reconstruction algorithm based on SpO2 is adopted, and its calculation formula is shown in the following formula (8).
[0083]
[0084] Among them, ΔOD λ is the change in the optical density of the outgoing light with a wavelength of λ, and the value of λ is 760nm and 850nm, D a is the differential path factor of the partial optical path, Δρ is the difference in distance between the two photodetectors and the light source, and It represents the molar extinction coefficient of oxygenated hemoglobin to 760nm and 850nm wavelength light sources. and It represents the molar extinction coefficient of reduced hemoglobin to 760nm and 850nm wavelength light sources. is the oxygenated hemoglobin concentration value at the monitoring site, C HbR is the reduced hemoglobin concentration value of the monitoring site, then SpO2 is calculated as shown in the following formula (9).
[0085]
[0086] In step S140, a corresponding analysis result graph is drawn for intuitive observation.
[0087] In this exemplary embodiment, sEMG time domain feature signal extraction is performed, and then SpO2 time domain feature signal extraction is performed, and then the original signals and analysis results of sEMG and SpO2 are displayed on the PC by jumping to the analysis interface.
[0088] In step S150, the IEMG, RMS, MF, MPF and SpO2 multi-dimensional data are trained and classified through the LSTM neural network model, and the model parameters of the LSTM are further optimized for real-time detection of corresponding different spinal postures.
[0089] In this example embodiment, Long Short-Term Memory (LSTM) is a special RNN that mainly solves the problem of gradient vanishing and gradient exploding during long sequence training. However, the general LSTM model is forward and cannot encode information from back to front, which led to the proposal and development of BiLSTM, combining forward LSTM and backward LSTM. In order to improve the accuracy of spinal posture recognition, this paper uses BiLSTM to classify sEMG feature values. LSTM contains a set of gates that can control sequence data, including forget gate, input gate and output gate.
[0090] The basic memory cell structure can be represented by the following mathematical models (10)-(15):
[0091] f t =σ(W f ·[h t-1 ,x t ]+b f ) (10)
[0092] i t =σ(W i ·[h t-1 ,x t ]+b i ) (11)
[0093]
[0094] O t =σ(W o ·[h t-1 ,x t ]+b o ) (14)
[0095] h t =O t *tanh(C t ) (15)
[0096] Among them, f t Represents the function of the forget gate, which determines the memory unit C at the previous moment t-1 The retention value of the state, h t-1 represents the output of the memory unit at the previous moment, x t Represents the input of the original signal, W f , W i , W c , W o Respectively represent the weight matrices corresponding to the forget gate, memory gate and output gate, b f 、b i 、b c 、b o is the bias vector corresponding to the forget gate, memory gate and output gate, i t The function representing the memory gate determines the current moment The retention value of C t Indicates the current state of the memory unit, C t-1 is the state of the memory unit at the previous moment, is the update function of the memory unit state at the current moment, tanh is the activation function, and σ(·) represents the Sigmoid function. t Represents the output gate signal, h t is the output of the memory unit at the current moment.
[0097] BiLSTM contains forward and reverse paths, not just the forward path. The total size of the dataset is 200, which is divided into four groups, including bending to the left, bending to the right, bending forward, and standing to the right. Then 80% of the data is the training set and 20% of the data is the test set. Here, 1s is used as the convolution window, and the 20s IEMG, RMS, MF, MPF and SpO feature vectors are calculated as the input of BiLSTM. The global feature vectors after the output of the forward LSTM and the backward LSTM are concatenated. The fully connected layer has 2 neurons, the activation function is the Softmax function, and Adam is selected as the training optimizer. The learning rate is set to 0.001, and 20 epochs are set to construct a higher-precision training network model. The final classification results are as follows Figure 4 As shown, the number of 0, 1, 2, and 3 are outputs, belonging to the 4 categories of the training set.
[0098] The model parameters of LSTM were further optimized, with an input vector dimension of 10×20, an optimizer of Adam, 10 training rounds, a learning rate of 0.01, 20 samples per batch, and an iterative approach to obtain the weight matrix and bias matrix in the optimal network for real-time detection of different corresponding spinal postures.
[0099] The present disclosure also provides a spinal posture detection device based on sEMG and SpO2, referring to Figure 3The medical data processing device may include a data acquisition module S310, a hardware processing module S320, an information transmission module S330, a feature analysis module S340, a posture analysis module S350 and a result display module S360. Among them:
[0100] The data acquisition module S310 collects sEMG signals and blood oxygen dual-wavelength signals through electromyography sensors and near-infrared sensors respectively;
[0101] The hardware processing module S320 obtains discrete values of the sEMG signal and the blood oxygen dual-wavelength signal through a modulation circuit that amplifies, filters and reduces noise;
[0102] The information transmission module S330 transmits the obtained electromyography sampling signal and blood oxygen sampling signal to the PC for analysis via the Bluetooth module;
[0103] The feature analysis module S340 extracts feature values from the electromyography sampling signal and processes and analyzes the blood oxygen sampling signal, and draws a corresponding analysis result graph for intuitive observation;
[0104] The posture analysis module S350 classifies the electromyographic sampling signals through the LSTM neural network model and combines them with the blood oxygen signal to detect and identify different spinal postures in real time;
[0105] The results of the analysis by the result display module S360 are displayed on the corresponding interface of the PC, which is convenient for real-time analysis and review of the spinal posture. In this example embodiment, the interface is built by pycharm to display the results of the analysis, which is convenient for real-time analysis and review of the spinal posture. An electronic device includes a processor and a display, and the processor has a built-in spinal posture detection method based on sEMG and SpO2.
[0106] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.
Claims
1. A spinal posture detection method based on sEMG and SpO2, characterized in that: The steps include: Step S110. Simultaneously collect the surface electromyography signal (sEMG) and the blood oxygen dual-wavelength signal of the object to be detected; Step S120. The surface electromyographic signal sEMG and the blood oxygen dual-wavelength signal are respectively obtained through a modulation circuit for amplification, filtering and noise reduction; the wavelet transform method is used to denoise the blood oxygen dual-wavelength signal; for the blood oxygen dual-wavelength signal, the db5 wavelet denoising is first used; based on the difference in the absorption spectra of oxygenated hemoglobin HbO2 and reduced hemoglobin HbR, a blood oxygen reconstruction algorithm based on blood oxygen saturation SpO2 is used, and its calculation formula is as follows: The calculation of blood oxygen saturation SpO2 is as follows: Among them, ΔOD λ is the change in the optical density of the outgoing light with a wavelength of λ, and the value of λ is 760nm and 850nm, D a is the differential path factor of the partial optical path, Δρ is the difference in distance between the two photodetectors and the light source, and It represents the molar extinction coefficient of oxygenated hemoglobin to 760nm and 850nm wavelength light sources. and It represents the molar extinction coefficient of reduced hemoglobin to 760nm and 850nm wavelength light sources. is the oxygenated hemoglobin concentration value at the monitoring site, C HbR is the reduced hemoglobin concentration value at the monitoring site; Step S130. Extract the time domain features and frequency domain features of sEMG within 20 seconds, including the integral value IEMG, the root mean square value RMS, the intermediate frequency MF and the mean power frequency MPF, and extract the time domain feature signal of SpO2 at the same time. All sEMG and blood oxygen signal data use the data within 1 second as the translation window and 0.5 second as the overlapping window. The corresponding feature values are obtained by translation calculation for 20 seconds, including IEMG, RMS, MF, MPF and SpO2; as shown below: Where EMG(t) is the sEMG signal, t is the time, T is the period, f is the frequency, and p(f) is the power spectral density of the sEMG signal; Step S140. Based on the time domain features and frequency domain features of sEMG and the time domain features of SpO2 obtained in S130, draw a waveform diagram corresponding to each feature; Step S150. Build and train a BiLSTM neural network model, which is composed of a forward LSTM and a backward LSTM. The feature time series information of the feature parameters IEMG, RMS, MF, MPF and SpO2 proposed in step S130 is obtained through the BiLSTM network. The attention weight of each of the above feature time series is calculated through the attention layer, and the output of the BiLSTM is weighted and summed using the attention weight. Finally, Softmax is used as the activation function to output the classification result of the spinal posture.
2. The method for detecting spinal posture based on sEMG and SpO2 according to claim 1, characterized in that: The method for collecting sEMG signals of the subject to be detected is as follows: patch electrodes are used to collect spinal electromyographic signals of the subject.
3. The method for detecting spinal posture based on sEMG and SpO2 according to claim 1, characterized in that: The method for collecting the dual-wavelength blood oxygen signal of the subject to be detected is as follows: using a blood oxygen probe to collect the dual-wavelength blood oxygen signal of the spine of the subject.
4. The method for detecting spinal posture based on sEMG and SpO2 according to claim 1, characterized in that: The memory unit structure of the BiLSTM neural network model is expressed as: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) The t =σ(W o ·[h t-1 ,x t ]+b o ) h t =O t *tanh(C t ) Among them, f t Represents the function of the forget gate, which determines the memory unit C at the previous moment t-1 The retention value of the state, h t-1 represents the output of the memory unit at the previous moment, x t Represents the input of the original signal, W f , W i , W c , W o Respectively represent the weight matrices corresponding to the forget gate, memory gate and output gate, b f , b i , b c , b o is the bias vector corresponding to the forget gate, memory gate and output gate, i t The function representing the memory gate determines the current moment The retention value of C t Indicates the current state of the memory unit, C t-1 is the state of the memory unit at the previous moment, is the update function of the memory unit state at the current moment, tanh is the activation function, σ(·) represents the Sigmoid function, O t Represents the output gate signal, h t is the output of the memory unit at the current moment.
5. A spinal posture detection device based on sEMG and SpO2, characterized in that: include: A data acquisition module (S310), wherein the data acquisition module (S310) collects sEMG signals and blood oxygen dual-wavelength signals through an electromyography sensor and a near-infrared sensor respectively; A hardware processing module (S320), wherein the hardware processing module (S320) obtains discrete values of the sEMG signal and the blood oxygen dual-wavelength signal through a modulation circuit that amplifies, filters and reduces noise; An information transmission module (S330), wherein the information transmission module (S330) transmits the obtained sEMG signal and the blood oxygen dual-wavelength signal to a PC for analysis; A feature analysis module (S340), wherein the feature analysis module (S340) extracts feature values from the sEMG signal and processes and analyzes the blood oxygen dual-wavelength signal, and draws a corresponding analysis result graph; A posture analysis module (S350), wherein the posture analysis module (S350) classifies sEMG signals and blood oxygen dual-wavelength signals through a BiLSTM neural network model, and detects and identifies different spinal postures in real time; a spinal posture detection method based on sEMG and SpO2 as described in claim 1 is built in the processor; The result display module (S360) displays the analysis results on a corresponding interface of the PC.
6. The spinal posture detection device based on sEMG and SpO2 according to claim 5, characterized in that: The information transmission module (S330) uses Bluetooth to realize information transmission.
