A low-power magnetic induction tomography target disturbance analysis method based on bioelectromagnetic principles

Abstract

The present invention discloses a low-power magnetic induction tomography target disturbance analysis method based on bioelectromagnetic principles. Two sinusoidal voltage signals with the same frequency and phase are sent through an excitation module. One of the signals is passed through a low-pass filter and enters a power amplifier module for signal conversion and amplification. The signal is then input into a coil module. The differential signal generated by the coil module is further amplified and input into a multiplexer MUX. The MUX selectively transmits the detection signal to an analog phase detection module and a digital phase detection module. Salt solutions of different conductivity are placed in the system field area of ​​the coil module. A series of modulated signals of different frequency bands and corresponding reference signals are collected by frequency sweeping to obtain a data set. The data set is trained based on an improved LSTM regression prediction model based on the distance correlation coefficient. The method of the present invention can detect different target disturbance state precision requirements with low power consumption, simple operation and accurate detection effect.

Description

Technical Field

[0001] The present invention belongs to the field of magnetic induction tomography detection, and in particular relates to a low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles. Background Art

[0002] Bioelectromagnetism is the study of electrical, magnetic, and electromagnetic phenomena in biological tissues. This includes the behavior of excitable tissues, currents and potentials within organisms, internal and external magnetic fields, the responses of biological cells to electromagnetic fields, and the electromagnetic properties of biological tissues. Charged ions and biomacromolecules exist within organisms. These charges and dipoles, driven by external electromagnetic fields or membrane potential, generate bioelectrical currents. Water plays a crucial role in living organisms, existing as a dipole. Ion channels and potential differences across cell membranes are the foundation of cellular electrical activity, and different tissues and organs possess varying electrical and magnetic conductivities. Bioelectromagnetism has a wide range of applications in medical diagnostics (such as electrocardiography, electroencephalography, and electromyography), body fat analysis, tumor detection, neuromodulation, and rehabilitation. Through experimental research and computational modeling, it has advanced medical technology and promoted the diagnosis and treatment of diseases.

[0003] Compared to technologies based on bioelectromagnetic principles, CT (computed tomography) and MRI (magnetic resonance imaging) are two distinct medical imaging technologies. For example, patent CN201880086909.X proposes a CT method for scanning objects in X-ray security inspection systems. CT utilizes the ability of X-rays to penetrate different tissues in the human body and detects the attenuation of the radiation to generate tomographic images of internal structures, providing high-resolution images. However, CT uses X-rays, which are ionizing radiation and pose certain radiation risks to the human body, making it unsuitable for frequent examinations. For example, patent CN202010794127.0 proposes a medical imaging diagnostic device. MRI utilizes strong magnetic fields and radiofrequency pulses to produce detailed images of the human body's internal structures through the nuclear magnetic resonance (NMR) phenomenon. It has excellent resolution, particularly for soft tissue imaging, and is widely used in the diagnosis of neurological and cardiovascular diseases. MRI does not use ionizing radiation and does not cause ionizing damage to human tissue, making it suitable for long-term and repeated examinations. However, the equipment is expensive, the scanning time is long, and it is not suitable for some patients.

[0004] Considering the close connection between electric and magnetic fields, and the fact that magnetic fields are the result of the movement of charges in matter and have virtually no effect on biological tissue, magnetic induction methods, compared to CT and MRI, require virtually no ambient light and are less susceptible to instrument performance and processing speed. Compared to CT, magnetic induction poses no risk of ionizing radiation to the human body; compared to MRI, magnetic induction is not limited by equipment cost and scan time, and consumes less power. Therefore, by leveraging the principles of magnetic induction and Maxwell's equations, and applying an external magnetic field to obtain relevant electrical property data, high-throughput and rapid analysis of conductivity, magnetic permeability, density, and other properties within the target area can be performed.

[0005] At present, most magnetic induction tomography devices on the market have the problem of high power consumption, which limits their application in portable and long-term monitoring. Therefore, the present invention proposes a low-power magnetic induction tomography target disturbance analysis method based on bioelectromagnetic principles, which obtains the conductivity data of the target area by using a single-channel magnetic induction tomography measurement device. The device uses a multiplexer (MUX) to select different phase discrimination methods. Analog phase discrimination has lower power consumption than digital phase discrimination, but its accuracy is not as high as digital phase discrimination. For different application scenarios, the most appropriate phase discrimination method can be selected through MUX to balance power consumption and accuracy. The low power consumption feature not only reduces the operating cost of the equipment, but also makes the device more practical and efficient in various application environments. Finally, by obtaining relevant physical and chemical indicators to construct a prediction model, the prediction model is used to perform corresponding analysis and detection, so as to achieve rapid and accurate detection of disturbances in the target area. Summary of the Invention

[0006] The purpose of the present invention is to propose a low-power magnetic induction tomography target disturbance analysis method based on bioelectromagnetic principles, so as to achieve low-power detection of different target disturbance state accuracy requirements, with simple operation and accurate detection effect.

[0007] The technical concept of the present invention is to first design an excitation module to ensure that the module can send two high-frequency and extremely low-phase-shift sinusoidal wave signals of the same frequency and phase voltage, and then pass one of the sinusoidal wave signals through a low-pass filter into the power amplifier module, so that the sinusoidal current is large enough to excite the coil module. The design of the coil module consists of two parts: simulation design and manufacturing. It is necessary to control relevant parameters to ensure that the simulated inductance is consistent with the measured inductance. Then, a differential amplifier circuit is designed to further amplify the differential signal generated by the coil module, and the amplified signal is input into the MUX. The MUX can selectively transmit the detection signal to the analog phase-locked module and the digital phase-locked module as needed, ultimately effectively ensuring that power consumption is reduced while meeting functional requirements, improving portability and long-term monitoring needs.

[0008] The technical solutions adopted by the present invention to achieve the above-mentioned purpose are as follows:

[0009] A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles comprises the following steps:

[0010] S1: Construct a high-frequency multi-channel excitation module with an extremely low phase offset rate. This module can send two single-ended sinusoidal voltage signals with the same frequency and phase at high frequency and extremely low phase offset rate. The first sinusoidal signal enters step S2, and the second sinusoidal signal is recorded as the reference signal.

[0011] S2: Build a power amplifier module with an integrated low-pass filter. After the first sine wave signal in step S1 passes through the low-pass filter to remove high-frequency aliasing and noise, it enters the power amplifier module, converts the sine wave signal into a differential signal, and amplifies the signal.

[0012] S3: Simulate and manufacture a coil module and a differential amplifier module. The coil module includes a receiving coil and a transmitting coil arranged in parallel and spaced apart. The differential signal amplified in step S2 is input to the transmitting coil. When the transmitting coil is energized, a magnetic field signal is generated. Under the action of the magnetic field signal, the receiving coil can achieve the effect of magnetoelectricity. The receiving coil outputs a sinusoidal voltage signal, which is further amplified by the differential amplifier module and converted into a single-ended signal to obtain a modulated signal.

[0013] S4: Prepare a series of salt solutions with different conductivities. Place the salt solutions in the system field area of ​​the coil module in step S3, that is, between the transmitting coil and the receiving coil. Follow the operations in steps S1-S3 to collect a series of modulated signals of different frequency bands under the interference of each conductive salt solution by frequency sweeping, as well as the corresponding reference signals, to obtain the corresponding data set.

[0014] S5: Build a low-power phase detection module. Use a multiplexer MUX to design a low-power phase detection module that combines analog phase detection and digital phase detection. Use analog phase detection or digital phase detection to calculate the phase difference between the two signals in the data set in step S4, and obtain a data set on the corresponding relationship between phase difference and conductivity.

[0015] S6: Construct an LSTM regression prediction model based on the improved distance correlation coefficient, divide the data set obtained in step S5 into a training set and a test set, and use the constructed prediction model for training; when the conductivity of the salt solution to be tested is subsequently tested, the phase difference is first tested according to the method of steps S1-S5, and then input into the trained prediction model to obtain the corresponding salt solution conductivity.

[0016] Preferably, the step S1 specifically includes a high-frequency multi-channel excitation module with an extremely low phase offset rate, including: a main control chip, a DDS (direct digital frequency synthesizer), and a low-pass filter. The main control chip uses STMF103RC, and the DDS uses the AD9959 chip with a maximum operating bandwidth of 200MHz, supporting linear frequency, amplitude, and phase scanning. The main control STMF103RC sends instructions for the frequency control word (FTW), phase offset word (POW), and amplitude scaling factor (ASF) to the DDS via the SPI interface to set the required output parameters. After completing the setting, by sending an I / O update signal, the DDS updates the register value to the internal buffer, thereby generating an accurate output signal. The DDS uses the following three formulas to calculate the output frequency, phase offset, and amplitude of the output signal:

[0017]

[0018] Where FTW is the frequency control word, f out is the desired output frequency, f sys is the system clock frequency of the DDS.

[0019]

[0020] where POW is the phase offset word and φ is the desired phase offset.

[0021] ASF=r I ×2 10 (3)

[0022] Where ASF is the amplitude scaling factor, r I It is the ratio of the required output current to the maximum output current.

[0023] Preferably, in step S1 , the phase shift rate of the sinusoidal wave signal sent by the high-frequency multi-channel excitation module is 0.001°-0.0001°.

[0024] Preferably, the step S2 specifically includes:

[0025] S2.1: Build a passive low-pass filter. A 200MHz Butterworth 9th-order low-pass filter is built after the direct digital frequency synthesizer (DDS) system. The main purpose is to remove high-frequency aliasing and noise and improve the spectral purity of the signal. The signal output by the DDS may contain unwanted high-frequency components and harmonics, which will cause spectral distortion. The 200MHz 9th-order Butterworth low-pass filter effectively filters out these high-frequency components and improves signal quality through its smooth frequency response (i.e., the amplitude is stable within the passband and rapidly decays after the cutoff frequency). Its transfer function is shown below:

[0026]

[0027] Among them, ω c =2πf c is the angular frequency of the cutoff frequency, f c is the cutoff frequency, n is the filter order which is 9, s is the complex frequency of the input signal entering the low-pass filter, and H(s) is the transfer function, which is the core of the filter and defines how the input signal changes in the frequency domain after passing through the filter.

[0028] S2.2: Build a two-stage power amplifier. The first stage uses the AD8138, and the second stage uses the AD8131. The AD8138 first converts the single-ended sinusoidal signal generated by the DDS into a differential signal and performs amplification (for example, by a factor of 10). The AD8131 then further amplifies the differential signal (for example, by a factor of 2), maintaining the differential signal's characteristics and increasing the output amplitude. This combination of two amplifier stages effectively amplifies the input signal by a significant factor (for example, by a factor of 20), ensuring that the amplified sinusoidal current is large enough to excite the coil while maintaining high accuracy and low distortion. The AD8138's gain is shown in Equation 5, and the AD8131's differential-mode gain is shown in Equation 6.

[0029]

[0030] Among them, G N is the gain, R F1 and R G1 are the feedback resistor and input resistor in the AD8138 peripheral circuit.

[0031]

[0032] Among them, V OUT and V IN are the differential mode output voltage and single-ended input voltage of AD8138, R F2 and R G2 are the feedback resistor and input resistor in the AD8131 peripheral circuit.

[0033] Preferably, the step S3 specifically includes: simulating and manufacturing a coil module to ensure that the inductance of the coil simulation can first meet the system requirements, and setting relevant parameters of the receiving coil and the transmitting coil according to the following formula:

[0034]

[0035] Where L is the inductance of the coil, r is the average radius of the coil, N is the number of turns in the coil, l is the length of the coil, and d is the thickness of the coil.

[0036] To ensure the coil module generates a sufficiently strong magnetic field, the actual inductance of the receiving coil must be measured. This process converts the receiving coil's differential input into a single-ended input using a loop connection. A network analyzer is then used to measure the inductance and impedance at different frequencies. By measuring the actual inductance, the power applied to the coil module can be adjusted.

[0037] S3.3: To improve the stability of the system, a specific device is 3D printed to fix the receiving coil and the transmitting coil.

[0038] Preferably, the differential amplifier module in step S3 employs a differential detection method that effectively avoids common-mode signal interference, reduces ground noise and system noise, and improves the reliability and anti-interference capability of weak signal detection by the magnetic induction coil. A differential amplifier chip with a very high common-mode rejection ratio is selected to convert the differential signal into a single-ended signal, facilitating subsequent processing.

[0039] Preferably, step S5 of constructing the low-power phase detection module specifically includes:

[0040] S5.1: Build a MUX selector with a fan-out of 2 and place it in the front stage of the module to enable the control signal to be output by the main control chip;

[0041] S5.2: The MUX selector connects the analog phase detector and the digital phase detector through circuitry. The MUX selector selects either analog or digital phase detection to calculate the phase difference based on the detection domain. Digital phase detection is used in domains with high detection accuracy, while analog phase detection is used in domains with low detection accuracy. For example, digital phase detection can be used to calculate the phase difference between two signals by executing a digital signal processing algorithm. This algorithm typically involves calculating the Fourier transform of the two signals to extract phase information, or using time-domain cross-correlation methods to estimate the phase difference. In this way, the digital phase detection module can accurately measure the phase difference of high-speed signals and use the result for further signal processing or analysis.

[0042] Preferably, in step S4, a series of salt solutions with different conductivities are prepared, which is a series of salt aqueous solutions with different concentrations, and the salt is sodium chloride.

[0043] Preferably, the LSTM regression prediction model with improved distance correlation coefficient in step S6 includes the following contents:

[0044] S6.1: Collect a series of modulated signals at different frequency bands under the interference of each conductive salt solution and the corresponding reference signal by frequency sweeping. Under each conductive salt solution interference, scan the voltage signals at i different frequency bands to obtain 30 frequency band phase difference data sets. Data is collected once every 1 second for each frequency band, for a total collection time of b seconds. Thus, each frequency band phase difference data set consists of b columns of phase difference data.

[0045] In order to obtain the frequency band data that can best represent the highest correlation between conductivity and phase difference data, it is necessary to analyze the distance correlation coefficient for each frequency band separately. The correlation calculation formula is as follows:

[0046]

[0047] Where dCor(X,Y) is the distance correlation coefficient, dCov(X,Y) is the distance covariance, dVar(X) and dVar(Y) are the distance variances of X and Y; X is a column of phase difference data in b columns of data under the corresponding frequency band, and Y is the different conductivities corresponding to a salt solutions; a and b are both positive integers between 10 and 1000.

[0048] The data set for the conductivity of the salt solution is defined as Y = [y1 y2 … y a ] T , define the phase difference data matrix of salt solution conductivity in frequency band i as W = [w i,1 w i,2 … w i,a ] T , where w i,1 It represents the phase difference data of the conductivity of salt solution No. 1 in frequency band i [Δσ i,1 Δσ i,2 … Δσ i,b ]; define the phase difference data of column b and column j of the conductivity of a salt solution in the corresponding frequency band as X=[x 1,j x 2,j … x a,j ] T ;

[0049] S6.2: The phase difference data corresponding to the frequency band with the highest total correlation value calculated and counted according to S6.1, i.e., the largest distance correlation coefficient, is used as the target phase difference data;

[0050] S6.3: Define the target phase difference data matrix obtained in S6.2 as:

[0051]

[0052] The above phase difference data D a，b As the data set input, the conductivity Y of the salt solution in S6 is output as the data set, μ X =E(X),μ Y =E(Y), and use LSTM to build a conductivity regression prediction model for training.

[0053] S6.4: The network structure of the LSTM regression prediction model begins with the input phase difference data, which is first embedded in the embedding layer. The main function of the embedding layer is to map the input high-dimensional discrete data into a low-dimensional continuous vector space, thereby converting the complex phase difference data into a form that the model can better process. Through data embedding, the model can capture potential relationships and patterns in the input data, reduce data sparsity, and make subsequent feature extraction more efficient. Furthermore, the embedding layer helps the model maintain consistency when processing data of different dimensions, improving the model's generalization ability. After data embedding, the data is input into a bidirectional LSTM (Bi-LSTM) layer for time series feature extraction. LSTM (Long Short-Term Memory) is a neural network architecture particularly well-suited for processing time series data, effectively capturing long-term dependencies in the data. By introducing two LSTM units, the Bi-LSTM, with its forward and backward feeds, can simultaneously consider the context of the time series, focusing not only on past data but also on future data. This design enables the model to better capture global contextual features when processing phase difference data, thereby extracting richer and more meaningful time series features. These extracted time series features serve as input to subsequent layers for further feature processing and model prediction. By combining the embedding layer with the Bi-LSTM, the network structure fully leverages the characteristics of time series data, improving its ability to process complex data and the accuracy of conductivity prediction.

[0054] S6.5: After being processed by the hidden layer, the extracted features are passed to the self-attention layer to capture global context and generate attention weights. These weights are applied to the features, and the predicted conductivity is output through the classification layer. This model combines bidirectional LSTM and self-attention mechanisms to improve feature extraction and prediction capabilities for time series data.

[0055] Preferably, the step S6 specifically includes: using the training set to train the prediction model, and using the test set to test the performance of the trained prediction model. And based on the determination coefficient R between the predicted value and the actual value of the test set 2 and RMSE are used as indicators to determine the effectiveness of the model.

[0056] The present invention provides a low-power magnetic induction tomography target disturbance analysis device and method based on bio-electromagnetic principles, comprising: a high-frequency multi-channel excitation module, a low-power phase detection module, and an improved LSTM regression prediction model.

[0057] The high-frequency multi-channel excitation module first constructs a high-frequency multi-channel excitation source with an extremely low phase offset rate to ensure signal accuracy and consistency. Next, a power amplifier circuit with an integrated low-pass filter is built to ensure that the amplified sinusoidal current is large enough to effectively excite the coil. Then, through simulation and manufacturing of the coil module, a specific 3D-printed device is designed to secure the receiving and transmitting coils, thereby improving the stability and accuracy of the system. These steps, combined, create a high-precision excitation module.

[0058] The low-power detection phase-lock module uses a multiplexer (MUX) adaptive control phase-lock circuit to address application scenarios with varying conductivities and precision requirements. This module can significantly reduce the high power consumption of current digital phase-locked magnetic induction tomography (MIT) devices, while also addressing the limited application scenarios of current analog phase-locked MIT devices. By using a low-power MIT measurement device, phase difference data from 200 salt solutions of varying conductivity is collected within a specified timeframe. This data consists of 30 frequency bands, ranging from 5MHz to 34MHz, with each band consisting of 256 values. This data is then used to construct a phase difference dataset.

[0059] The improved LSTM regression prediction model. The correlation between the phase difference and the conductivity is calculated to obtain the frequency band data with the highest correlation. Specifically, since each conductivity salt solution has corresponding 30 frequency bands of conductivity data, in order to obtain the frequency band data that can best represent the strongest correlation between the phase difference data and the conductivity. It is necessary to perform a distance correlation coefficient correlation analysis on each frequency band separately, calculate and count the frequency band data with the highest total correlation value as the target conductivity data, and divide the training set and test set into 7:3 ratios according to interval sampling. LSTM is used to construct a conductivity regression prediction model, and the phase difference data of the corresponding target obtained by the detection device is input as the test set to obtain the conductivity of the corresponding numbered salt solution. On this basis, the R between the actual value and the measured value is calculated. 2 and RMSE.

[0060] The beneficial effects of the present invention are:

[0061] (1) A high-frequency multi-channel excitation module with an extremely low phase shift rate is used, combined with a power amplifier circuit with an integrated low-pass filter to achieve high-precision signal excitation, ensuring signal accuracy and consistency. The stability and accuracy of the system are improved through the fixed design of the simulated and manufactured coil module and the 3D printing device.

[0062] (2) By adaptively controlling the phase detection circuit through the MUX, analog and digital phase detection methods can be flexibly selected, significantly reducing system power consumption and being applicable to a variety of application scenarios with different conductivity and accuracy requirements. Analog phase detection consumes less power, while digital phase detection has higher accuracy. The choice of multiplexer allows finding the optimal balance between power consumption and accuracy in different application environments.

[0063] (3) Conductivity prediction is performed using an improved LSTM regression prediction model based on the distance correlation coefficient. By calculating the correlation between phase difference and conductivity, the frequency band data with the highest correlation with conductivity is accurately extracted, and an LSTM regression prediction model is constructed, significantly improving the accuracy of conductivity prediction. This model performs well in processing complex, multi-band data, helping to achieve accurate analysis and detection of disturbance states within the target area.

[0064] (4) Through low-power design and efficient data processing, the device is suitable for portable and long-term monitoring applications, reducing equipment operating costs and improving application convenience and sustainability. The overall system design not only performs well in laboratory environments but also has strong practicality and is widely used in fields such as medical diagnosis and environmental monitoring, providing an effective solution for low-power, high-precision target disturbance analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a schematic diagram of the process structure of a low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles of the present invention;

[0066] Figure 2 It is a scatter plot of the predicted and actual values ​​of the conductivity of the physical and chemical indicators;

[0067] Figure 3 Schematic diagram of the prediction model network structure of the present invention. DETAILED DESCRIPTION

[0068] The specific embodiments of the present invention are described in further detail below with reference to the accompanying drawings.

[0069] Reference Figures 1 to 3 , a low-power magnetic induction tomography target disturbance analysis device and method based on bio-electromagnetic principles, comprising the following steps:

[0070] S1: Construct a high-frequency multi-channel excitation module with extremely low phase offset rate, specifically including:

[0071] Build a high-frequency multi-channel excitation module with an extremely low phase offset rate, including: a main control chip, a DDS (direct digital frequency synthesizer), and a low-pass filter. The main control chip uses the STMF103RC, and the DDS uses the AD9959 chip, with a maximum operating bandwidth of 200MHz and support for linear frequency, amplitude, and phase scanning. The main control chip STMF103RC sends instructions for the frequency control word (FTW), phase offset word (POW), and amplitude scaling factor (ASF) to the DDS via the SPI interface to set the required output parameters. After the settings are completed, the DDS updates the register values ​​to the internal buffer by sending an I / O update signal, thereby generating an accurate output signal. The DDS uses the following three formulas to calculate the output frequency, phase offset, and output signal amplitude:

[0072]

[0073] Where FTW is the frequency control word, f out is the desired output frequency, f sys is the system clock frequency of the DDS.

[0074]

[0075] where POW is the phase offset word and φ is the desired phase offset.

[0076] ASF=r I ×2 10 (3)

[0077] Where ASF is the amplitude scaling factor, r I It is the ratio of the required output current to the maximum output current.

[0078] The phase shift rate of the sine wave signal sent by the high-frequency multi-channel excitation module is 0.0001°.

[0079] Step S1: The high-frequency multi-channel excitation module can send two single-ended sinusoidal signals of the same frequency and phase with a high frequency and extremely low phase offset rate. The first sinusoidal signal enters step S2, and the second sinusoidal signal is recorded as a reference signal.

[0080] S2: Build a power amplifier module with an integrated low-pass filter so that the amplified sinusoidal current is large enough to excite the coil. Specifically, it includes:

[0081] A passive low-pass filter is built, and a 9th-order 200MHz Butterworth low-pass filter is built after the direct digital synthesizer (DDS) system. This is mainly to remove high-frequency aliasing and noise and improve the spectral purity of the signal. The signal output by the DDS may contain unwanted high-frequency components and harmonics, which can cause spectral distortion. The 9th-order 200MHz Butterworth low-pass filter effectively filters out these high-frequency components and improves signal quality through its smooth frequency response (i.e., the amplitude is stable within the passband and rapidly decays after the cutoff frequency). Its transfer function is shown below:

[0082]

[0083] Among them, ω c =2πf c is the angular frequency of the cutoff frequency, f c is the cutoff frequency, n is the filter order which is 9, s is the complex frequency of the input signal entering the low-pass filter, and H(s) is the transfer function, which is the core of the filter and defines how the input signal changes in the frequency domain after passing through the filter.

[0084] A two-stage power amplifier is constructed, with the AD8138 used in the first stage and the AD8131 in the second. The AD8138 first converts the single-ended sinusoidal signal generated by the DDS into a differential signal and amplifies it with a gain of 10. The AD8131 then further amplifies the differential signal with a gain of 2, maintaining the differential signal's characteristics while increasing the output amplitude. This combination of two amplifier stages effectively amplifies the input signal by 20 times, ensuring the amplified sinusoidal current is high enough to excite the coil while maintaining high accuracy and low distortion. The AD8138's gain is shown in Equation 5, and the AD8131's differential-mode gain is shown in Equation 6.

[0085]

[0086] Among them, G N is the gain, R F1 and R G1 are the feedback resistor and input resistor in the AD8138 peripheral circuit.

[0087]

[0088] Among them, V OUT and V IN are the differential mode output voltage and single-ended input voltage of AD8138, R F2 and R G2 are the feedback resistor and input resistor in the AD8131 peripheral circuit.

[0089] In step S2, the first sine wave signal in step S1 is passed through a low-pass filter to remove high-frequency aliasing and noise, and then enters a two-stage power amplifier to convert the sine wave signal into a differential signal and amplify the signal.

[0090] S3: Simulate and manufacture the coil module, and design a specific 3D printing device to fix the receiving coil and the receiving coil to improve system stability. Specifically, it includes:

[0091] Simulate and manufacture the coil module to ensure that the simulated coil inductance can first meet the system requirements. The relevant parameters of the receiving coil and the transmitting coil are set using the following formula:

[0092]

[0093] Where L is the inductance of the coil, r is the average radius of the coil, N is the number of turns in the coil, l is the length of the coil, and d is the thickness of the coil.

[0094] To determine the actual inductance of the completed receiving coil, the differential input coil must first be looped to create a single-ended input coil (this loop connection converts the differential input of the receiving coil into a single-ended input coil). A network analyzer is then used to measure the inductance and impedance at different frequency domains. Both the receiving and transmitting coils are PCBs, and coil radiation is maximized when placed perpendicular to the ground. To improve system stability, a specific 3D-printed device is used to secure the receiving and transmitting coils.

[0095] The differential signal amplified in step S2 is input to the transmitting coil in S3. When the transmitting coil is energized, a magnetic field signal is generated. Under the action of the magnetic field signal, the receiving coil can achieve the effect of magnetoelectricity and output a sinusoidal wave signal of voltage.

[0096] S4: Build a differential amplifier module to input the differential amplification signal generated by the detection coil into the subsequent module, namely the low-power phase detection module. Specifically, it includes:

[0097] A differential amplifier module uses a differential detection method that effectively avoids common-mode signal interference, reduces ground noise and system noise, and improves the reliability and anti-interference capability of weak signal detection in the magnetic induction coil. A differential amplifier chip with a very high common-mode rejection ratio is selected to convert the S3 differential signal into a single-ended signal, facilitating subsequent processing.

[0098] S5: Prepare a series of NaCl aqueous solutions with different mass concentrations to obtain a series of salt solutions with different conductivities (use a conductivity meter to measure the conductivity of the salt solutions to obtain actual conductivity data of the salt solutions). The salt solutions are placed in the system field area of ​​the coil module in step S3, that is, between the transmitting coil and the receiving coil. According to the operations in steps S1-S4, a series of modulated signals of different frequency bands under the interference of each conductive salt solution are collected by frequency sweeping, as well as the corresponding reference signals, to obtain the corresponding data sets;

[0099] When acquiring the modulation signal in step S5, two modulation signals are acquired in the same frequency band. The first is an empty field modulation signal when no saline solution is placed in the system physical field region, and the second is a conductivity interference modulation signal when a saline solution is placed in the system physical field region.

[0100] S6: Build a low-power phase detection module. Use MUX to design a low-power phase detection module that combines analog and digital phase detection. Specifically, it includes:

[0101] A MUX selector with a fan-out of 2 is built and placed before the analog and digital phase detectors. The MUX selector connects the analog and digital phase detectors via circuitry. Depending on the detection domain, the MUX selector selects either analog or digital phase detection to estimate the phase difference. Digital phase detection is used in domains with high detection accuracy, while analog phase detection is used in domains with low detection accuracy. The digital phase detector calculates the phase difference between the two signals by executing a digital signal processing algorithm. This algorithm typically involves calculating the Fourier transform of the two signals to extract phase information, or using time-domain cross-correlation methods to estimate the phase difference. In this way, the digital phase detector module can accurately measure the phase difference of high-speed signals and use the result for further signal processing or analysis.

[0102] S7: Build an LSTM (Long Short-Term Memory) regression prediction model based on the improved distance correlation coefficient, including:

[0103] S7.1: A series of modulation signals of different frequency bands under the interference of each conductive salt solution and the corresponding reference signals are collected by frequency sweeping. Under the interference of each conductive salt solution, 30 voltage signals of different frequency bands are scanned. The 30 frequency bands are from 5 MHz to 34 MHz, and phase difference data sets of 30 frequency bands are obtained. When collecting data under the interference of each conductive salt solution, data is collected once every 1 second in each frequency band. The total collection time is 256 seconds, that is, a total of 256 data collection times. Therefore, the phase difference data set of each frequency band consists of 256 columns of phase difference data.

[0104] The following two groups of experimental data were collected in each frequency band: 1) experimental group, collecting the modulation signal when saline solution was placed in the system field area; 2) blank control group: no saline solution was placed in the system field area, and the empty field modulation signal was collected.

[0105] The phase difference is the phase difference between the phase of the conductivity interference modulation signal minus the phase of the empty field modulation signal and the phase of the reference signal.

[0106] Since the system scans the phase difference data of 30 frequency bands, in order to obtain the frequency band data that can best represent the highest correlation between conductivity and phase difference data, it is necessary to analyze the distance correlation coefficient for each frequency band separately. The correlation calculation formula is as follows:

[0107]

[0108] Where dCor(X,Y) is the distance correlation coefficient, dCov(X,Y) is the distance covariance, dVar(X) and dVar(Y) are the distance variances of X and Y. X is a column of phase difference data in the 256 columns of data under the corresponding frequency band, and Y is the different conductivities corresponding to the 200 salt solutions.

[0109] Define the conductivity of the salt solution as Y = [y1 y2 … y 200 ] T , define the phase difference data matrix of salt solution conductivity in frequency band i as W = [w i,1 w i,2 … w i,200 ] T , where w i,1 It represents the phase difference data of the conductivity of salt solution No. 1 in frequency band i [Δσ i,1 Δσ i,2 … Δσ i,256 The phase difference data of the jth column of the 256 columns of data of the conductivity of 200 salt solutions in the corresponding frequency band is defined as X=[x 1,j x 2,j … x 200,j ] T .

[0110] The frequency band data with the highest calculated and statistical total correlation value is used as the target phase difference data, that is, the phase difference data corresponding to the frequency band with the largest distance correlation coefficient is calculated as the target phase difference data.

[0111] The acquired target phase difference data is defined as:

[0112]

[0113] The above phase difference data D200,256 As the data set input, the salt solution conductivity Y in S6 is used as the data set output. The training set and the test set are divided into 7:3, and 10 consecutive samples are regarded as an independent subset. The 1st, 5th, and 10th samples in each subset are selected as the test set. X =E(X),μ Y =E(Y), and the rest are used as training sets.

[0114] The network structure specifically includes a bidirectional LSTM (Bi-LSTM) layer to extract the phase difference features corresponding to the time series. After the extracted features are processed by the hidden layer, they are passed to the self-attention layer to capture global context information and generate attention weights. These weights are applied to the features, and finally the predicted conductivity results are output through the classification layer. This model combines bidirectional LSTM and self-attention mechanisms to improve the feature extraction and prediction capabilities of phase difference data. Figure 3 shown.

[0115] The low-power detection phase-lock module uses a multiplexer (MUX) adaptively controlled phase-lock circuit to address application scenarios with different conductivities and varying precision requirements. This module can significantly reduce the high power consumption of current digital phase-locked magnetic induction tomography equipment, while also addressing the limited application scenarios of current analog phase-locked magnetic induction tomography equipment. By using a low-power magnetic induction tomography measurement device, phase difference data of 200 salt solutions with different conductivities is collected within a specified timeframe. This data consists of data from 30 frequency bands, ranging from 5MHz to 34MHz, with each frequency band consisting of 256 values. This data is then used to construct a phase difference dataset.

[0116] S8: Based on the prediction model in S7, input the phase difference data of the corresponding target obtained by the detection device to obtain the corresponding salt solution conductivity, specifically including:

[0117] The training set in S7 is used to train the prediction model in S7, and the test set in S7 is used to test the performance of the model in S7. The coefficient of determination R between the predicted value and the actual value of the test set is calculated. 2 and RMSE are used as indicators to determine the effectiveness of the model.

[0118] The improved LSTM regression prediction model. The correlation between the phase difference and the conductivity is calculated to obtain the frequency band data with the highest correlation. Specifically, since each conductivity salt solution has corresponding 30 frequency bands of conductivity data, in order to obtain the frequency band data that can best represent the strongest correlation between the phase difference data and the conductivity. It is necessary to perform a distance correlation coefficient correlation analysis on each frequency band separately, calculate and count the frequency band data with the highest total correlation value as the target conductivity data, and divide the training set and test set into 7:3 ratios according to interval sampling. LSTM is used to construct a conductivity regression prediction model, and the phase difference data of the corresponding target obtained by the detection device is input as the test set to obtain the conductivity of the corresponding numbered salt solution. On this basis, the R between the actual value and the measured value is calculated. 2 and RMSE.

[0119] The relationship between the predicted phase difference and conductivity can be visualized by using the prediction model of the present invention, and a scatter plot between the predicted value and the actual value of the conductivity of the physical and chemical indicators can be obtained. Figure 2 .

[0120] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The scope of protection of the present invention should not be regarded as limited to the specific forms described in the embodiments. The scope of protection of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles, characterized by The following steps are involved: S1: Construct a high-frequency multi-channel excitation module with an extremely low phase offset rate. This module can send two single-ended sinusoidal voltage signals with the same frequency and phase at high frequency and extremely low phase offset rate. The first sinusoidal signal enters step S2, and the second sinusoidal signal is recorded as the reference signal. S2: Build a power amplifier module with an integrated low-pass filter. After the first sine wave signal in step S1 passes through the low-pass filter to remove high-frequency aliasing and noise, it enters the power amplifier module, converts the sine wave signal into a differential signal, and amplifies the signal. S3: Simulate and manufacture a coil module and a differential amplifier module. The coil module includes a receiving coil and a transmitting coil arranged in parallel and spaced apart. The differential signal amplified in step S2 is input to the transmitting coil. When the transmitting coil is energized, a magnetic field signal is generated. Under the action of the magnetic field signal, the receiving coil can achieve the effect of magnetoelectricity. The receiving coil outputs a sinusoidal voltage signal, which is further amplified by the differential amplifier module and converted into a single-ended signal to obtain a modulated signal. S4: Prepare a series of salt solutions with different conductivities. Place the salt solutions in the system field area of ​​the coil module in step S3, that is, between the transmitting coil and the receiving coil. Follow the operations in steps S1-S3 to collect a series of modulated signals of different frequency bands under the interference of each conductive salt solution by frequency sweeping, as well as the corresponding reference signals, to obtain the corresponding data set. S5: Build a low-power phase detection module. Use a multiplexer MUX to design a low-power phase detection module that combines analog phase detection and digital phase detection. Use analog phase detection or digital phase detection to calculate the phase difference between the two signals in the data set in step S4, and obtain a data set on the corresponding relationship between phase difference and conductivity. S6: Construct an LSTM regression prediction model based on the improved distance correlation coefficient, divide the data set obtained in step S5 into a training set and a test set, and use the constructed prediction model for training; when the conductivity of the salt solution to be tested is subsequently tested, the phase difference is first tested according to the method of steps S1-S5, and then input into the trained prediction model to obtain the corresponding salt solution conductivity.

2. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles as claimed in claim 1, characterized in that The high-frequency multi-channel excitation module with extremely low phase offset rate described in step S1 includes a main control chip, a direct digital frequency synthesizer (DDS), and a low-pass filter. The main control chip sends instructions for the frequency control word FTW, the phase offset word POW, and the amplitude scaling factor ASF to the DDS to set the required output parameters. The DDS updates the register values ​​to the internal buffer to generate an accurate output signal. The DDS uses the following three formulas to calculate the output frequency, phase offset, and amplitude of the output signal: Where FTW is the frequency control word, f out is the desired output frequency, f sys is the system clock frequency of the DDS; Where POW is the phase offset word and φ is the desired phase offset; ASF=r I ×2 10 (3) Where ASF is the amplitude scaling factor, r I It is the ratio of the required output current to the maximum output current.

3. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles as claimed in claim 1, characterized in that In step S1 , the phase shift rate of the sinusoidal wave signal sent by the high-frequency multi-channel excitation module is 0.001°-0.0001°.

4. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles as claimed in claim 1, characterized in that In step S2, the transfer function of the low-pass filter to remove high-frequency aliasing and noise is as follows: Among them, ω c =2πf c is the angular frequency of the cutoff frequency, f c is the cutoff frequency, n is the order of the filter, and s is the complex frequency of the input signal entering the low-pass filter; H(s) is the transfer function, which is the core of the filter and defines the changes in the frequency domain of the input signal after passing through the filter; The construction of the power amplifier module includes building a front and back two-stage power amplifier. The first stage power amplifier converts the single-ended sinusoidal wave signal generated in step S1 into a differential signal and performs gain amplification. The second stage power amplifier further amplifies the differential signal of the previous stage, maintaining the characteristics of the differential signal and increasing the output amplitude. Through the combination of these two stages of power amplifiers, the current of the amplified sinusoidal wave is large enough to excite the coil. The gain amplification function of the first-stage power amplifier is shown in Formula 5, and the differential-mode gain function of the second-stage power amplifier is shown in Formula 6: In formula 5, G N is the gain, R F1 and R G1 They are respectively the feedback resistor and input resistor in the peripheral circuit of the first-stage power amplifier; In formula 6, V OUT and V IN are the differential mode output voltage and single-ended input voltage of the first stage power amplifier, R F2 and R G2 They are respectively the feedback resistor and input resistor in the peripheral circuit of the second-stage power amplifier.

5. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles as claimed in claim 1, characterized in that The specific steps of step S3 include: simulating and manufacturing the coil module to ensure that the inductance of the coil simulation can first meet the system requirements, and setting the relevant parameters of the receiving coil and the transmitting coil according to the following formula: Where L is the inductance of the coil, r is the average radius of the coil, N is the number of turns in the coil, l is the length of the coil, and d is the thickness of the coil.

6. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles as claimed in claim 1, characterized in that When acquiring the modulation signal in step S4, two modulation signals are acquired in the same frequency band. The first is an empty field modulation signal when no saline solution is placed in the system physical field region, and the second is a conductivity interference modulation signal when a saline solution is placed in the system physical field region. The phase difference in step S5 is the difference between the phase of the conductivity interference modulation signal minus the phase of the empty field modulation signal and the phase of the reference signal.

7. The low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles according to claim 1, characterized in that Step S5: Building a low-power phase detection module specifically includes: S5.1: Build a MUX selector with a fan-out of 2 and place it in the front stage of the module to enable the control signal to be output by the main control chip; S5.2: The MUX selector connects the analog phase detector and the digital phase detector through a circuit. The MUX selector selects the analog phase detector or the digital phase detector to calculate the phase difference according to the detection field. The digital phase detector is selected in the field with high detection accuracy, and the analog phase detector is selected in the field with low detection accuracy.

8. The low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles according to claim 1, characterized in that In step S4, a series of salt solutions with different conductivities are prepared, which is a series of salt aqueous solutions with different concentrations, wherein the salt is sodium chloride.

9. A low-power magnetic induction tomography target disturbance analysis method based on bio-electromagnetic principles as claimed in claim 1, characterized in that The LSTM regression prediction model improved by the distance correlation coefficient includes the following contents: S6.1: Collect a series of modulated signals at different frequency bands under the interference of each conductive salt solution and the corresponding reference signal by frequency sweeping. Under each conductive salt solution interference, scan the voltage signals at i different frequency bands to obtain 30 frequency band phase difference data sets. Data is collected once every 1 second for each frequency band, for a total collection time of b seconds. Thus, each frequency band phase difference data set consists of b columns of phase difference data. In order to obtain the frequency band data that can best represent the highest correlation between conductivity and phase difference data, it is necessary to analyze the distance correlation coefficient for each frequency band separately. The correlation calculation formula is as follows: Where dCor(X,Y) is the distance correlation coefficient, dCov(X,Y) is the distance covariance, dVar(X) and dVar(Y) are the distance variances of X and Y; X is a column of phase difference data in b columns of data under the corresponding frequency band, and Y is the different conductivities corresponding to a salt solutions; a and b are both positive integers between 10 and 1000. The data set for the conductivity of the salt solution is defined as Y = [y1 y2…y a ] T , define the phase difference data matrix of salt solution conductivity in frequency band i as W = [w i,1 w i,2 …w i,a ] T , where w i,1 It represents the phase difference data of the conductivity of salt solution No. 1 in frequency band i [Δσ i,1 Δσ i,2 …Δσ i,b ]; define the phase difference data of column b and column j of the conductivity of a salt solution in the corresponding frequency band as X=[x 1,j x 2,j … x a,j ] T ; S6.2: The phase difference data corresponding to the frequency band with the highest total correlation value calculated and counted according to S6.1, i.e., the largest distance correlation coefficient, is used as the target phase difference data; S6.3: Define the target phase difference data matrix obtained in S6.2 as: The above phase difference data D a，b As the data set input, the conductivity Y of the salt solution in S6 is output as the data set, μ X =E(X),μ Y =E(Y), and use LSTM to build a conductivity regression prediction model for training.

Images