A method for quality detection of GABA extract in peanut sprouts

Abstract

This application relates to the field of quality testing technology and provides a method for quality testing of GABA extract in peanut sprouts. It utilizes a molecularly imprinted sensor for specific identification of GABA, combined with a pre-set peanut sprout interference database and partial least squares regression to calculate impurity contribution. This method can specifically eliminate interference from complex matrix impurities in peanut sprout extracts, solving the problem of insufficient specificity in traditional methods for complex samples. The molecularly imprinted sensor is connected in series with a quartz crystal microbalance via a microfluidic channel. Capacitance changes trigger resonant frequency detection, creating a coherent process from identification to quality analysis and improving detection efficiency. By inputting corrected frequency offset, phase angle, and temperature into a recurrent neural network, the method integrates multi-dimensional detection information to achieve accurate output of GABA concentration. Compared to single-parameter detection methods, this method more comprehensively reflects the true content of GABA in the extract, providing a reliable basis for quality judgment.

Description

Technical Field

[0001] This application relates to the field of quality testing technology, and in particular to a method for quality testing of GABA extract in peanut sprouts. Background Technology

[0002] In the field of GABA detection, with the advancement of technology, detection methods are constantly being iterated and updated. Early colorimetric methods, enzyme-linked immunosorbent assay (ELISA) methods, and thin-layer chromatography (TLC) methods had low sensitivity and could not meet the requirements for accurate detection. Later, amino acid analyzer methods and ion chromatography methods were developed, but these methods have great limitations on the types of samples, strict analytical conditions, and limited application scenarios.

[0003] Currently, mainstream detection methods can be divided into three main categories: chromatography, spectroscopy, and biosensors. Among them, high-performance liquid chromatography (HPLC) has become the gold standard in the industry due to its high separation efficiency. It typically uses a C18 reversed-phase column with methanol-water or acetonitrile-water as the mobile phase, combined with ultraviolet or fluorescence detectors. After derivatization with phthalaldehyde, the detection sensitivity is improved. The emerging liquid chromatography-mass spectrometry (LC-MS) technology shows stronger specificity. Through multiple reaction monitoring (MRM) mode, it can effectively distinguish between GABA and its isomers.

[0004] Despite some progress in existing technologies, numerous problems remain. For example, high-performance liquid chromatography (HPLC) requires sample derivatization, which uses toxic reagents and involves cumbersome experimental procedures. Furthermore, different detection standards exhibit significant differences. Balancing rapid detection with accuracy is challenging; for instance, while a portable GABA detector can produce results within 5 minutes, its correlation coefficient with HPLC-MS is low. Moreover, there is a lack of a unified method for detecting the release rate of GABA in novel sustained-release formulations, and existing dialysis bag methods struggle to simulate the actual in vivo release environment. Most existing technologies have not yet addressed how to eliminate interference from complex impurities during the detection process to achieve quality detection of GABA extracts from peanut sprouts. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a method for quality detection of GABA extract in peanut sprouts. The method includes: preprocessing a molecularly imprinted sensor, filtering the GABA extract in peanut sprouts through a filter membrane and injecting it into the molecularly imprinted sensor, applying an AC perturbation signal, and monitoring and recording the capacitance change of the molecularly imprinted sensor in real time.

[0006] The molecular imprint sensor is connected in series with the quartz crystal microbalance through a microfluidic channel. When the recorded capacitance change is greater than the preset change threshold, the resonant frequency detection of the quartz crystal microbalance is triggered based on the pre-calibrated linear relationship between capacitance change and mass change, and the original frequency offset of the crystal is recorded.

[0007] The electrochemical impedance spectroscopy of the extract was obtained and the phase angle was extracted. A preset peanut sprout interference database was called and the contribution of impurities was calculated by partial least squares regression to obtain the corrected frequency offset. The frequency offset, phase angle and temperature were then input into a recurrent neural network to output the concentration of GABA to determine the quality of the GABA extract in peanut sprouts.

[0008] As an optional implementation, the capacitance change of the recorded molecular imprint sensor specifically includes:

[0009] The molecularly imprinted sensor is pretreated, including eluting template molecules and activation treatment, to form a detection cell for recognizing GABA. GABA extract from peanut sprouts is filtered through a filter membrane and then injected into the detection cell of the molecularly imprinted sensor.

[0010] An AC disturbance signal is applied, covering multiple frequency bands, to monitor and extract capacitance signals in different frequency bands in real time. The capacitance signal is then decomposed by a fast Fourier transform to extract the fundamental frequency capacitance value and the high frequency capacitance value to determine the capacitance baseline value. The difference between the real-time capacitance value and the capacitance baseline value is calculated to obtain the instantaneous capacitance change.

[0011] When the fluctuation amplitude of the instantaneous capacitance change within multiple consecutive detection cycles is less than the set amplitude threshold, the capacitance signal is determined to have reached a stable state, and the instantaneous capacitance change in the stable state is recorded as the capacitance change of the molecular imprint sensor.

[0012] As an optional implementation, the step of connecting the molecularly imprinted sensor and the quartz crystal microbalance in series via a microfluidic channel specifically includes:

[0013] Prepare a tubular microfluidic channel, connect the detection cell of the molecularly imprinted sensor to the input end of the microfluidic channel, so that the detection cell is inside the microfluidic channel and perpendicular to the axis of the microfluidic channel;

[0014] The detection area of ​​the quartz crystal microbalance is connected to the output end of the microfluidic channel, so that the detection area is inside the microfluidic channel and parallel to the detection cell of the molecular imprint sensor.

[0015] Adjust the positions of the molecular imprint sensor and the quartz crystal microbalance in the microfluidic channel so that the extract flows out of the detection cell and into the detection area of ​​the quartz crystal microbalance within a set time.

[0016] As an optional implementation, the recording logic for the original frequency offset includes:

[0017] After receiving the detection start signal, the quartz crystal microbalance acquires the resonance frequency and uses the resonance frequency at the initial time as the reference frequency.

[0018] Calculate the difference between the resonant frequency and the reference frequency each time to obtain the instantaneous frequency offset;

[0019] Continuously record the instantaneous frequency offset until the difference between the instantaneous frequency offsets of a preset number of consecutive times is less than the set fluctuation threshold, and record it as the original frequency offset of the crystal.

[0020] As an optional implementation, the triggering sub-logic for resonant frequency detection includes:

[0021] By repeatedly detecting the capacitance change of blank peanut sprout extract with a preset change threshold, and performing linear fitting with capacitance change as the x-axis and mass change as the y-axis, a pre-calibrated linear relationship between capacitance change and mass change was obtained.

[0022] The recorded capacitance change is compared with a preset change threshold. When the recorded capacitance change is greater than the preset change threshold, the mass change value is determined based on the pre-calibrated linear relationship between capacitance change and mass change.

[0023] A preset sensitivity threshold for the quartz crystal microbalance is established. When the determined mass change value is greater than or equal to the sensitivity threshold, a detection start signal is sent to the quartz crystal microbalance to trigger the resonance frequency detection.

[0024] As an optional implementation, the corrected frequency offset specifically includes:

[0025] Temperature compensation is applied to the contribution of impurities based on the temperature during the detection process.

[0026] Based on the linear relationship between capacitance change and mass change, the influence weight of capacitance change on impurity contribution is determined.

[0027] The corrected frequency offset is obtained by subtracting the impurity contribution after temperature compensation and influence weighting from the original frequency offset.

[0028] As an optional implementation, the phase angle extraction sub-logic includes:

[0029] The electrochemical workstation was started to apply a frequency sweep excitation signal to the extract, and the electrochemical impedance spectrum of the extract was obtained. The electrochemical impedance spectrum was processed to obtain the phase angle at different frequency points.

[0030] Characteristic frequency points were selected based on the characteristic changes in the electrochemical impedance spectroscopy, and the phase angle of the characteristic frequency points was extracted.

[0031] The phase angle at the characteristic frequency point is sampled multiple times and the average value is taken as the final phase angle.

[0032] As an optional implementation, the sub-logic for calculating the impurity contribution includes:

[0033] Call the preset peanut sprout interference database, which includes the contribution coefficients of the phase angle and frequency shift of the electrochemical impedance spectrum of impurities;

[0034] The extracted phase angles were matched with the phase angles of different impurities in the peanut sprout interference database to screen out interfering impurities.

[0035] Using the screened interfering impurities as variables and the original frequency shift of the crystal as the dependent variable, the frequency contribution weights of different interfering impurities are determined through cross-validation.

[0036] The contribution of each interfering impurity to the original frequency offset is calculated by partial least squares regression based on the frequency contribution weight, and the impurity contribution is obtained by summing them up.

[0037] As an optional implementation, determining the quality of the GABA extract in peanut sprouts specifically includes:

[0038] Set the acceptable threshold for GABA concentration and the deviation range for test results;

[0039] The absolute value of the output GABA concentration is compared with the qualified threshold, and the relative standard deviation of the GABA concentration of three consecutive tests is determined to be within the deviation range.

[0040] Based on the comparison and judgment results, determine whether the quality of the GABA extract in peanut sprouts is up to standard. If it is not up to standard, output the interfering impurities and analysis report.

[0041] As an optional implementation, the output sub-logic for the GABA concentration includes:

[0042] Construct a recurrent neural network in which the input layer includes the corrected frequency offset, phase angle, and temperature;

[0043] A recurrent neural network was trained using standard sample data containing different concentrations of GABA and impurity combinations, and the parameters of the recurrent neural network were optimized using a time backpropagation algorithm.

[0044] The corrected frequency offset, phase angle, and temperature obtained from real-time detection are input into the trained recurrent neural network, which outputs the concentration of GABA and simultaneously calculates the confidence level of the recurrent neural network output.

[0045] Compared with existing technologies, the beneficial effects of this application are as follows: By using a molecularly imprinted sensor to specifically identify GABA, combined with a pre-set peanut sprout interference database and partial least squares regression to calculate the contribution of impurities, interference from complex matrix impurities in peanut sprout extracts can be specifically eliminated, solving the problem of insufficient specificity in the detection of complex samples by traditional methods; By connecting the molecularly imprinted sensor and a quartz crystal microbalance in series through a microfluidic channel, and using capacitance changes to trigger resonant frequency detection, a coherent process from identification to quality analysis is formed in the sample detection, avoiding errors caused by sample transfer in traditional step-by-step detection and improving detection efficiency; By inputting the corrected frequency offset, phase angle, and temperature into a recurrent neural network, the accurate output of GABA concentration is achieved by integrating multi-dimensional detection information. Compared with single-parameter detection methods, this can more comprehensively reflect the true content of GABA in the extract, providing a reliable basis for quality judgment; The detection logic is designed specifically for the characteristics of GABA extracts in peanut sprouts, and the entire process from sample pretreatment to quality judgment is consistent with its matrix characteristics, solving the problem of insufficient adaptability of general detection methods in the detection of specific samples. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0047] Figure 1 This is a system flowchart of a method for quality detection of GABA extract in peanut sprouts provided in an embodiment of this application;

[0048] Figure 2 The trigger logic diagram for the resonance frequency detection of a method for quality detection of GABA extract in peanut sprouts provided in this application embodiment;

[0049] Figure 3 A sub-logic diagram for calculating the impurity contribution of a quality detection method for GABA extract in peanut sprouts provided in this application embodiment;

[0050] Figure 4 This is a logic diagram for judging the quality of GABA extract in peanut sprouts, provided in an embodiment of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of this application more apparent and understandable, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0052] like Figure 1 The diagram shown is a flowchart of a method for quality detection of GABA extract in peanut sprouts, provided in an embodiment of this application. The method includes:

[0053] S1. Preprocess the molecularly imprinted sensor by filtering the aminobutyric acid extract from peanut sprouts through a filter membrane and injecting it into the molecularly imprinted sensor. Apply an AC perturbation signal and monitor and record the capacitance change of the molecularly imprinted sensor in real time.

[0054] The specific changes in capacitance recorded by the molecularly imprinted sensor include:

[0055] The molecularly imprinted sensor is pretreated, including eluting template molecules and activation treatment, to form a detection cell for recognizing GABA. GABA extract from peanut sprouts is filtered through a filter membrane and then injected into the detection cell of the molecularly imprinted sensor.

[0056] An AC disturbance signal is applied, covering multiple frequency bands, to monitor and extract capacitance signals in different frequency bands in real time. The capacitance signal is then decomposed by a fast Fourier transform to extract the fundamental frequency capacitance value and the high frequency capacitance value to determine the capacitance baseline value. The difference between the real-time capacitance value and the capacitance baseline value is calculated to obtain the instantaneous capacitance change.

[0057] When the fluctuation amplitude of the instantaneous capacitance change within multiple consecutive detection cycles is less than the set amplitude threshold, the capacitance signal is determined to have reached a stable state, and the instantaneous capacitance change in the stable state is recorded as the capacitance change of the molecular imprint sensor.

[0058] During the fabrication of molecularly imprinted sensors, template molecules occupy specific recognition sites on the polymer membrane. If these sites are not removed, they will directly affect the binding efficiency of para-aminobutyric acid (GABA). The presence of unactivated functional groups on the surface of newly fabricated sensors will reduce the sensitivity of the detection signal. Peanut sprout extract contains impurities such as plant tissue debris and macromolecular polymers, which can adhere to the sensor surface or block the flow channels, interfering with the stability of capacitive detection.

[0059] A molecularly imprinted polymer membrane modified with glutamic acid as a template was selected as the main body of the sensor. The sensor was placed in a mixture of ethanol and acetic acid, and the template molecules were eluted by isothermal oscillation with the oscillation speed kept stable. After elution, the electrode surface was repeatedly rinsed with ultrapure water. Then, the sensor was placed in a buffer solution and connected to an electrochemical workstation. Activation was performed by applying a constant potential. During the activation process, the change in current on the electrode surface was continuously monitored until the current stabilized, indicating that activation was complete and a detection cell that could specifically recognize GABA was formed. GABA extract from peanut sprouts was taken and filtered under pressure using an organic filter membrane. During the filtration process, the residue on the filter membrane surface was observed. After ensuring that there were no obvious impurities, the filtrate was injected into the detection cell through a micro-syringe. The injection speed was controlled to ensure that the liquid flowed in slowly and to avoid the generation of bubbles.

[0060] The elution process thoroughly removes residual template molecules, fully exposing the sensor's specific recognition sites and enhancing its selective binding ability to GABA. The activation process activates functional groups on the sensor surface, improving electrochemical response activity and making it easier to capture subtle capacitance changes. The membrane filtration effectively removes solid impurities and macromolecular interferences from the extract, preventing their non-specific adsorption on the sensor surface and ensuring the cleanliness of the detection environment.

[0061] The binding of GABA on the sensor surface causes changes in the capacitance characteristics of the interface at different depths. A single-frequency perturbation signal can only reflect local capacitance changes and cannot fully characterize the binding process. The capacitance signal contains multiple frequency components, among which the fundamental frequency signal reflects the capacitance characteristics of the sensor body, and the high-frequency signal reflects the capacitance changes of the interface double layer. By decomposing the signal, the initial capacitance state can be accurately determined. The accuracy of the baseline value affects the calculation accuracy of the capacitance change. The capacitance characteristics of different frequency bands are combined to determine the capacitance state.

[0062] The frequency range of the AC perturbation signal is set using an electrochemical workstation, covering multiple frequency bands from low to high frequency. The AC perturbation signal maintains stable amplitude during application. The capacitance signal of the molecularly imprinted sensor is monitored and extracted in real time. The extracted complex capacitance signal is decomposed using a fast Fourier transform to separate the capacitance components corresponding to different frequencies. From these components, the fundamental frequency capacitance value that reflects the sensor characteristics and the high-frequency capacitance value that reflects the interface state are selected. The two capacitance values ​​are weighted and averaged to determine the capacitance baseline value in the initial detection stage. During the binding of GABA with the sensor, real-time capacitance values ​​in different frequency bands are continuously acquired. By comparing with the capacitance baseline value, the instantaneous capacitance change at each time point is calculated.

[0063] Multi-frequency AC disturbance signals can fully excite the sensor's capacitive response, capturing the capacitance changes at various levels caused by GABA binding, avoiding the information limitations of single-frequency signals. Fast Fourier transform enables effective decomposition of complex signals, making the extraction of fundamental and high-frequency capacitance values ​​more accurate. The capacitance baseline value determined based on this can truly reflect the initial detection state. Real-time calculation of instantaneous capacitance changes dynamically presents the binding process of GABA and the sensor, providing data support for judging whether the binding has reached equilibrium.

[0064] The specific binding of GABA to the molecularly imprinted sensor is a dynamic equilibrium process. In the initial stage, the binding rate is rapid, and the capacitance changes drastically. At this stage, the capacitance change cannot represent the true binding amount. Only when the binding reaches equilibrium does the capacitance change tend to stabilize, and only then can the capacitance change accurately reflect the concentration of GABA. Therefore, it is necessary to set a stability judgment condition. The duration of each detection cycle is set, and a reasonable cycle interval is determined based on the rate of change of the capacitance signal. An amplitude threshold is also set, which is determined based on the sensor's inherent noise level and detection accuracy requirements. The instantaneous capacitance change is monitored in real time, and the change amplitude is continuously statistically analyzed over multiple detection cycles. When the fluctuation amplitude of the instantaneous capacitance change in all detection cycles is less than the set amplitude threshold, the capacitance signal is determined to have reached a stable state. At this point, the instantaneous capacitance change in the stable state is selected as the final recorded capacitance change, and the capacitance change value, along with the corresponding detection time and environmental conditions, is stored.

[0065] By monitoring fluctuations over multiple consecutive cycles, misjudgments caused by single-time signal fluctuations are avoided, ensuring more reliable determination of the stable state. The capacitance change recorded in the stable state can accurately reflect the binding equilibrium state of GABA and the sensor, reducing error interference in the dynamic process and improving the accuracy of the detection results. The stable and reliable capacitance change provides an accurate basis for subsequent comparison with the preset threshold, ensuring more precise timing for triggering the resonance frequency detection of the quartz crystal microbalance, and ensuring more reliable calculation of subsequent mass change.

[0066] S2. Connect the molecular imprint sensor and the quartz crystal microbalance in series through a microfluidic channel. When the recorded capacitance change is greater than the preset change threshold, the resonant frequency detection of the quartz crystal microbalance is triggered based on the pre-calibrated linear relationship between capacitance change and mass change, and the original frequency offset of the crystal is recorded.

[0067] Connecting a molecularly imprinted sensor to a quartz crystal microbalance via a microfluidic channel specifically includes:

[0068] Prepare a tubular microfluidic channel, connect the detection cell of the molecularly imprinted sensor to the input end of the microfluidic channel, so that the detection cell is inside the microfluidic channel and perpendicular to the axis of the microfluidic channel;

[0069] The detection area of ​​the quartz crystal microbalance is connected to the output end of the microfluidic channel, so that the detection area is inside the microfluidic channel and parallel to the detection cell of the molecular imprint sensor.

[0070] Adjust the positions of the molecular imprint sensor and the quartz crystal microbalance in the microfluidic channel so that the extract flows out of the detection cell and into the detection area of ​​the quartz crystal microbalance within a set time.

[0071] The tubular structure constrains the fluid flow direction, preventing turbulence or dead zones during extract transport. The perpendicularity of the detection cell to the channel axis ensures the extract impacts the sensing surface at a vertical angle, increasing the probability of molecular collisions and enhancing specific binding. A PTFE tube with a consistent inner diameter is selected as the microfluidic channel. The inner wall is mirror-polished using precision grinding equipment. A mounting hole matching the molecular imprint sensor detection cell is created at the input end of the microfluidic channel, with a corrosion-resistant rubber sealing ring embedded in the hole wall. After embedding the detection cell, it is fixed with a flange. The detection cell is rotated to ensure the sensing surface is perpendicular to the channel axis. Calibration with a laser goniometer ensures the perpendicularity meets requirements. After installation, a colored solution is injected into the channel, and the area around the detection cell is observed for leaks or vortices to ensure a sealed installation and stable flow. The mirror-polished inner wall reduces the fluid boundary layer thickness, minimizing the adsorption and deposition of macromolecules from the extract on the wall surface. The vertically installed detection cell allows GABA molecules to bind to the recognition site from multiple directions, improving binding efficiency. Strict sealing and perpendicularity control prevent signal distortion due to installation defects.

[0072] The output position ensures that the extract directly enters the quality detection stage after being processed by the molecular imprinted sensor, reducing intermediate transmission losses. The parallelism between the detection area and the detection cell ensures consistent fluid pressure and shear force, avoiding detection deviations caused by force differences. Symmetrical mounting holes of the same specifications as the input end are opened at the output end of the microfluidic channel. The detection crystal of the quartz crystal microbalance is embedded in the hole, and the crystal angle is adjusted by a fine-tuning bracket to keep it parallel to the detection cell of the molecular imprinted sensor. Parallelism is verified using a parallelism tester. The surface of the detection crystal and the inner wall of the channel are seamlessly connected using laser welding technology, forming a continuous and smooth inner wall, eliminating flow disturbance points. The seamless channel design prevents the extract from forming eddies in front of the detection area, ensuring uniform sample flow across the crystal surface. The parallel arrangement ensures that capacitance detection and quality detection are in the same flow field environment. The flush alignment of the electrode surface with the channel reduces interference from non-specific adsorption on the frequency signal.

[0073] Setting the flow time ensures that unbound GABA molecules in the extract enter the quality detection area before diffusion, avoiding concentration gradients caused by diffusion, and providing sufficient stabilization time for the capacitive response of the molecularly imprinted sensor. Setting the flow time of the extract from the detection cell to the detection area to an appropriate multiple of the sensor response time is achieved by adjusting the distance between the two by moving the mounting bracket and calibrating the flow rate using a flow meter to ensure the extract flows through the channel at a constant rate. High-speed imaging technology is used to record the trajectory of tracer particles, verifying that the flow time meets the set requirements. A reasonable spacing design balances response time and diffusion effects, ensuring specific binding by the molecularly imprinted sensor while preventing excessive diffusion of GABA during transport. Constant flow rate and precise timing ensure consistent testing conditions for different batches. Controllable flow time allows for precise matching of capacitance and mass changes in the time dimension, improving the prediction accuracy of linear relationships and providing a time reference for accurate determination of trigger thresholds.

[0074] Furthermore, such as Figure 2 As shown, the triggering sub-logic for resonant frequency detection includes:

[0075] By repeatedly detecting the capacitance change of blank peanut sprout extract with a preset change threshold, and performing linear fitting with capacitance change as the x-axis and mass change as the y-axis, a pre-calibrated linear relationship between capacitance change and mass change was obtained.

[0076] The recorded capacitance change is compared with a preset change threshold. When the recorded capacitance change is greater than the preset change threshold, the mass change value is determined based on the pre-calibrated linear relationship between capacitance change and mass change.

[0077] A preset sensitivity threshold for the quartz crystal microbalance is established. When the determined mass change value is greater than or equal to the sensitivity threshold, a detection start signal is sent to the quartz crystal microbalance to trigger the resonance frequency detection.

[0078] The capacitance change of blank peanut sprout extract contains matrix interference signals. Setting a threshold based on this can effectively distinguish specific signals from interference signals. The pre-calibrated linear relationship needs to cover the detection concentration range to ensure the accuracy of conversion at different concentrations. Multiple batches of blank peanut sprout extract from different sources were selected, and the capacitance change of each batch was repeatedly measured multiple times. The process capability index was calculated using statistical process control methods, and the value corresponding to an appropriate multiple standard deviation was set as the change threshold. Multiple concentration gradients of GABA standard solutions covering the detection range were prepared, and each concentration was repeatedly measured multiple times. The capacitance change and the mass change value detected by the quartz crystal microbalance were recorded simultaneously. Linear fitting was performed using the weighted least squares method, with high concentration points assigned high weights. The fitting curve was obtained, and the goodness of fit was verified by residual analysis to obtain the pre-calibrated linear relationship between capacitance change and mass change. The threshold set by the statistical method is statistically significant, reducing the probability of false triggering. The weighted fitting takes into account the difference in detection accuracy at different concentrations, improving the conversion accuracy in the high concentration region. Residual analysis ensures the reliability of the linear relationship.

[0079] Threshold comparison is a crucial step in distinguishing effective signals from noise. Only signals exceeding the threshold are worth further processing. Through linear relationship conversion, capacitance signals can be quantized into mass signals, providing a clear trigger condition for the quartz crystal microbalance's detection. The recorded capacitance change is dynamically compared with a preset threshold. When the capacitance change exceeds the threshold for multiple detection cycles, the linear relationship is invoked. Interpolation is used to find the corresponding mass change value in the pre-calibrated linear relationship fitting curve. Dynamic comparison and the judgment of continuous exceedance of the threshold avoid false triggering caused by instantaneous pulse interference. Interpolation improves the calculation accuracy of the mass change value. The accurate mass change value provides a direct basis for the sensitivity threshold judgment of the quartz crystal microbalance, ensuring that the detection start timing matches the sample concentration.

[0080] Quartz crystal microbalances have a lower sensitivity limit; mass changes below this limit cannot be accurately detected. Setting a sensitivity threshold avoids invalid detection operations, protects the equipment, and improves efficiency. By pre-setting the sensitivity threshold of the quartz crystal microbalance based on the actual detection environment, the determined mass change value is compared with this threshold. When the mass change value is greater than or equal to the sensitivity threshold, a detection start signal is sent to the quartz crystal microbalance via an opto-isolation circuit to avoid electrical interference and trigger the quartz crystal microbalance to enter resonant frequency detection. The environmentally corrected sensitivity threshold more closely matches the actual detection conditions, improving the accuracy of trigger judgment. The opto-isolation circuit avoids electromagnetic interference between different devices, and the resonant frequency detection trigger ensures the capture of instantaneous frequency changes. Precise trigger control allows the quartz crystal microbalance to initiate resonant frequency detection at the optimal time, providing high-time-resolution data support for recording the original frequency offset.

[0081] Specifically, the recording logic for the original frequency offset includes:

[0082] After receiving the detection start signal, the quartz crystal microbalance acquires the resonance frequency and uses the resonance frequency at the initial time as the reference frequency.

[0083] Calculate the difference between the resonant frequency and the reference frequency each time to obtain the instantaneous frequency offset;

[0084] Continuously record the instantaneous frequency offset until the difference between the instantaneous frequency offsets of a preset number of consecutive times is less than the set fluctuation threshold, and record it as the original frequency offset of the crystal.

[0085] The reference frequency serves as the zero point for calculating the frequency offset. Using the start signal reception time as the time origin ensures that the starting point of the offset calculation is synchronized with the detection start, avoiding errors caused by time differences. Upon receiving the detection start signal, the quartz crystal microbalance immediately triggers its internal clock synchronization, marking the instant the detection start signal is received as the time zero point. Simultaneously, the resonant frequency at that time is acquired as the reference frequency, controlling the synchronization error between the detection start signal and the frequency acquisition. The time difference between the synchronization pulse and the frequency sampling pulse is monitored in real-time using an oscilloscope. High-precision synchronization ensures the time accuracy of the reference frequency, avoiding initial errors caused by delays. The extracted frequency values ​​exhibit high stability, reducing reference noise. The time zero point marking provides a unified reference for the time correlation of subsequent data.

[0086] Instantaneous frequency offset can reflect the dynamic changes in mass on the crystal surface in real time. Through continuous calculation, the adsorption process of GABA can be tracked, providing a basis for judging the adsorption equilibrium. The quartz crystal microbalance continuously acquires the resonant frequency at a certain sampling rate. After acquiring a certain number of data points, a moving average is performed to reduce high-frequency noise. Then, the difference with the reference frequency is calculated to obtain the instantaneous frequency offset. The instantaneous frequency offset is stored in time series and marked with the timestamp of each calculation. High-frequency sampling and moving average take into account the temporal resolution and signal-to-noise ratio of the data. The timestamp marking facilitates the tracking of the dynamic changes in the offset. The continuously recorded instantaneous frequency offset fully presents the kinetic characteristics of mass adsorption.

[0087] The adsorption process involves a dynamic equilibrium phase. Only when the instantaneous frequency offset is stable can its value truly reflect the quality of GABA. Continuous compliance judgment can avoid misjudging transient stability as equilibrium. A fluctuation threshold is set as an appropriate multiple of the minimum resolution of the quartz crystal microbalance. Based on the equipment's technical parameters, the number of consecutive compliance attempts is set, and the instantaneous frequency offset is monitored in real time. The difference between two adjacent data points is calculated. When the difference is less than the fluctuation threshold for multiple consecutive times, a stable state is determined. The offset at this point is recorded as the original frequency offset, and a complete data report containing sampling time, environmental parameters, and the basis for stability judgment is automatically generated. The threshold setting based on the equipment resolution ensures the scientific nature of the judgment standard. Continuous compliance judgment improves the reliability of stable state identification. The complete data report provides a basis for subsequent traceability and error analysis. Stable and reliable original frequency offset provides high-quality raw data for subsequent impurity deduction and concentration calculation, which is a key step in ensuring the final detection accuracy.

[0088] S3. Obtain the electrochemical impedance spectrum of the extract and extract the phase angle. Call the preset peanut sprout interference database and calculate the impurity contribution through partial least squares regression to obtain the corrected frequency offset. Input the frequency offset, phase angle and temperature into the recurrent neural network to output the GABA concentration to determine the quality of the GABA extract in peanut sprouts.

[0089] Furthermore, the phase angle extraction sub-logic includes:

[0090] The electrochemical workstation was started to apply a frequency sweep excitation signal to the extract, and the electrochemical impedance spectrum of the extract was obtained. The electrochemical impedance spectrum was processed to obtain the phase angle at different frequency points.

[0091] Characteristic frequency points were selected based on the characteristic changes in the electrochemical impedance spectroscopy, and the phase angle of the characteristic frequency points was extracted.

[0092] The phase angle at the characteristic frequency point is sampled multiple times and the average value is taken as the final phase angle.

[0093] The frequency sweep excitation signal can excite the electrochemical response of the extract at different frequencies. The electrochemical impedance spectroscopy contains rich interfacial reaction information, and the phase angle reflects the kinetic characteristics of the electrode process, providing a basis for subsequent impurity identification. The electrochemical workstation is started, and the frequency range of the frequency sweep excitation signal is set from low frequency to high frequency. The frequency points are divided into logarithmic intervals to cover the key response frequency band. The three electrodes are immersed in the extract after flowing through the detection area of ​​the quartz crystal microbalance, ensuring that the electrode spacing is fixed and completely submerged. The working electrode is a molecularly imprinted sensor modified electrode, the reference electrode is a saturated calomel electrode, and the counter electrode is a platinum sheet electrode. The amplitude of the sinusoidal frequency sweep signal is applied. After the signal is stabilized at each frequency point, multiple sets of voltage and current data are acquired. The average value is taken as the response signal at that frequency point. The voltage and current signals are converted into electrochemical impedance spectra by the complex impedance algorithm built into the electrochemical workstation. The electrochemical impedance spectra are smoothed by the moving average filtering method. By jointly analyzing the Nyquist plot and Bode plot, the real and imaginary parts of the complex impedance data are converted into the phase angles corresponding to each frequency point, that is, the arctangent value of the ratio of the imaginary part to the real part.

[0094] Wideband sweep frequency combined with logarithmic intervals ensures the capture of characteristic responses of different substances, especially diffusion control processes in the low-frequency band and charge transfer processes in the high-frequency band. Averaging and smoothing of multiple sets of data significantly reduces electronic noise and environmental interference. Three electrodes ensure the stability and repeatability of electrochemical responses. The finely processed phase angle retains the characteristic differences of impurities and GABA, providing high-resolution raw materials for subsequent screening of characteristic frequency points.

[0095] Different substances exhibit unique impedance change characteristics at specific frequencies. Screening characteristic frequency points can focus on key frequency bands that can distinguish GABA from impurities, reducing interference from irrelevant information. Electrochemical impedance spectroscopy is plotted, showing the impedance magnitude as a function of frequency on the amplitude-frequency curve and the phase angle as a function of frequency on the phase-frequency curve. Characteristic inflection points are identified using the first derivative method. On the amplitude-frequency curve, the two frequency points with the largest and smallest absolute values ​​of the first derivative of the impedance magnitude are marked as characteristic frequency points. On the phase-frequency curve, the frequency point corresponding to a phase angle of -45° is marked as a characteristic frequency point. For each characteristic frequency point, the corresponding phase angle is extracted from the processed impedance spectrum data, establishing a correlation dataset including frequency, phase angle, and acquisition time. The first derivative method improves the accuracy of characteristic inflection point identification, avoiding subjective judgment errors. Multi-curve joint screening ensures that characteristic frequency points possess both amplitude-frequency and phase-frequency characteristics, enhancing their correlation with the target substance. Accurately screened characteristic frequency points and their phase angles make subsequent comparisons with interference databases more targeted, significantly improving the specificity of impurity identification.

[0096] The phase angle of a single sample can be affected by instantaneous noise or fluctuations in the state of the electrode surface. Taking the average of multiple samples can reduce random errors and improve data stability. At each characteristic frequency point, the electrochemical workstation parameters and extract flow rate are kept stable, and the phase angle is continuously acquired multiple times. The acquired phase angles are then outlier detected by the Grubbs test. After removing the phase angles identified as outliers, the arithmetic mean of the remaining phase angles is calculated. This average is taken as the final phase angle at that characteristic frequency point, and the standard deviation is recorded as a data reliability indicator. Multiple samples at fixed intervals capture the dynamic stability of the phase angle. The Grubbs test scientifically removes extreme value interference. The joint recording of the average and standard deviation ensures both data accuracy and provides a basis for reliability assessment. The highly stable final phase angle makes the similarity matching results with the interference database more reliable, providing a solid foundation for accurately screening interfering impurities.

[0097] Specifically, such as Figure 3 As shown, the sub-logic for calculating the impurity contribution includes:

[0098] Call the preset peanut sprout interference database, which includes the contribution coefficients of the phase angle and frequency shift of the electrochemical impedance spectrum of impurities;

[0099] The extracted phase angles were matched with the phase angles of different impurities in the peanut sprout interference database to screen out interfering impurities.

[0100] Using the screened interfering impurities as variables and the original frequency shift of the crystal as the dependent variable, the frequency contribution weights of different interfering impurities are determined through cross-validation.

[0101] The contribution of each interfering impurity to the original frequency offset is calculated by partial least squares regression based on the frequency contribution weight, and the impurity contribution is obtained by summing them up.

[0102] Peanut sprout extract contains various natural impurities that can interfere with frequency shifts. A pre-defined interference database provides characteristic information for each impurity, offering a basis for identifying and quantifying interference. This database, constructed through prior experiments, identifies 12 common impurities in peanut sprouts, including glutamic acid, aspartic acid, alanine, sucrose, glucose, and citric acid. Single-impurity standard solutions and mixed-impurity solutions were prepared. Under conditions identical to actual detection (e.g., same electrodes, flow rates, and temperatures), the electrochemical impedance spectroscopy (EIS) of each impurity was measured in the corresponding frequency bands from low to high frequencies. The phase angles of characteristic frequency points were extracted and a feature library was established. Simultaneously, the frequency shift of each impurity was measured on a quartz crystal microbalance. The contribution coefficient of the frequency shift per unit concentration (i.e., the change in frequency shift per unit increase in concentration) was calculated and associated with the phase angle for storage. The database comprehensively covers common impurities, and all characteristic parameters were measured under the same detection conditions, ensuring comparability with actual detection data. The introduction of the unit contribution coefficient makes the quantitative calculation of impurity interference possible.

[0103] Phase angle similarity matching identifies interfering impurities in the extract that are similar in characteristics to those in the peanut sprout interference database, providing a target for subsequent contribution calculation. Using a cosine similarity algorithm, the phase angles of the three extracted feature frequency points are combined to form a feature vector, which is compared with the phase angle feature vectors of each impurity in the peanut sprout interference database. The cosine similarity value is calculated, with a range of 0-1; a higher value indicates higher similarity. A similarity threshold is set, and impurities with similarity values ​​greater than the threshold are listed as candidate interfering impurities. These candidate impurities undergo secondary verification by calculating the sum of squares of their phase angle differences. Impurities with sums greater than the set threshold are eliminated, ultimately determining the interfering impurities. The cosine similarity algorithm effectively measures the overall matching degree of feature vectors, and joint matching of multiple feature frequency points improves identification specificity. Secondary verification further eliminates impurities with low similarity, reducing the false positive rate. Accurately selected interfering impurities clearly define the objects of subsequent contribution calculation, ensuring that the calculation process focuses on truly existing interfering substances.

[0104] Different interfering impurities have varying degrees of influence on the original frequency offset. Determining the frequency contribution weight quantifies the relative influence of each impurity, providing coefficients for accurately calculating the impurity contribution. Using the concentration of each selected interfering impurity as the independent variable, and estimating it based on the calibration relationship between phase angle and concentration in the previous peanut sprout interference database, and using the original frequency offset of the quartz crystal microbalance as the dependent variable, a partial least squares regression model is constructed. The training and validation sets are distinguished, and the model parameters are iteratively optimized using the least squares method to obtain the frequency contribution weight of each interfering impurity—that is, the influence coefficient on the original frequency offset when the impurity concentration changes by one unit. The reliability of the weight is assessed using the mean square error of the validation set, retaining frequency contribution weights with a mean square error less than a set value. The partial least squares regression model can simultaneously handle the combined effects of multiple impurities, ensuring the stability of the frequency contribution weight across different datasets. Error assessment provides a quantitative basis for the reliability of the weight. Accurate frequency contribution weights enable precise quantification of the influence of each impurity on the original frequency offset, providing key coefficients for subsequent calculation of the specific impurity contribution value.

[0105] Partial least squares regression (PLR) effectively handles situations where variables are correlated, making it suitable for analyzing the contribution of multiple impurities. The total interference level is obtained by summing the contributions of each impurity. Using the concentrations of selected interfering impurities as the independent variable matrix and the original frequency offset as the dependent variable matrix, PLR, combined with the determined frequency contribution weights, calculates the individual contribution of each interfering impurity to the original frequency offset at the current concentration. This contribution is the product of the impurity concentration and its contribution weight. All individual contribution values ​​are algebraically summed to obtain the final impurity contribution. Simultaneously, the explanatory power of the PLR ​​model is calculated, ensuring it exceeds a set threshold to verify the reliability of the impurity contribution calculation. PLR effectively eliminates collinearity interference among impurities, improving the accuracy of impurity contribution calculation. The evaluation of the PLR ​​model's explanatory power ensures the reliability of the results. The joint recording of individual contribution values ​​and the total contribution provides data support for subsequent analysis of the interference proportion of each impurity. Accurate impurity contribution provides a clear basis for subsequent correction of the original frequency offset, making it the core data for achieving precise interference removal.

[0106] The corrected frequency offset specifically includes:

[0107] Temperature compensation is applied to the contribution of impurities based on the temperature during the detection process.

[0108] Based on the linear relationship between capacitance change and mass change, the influence weight of capacitance change on impurity contribution is determined.

[0109] The corrected frequency offset is obtained by subtracting the impurity contribution after temperature compensation and influence weighting from the original frequency offset.

[0110] Temperature affects electrode reaction kinetics and solution physicochemical properties, thereby altering the frequency contribution characteristics of impurities. Temperature compensation can eliminate the influence of ambient temperature fluctuations on the calculation of impurity contribution. A temperature sensor installed on the outer wall of the microfluidic channel records the temperature in real time during the detection process. The average temperature during the original frequency offset recording period is used as the compensation benchmark. The temperature response curves of the frequency offset contribution coefficients of each impurity stored in the peanut sprout interference database are used, i.e., the variation law of the contribution coefficients at different temperatures. Based on the difference between the actual average temperature and the standard temperature of 25℃, the frequency offset contribution coefficient of each impurity is corrected using linear interpolation. Then, the contribution value of each impurity and the final impurity contribution are recalculated to obtain the temperature-compensated impurity contribution. Real-time temperature monitoring captures temperature changes in the actual detection environment, and linear interpolation achieves smooth correction of the contribution coefficients. The temperature-compensated impurity contribution is closer to the true interference level under actual detection conditions.

[0111] The change in capacitance reflects the degree of binding between GABA and the sensor, while the contribution of impurities varies with the concentration of GABA. Determining the weights based on the linear relationship between the two allows for adjustments to the impurity contribution to better reflect actual detection scenarios. By retrieving the pre-calibrated linear relationship between capacitance and mass change, and substituting the currently detected capacitance change into this relationship, the corresponding mass change value of GABA is obtained. Based on the correlation between capacitance change and impurity interference intensity established from historical data—that is, the relative change pattern of impurity interference under different GABA concentrations—the influence weight of the impurity contribution corresponding to the current capacitance change is determined. When the GABA concentration is high, impurity interference is relatively weaker, and the influence weight decreases; conversely, the influence weight increases. Adjusting the influence weights based on the actual capacitance change considers the impact of the interaction between GABA and impurities on interference intensity, making the correction of the impurity contribution more dynamic and adaptable, avoiding errors caused by fixed weights. The weighted impurity contribution more accurately reflects the actual interference situation, providing reliable parameters for accurately calculating the corrected frequency offset.

[0112] The original frequency offset includes contributions from both GABA and impurities. Subtracting the compensated and adjusted impurity contribution removes interference, resulting in a frequency offset that reflects only the mass change of GABA. Using the original frequency offset recorded by the quartz crystal microbalance as the base value, the impurity contribution after temperature compensation and influence weight adjustment is subtracted (i.e., impurity contribution multiplied by influence weight) to obtain the corrected frequency offset. Error propagation analysis is performed on the correction process to calculate the uncertainty of the corrected frequency offset, ensuring that the uncertainty is within the allowable range. If it exceeds the allowable range, the compensation and adjustment process is re-examined. The difference calculation directly removes impurity interference, and error propagation analysis ensures the reliability of the correction result. The corrected frequency offset retains the characteristic signal of GABA to the greatest extent, eliminating interfering components and environmental influences. The high-precision corrected frequency offset provides the core parameters for inputting the recurrent neural network to calculate the GABA concentration, which is a key step in ensuring the accuracy of concentration calculation.

[0113] Furthermore, the output sub-logic for GABA concentration includes:

[0114] Construct a recurrent neural network in which the input layer includes the corrected frequency offset, phase angle, and temperature;

[0115] A recurrent neural network was trained using standard sample data containing different concentrations of GABA and impurity combinations, and the parameters of the recurrent neural network were optimized using a time backpropagation algorithm.

[0116] The corrected frequency offset, phase angle, and temperature obtained from real-time detection are input into the trained recurrent neural network, which outputs the concentration of GABA and simultaneously calculates the confidence level of the recurrent neural network output.

[0117] The corrected frequency offset is directly related to the mass of GABA, the phase angle reflects the differences in electrochemical properties of different substances, and temperature affects the electrochemical environment and molecular motion state of the entire detection process. These three factors together constitute the core parameters describing the mass detection scenario. Recurrent neural networks (RNNs) process time-series data through a memory mechanism, capturing the dynamic correlation of parameters during the detection process, making them more suitable for handling complex nonlinear mapping relationships than traditional methods. A three-layer RNN structure consisting of an input layer, hidden layers, and an output layer is used. The input layer has three nodes, corresponding to the corrected frequency offset, phase angle, and temperature, respectively. The hidden layer has several neurons, using activation functions suitable for processing time-series data, and retains the calculation state from the previous moment through feedback connections. The output layer is a single node, corresponding to the predicted GABA concentration. During RNN initialization, random initial values ​​are assigned to the weights and biases of each layer, and hyperparameters such as the number of training iterations and the learning rate are set. Multi-parameter collaborative input covers dimensions such as mass, electrochemical properties, and environmental factors, ensuring that the RNN comprehensively perceives the detection state. The memory characteristics of the recurrent structure can mine the changing patterns of parameters over time, improving the dynamic adaptability of concentration prediction. The combination of layered design and activation functions enhances the RNN's ability to handle nonlinear relationships.

[0118] In actual detection, the concentration of GABA and its combinations with impurities are diverse. Recurrent neural networks trained on a single sample lack generalization ability. Standard sample data needs to cover a wide range of concentrations and impurity combinations for the recurrent neural network to learn general patterns. The time-backpropagation algorithm can calculate errors backward along the time axis, allowing for targeted adjustment of the parameters of each time-based recurrent neural network, making it an effective method for optimization. Standard samples are prepared, covering multiple GABA concentration gradients within the target concentration range. Each concentration gradient is paired with different types and proportions of common impurities, such as glutamic acid and sucrose, to simulate the complex components in actual extracts under conditions completely consistent with actual detection, i.e., the same flow rate, temperature, and electrodes. The corrected frequency offset, phase angle, and temperature of each standard sample are obtained. The training set and validation set are divided proportionally. The training set is input into the recurrent neural network (RNN). The error between the predicted concentration and the actual concentration is calculated using the time backpropagation algorithm. The error is propagated backward along the network layers and time axis, and the weights and biases of each layer are dynamically adjusted. After each iteration, the performance of the RNN is evaluated using the validation set until the validation set error is stable and reaches the preset accuracy. Training with diverse standard samples enables the RNN to adapt to complex compositional changes and reduces the risk of overfitting. The time backpropagation algorithm accurately optimizes the parameters and improves the prediction accuracy of the RNN for unknown samples. Validation with different datasets ensures that the RNN maintains stable performance under different data distributions.

[0119] Real-time detection data is input into the trained network to obtain targeted concentration predictions. Confidence level quantifies the reliability of the prediction results, avoiding misjudgments due to the uncertainty of the recurrent neural network. This is especially important when the sample composition is complex, providing crucial reference for result interpretation. The corrected frequency offset, phase angle, and temperature, acquired in real-time, are fed into the trained recurrent neural network in the order of the input layer nodes. The recurrent neural network outputs a predicted GABA concentration through forward calculation. Simultaneously, based on the probability distribution and prediction error distribution of the output layer, it calculates the confidence level of the predicted value, reflecting the probability that the predicted result falls within a certain interval near the true value. An alert is triggered when the confidence level is less than a set confidence threshold. Real-time data input automates the process from detection to concentration output, improving detection efficiency. Confidence quantification provides an intuitive indicator of result reliability, helping testing personnel scientifically judge the credibility of the predicted value and reduce decision-making risks. The output GABA concentration and confidence level provide core evidence for quality judgment, ensuring that the judgment process is based on accurate data and considers the reliability of the results.

[0120] like Figure 4 As shown, determining the quality of GABA extract in peanut sprouts specifically includes:

[0121] Set the acceptable threshold for GABA concentration and the deviation range for test results;

[0122] The absolute value of the output GABA concentration is compared with the qualified threshold, and the relative standard deviation of the GABA concentration of three consecutive tests is determined to be within the deviation range.

[0123] Based on the comparison and judgment results, determine whether the quality of the GABA extract in peanut sprouts is up to standard. If it is not up to standard, output the interfering impurities and analysis report.

[0124] The pass threshold is a key standard for distinguishing whether an extract meets quality standards. It needs to be determined in conjunction with the application scenario of the extract, such as food, health products, and industry quality specifications. The deviation range of the test results reflects the repeatability of the method. If the deviation of multiple test results is too large, even if the concentration meets the standard in a single test, there will be instability in the test. It is necessary to constrain the deviation range. Referring to the quality requirements of peanut sprout GABA extract in relevant industry standards, and combining the test data of a large number of actual qualified samples, the pass threshold of GABA concentration is comprehensively determined. By repeatedly testing the same standard sample, the relative standard deviation of the test results is calculated. Combined with the allowable error range of the method, the deviation range of the relative standard deviation of the results of consecutive tests is set to ensure that the deviation range can tolerate reasonable random errors and effectively identify test instability. The pass threshold provides a clear quantitative standard for quality judgment, avoiding the ambiguity of subjective judgment. The deviation range constrains the stability of the test results, ensuring that qualified samples not only meet the standard in a single test, but also maintain consistency in multiple tests, improving the rigor of quality judgment. Clear thresholds and deviation ranges provide a unified scale for subsequent comparison and judgment, ensuring that the quality assessment process is standardized and repeatable.

[0125] The comparison of the absolute concentration value with the acceptable threshold directly determines whether the content meets the standard, and is a core indicator for quality assessment. The relative standard deviation of multiple consecutive tests reflects the repeatability of the method and the homogeneity of the sample. A single test result can be affected by random factors, and the stability verification of multiple tests can improve the reliability of the judgment. The combination of the two achieves a comprehensive quality assessment. The concentration value of GABA output by the recurrent neural network is extracted and compared with the set acceptable threshold. If the concentration value is greater than or equal to the acceptable threshold, the content is judged to meet the standard. At the same time, the same sample is continuously tested three times independently. Before each test, the sample and sensor are reprocessed, the three concentration results are recorded, the relative standard deviation is calculated, and it is judged whether the relative standard deviation is within the preset deviation range. If it is within the deviation range, the test result is judged to be stable. The dual judgment criteria take into account both the content compliance and the stability of the results, avoiding misjudgment caused by single test error or sample inhomogeneity. Three consecutive tests improve the statistical significance of the data and enhance the reliability of the quality assessment. The comparison and judgment results provide a direct basis for the final quality conclusion, ensuring that subsequent compliance judgment or problem analysis has clear data support.

[0126] A comprehensive quality conclusion can only be drawn by considering both content and stability. If the result is unqualified, it is insufficient to solve the problem; the reasons for the unqualified result and the main interfering factors must be identified to provide direction for process improvement. The analysis report should systematically present the testing process and the root cause of the problem, enhancing the application value of the results. If the GABA concentration meets the standard and the relative standard deviation of three consecutive tests is within the deviation range, the extract is considered qualified, and a qualified conclusion and the corresponding GABA concentration are output. If the GABA concentration does not meet the standard or the relative standard deviation exceeds the deviation range, it is considered unqualified, and the interfering impurities and their contribution amounts identified in the previous screening are called in. According to the report, an analysis report is generated, which includes information on non-conforming items such as insufficient concentration or poor stability, major interfering impurities and their impact on the test results, and suggestions for process improvement, such as optimizing extraction conditions to reduce specific impurities. A qualified conclusion provides quality certification for the application and distribution of the extract, while a detailed report for non-conforming items can accurately pinpoint the problem, guide the production end to optimize the extraction process, improve extract quality from the source, and achieve closed-loop improvement between testing and production. The quality conclusion and analysis report provide decision-making basis for the subsequent processing of the extract, such as releasing qualified products and reworking non-qualified products, thus completing the entire process support from testing to application improvement.

Claims

1. A method for detecting the quality of an amino butyric acid extract from peanut sprouts, characterized by, The method comprises the following steps: The amino butyric acid extract in peanut sprouts is filtered through a filter membrane and injected into the molecular imprinting sensor, an alternating current disturbance signal is applied, and the change in the capacitance of the molecular imprinting sensor is monitored and recorded in real time; The step of recording the change in the capacitance of the molecular imprinting sensor specifically comprises: The molecular imprinting sensor is pretreated, and the pretreatment comprises eluting a template molecule and an activation treatment to form a detection pool for recognizing amino butyric acid, and the amino butyric acid extract in peanut sprouts is filtered through a filter membrane and injected into the detection pool of the molecular imprinting sensor; An alternating current disturbance signal is applied, the alternating current disturbance signal covers multiple frequency bands, the capacitance signals at different frequency bands are monitored and extracted in real time, the capacitance signals are decomposed by fast Fourier transform, the fundamental frequency capacitance value and the high frequency capacitance value are extracted respectively to determine the capacitance baseline value, the difference between the real-time capacitance value and the capacitance baseline value is calculated to obtain the instantaneous capacitance change; When the fluctuation amplitude of the instantaneous capacitance change in a plurality of consecutive detection cycles is less than a set amplitude threshold, it is determined that the capacitance signal reaches a stable state, and the instantaneous capacitance change in the stable state is recorded as the change in the capacitance of the molecular imprinting sensor; The molecular imprinting sensor and the quartz crystal microbalance are connected in series through a microfluidic channel, when the recorded change in the capacitance is greater than a preset change threshold, the resonance frequency detection of the quartz crystal microbalance is triggered based on a pre-calibrated linear relationship between the change in the capacitance and the change in the mass, and the original frequency offset of the crystal is recorded; The electrochemical impedance spectrum of the extract is obtained, the phase angle is extracted, a preset peanut sprout interference database is called, the impurity contribution amount is calculated by partial least squares regression to obtain a corrected frequency offset, and the frequency offset, the phase angle and the temperature are input into a recurrent neural network to output the concentration of amino butyric acid to determine the quality of the amino butyric acid extract in peanut sprouts; the step of obtaining the corrected frequency offset specifically comprises: temperature compensation of the impurity contribution amount in combination with the temperature in the detection process; determining the influence weight of the change in the capacitance on the impurity contribution amount based on the linear relationship between the change in the capacitance and the change in the mass; subtracting the impurity contribution amount adjusted by the temperature compensation and the influence weight from the original frequency offset to obtain the corrected frequency offset; The extraction sub-logic of the phase angle comprises: An electrochemical workstation is started to apply a sweep excitation signal to the extract, the electrochemical impedance spectrum of the extract is obtained, the electrochemical impedance spectrum is processed to obtain the phase angle at different frequency points; characteristic frequency points are selected according to the characteristic changes of the electrochemical impedance spectrum, and the phase angle at the characteristic frequency points is extracted; the phase angle at the characteristic frequency points is sampled multiple times and averaged to obtain the final phase angle; The calculation sub-logic of the impurity contribution amount comprises: Call the preset peanut bud interference database, the peanut bud interference database includes the phase angle and the frequency offset of the electrochemical impedance spectrum of the impurity; the extracted phase angle is similarity matched with the phase angle of different impurities in the peanut bud interference database to screen out interfering impurities; the screened interfering impurities are variables, and the original frequency offset of the crystal is dependent variable; the frequency contribution weight of different interfering impurities is determined through cross-validation; the contribution value of each interfering impurity to the original frequency offset is calculated based on the frequency contribution weight through partial least squares regression, and the impurity contribution amount is calculated by accumulation.

2. The method for quality detection of GABA extract in peanut sprouts as described in claim 1, characterized in that, The series connection of the molecular imprinting sensor and the quartz crystal microbalance through the microfluidic channel specifically includes: Prepare a tubular microfluidic channel, connect the detection pool of the molecular imprinting sensor to the input end of the microfluidic channel, so that the detection pool is inside the microfluidic channel and perpendicular to the axis of the microfluidic channel; The detection area of the quartz crystal microbalance is connected to the output end of the microfluidic channel, so that the detection area is inside the microfluidic channel and parallel to the detection pool of the molecular imprinting sensor; Adjust the positions of the molecular imprinting sensor and the quartz crystal microbalance in the microfluidic channel, so that after the extract flows out of the detection pool, it can flow into the detection area of the quartz crystal microbalance within a set time.

3. The method for quality detection of GABA extract in peanut sprouts as described in claim 2, characterized in that, The recording logic of the original frequency offset includes: After receiving the detection start signal, the quartz crystal microbalance acquires the resonance frequency, and takes the resonance frequency at the initial time as the reference frequency; Calculate the difference between the resonance frequency acquired each time and the reference frequency to obtain the instantaneous frequency offset; Continuously record the instantaneous frequency offset until the difference between the consecutive preset number of instantaneous frequency offsets is less than the set fluctuation threshold, and record it as the original frequency offset of the crystal.

4. The method for quality detection of GABA extract in peanut sprouts as described in claim 3, characterized in that, The trigger sub-logic of the resonance frequency detection includes: Pre-calibrate the linear relationship between the capacitance change and the mass change by detecting the capacitance change of the blank peanut bud extract multiple times to pre-set the change threshold, and performing linear fitting with the capacitance change as the abscissa and the mass change as the ordinate; Compare the recorded capacitance change with the pre-set change threshold, and when the recorded capacitance change is greater than the pre-set change threshold, determine the mass change value based on the pre-calibrated linear relationship between the capacitance change and the mass change; Pre-set the sensitivity threshold of the quartz crystal microbalance, and when the determined mass change value is greater than or equal to the sensitivity threshold, send a detection start signal to the quartz crystal microbalance to trigger the resonance frequency detection.

5. The method for quality detection of GABA extract in peanut sprouts as described in claim 4, characterized in that, The judgment of the quality of the butyrobetaine extract in the peanut bud specifically includes: Set the qualified threshold of the butyrobetaine concentration and the deviation range of the detection result; Compare the absolute value of the output butyrobetaine concentration with the qualified threshold, and at the same time, judge whether the relative standard deviation of the butyrobetaine concentration of the consecutive 3 detections is within the deviation range; According to the comparison and judgment results, determine whether the quality of the butyrobetaine extract in the peanut bud is qualified, and if not, output the interfering impurities and the analysis report.

6. The method for quality detection of GABA extract in peanut sprouts as described in claim 5, characterized in that, The output sub-logic of the butyrobetaine concentration includes: Build a recurrent neural network, wherein the input layer includes the corrected frequency offset, phase angle and temperature; The recurrent neural network is trained by standard sample data containing different concentrations of aminobutyric acid and impurity combinations, and the parameters of the recurrent neural network are optimized by a time reverse propagation algorithm; The real-time detected corrected frequency offset, phase angle and temperature are input into the trained recurrent neural network, the aminobutyric acid concentration is output, and the confidence of the recurrent neural network output is calculated synchronously.

Images