A method for determining the residue of a bactericide in fruit juice

Abstract

The application discloses a method for determining the residue of a sterilizing agent in fruit juice, and belongs to the technical field of food analysis and detection. The method comprises the following steps: applying a preset frequency pulse electric field excitation to a fruit juice sample to be detected, collecting an initial conductivity relaxation spectrum sequence generated in a relaxation stage; separating a characteristic spectrum sub-sequence contributed by the directional polarization of the sterilizing agent molecules from the spectrum sequence, and extracting a decay time constant distribution vector; performing phase space reconstruction on the vector to generate a high-dimensional phase space trajectory matrix; calculating recursive quantitative analysis characteristic parameters of the matrix, including a recursive rate, certainty and laminar flow; inputting the characteristic parameters into a pre-constructed concentration inversion model to output the concentration of the sterilizing agent residue. The method can accurately extract the characteristic response of the sterilizing agent from a complex fruit juice matrix, and realizes high-sensitivity concentration inversion.

Description

Technical Field

[0001] This invention relates to the field of food analysis and testing technology, specifically to a method for determining bactericide residues in fruit juices. Background Technology

[0002] The determination of fungicide residues in fruit juices often relies on chromatographic-mass spectrometry (GC-MS) and liquid-chromatographic-tandem mass spectrometry (LC-MS / MS) techniques. These methods require cumbersome pretreatment processes such as liquid-liquid extraction, solid-phase extraction, and derivatization to separate and enrich the target analytes. This results in long analysis cycles, high detection costs, and the need for professional personnel to perform the procedures, and the detection is destructive to the sample. While conductivity relaxation spectroscopy based on pulsed electric field excitation offers the potential for rapid and non-destructive detection, fruit juice samples contain a large amount of inherently charged or polar components such as sugars, organic acids, vitamins, and pectin. The overall current decay signal measured after the applied pulsed electric field is removed represents a complex convolution of the relaxation responses of various matrix components and fungicide molecules. The weak characteristic signals contributed by trace fungicide molecules due to directional polarization dissipation are completely submerged by the matrix background, making it extremely difficult to extract characteristic components reflecting only fungicide behavior from this mixed spectrum. Conventional relaxation signal analysis often employs single or multiple discrete exponential decay models to directly fit the overall sequence. This results in a limited number of relaxation time parameters and low information dimensionality, making it difficult to characterize the multimodal and nonlinear orientation polarization relaxation dynamics of fungicide molecules in the complex microenvironment of fruit juice. This low-dimensionality leads to significant overlap between samples of different concentrations in the feature space, causing overfitting or underfitting in concentration inversion models, and making it difficult to meet practical detection requirements in terms of quantitative accuracy and robustness. Overcoming these limitations requires solving the problem of accurately separating fungicide characteristic relaxation signals under the interference of fruit juice matrix, and mining high-dimensional systemic dynamic characteristics that are highly sensitive to concentration changes from the separated signals. Summary of the Invention

[0003] This invention provides a method for determining bactericide residues in fruit juices, addressing the problems of strong background signal interference from the juice matrix and poor accuracy in inverting bactericide residue concentrations due to the low dimension of conventional relaxation features.

[0004] The objective of this invention can be achieved through the following technical solutions:

[0005] This invention provides a method for determining fungicide residues in fruit juices. By combining pulsed electric field excitation and relaxation spectroscopy with nonlinear kinetic feature mining, it achieves highly sensitive, label-free quantitative detection of trace fungicide residues in complex fruit juice matrices. The method includes the following steps:

[0006] A pulsed electric field of a preset frequency is applied to the fruit juice sample to be tested, and the initial conductivity relaxation spectrum sequence generated during the relaxation phase is acquired. Under the action of the pulsed electric field, different dipole components in the fruit juice produce differentiated polarization responses. The relaxation process carries rich molecular dynamics information, providing a raw signal basis for subsequently distinguishing the contributions of fungicide molecules from other background components.

[0007] Characteristic spectral subsequences contributed by the directional polarization of fungicide molecules are separated from the initial conductivity relaxation spectral sequence, and the decay time constant distribution vector of the characteristic spectral subsequences is extracted. By removing the dielectric relaxation behavior unique to fungicide molecules from complex background noise, the detection selectivity of target pesticide components can be significantly improved, and the decay time constant distribution vector concentrates the kinetic characteristics closely related to the type and concentration of fungicide.

[0008] Phase space reconstruction is performed on the decay time constant distribution vector to generate a high-dimensional phase space trajectory matrix. Mapping the one-dimensional feature sequence to the high-dimensional phase space fully reveals the nonlinear deterministic structure implicit in the relaxation process, improving the sensitivity and noise resistance of subsequent quantitative analysis to minute changes in concentration.

[0009] The recursive quantitative analysis characteristic parameters of the high-dimensional phase space trajectory matrix are calculated. These parameters include recursion rate, determinism, and laminarity. The recursion rate characterizes the probability of the system state recurring in phase space, determinism reflects the regularity of the dynamic process, and laminarity characterizes the layered structure resulting from the alternating dominance of different relaxation modes in the relaxation process. These parameters quantify the kinetic fingerprint changes introduced by the bactericide from different perspectives, significantly enhancing the robustness of the characteristic characterization.

[0010] The recursive quantitative analysis feature parameters are input into a pre-constructed concentration inversion model, which outputs the residual concentration of fungicide in the fruit juice sample. The constructed concentration inversion model uses recursion rate, determinism, and laminarity as multi-dimensional inputs, which can effectively approximate the nonlinear mapping relationship between feature parameters and concentration, and the output concentration value has high prediction accuracy and repeatability.

[0011] As a preferred embodiment of the present invention, before applying pulsed electric field excitation, the juice sample to be tested is first centrifuged, and the supernatant is collected. The supernatant is then filtered through a filter membrane to remove suspended particles with a diameter larger than the pore size of the filter membrane. The filtered juice sample is then adjusted to a preset conductivity value and a preset temperature value to obtain a pretreated juice sample. This pretreatment step, by removing large suspended particles and standardizing the conductivity and temperature of the medium, reduces the interference of conduction current fluctuations and thermal drift on the relaxation spectrum, ensuring the comparability and stability of the detection results.

[0012] As another preferred technical solution of the present invention, the process of applying pulsed electric field excitation and acquiring the initial conductivity relaxation spectrum sequence specifically involves: placing the juice sample to be tested between parallel plate electrodes, controlling a pulse generator to output a square wave pulse voltage of a preset frequency to the parallel plate electrodes; during the relaxation period after the square wave pulse voltage is turned off, continuously recording the current decay signal between the parallel plate electrodes at a preset sampling rate using a high-speed data acquisition card; arranging the current decay signal in chronological order to generate the initial conductivity relaxation spectrum sequence. Preferably, the high-speed data acquisition card is equipped with a transconductance amplifier and an anti-aliasing low-pass filter connected in series at its front end, thereby effectively suppressing high-frequency noise and spectral aliasing while ensuring high time resolution and improving the fidelity of the relaxation signal.

[0013] As a further preferred technical solution of the present invention, the separation of characteristic spectral subsequences from the initial conductivity relaxation spectral sequence is achieved as follows: wavelet packet decomposition is performed on the initial conductivity relaxation spectral sequence to obtain multiple frequency band components; the sample entropy value of each frequency band component is calculated, and target frequency band components whose sample entropy values ​​are located within a preset entropy value interval are identified; wavelet packet reconstruction is performed on the target frequency band components to obtain the characteristic spectral subsequences. Wavelet packet decomposition provides fine time-frequency localization capabilities, and the sample entropy value can sensitively reflect the ordered changes in the signal caused by the polarization contribution of the fungicide. By filtering the entropy value interval, frequency bands closely related to fungicide relaxation can be automatically locked, achieving adaptive signal separation. The preset entropy value interval is calibrated by analyzing the sample entropy distribution range of the wavelet packet decomposition frequency band components of a known blank fruit juice sample without fungicide, further ensuring the specificity of feature extraction.

[0014] As a preferred embodiment of the present invention, the step of extracting the decay time constant distribution vector includes: performing a first-order exponential decay model fitting on the characteristic spectral subsequence to obtain the principal decay time constant; calculating the residuals of the exponential decay curves corresponding to the principal decay time constants on the characteristic spectral subsequence to obtain the residual sequence; continuing to perform a first-order exponential decay model fitting on the residual sequence to obtain the secondary decay time constants; iterating the above steps until the root mean square of the residual sequence is less than a preset threshold; and arranging all the obtained decay time constants in order to form the decay time constant distribution vector. This hierarchical stripping fitting method can decompose the time constants of multiple relaxation components step by step, completely preserving the time scale information of multiple relaxation modes of the fungicide in the system, and avoiding parameter aliasing and accuracy loss caused by single model fitting.

[0015] As a preferred technical solution of the present invention, the specific steps of phase space reconstruction are as follows: determining the embedding dimension and delay time of the decay time constant distribution vector, wherein the embedding dimension is determined by the pseudo-nearest neighbor method and the delay time is determined by the mutual information method; reconstructing phase points for each element in the decay time constant distribution vector according to the embedding dimension and delay time to generate multiple phase point vectors; arranging each phase point vector as a row in chronological order to construct a high-dimensional phase space trajectory matrix. By using the pseudo-nearest neighbor method and the mutual information method to optimize the embedding dimension and delay time respectively, the dynamic structure in the true space can be preserved to the greatest extent, and trajectory folding or noise amplification caused by redundant or insufficient embedding can be avoided, providing a high-quality trajectory matrix that can directly characterize the inherent regularity of the system for recursive quantitative analysis.

[0016] As a preferred technical solution of the present invention, the process of obtaining the recursive quantitative analysis feature parameters includes: calculating the Euclidean distance between each row vector in the high-dimensional phase space trajectory matrix to form a distance matrix; binarizing the distance matrix according to a preset recursion threshold to generate a recursive matrix, wherein the elements equal to the preset value correspond to row vector pairs whose Euclidean distance is less than or equal to the recursion threshold; statistically analyzing the length distribution of the diagonal structure and the length distribution of the vertical structure in the recursive matrix; using the ratio of the number of all elements equal to the preset value to the total number of elements in the recursive matrix as the recursion rate; using the ratio of the number of diagonal elements whose length is greater than or equal to the first length threshold to the total number of diagonal elements as determinism; and using the ratio of the number of vertical elements whose length is greater than or equal to the second length threshold to the total number of vertical elements as laminarity. The three quantitative parameters extracted from the diagonal and vertical structures of the recursive graph independently describe different aspects of relaxation dynamics, collectively constituting a comprehensive feature set highly sensitive to changes in fungicide concentration, effectively overcoming the defect of single indicators being easily interfered with in complex matrices.

[0017] As a further preferred technical solution of the present invention, the concentration inversion model is constructed as follows: Multiple standard fruit juice samples with known fungicide concentrations are acquired; for each standard fruit juice sample, the aforementioned pulsed electric field excitation, acquisition, separation, extraction, reconstruction, and calculation steps are performed to obtain the corresponding standard recursion rate, standard determinism, and standard laminarity; using the standard recursion rate, standard determinism, and standard laminarity as inputs and the corresponding known fungicide concentration as output, a support vector regression model is trained to obtain the concentration inversion model; the recursion rate, determinism, and laminarity calculated for the fruit juice sample to be tested are input into the concentration inversion model, and the concentration inversion model outputs the fungicide residue concentration. The support vector regression model, with its structural risk minimization and kernel trick, can stably establish a multidimensional nonlinear regression relationship between recursive quantitative analysis feature parameters and fungicide concentration under conditions of few samples, balancing the model's generalization ability and prediction accuracy.

[0018] As another preferred embodiment of the present invention, after outputting the fungicide residue concentration, the method further includes: obtaining the variety and origin information of the fruit juice sample to be tested; querying the corresponding matrix correction coefficient from the database based on the variety and origin information; and multiplying the fungicide residue concentration output by the concentration inversion model by the matrix correction coefficient to obtain the final corrected fungicide residue concentration. By introducing a matrix correction coefficient related to variety and origin, the differences in dielectric background of natural products in fruit juice samples from different sources can be compensated, so that the quantitative results can maintain high accuracy and consistency in fruit juices of different varieties and origins.

[0019] The beneficial effects of this invention are:

[0020] Wavelet packet decomposition was performed on the initial conductivity relaxation spectral sequence to obtain multiple frequency band components corresponding to different relaxation rates. The sample entropy value of each frequency band component was calculated sequentially. Based on a preset entropy value range determined on a blank juice sample, target frequency band components whose sample entropy falls within this range were identified and retained, while other frequency band components dominated by the inherent components of the juice were discarded. Wavelet packet reconstruction was performed on the retained target frequency band components to obtain a characteristic spectral subsequence contributed solely by the directional polarization of fungicide molecules. Subsequently, a first-order exponential decay model was iteratively applied to fit this characteristic spectral subsequence to extract the principal decay time constant. The fitted curve was subtracted step-by-step to calculate the residual sequence. The residual sequence was then fitted again to obtain secondary decay time constants. This process was repeated until the root mean square value of the residuals was less than a preset threshold. All sequentially extracted decay time constants were arranged to form a decay time constant distribution vector. This separation and multi-level fitting strategy eliminates the response masking of large background components in the juice matrix, so that each time constant in the distribution vector accurately corresponds to the relaxation decay rate of the fungicide molecule in different orientation polarization modes in the complex liquid environment. It truly reflects the nonlinear multi-level relaxation behavior of its interaction with the microenvironment, providing a high-purity information source for subsequent characterization.

[0021] The pseudo-nearest neighbor method is used to determine the embedding dimension of the decay time constant distribution vector, and the mutual information method is used to determine the delay time. Based on this, the elements in the distribution vector are reconstructed into phase points, expanding the one-dimensional time constant sequence into multiple rows of phase point vectors, which are then arranged in order to form a high-dimensional phase space trajectory matrix. The Euclidean distance between each row vector in the trajectory matrix is ​​calculated, and the matrix is ​​binarized by a preset threshold to form a recursive matrix. The length distributions of the diagonal and vertical structures in the recursive matrix are further statistically analyzed to obtain three recursive quantitative analysis characteristic parameters: recursion rate, determinism, and laminarity. By embedding the low-dimensional relaxation time distribution into a high-dimensional nonlinear dynamic space, the relaxation trajectory shift caused by small changes in fungicide concentration is amplified in the recursive graph into significant changes in deterministic line segments and laminar structure. The recursion rate reflects the degree of clustering of phase space trajectories, determinism characterizes the predictability of system dynamics, and laminarity reveals intermittent stagnation behavior during relaxation. This combination of feature parameters constitutes a nonlinear fingerprint that is highly sensitive to fungicide residue concentration, effectively avoiding the limitation of insufficient identification power of linear relaxation parameters in the low concentration range, while also exhibiting natural robustness to random measurement perturbations. Using this high-dimensional feature parameter as input to train a support vector regression concentration inversion model, the fungicide residue concentration value in the tested juice sample can be accurately and stably output, significantly reducing the concentration quantification bias caused by matrix variation and feature overlap. Attached Figure Description

[0022] The invention will now be further described with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart of a method for determining bactericide residues in fruit juice;

[0024] Figure 2 This is a schematic diagram of the conductivity relaxation spectroscopy measurement device;

[0025] Figure 3 This is a flowchart of the iterative extraction process for the decay time constant of the characteristic spectral subsequence;

[0026] Figure 4 This is a flowchart of the juice sample pretreatment and bactericide residue detection process based on support vector regression concentration inversion. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See Figure 1This invention provides a method for determining bactericide residues in fruit juice, comprising: applying a pulsed electric field of a preset frequency to a fruit juice sample to be tested, and acquiring the initial conductivity relaxation spectrum sequence generated by the fruit juice sample during the relaxation phase; separating the characteristic spectral sub-sequence contributed by the directional polarization of bactericide molecules from the initial conductivity relaxation spectrum sequence, and extracting the decay time constant distribution vector of the characteristic spectral sub-sequence; performing phase space reconstruction on the decay time constant distribution vector to generate a high-dimensional phase space trajectory matrix; calculating the recursive quantitative analysis characteristic parameters of the high-dimensional phase space trajectory matrix, wherein the recursive quantitative analysis characteristic parameters include recursion rate, determinism, and laminarity; inputting the recursive quantitative analysis characteristic parameters into a pre-constructed concentration inversion model, and outputting the bactericide residue concentration in the fruit juice sample to be tested.

[0029] In specific implementation, please refer to Figure 2 The fruit juice sample to be tested is placed between parallel plate electrodes. The parallel plate electrodes are made of an inert conductive material to avoid electrochemical reactions with the fruit juice sample. The spacing between the parallel plate electrodes is adjusted according to the volume of the fruit juice sample to ensure sufficient contact between the electrodes and the sample. A control pulse generator outputs a square wave pulse voltage of a preset frequency to the parallel plate electrodes. The square wave pulse voltage output by the pulse generator has a set amplitude, preset frequency, and duty cycle. The preset frequency is determined based on the polarization relaxation time constant of the target bactericide molecules in the fruit juice sample. During the relaxation period after the square wave pulse voltage is turned off, the current decay signal between the parallel plate electrodes is continuously recorded at a preset sampling rate using a high-speed data acquisition card. The relaxation period begins at the falling edge of the square wave pulse voltage, and the duration of the relaxation period is greater than the time required for the polarization state of ions and molecules in the fruit juice sample to return to equilibrium. The preset sampling rate is set based on the highest frequency component of interest in the current decay signal, satisfying the Nyquist sampling theorem. The high-speed data acquisition card converts the recorded current decay signal into a discrete digital signal sequence, and arranges the digital signal sequence in chronological order to generate an initial conductivity relaxation spectrum sequence. A transconductance amplifier and an anti-aliasing low-pass filter are connected in series at the front end of the high-speed data acquisition card. The transconductance amplifier converts the weak current signal output from the parallel plate electrodes into a voltage signal and amplifies it. The transconductance gain of the transconductance amplifier is set according to the amplitude range of the current decay signal. The anti-aliasing low-pass filter performs low-pass filtering on the amplified voltage signal. The cutoff frequency of the anti-aliasing low-pass filter is set to be lower than half of the preset sampling rate to filter out signal components with frequencies higher than half of the preset sampling rate.

[0030] Optionally, the spacing between the parallel plate electrodes is set to 2 mm. The square wave pulse voltage output by the pulse generator has an amplitude of 10 volts, a preset frequency of 10 kHz, and a duty cycle of 50%. The relaxation time duration is set to 50 milliseconds. The preset sampling rate is set to 5 MHz. The transconductance gain of the transconductance amplifier is 10^4 volts per ampere. The cutoff frequency of the anti-aliasing low-pass filter is set to 2 MHz.

[0031] In some embodiments, the spacing between the parallel plate electrodes is between 1 mm and 5 mm. The amplitude of the square wave pulse voltage output by the pulse generator is between 5 V and 20 V, and the preset frequency is between 1 kHz and 100 kHz. The duration of the relaxation period is set between 20 ms and 100 ms. The preset sampling rate is set between 1 MHz and 10 MHz. The transconductance gain of the transconductance amplifier is selected in the range of 10^3 volts per ampere to 10^5 volts per ampere based on the impedance characteristics of the parallel plate electrode gap. The cutoff frequency of the anti-aliasing low-pass filter is set to 0.4 times the preset sampling rate.

[0032] In practice, wavelet packet decomposition is performed on the initial conductivity relaxation spectrum sequence to obtain multiple frequency band components. The number of decomposition levels is set to 5, and the wavelet basis functions used are the Daubechies4 wavelet basis functions. After 5 levels of wavelet packet decomposition, the initial conductivity relaxation spectrum sequence is decomposed into 32 frequency band components, each corresponding to a signal component within a specific frequency range in the initial conductivity relaxation spectrum sequence.

[0033] Calculate the sample entropy value for each frequency band component. For a single frequency band component, reconstruct the frequency band component into a one-dimensional time series signal, denoted as . ,in This represents the total number of data points in the frequency band components. (From frequency band components) Extract length in sequence from the middle. The vectors form a vector sequence. ,in The value is arrive Vector sequence The dimension is Each vector contains Each element.

[0034] Define vector with vector Distance between It is the maximum absolute value of the difference between the corresponding elements of the two, that is:

[0035]

[0036] in, and The index is the ordinal number of the vector, and its value range is [value missing]. to ,and ; This represents the offset of an element within the vector, with a value range of... to ; For frequency band components The Middle The value of each data point; For frequency band components The Middle The value of each data point; This indicates the operation of finding the maximum value. This indicates the absolute value operation. Vector length. Set to 2, based on: when When the value is 2, the sample entropy calculation is highly sensitive to changes in short-range patterns in the time series, and can achieve a balance between computational efficiency and pattern recognition accuracy.

[0037] Set similarity tolerance Similarity tolerance The value is 0.15 multiplied by the frequency band component. The standard deviation is set based on the following: 0.15 times the standard deviation is widely used as a similarity tolerance in the time series complexity analysis of bioelectrical and chemical signals. It can effectively distinguish the differences in the orderliness of signals from different components, avoiding a decrease in pattern discrimination ability due to an excessively large similarity tolerance or the loss of effective patterns due to an excessively small value. For each... Value, statistically satisfied conditional vector The number of , denoted as ,in The range of values ​​is to and Define intermediate variables and The average of the ratios is:

[0038]

[0039] in, The sample entropy value is the frequency band component. The length of the vector is 2; For similarity tolerance; This represents the total number of data points for the frequency band components. It is the natural logarithm function; The vector length is The number of vectors that satisfy the distance condition; The vector length is The number of vectors that satisfy the distance condition; the two summation operations respectively on from Accumulate to The entropy values ​​of the 32 frequency band components were calculated one by one according to the above formula, resulting in 32 sample entropy values.

[0040] The target frequency band component whose sample entropy value falls within a preset entropy value interval is identified. This preset entropy value interval is calibrated by analyzing the sample entropy distribution range of the frequency band components from wavelet packet decomposition of known blank fruit juice samples free of fungicides. The calibration process is as follows: Multiple batches of known blank fruit juice samples free of fungicides are taken. For each batch, pulsed electric field excitation, acquisition, and 5-layer wavelet packet decomposition are performed to obtain each frequency band component, and the sample entropy value of each frequency band component is calculated. The sample entropy values ​​of all batches of blank fruit juice samples in each frequency band component are summarized, and the interval formed by the minimum to maximum value of the sample entropy value in each frequency band component is used as the preset entropy value interval. Each of the 32 sample entropy values ​​is compared with the preset entropy value interval, and the frequency band component whose sample entropy value falls within the preset entropy value interval is determined to be the target frequency band component contributed by the directional polarization of fungicide molecules.

[0041] Wavelet packet reconstruction is performed on the target frequency band components. The wavelet packet coefficients of all frequency band components identified as target frequency band components are retained, while the wavelet packet coefficients of other frequency band components are set to zero. The retained wavelet packet coefficients are then reconstructed into a one-dimensional time series signal through inverse wavelet packet transform. The reconstructed one-dimensional time series signal is the characteristic spectral subsequence.

[0042] Optionally, the number of wavelet packet decomposition levels is set to 4, and the wavelet basis functions used for wavelet packet decomposition are Daubechies6 wavelet basis functions, with a vector length of... Set to 3, similarity tolerance The value is 0.2 multiplied by the standard deviation of the frequency band components.

[0043] In some embodiments, the number of decomposition layers in wavelet packet decomposition is selected between 3 and 7 layers based on the data length of the initial conductivity relaxation spectrum sequence. The wavelet basis functions used in wavelet packet decomposition are selected from the Daubechies series of wavelet basis functions, and the vector length... Values ​​can be 1, 2, or 3, similarity tolerance The value is between 0.1 and 0.25 times the standard deviation of the frequency band components.

[0044] In specific implementation, please refer to Figure 3 A first-order exponential decay model is applied to the characteristic spectral subsequence to obtain its principal decay time constant. The characteristic spectral subsequence is a one-dimensional time series signal, denoted as... ,in For time variables, In time The signal amplitude of the characteristic spectral subsequence. The expression for the first-order exponential decay model is:

[0045]

[0046] in, For a first-order exponential decay model in time The fitted value at that point; These are the initial amplitude parameters for the first-order exponential decay model; For the decay time constant parameter of the first-order exponential decay model; The base of the natural logarithm is used. A least-squares fitting algorithm is employed, with the characteristic spectral subsequence as the basis. Compared with the first-order exponential decay model The objective is to minimize the sum of squared residuals between the two values, and to solve for the initial magnitude parameter. and decay time constant parameter The optimal estimate is obtained. The least squares fitting algorithm uses the Levenberg-Marquardt algorithm for optimization. The optimal decay time constant parameters obtained are... This is the main decay time constant, denoted as .

[0047] The residual sequence is obtained by calculating the residual between the characteristic spectral subsequence and the exponential decay curve corresponding to the principal decay time constant. The optimal estimate and the main decay time constant The fitted curve obtained by substituting the first-order exponential decay model expression is denoted as... The residual sequence is calculated as follows: at each time sampling point of the characteristic spectral subsequence... At this point, the signal amplitude of the characteristic spectral subsequence is used. Subtract the exponential decay curve corresponding to the main decay time constant The fitted values ​​at the same sampling points at the same time point yield the residual sequence. .

[0048] Continue fitting the residual sequence with a first-order exponential decay model to obtain the secondary decay time constant. Then, apply the residual sequence... As the new signal to be fitted, the same first-order exponential decay model and least-squares fitting algorithm used to obtain the main decay time constant are employed to fit the residual sequence. By fitting the model, the expression for the first-order exponential decay model is: ,in For the residual sequence The initial magnitude parameters of the first-order exponential decay model during fitting. For the residual sequence The decay time constant parameters of the first-order exponential decay model during fitting. The optimal decay time constant parameters obtained by solving. This is the secondary decay time constant, denoted as . After completing this fitting, calculate the residual sequence. Exponential decay curve corresponding to the secondary decay time constant The residuals between them yield a new residual sequence. .

[0049] Repeat the above process: obtain the first... When the decay time constant is , for the , sub-residual sequence Perform a first-order exponential decay model fitting to obtain the... A decay time constant And calculate the first sub-residual sequence After each successful fitting of the first-order exponential decay model to the residual sequence, the current residual sequence is calculated. root mean square value The formula for calculating the root mean square value is: ,in This represents the total number of time sampling points in the residual sequence. The root mean square value... Compared with a preset threshold, when the root mean square value The iteration stops when the value is less than a preset threshold. The preset threshold is set to the characteristic spectral subsequence. The maximum amplitude is set to 0.01 times, based on the following: when the root mean square value of the residual sequence is less than one percent of the maximum amplitude of the characteristic spectral subsequence, the proportion of residual signal energy in the residual sequence to the total energy of the original signal is less than one ten-thousandth. Therefore, the contribution of the extracted decay time constants to subsequent concentration inversion can be ignored, and it avoids introducing noise components that are incorrectly identified as effective decay modes. All decay time constants obtained during the iteration process are arranged in the order of extraction to form a decay time constant distribution vector, i.e. ,in This represents the number of iterations.

[0050] Optionally, the least squares fitting algorithm uses gradient descent as the optimization method. The preset threshold is set to 0.05 times the root mean square value of the feature spectral subsequence.

[0051] In some embodiments, the optimization method used in the least squares fitting algorithm is selected from the Levenberg-Marquardt algorithm, gradient descent method, and quasi-Newton method. The preset threshold is set to a value between 0.005 and 0.02 times the maximum amplitude of the characteristic spectral subsequence, or a value between 0.01 and 0.1 times the root mean square value of the characteristic spectral subsequence.

[0052] In the specific implementation, the embedding dimension and delay time of the decay time constant distribution vector are determined. The decay time constant distribution vector is an ordered sequence of real numbers obtained in Example 3, denoted as the decay time constant distribution vector. ,in The first element in the decay time constant distribution vector represents the... One element, The embedding dimension is the length of the decay time constant distribution vector, i.e., the total number of decay time constants. The embedding dimension is determined using the pseudo-nearest neighbor method. The implementation process of the pseudo-nearest neighbor method is as follows: set the embedding dimension. The initial value is 1, and it is determined based on the current embedding dimension. The delay time obtained from the following delay time determination steps The decay time constant distribution vector Refactor to In the phase space, we obtain indivual A dimensional phase point vector. For each 3D phase point vector, in In the 3D phase space, find the nearest neighboring phase vector and calculate the Euclidean distance between them, denoted as . Increase the embedding dimension to The decay time constant distribution vector Refactor to In 3D phase space, calculating the phase point in the same phase space... The Euclidean distance between vectors of the same neighboring phase points in 3D space is denoted as . Define the pseudo-nearest neighbor criterion. When the false nearest neighbor criterion exceeds the false nearest neighbor threshold, the neighboring phase vector is determined to be a false neighbor in the low-dimensional space. The proportion of false neighbors among all phase vectors is calculated; when the proportion of false neighbors drops below 5%, the current embedding dimension is adjusted. The final embedding dimension is determined. The pseudo-nearest neighbor threshold is set to 10. The rationale is that when the relative change between the high-dimensional distance and the low-dimensional distance exceeds one order of magnitude, it indicates that the proximity relationship in the low-dimensional space is a spurious phenomenon caused by projection. A value of 10 can strike a balance between identifying spurious neighbors and avoiding overestimation of the embedding dimension.

[0053] The delay time is determined using the mutual information method. The implementation process of the mutual information method is as follows: the decay time constant distribution vector... Treat it as a set of observation sequences, and define the delay time variable. From the decay time constant distribution vector Construct two sequences, one of which is the original sequence. and delayed sequence . convert the original sequence and delayed sequence The range of values ​​is divided into equal parts. Given intervals, statistically analyze the original sequence. Falling in The probability of each interval Delayed sequences Falling in The probability of each interval and the original sequence Falling in intervals and delayed sequence At the same time, it fell into the first The joint probability of each interval ,in and All are interval indices, with a value range of 100. to Set the delay time variable. from Start incrementing for each delay time variable. The value of is used to calculate the original sequence. With delayed sequence Mutual information values ​​between Mutual information value The calculation formula is:

[0054]

[0055] in, The delay time variable takes the value of Mutual information value at time; The number of equal divisions for the value interval is set to 10. The rationale for this setting is that 10 equal divisions can provide sufficient details of the probability distribution, while ensuring that each interval has enough data points for probability estimation. The original sequence falls into the first position. The interval and the delayed sequence fall into the first interval. Joint probability of intervals; The original sequence falls into the first position. Marginal probabilities of the interval; For the delayed sequence to fall into the first Marginal probabilities of the interval; Represents a base-2 logarithmic operation. Plot the mutual information values. With delay time variable The changing curve represents the mutual information value. The time delay variable corresponding to when the curve first drops to the minimum of mutual information and tends to plateau. The value of is determined as the final delay time.

[0056] Based on the determined embedding dimension and delay time, phase point reconstruction is performed on each element of the decay time constant distribution vector to generate multiple phase point vectors. The final determined embedding dimension is set to... The final determined delay time is For the decay time constant distribution vector From the Starting from the element to the th element Each time index position up to the element Construct a dimensional phase vector ,in The range of values ​​is arrive Integer values ​​between [aspects]. For each phase point vector... As a single line, arranged by time index position Arrange them in ascending order to construct a high-dimensional phase space trajectory matrix. The number of rows in the high-dimensional phase space trajectory matrix is... The number of columns is .

[0057] Calculate the Euclidean distances between the row vectors in the high-dimensional phase space trajectory matrix to form a distance matrix. The high-dimensional phase space trajectory matrix is ​​denoted as... Its number of rows is The number of columns is .right The first in row vectors and the row vectors Calculate the Euclidean distance between the two. ,in and All are index numbers of the row vectors, with values ​​ranging from 1 to 2. to ; This is the column index, and its value range is... to ; For high-dimensional phase space trajectory matrix The Middle Line 1 The numerical values ​​of the column elements; For high-dimensional phase space trajectory matrix The Middle Line 1 The element values ​​of the column. The Euclidean distance between all row vector pairs. By row index and column indexes Arranged, forming a size of The distance matrix, the first in the distance matrix Line 1 Column elements are .

[0058] The distance matrix is ​​binarized according to a preset recursion threshold to generate a recursive matrix. The recursion threshold is denoted as... The recursive threshold is taken as the high-dimensional phase space trajectory matrix. The recursion threshold is set to 0.1 times the mean Euclidean distance of all row vectors. The rationale is that using 0.1 times the mean distance as a recursion threshold effectively distinguishes between neighboring and non-neighboring states in the phase space trajectory. When the recursion threshold is set to this value, the recursion matrix can capture approximately 5% to 20% of the recursion points in the phase space trajectory, forming a clear diagonal and vertical structure. For each element in the distance matrix... Perform binarization determination: if Then the corresponding position in the recursive matrix will be... Line 1 Column elements Set to 1; if Then Set to 0. The elements in the recursive matrix that are equal to the preset value correspond to row vector pairs whose Euclidean distance is less than or equal to the recursion threshold. The preset value is 1.

[0059] This study analyzes the length distributions of diagonal and vertical line structures in a recursive matrix. A diagonal structure is a line segment formed by consecutive elements with a value of 1 along the diagonal direction of the recursive matrix. For each diagonal line in the recursive matrix, the length of the segment containing consecutive elements with a value of 1 is scanned and recorded; this length represents the length of the diagonal structure. The lengths of all diagonal structures constitute the diagonal structure length distribution. Similarly, a vertical line structure is a line segment formed by consecutive elements with a value of 1 along the vertical direction of the recursive matrix. For each column of the recursive matrix, the length of the segment containing consecutive elements with a value of 1 is scanned and recorded; this length represents the length of the vertical line structure. The lengths of all vertical line structures constitute the vertical line structure length distribution.

[0060] The recursion rate, determinism, and laminarity are calculated based on the length distribution. The recursion rate is the ratio of the number of elements equal to 1 in the recursion matrix to the total number of elements in the recursion matrix. The total number of elements in the recursion matrix is... The ratio of the number of diagonal elements in the recursion matrix whose length is greater than or equal to a first length threshold to the total number of diagonal elements is used as determinism. The first length threshold is set to 2, based on the following: a diagonal length of at least 2 indicates that there are at least two consecutive recursive points in the phase space trajectory, reflecting the trajectory evolution law in a deterministic system. Excluding isolated recursive points with a length of 1 avoids noise interference with the deterministic measure. The total number of diagonal elements is the total number of elements in the recursion matrix that are 1 along the diagonal direction. The ratio of the number of vertical line elements in the recursion matrix whose length is greater than or equal to a second length threshold to the total number of vertical elements is used as laminarity. The second length threshold is set to 2, based on the following: a vertical line length of at least 2 indicates that the phase space trajectory is in an adjacent state for at least two consecutive moments, reflecting the laminar evolution characteristics of the system state. Excluding isolated vertical line recursive points with a length of 1 can more accurately measure laminarity. The total number of vertical line elements is the total number of elements in the recursive matrix that are vertically aligned and have a value of 1.

[0061] Optionally, the pseudo-nearest neighbor threshold is set to 15. The first length threshold is set to 3, and the second length threshold is set to 3.

[0062] In some embodiments, the pseudo-nearest neighbor threshold is set to an integer value between 5 and 20. The recursive threshold is set to a value between 0.05 and 0.2 times the mean of the Euclidean distances of all row vectors in the high-dimensional phase space trajectory matrix. The first length threshold is set to an integer value between 2 and 5, and the second length threshold is set to an integer value between 2 and 5.

[0063] In specific implementation, please refer to Figure 4 The fruit juice sample to be tested was centrifuged, and the supernatant was collected. Centrifugation was performed in a centrifuge at a speed of 4000 rpm for 15 minutes. The centrifugation speed and duration were set based on the following principle: a centrifugal force of 4000 rpm for 15 minutes ensures sufficient sedimentation of suspended particles larger than 1 micrometer in diameter in the fruit juice sample, while avoiding excessive centrifugation force or time that could lead to the loss of target bactericide molecules due to adsorption on the precipitate surface. After centrifugation, the supernatant, comprising 80% to 90% of the original total volume of the fruit juice sample, was removed from the centrifuge tube using a pipette.

[0064] The supernatant was filtered through a filter membrane to remove suspended particulate matter larger than the membrane pore size, yielding the filtered fruit juice sample. The filter membrane was made of polyethersulfone with a pore size of 0.45 micrometers. The pore size of 0.45 micrometers was chosen because suspended particulate matter larger than 0.45 micrometers can cause uneven local current distribution during pulsed electric field excitation, resulting in low-frequency drift interference on the baseline of the initial conductivity relaxation spectrum sequence. Colloidal particles smaller than 0.45 micrometers, however, have an impact on the macroscopic conductivity relaxation characteristics of the fruit juice sample within the measurement error range and are therefore retained. The filtration operation was performed under negative pressure conditions of -0.05 MPa.

[0065] The filtered juice sample was adjusted to preset conductivity and temperature values ​​to obtain a pre-treated juice sample, which was used to replace the juice sample to be tested in subsequent steps. The preset conductivity value was set to 1.5 mSiemens per centimeter, and the preset temperature value was set to 25 degrees Celsius. The conductivity value was adjusted as follows: when the measured conductivity value of the filtered juice sample was higher than 1.5 mSiemens per centimeter, deionized water was added to dilute the filtered juice sample; when the measured conductivity value of the filtered juice sample was lower than 1.5 mSiemens per centimeter, concentrated juice extract with a known conductivity value was added to the filtered juice sample. The temperature value was adjusted as follows: the filtered juice sample was placed in a constant temperature water bath with a temperature control accuracy of ±0.1 degrees Celsius, and removed after the measured temperature value stabilized at 25 degrees Celsius. The preset conductivity value of 1.5 mSiemens per centimeter and the preset temperature value of 25 degrees Celsius were set based on the following: under these conductivity and temperature conditions, the separation degree between the directional polarization relaxation signal of the bactericide molecules and the background signal of the fruit juice matrix is ​​optimal, and these conditions are within the controllable range of common laboratory environments, making them easy to operate and reproduce. The pretreated fruit juice samples underwent pulsed electric field excitation, acquisition, separation, extraction, reconstruction, and calculation steps to obtain the recurrence rate, determinism, and laminar flow properties of the pretreated fruit juice samples.

[0066] Multiple standard fruit juice samples with known fungicide concentrations were obtained. These standard fruit juice samples were prepared by adding a fungicide standard solution to a known blank fruit juice of the same variety that did not contain fungicide. The concentration of the fungicide standard solution was calibrated using a national metrological certification standard substance certificate. The fungicide concentration of the standard fruit juice samples covered a range from 0.01 mg / kg to 10 mg / kg, with a total of 8 concentration gradients. Each concentration gradient of standard fruit juice samples was packaged into 3 parallel samples. The pulsed electric field excitation, acquisition, separation, extraction, reconstruction, and calculation steps were performed on each standard fruit juice sample to obtain the standard recursion rate, standard determinism, and standard laminarity corresponding to each standard fruit juice sample. The pulsed electric field excitation step was completely consistent with the step used when processing the fruit juice samples to be tested; the number of decomposition layers in wavelet packet decomposition, the parameters for calculating sample entropy, the recursion threshold, and the first and second length thresholds were all kept at the same settings.

[0067] Using standard recursion rate, standard determinism, and standard laminar flow as inputs, and the corresponding known fungicide concentration as output, a support vector regression model is trained to obtain a concentration inversion model. The core architecture of the support vector regression model includes: an input layer that receives three-dimensional feature vectors, which are composed of standard recursion rate, standard determinism, and standard laminar flow in sequence; a kernel function layer that uses a radial basis function (RBF) kernel to map the three-dimensional feature vectors from the original three-dimensional space to a higher-dimensional feature space; a support vector layer composed of support vectors selected from all standard juice samples during training, with each support vector corresponding to a three-dimensional feature vector of a standard juice sample; and an output layer that uses a decision function to perform a weighted summation of the mapping results in the higher-dimensional feature space, outputting the predicted fungicide concentration. The mathematical form of the RBF kernel is: ,in The result is the calculation result of the radial basis kernel function. For the first The three-dimensional feature vector of a standard juice sample, For the first The three-dimensional feature vector of a standard juice sample, The square of the Euclidean distance between two three-dimensional eigenvectors is given. The width parameter of the radial basis kernel function. For the natural constant The decision function expression for the support vector regression model is: (The expression is an exponential function operation with base 0.05).

[0068]

[0069] in, For the support vector regression model, the input three-dimensional feature vector The output predicted value of the bactericide concentration; The input is a three-dimensional feature vector; This represents the total number of support vectors. For the first The three-dimensional feature vectors corresponding to each support vector; For the first The radial basis kernel function values ​​between each support vector and the input 3D feature vector; and The first result obtained by solving a quadratic optimization problem during the training process. The Lagrange multipliers corresponding to each support vector are both non-negative real numbers, and at least one of them is zero; This is the bias term, which is obtained by solving it together with the Lagrange multiplier coefficients during the training process.

[0070] The training steps for the support vector regression model are as follows: First, construct a training dataset using the three-dimensional feature vectors of all standard fruit juice samples and their corresponding known fungicide concentrations. Second, normalize each dimension of the three-dimensional feature vectors by subtracting the mean of all samples in that dimension and then dividing by the standard deviation of all samples in that dimension, ensuring that the numerical distribution of each dimension has zero mean and unit variance. Third, employ an insensitive loss function. Insensitive loss function The value of 0.01 is set based on the following: when the absolute deviation between the predicted concentration and the known fungicide concentration is less than 0.01 mg / kg, this prediction error does not constitute actual loss within the requirements of quantitative analysis, and helps to improve the sparsity and generalization ability of the support vector regression model; the optimal penalty coefficient of the support vector regression model is determined by five-fold cross-validation and grid search method. and optimal kernel width parameters Penalty coefficient The search scope is kernel width parameter The search scope is The grid search step size is an exponential step size with a base of 10; the penalty coefficient is the minimum root mean square error of the cross-validation. and kernel width parameter The optimal penalty coefficient and kernel width parameters are combined; using these parameters, a quadratic optimization problem is solved on the entire training dataset to obtain all support vectors. The corresponding Lagrange multipliers , and bias terms The concentration inversion model was constructed.

[0071] The recurrence rate, determinism, and laminar flowability calculated for the pretreated juice samples are input into the concentration inversion model, which outputs the fungicide residue concentration. During input, the normalized parameters used in training with standard juice samples are applied. The recurrence rate, determinism, and laminar flowability of the pretreated juice samples are subtracted from their corresponding training set mean, and then divided by their corresponding training set standard deviation to obtain a normalized three-dimensional feature vector, which is then substituted into the decision function. The residual concentration of the bactericide was calculated and output.

[0072] The concentration inversion model is constructed as follows: Multiple standard fruit juice samples with known fungicide concentrations are acquired. For each standard fruit juice sample, the aforementioned pulsed electric field excitation, acquisition, separation, extraction, reconstruction, and calculation steps are performed to obtain the corresponding standard recursion rate, standard determinism, and standard laminarity. Using the standard recursion rate, standard determinism, and standard laminarity as inputs and the corresponding known fungicide concentration as output, a support vector regression model is trained to obtain the concentration inversion model. The recursion rate, determinism, and laminarity calculated for the fruit juice sample to be tested are input into this concentration inversion model, and the concentration inversion model outputs the fungicide residue concentration.

[0073] In practice, the training process for the concentration inversion model is as follows: Standardization preprocessing is performed on the obtained standard recursion rate, standard determinism, and standard laminarity. The mean of all standard samples across the three feature dimensions is calculated. , , with standard deviation , , ; Apply the formula to each characteristic value Normalization is performed to ensure the model follows a zero-mean, unit-variance distribution, eliminating the interference of dimensional differences on model training. A training dataset is constructed, and a Support Vector Regression (SVR) model is used for training (the kernel function follows the radial basis function defined earlier). The hyperparameters were optimized using a five-fold cross-validation and grid search method, with a penalty coefficient set. The search scope is nuclear parameters The search scope is Insensitive loss coefficient The parameter set is fixed at 0.01; the optimal parameter combination is determined with the goal of minimizing the root mean square error (RMSE) of the validation set. The model is constructed by solving a quadratic programming problem on the full training set.

[0074] In another embodiment, to further enhance the model's generalization ability and anti-interference performance, an ensemble learning strategy can be used to construct alternative models. A random forest regressor (RF) and a gradient boosting decision tree (GBDT) are trained in parallel; the number of trees in the random forest is set to 100, and the maximum depth is set to 8; the learning rate of the GBDT is set to 0.1, and the maximum number of iterations is set to 200. The final output fungicide residue concentration is a weighted average of the predictions from the SVR, RF, and GBDT models, with the weights determined by the coefficient of determination of each model on the validation set. Dynamically determined.

[0075] When inputting the recursion rate, determinism, and laminarity calculated for the juice sample to be tested into the concentration inversion model, the mean and standard deviation from the training phase must first be normalized in the same way, and then the concentration inversion model outputs the residual concentration of fungicide.

[0076] The process involves obtaining the variety and origin information of the fruit juice samples to be tested, and then retrieving the corresponding matrix correction coefficients from the database based on these information. Variety information is represented by the fruit variety name string, and origin information by the geographical production area name string. The database is a pre-constructed matrix correction coefficient library, where the key is a combination identifier of the variety and origin information, and the value is the corresponding matrix correction coefficient. The matrix correction coefficients are obtained as follows: Blank fruit juice samples of the same variety and origin are selected, and a known concentration of fungicide standard solution is added to the blank samples to prepare a validation fruit juice sample with a known spiking concentration. The validation fruit juice sample undergoes the same pretreatment, pulsed electric field excitation, acquisition, separation, extraction, reconstruction, calculation, and concentration inversion steps as the test samples to obtain the predicted concentration of the validation fruit juice sample. The known spiking concentration is divided by the predicted concentration to obtain the matrix correction coefficient. Each combination of variety and origin information is repeated three times in parallel, and the average of the three matrix correction coefficients is entered into the database.

[0077] The final corrected fungicide residual concentration is obtained by multiplying the fungicide residual concentration output by the concentration inversion model by the matrix correction coefficient.

[0078] Optionally, the centrifugation speed is set to 3500 rpm, the centrifugation time to 20 minutes; the filter membrane pore size is 0.22 micrometers; the preset conductivity is set to 1.8 millisiemens per centimeter, and the preset temperature is set to 20 degrees Celsius; insensitive loss function. The value is 0.02; penalty coefficient The search scope is kernel width parameter The search scope is .

[0079] In some embodiments, the centrifugation speed is set to a value between 3000 rpm and 5000 rpm, and the centrifugation time is set to a value between 10 minutes and 25 minutes; the filter membrane pore size is selected as 0.22 micrometers or 0.45 micrometers; the preset conductivity value is selected in the range of 1.0 millisiemens per centimeter to 2.0 millisiemens per centimeter, and the preset temperature value is selected in the range of 20 degrees Celsius to 30 degrees Celsius; an insensitive loss function is used. The value is selected within the range of 0.005 to 0.05; penalty coefficient Search range and kernel width parameters The search scope is adjusted based on the number of standard juice samples and the distribution of feature values.

[0080] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for determining bactericide residues in fruit juice, characterized in that, Includes the following steps: A pulsed electric field of a preset frequency is applied to the juice sample to be tested, and the initial conductivity relaxation spectrum sequence of the juice sample during the relaxation phase is collected. From the initial conductivity relaxation spectral sequence, the characteristic spectral subsequence contributed by the directional polarization of the bactericide molecules is separated, and the decay time constant distribution vector of the characteristic spectral subsequence is extracted; Phase space reconstruction is performed on the decay time constant distribution vector to generate a high-dimensional phase space trajectory matrix; Calculate the recursive quantitative analysis characteristic parameters of the high-dimensional phase space trajectory matrix, wherein the recursive quantitative analysis characteristic parameters include recursion rate, determinism, and laminarity; The recursive quantitative analysis feature parameters are input into a pre-constructed concentration inversion model, which outputs the residual concentration of bactericide in the fruit juice sample to be tested.

2. The method for determining bactericide residues in fruit juice according to claim 1, characterized in that, The process of applying a pulsed electric field of a preset frequency to the juice sample to be tested and acquiring the initial conductivity relaxation spectrum sequence of the juice sample during the relaxation phase specifically includes: The juice sample to be tested is placed between parallel plate electrodes, and the pulse generator is controlled to output a square wave pulse voltage of the preset frequency to the parallel plate electrodes; During the relaxation period after the square wave pulse voltage is turned off, the current attenuation signal between the parallel plate electrodes is continuously recorded by a high-speed data acquisition card at a preset sampling rate. The current decay signals are arranged in chronological order to generate the initial conductivity relaxation spectrum sequence.

3. The method for determining bactericide residues in fruit juice according to claim 2, characterized in that, The high-speed data acquisition card has a transconductance amplifier and an anti-aliasing low-pass filter connected in series at its front end.

4. The method for determining bactericide residues in fruit juice according to claim 1, characterized in that, The characteristic spectral subsequence contributed by the directional polarization of bactericide molecules is separated from the initial conductivity relaxation spectral sequence, specifically including: Wavelet packet decomposition is performed on the initial conductivity relaxation spectrum sequence to obtain multiple frequency band components; Calculate the sample entropy value of each frequency band component, and identify the target frequency band component whose sample entropy value is located within a preset entropy value range; Wavelet packet reconstruction is performed on the target frequency band components to obtain the characteristic spectral subsequence.

5. The method for determining bactericide residues in fruit juice according to claim 4, characterized in that, The preset entropy range is determined by analyzing the sample entropy distribution range of the wavelet packet decomposition frequency band components of a known blank juice sample that does not contain bactericides.

6. The method for determining bactericide residues in fruit juice according to claim 1, characterized in that, Extracting the decay time constant distribution vector of the characteristic spectral subsequence specifically includes: A first-order exponential decay model is applied to the characteristic spectral subsequence to obtain the principal decay time constant of the characteristic spectral subsequence; The residual sequence is obtained by performing residual calculation on the exponential decay curves corresponding to the characteristic spectral subsequence and the principal decay time constant; Continue to fit the first-order exponential decay model to the residual sequence to obtain the secondary decay time constant. Repeat this step until the root mean square of the residual sequence is less than a preset threshold. Arrange all the obtained decay time constants in order to form the decay time constant distribution vector.

7. The method for determining bactericide residues in fruit juice according to claim 1, characterized in that, Perform phase space reconstruction on the decay time constant distribution vector to generate a high-dimensional phase space trajectory matrix, specifically including: The embedding dimension and delay time of the decay time constant distribution vector are determined by means of the pseudo-nearest neighbor method and the delay time is determined by means of the mutual information method. Based on the embedding dimension and the delay time, each element in the decay time constant distribution vector is reconstructed into a phase point to generate multiple phase point vectors. Each phase point vector is arranged as a row in chronological order to construct the high-dimensional phase space trajectory matrix.

8. The method for determining bactericide residues in fruit juice according to claim 1, characterized in that, The recursive quantitative analysis characteristic parameters for calculating the high-dimensional phase space trajectory matrix include recursion rate, determinism, and laminarity, specifically including: Calculate the Euclidean distance between each row vector in the high-dimensional phase space trajectory matrix to form a distance matrix; The distance matrix is ​​binarized according to a preset recursion threshold to generate a recursive matrix; The length distribution of the diagonal structure and the length distribution of the vertical structure in the recursion matrix are statistically analyzed, and the recursion rate, the determinism, and the laminarity are calculated based on the length distribution.

9. A method for determining bactericide residues in fruit juice according to claim 8, characterized in that, The elements in the recursive matrix that are equal to a preset value correspond to row vector pairs whose Euclidean distance is less than or equal to the recursive threshold.

10. A method for determining bactericide residues in fruit juice according to claim 8, characterized in that, The calculation of the recursion rate, the determinism, and the laminarity based on the length distribution specifically includes: The recursion rate is the ratio of the number of all elements in the recursion matrix that are equal to a preset value to the total number of elements in the recursion matrix. The ratio of the number of diagonal elements in the recursive matrix whose length is greater than or equal to the first length threshold to the total number of diagonal elements is used as the determinism. The ratio of the number of vertical line elements in the recursive matrix whose length is greater than or equal to the second length threshold to the total number of vertical line elements is taken as the laminarity.

Images