Intelligent analysis method for electrochemical recognition of flavonoid compounds

Abstract

The invention relates to the technical field of intelligent analysis of flavonoid compounds, in particular to an intelligent analysis method for electrochemical recognition of flavonoid compounds. The method comprises the following steps: acquiring a response signal matrix of a time-frequency dimension, and performing frequency domain compression and nonlinear interactive mapping to obtain a one-dimensional nonlinear compression current response sequence; based on the one-dimensional nonlinear compression current response sequence, a variable window integral singular response recognition mechanism is introduced, and singular response intensity is calculated; and determining a singular response occurrence point based on the singular response intensity, carrying out weighted fusion on the local window tensor, carrying out asynchronous reconstruction processing on the obtained singular weighted fusion tensor to obtain feature representation of the flavone target object after asynchronous reconstruction processing, and carrying out intelligent analysis on the flavone compound. The problems that a traditional electrochemical identification analysis method for the flavonoid compounds is low in resolution ratio and poor in stability and repeatability, and the analysis precision is low due to the fact that the characteristics of various target objects in the flavonoid compounds cannot be obtained are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent analysis of flavonoids, and particularly relates to an intelligent analysis method for electrochemical recognition of flavonoids. BACKGROUND

[0002] With the increasing demand for the recognition accuracy of natural active substances in the application fields such as food functional detection, traditional Chinese medicine component identification and nutritional health evaluation, flavonoids, as a kind of functional component widely existing in plant-derived substances, have become a research hotspot for qualitative recognition and quantitative analysis. At present, electrochemical analysis is widely used in the detection of flavonoids due to its high sensitivity, rapid response and no need for complex sample pretreatment. However, in practical application, the traditional electrochemical analysis technology still faces many technical bottlenecks in dealing with samples containing multiple flavonoids with similar structures. For example, the response signals obtained by conventional voltammetry, potential step method or cyclic voltammetry are mostly one-dimensional current-time or current-voltage curves, which lack the structural expression ability of complex redox behavior under multi-frequency electric excitation, resulting in serious overlapping of response curve peak positions of different flavonoid components and unstable response amplitude, which makes it impossible to achieve high-resolution effective recognition. On the other hand, the existing signal analysis methods mostly rely on manual feature extraction or low-dimensional statistical parameters (such as peak height, half-peak width, potential shift, etc.), which are difficult to capture the high-order features such as frequency interaction coupling, response mutation and nonlinear transition contained in the response signal. In addition, different flavonoids have different starting time delays and response rates when undergoing redox reactions, which makes it difficult for traditional machine learning methods based on static feature models to effectively model the asynchronous behavior, further reducing the classification accuracy and model generalization ability.

[0003] Therefore, in the traditional electrochemical recognition analysis of flavonoids, there are still problems of low resolution, poor stability and repeatability, and low analysis accuracy due to the inability to obtain the characteristics of various target substances in flavonoids. SUMMARY

[0004] The present application provides an intelligent analysis method for electrochemical recognition of flavonoids to solve the technical problems of low resolution, poor stability and repeatability, and low analysis accuracy due to the inability to obtain the characteristics of various target substances in flavonoids in the traditional electrochemical recognition analysis of flavonoids.

[0005] The intelligent analysis method for electrochemical recognition of flavonoids of the present application specifically includes the following technical solutions: The intelligent analysis method for electrochemical recognition of flavonoids includes the following steps: S1. Obtain a response signal matrix in time-frequency dimension, and perform frequency domain compression and nonlinear interaction mapping on the response signal matrix by a nonlinear two-frequency cross spectrum compression algorithm to obtain a one-dimensional nonlinear compressed current response sequence; based on the one-dimensional nonlinear compressed current response sequence, introduce a variable window integral singular response identification mechanism to calculate a singular response strength; S2. Based on the singular response strength, construct a singular response strength sequence, determine a singular response occurrence point, and obtain a singular response weighting coefficient; construct a local window tensor, combine the singular response weighting coefficient, and obtain a singular weighted fusion tensor; introduce an inversion mechanism based on current response gradient and time delay mapping to perform asynchronous reconstruction processing on the singular weighted fusion tensor to obtain a feature representation of the flavone target after asynchronous reconstruction processing; based on the feature representation of the flavone target after asynchronous reconstruction processing, intelligently analyze the flavonoid compound.

[0006] Preferably, S1 specifically comprises: In the implementation process of the nonlinear two-frequency cross spectrum compression algorithm, based on the response signal matrix in time-frequency dimension, a current derivative cooperative term is constructed to quantify the consistency and cooperativity of the flavonoid compound reaction trend under different excitation frequencies, and a compression kernel matrix is constructed by combining the frequency domain distance between the excitation frequencies and calculating the frequency interaction weight.

[0007] Preferably, S1 specifically comprises: Based on the compression kernel matrix, the response signal matrix in time-frequency dimension is weighted in the frequency dimension to generate a frequency spectrum compression representation, and a one-dimensional nonlinear compressed current response sequence is obtained.

[0008] Preferably, S1 specifically comprises: In the implementation process of the variable window integral singular response identification mechanism, based on the one-dimensional nonlinear compressed current response sequence, a local stationarity coefficient and a normalized periodic sinusoidal mapping term are constructed, and a steady-state jump response score of a time segment is calculated to obtain a singular response strength.

[0009] Preferably, S2 specifically comprises: The singular response strength sequence is processed to obtain a maximum value point set, and based on the maximum value point set, a threshold is determined to determine the singular response occurrence point.

[0010] Preferably, S2 specifically comprises: After normalizing the singular response strength corresponding to the singular response occurrence point, a singular response weighting coefficient is obtained; based on the singular response weighting coefficient, a local window tensor is weighted and fused to obtain a singular weighted fusion tensor; the local window tensor is obtained by feature extraction on the one-dimensional nonlinear compressed current response corresponding to the singular response occurrence point.

[0011] Preferably, S2 specifically comprises: In the implementation process of the inversion mechanism based on the current response gradient and time delay mapping, a time diffusion scale factor is introduced based on a singularly weighted fusion tensor, a gradient penalty term is combined, sliding integral reconstruction is performed for each time step, and an asynchronous response window vector of the flavone target object is obtained.

[0012] Preferably, S2 specifically comprises: Based on the asynchronous response window vector of the flavone target object, an association weight between the flavone target objects is obtained by combining an attention mechanism; and the asynchronous response window vector of the flavone target object is weighted and accumulated based on the association weight between the flavone target objects, so as to obtain a feature representation of the flavone target object after asynchronous reconstruction processing.

[0013] Preferably, S2 specifically comprises: All the feature representations of the flavonoid target objects after asynchronous reconstruction processing are combined to generate a feature representation tensor in a unified semantic space, and the feature representation tensor is input into an inference network to output a classification probability of each flavone target object, so as to realize intelligent analysis of flavonoid compounds.

[0014] The beneficial effects of the technical scheme of the present application are: 1. The nonlinear double-frequency cross-spectrum compression algorithm introduced in the present application fully utilizes the internal synergistic trend between the current responses of the flavone target objects under different excitation frequencies by constructing a compression kernel matrix regulated by a logistic normalization function, effectively retains the structural information contained in the cross-frequency coupling, and eliminates the interference of the low correlation frequency bands between frequencies. Compared with the traditional Fourier or wavelet transform method, the present application constructs the frequency interaction weight by introducing the square of the frequency difference and the current derivative synergy term, which significantly improves the structural consistency between the spectrum mapping and the physical response, suppresses the redundant channel noise while maintaining the integrity of the information, and improves the distinguishability of the response behavior in the subsequent analysis task.

[0015] 2. The present application introduces a variable window integral singular response recognition mechanism based on the one-dimensional nonlinear compressed current response sequence, which is characterized by constructing an integral combination of local stationarity coefficients and normalized periodic sine mapping terms to quantize and intensively express the transient current peaks formed by the short-time strong response of the flavonoid substances. Not only can the time segment of the redox transition be accurately judged, but also the problem of noise sensitivity in the traditional differential-threshold jump detection can be effectively avoided, and the actual situation that the response time scales of different flavonoid substances are inconsistent can be adapted, so as to ensure the robust capture ability of the short-time response behavior of various flavone target objects.

[0016] 3. In view of the response delay difference of flavonoid compounds under electrochemical excitation, the present application designs an inversion mechanism based on current response gradient and time delay mapping, constructs a kernel function with time diffusion scale factor centered on response delay, and performs nonlinear reconstruction on different response time windows. By introducing an independent reaction start time delay for each type of flavonoid target, the aggregation mode of different response behaviors on the time axis can be automatically adjusted, thereby unifying the reaction trajectory in logical time and breaking the limitation of "fixed window and synchronous sampling" in traditional convolutional neural networks. A highly adaptive feature reconstruction mechanism is established for time-asynchronous electrochemical detection tasks. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The intelligent analysis method for electrochemical recognition of flavonoid compounds described in the present application is shown in the flowchart. DETAILED DESCRIPTION

[0018] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined object, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0020] The specific scheme of the intelligent analysis method for electrochemical recognition of flavonoid compounds provided by the present application will be described below with reference to the drawings.

[0021] Referring to the drawings Figure 1 , which shows the flowchart of the intelligent analysis method for electrochemical recognition of flavonoid compounds provided by an embodiment of the present application, the method comprises the following steps: S1. Obtain a time-frequency dimension response signal matrix, and perform frequency domain compression and nonlinear interaction mapping by a nonlinear double-frequency cross spectrum compression algorithm to obtain a one-dimensional nonlinear compressed current response sequence; based on the one-dimensional nonlinear compressed current response sequence, introduce a variable window integral singular response recognition mechanism to calculate the singular response intensity; Obtain a time-frequency dimension response signal matrix from the flavonoid compound solution by an electrochemical sensor according to the set excitation frequency The time-frequency dimension response signal matrix is a two-dimensional dynamic current response data structure, wherein, is a sampling time sequence, is the th excitation potential corresponding frequency, i.e., excitation frequency; the electrochemical sensor is a microelectrode sensing platform based on a three-electrode system, including a working electrode, a reference electrode and an auxiliary electrode, having excitation control capability, response acquisition capability and frequency resolution capability.

[0022] Due to the weak but specific redox response peaks of different flavonoids in a specific frequency range, there is a cross-coupling in the frequency dimension and a drift randomness in the time dimension between the response patterns implied in the time-frequency dimension response signal matrix, therefore, a nonlinear double-frequency cross-spectrum compression algorithm is needed to compress and nonlinearly interact map the time-frequency dimension response signal matrix to retain the frequency spectrum collaborative patterns with physical significance, so as to extract the physically recognizable characteristic time sequence, and obtain a one-dimensional nonlinear compressed current response sequence; the specific implementation process of the nonlinear double-frequency cross-spectrum compression algorithm is: based on the logistic normalization function, a compression kernel matrix is constructed, wherein any element is expressed as:

[0023] is the frequency interaction weight, representing the coupling strength between the excitation frequency and the excitation frequency corresponding current response signal, for compression mapping, is an element of the compression kernel matrix; is the th excitation potential corresponding frequency, i.e., excitation frequency; is the square of the frequency difference, used to measure the frequency domain distance between the excitation frequencies; is the frequency compression sensitivity coefficient, used to control the influence degree of the frequency difference on the compression result, determined by cross-validation method, unit: Hz, reference value range is , which can be 500, and the cross-validation method is a technology known to those skilled in the art, which will not be described here; , respectively represent the current response signals collected at time and under the excitation frequencies and , is the first derivative of the current response signal, respectively representing the change rate of the current response signal with time under the excitation frequencies and , which can reflect the redox dynamic trend, and is obtained by numerical difference calculation; ​is the square term of normalized frequency difference, which is essentially a frequency space distance factor of Gaussian kernel function, and its physical meaning is to judge whether the excitation frequency and the "proximity" in the frequency domain; is the current derivative synergy term, which reflects the change trend of electrochemical response in time , under two different excitation frequencies and whether the change trend is fast and consistent at the same time, which realizes the measurement of the consistency and synergy of flavonoids under different excitation frequencies; is the criterion term of weight, which is used to control the compression correlation degree between frequency pairs, and is converted into the final frequency interaction weight by sigmoid function; Further, the response signal matrix in time-frequency dimension is weighted in frequency dimension using the compression kernel matrix, and a spectral compression representation is generated, obtaining a one-dimensional nonlinear compressed current response sequence:

[0024] wherein, is the compressed and fused current response time sequence, i.e. one-dimensional nonlinear compressed current response sequence, which is a one-dimensional time function obtained by fusing the current response signals under all frequency channels according to the frequency interaction weight; is the total number of excitation frequencies; is the response signal matrix under excitation frequency ; The significance of the above formula is to map the response signal matrix in time-frequency dimension to a one-dimensional nonlinear compressed current response sequence , the value of each time point of which is the combined strength of the self-correlation and cross-correlation behavior between multiple frequencies, which has the ability to remove non-structural interference while retaining the physical response characteristics, and serves as the input basis for subsequent singular steady-state extraction.

[0025] Based on the one-dimensional nonlinear compressed current response sequence , the current mutation point of redox reaction occurrence, i.e. steady-state jump point, is found. The steady-state jump behavior is a phenomenon of short-term high-response current produced by most flavones under certain excitation, which usually lasts for a very short time and is easily covered by conventional smoothing processing. Therefore, a variable window integral singular response identification mechanism is introduced, which uses a variable window integral method to calculate the singular response strength at each time point, and the integral formula is as follows:

[0026] wherein, is the steady-state step response score of the time segment, i.e. singularity response intensity; is the current integral starting time, representing the sampling moment; is the integral time window length, used to envelop a local signal, and is extracted based on the rise time judged by the extreme value derivative, with a reference value range of , which can be taken as 0.5; is the one-dimensional nonlinearly compressed current response at time ; is the time integral variable; is the first-order derivative of the one-dimensional nonlinearly compressed current response with respect to time, i.e. the current change rate, which is calculated by numerical difference; represents the maximum value of the one-dimensional nonlinearly compressed current response of the entire signal segment, used for normalization processing; is the exponential decay coefficient, used to control the weight decay rate of the far time point, and is determined by automatic optimization through gradient descent method, with a reference value range of , which can be taken as 5; is the time decay kernel, used to weaken the contribution of the time points far away from the center time point, so as to ensure that the closer to the integral starting point, the greater the contribution; is the local stationarity coefficient, which is close to 1 when the current changes gently, and tends to 0 when the current changes abruptly; is the normalized periodic sine mapping item, used to detect the response of periodic or short-time surge signals; is the time-domain gradient suppression item, used to suppress the burst interference at the place of sharp change.

[0027] S2. Based on the singularity response intensity, a singularity response intensity sequence is constructed, the singularity response occurrence points are determined, and the singularity response weighting coefficients are obtained; a local window tensor is constructed, combined with the singularity response weighting coefficients, to obtain a singularity weighted fusion tensor; an inversion mechanism based on current response gradient and time delay mapping is introduced to perform asynchronous reconstruction processing on the singularity weighted fusion tensor, to obtain the characteristic representation of the flavone target object after asynchronous reconstruction processing; based on the characteristic representation of the flavone target object after asynchronous reconstruction processing, the flavone compounds are intelligently analyzed.

[0028] Based on the singularity response intensity, the singularity response intensity sequence is processed by a local maximum value detection algorithm, such as non-maximum suppression with suppression threshold, to obtain a maximum value point set, and the points in the maximum value point set greater than the threshold are taken as the singularity response occurrence points, and the threshold is set according to the mean value of the historical singularity response intensity, and the total number of singularity response occurrence points is recorded as ; The sum of the singular response intensities corresponding to all singular response occurrence points is taken as a normalization term, and the singular response intensity corresponding to each singular response occurrence point is normalized to obtain a singular response weighting coefficient and finally generates a singular weighted fusion tensor:

[0029] wherein, is the feature representation tensor after fusion at time point , i.e., the singular weighted fusion tensor; is the index of the singular response occurrence point; is obtained by performing feature extraction on the one-dimensional nonlinearly compressed current response of the section corresponding to the singular response occurrence point, including the current response signal, the rate of change (first derivative), the local energy, the peak symmetry index, the positive and negative direction jump marker, etc. The length of the local window is obtained by calculating half of the ratio of the total length of the sampling time sequence to the total number of singular response occurrence points, and then rounding down.

[0030] Further, in order to avoid the problem of different flavonoids responding at different times, i.e., due to the time delay of electrochemical reaction, different substances may respond at different delays after excitation, it is necessary to perform asynchronous reconstruction processing on the above singular weighted fusion tensor; an inversion mechanism based on current response gradient and time delay mapping is introduced, and sliding integral reconstruction is performed at each time step to obtain asynchronous response window vectors of various flavonoid target substances, and combined with the existing attention mechanism, the feature representation of the flavonoid target substance after asynchronous reconstruction processing is obtained. The specific calculation formula of the asynchronous response window vector of the flavonoid target substance is as follows:

[0031] wherein, is the feature representation tensor after fusion at time point is the asynchronous response window vector of the flavonoid target substance, and its physical meaning is that, under the premise of considering the reaction start-up delay , the time-aggregated inversion value of the electrochemical expression of the flavonoid target substance at a certain moment; is the estimated reaction start-up time delay of the flavonoid target substance, which is determined by backpropagation optimization, and the reference value range is , and can be 1.0; is a time diffusion scale factor, which is used to control the sharpness or flatness of the integral kernel function within the time window, and is determined based on the existing kernel weighted sliding integral model, and the reference value range is , and can be 0.1; It is the weighting function in the integral kernel, used to simulate the maximum response of flavonoids near the reaction initiation time, to ensure decay at a more distant time. It is a one-dimensional nonlinear compressive current response at time point The rate of change of current, used to reflect the drasticness of current change, is determined by numerical difference; It is a gradient penalty term, which avoids gradient explosion caused by drastic current jumps by performing a logarithmic smoothing transformation on the rate of change of current. At a certain point in time The singular weighted fusion tensor on; It is the weighted core part of the asynchronous time-compensated integral structure, which describes the electrochemical response energy of the flavonoid target near its characteristic reaction initiation time.

[0032] Furthermore, based on the asynchronous response window vectors of flavonoid targets, an existing attention mechanism is used to determine the association weights between the two classes of flavonoid targets. Then, based on these association weights, the asynchronous response window vectors of the flavonoid targets are weighted and accumulated to obtain the [followed by the previous step]. Flavonoid target at time point Feature representation after asynchronous reconstruction Furthermore, the feature representations of all flavonoid target compounds after asynchronous reconstruction are merged to generate a feature representation tensor in a unified semantic space. The feature representation tensor is then input into an existing inference network, such as a temporal convolutional network, to output classification probabilities and obtain the classification probability of each type of flavonoid target, thereby achieving intelligent analysis of flavonoid compounds.

[0033] In summary, an intelligent analytical method for the electrochemical recognition of flavonoids has been developed.

[0034] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0035] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0036] The above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An intelligent analysis method for electrochemical recognition of flavonoids, characterized in that, The method comprises the following steps: S1. Obtain a response signal matrix in time-frequency dimension, and perform frequency domain compression and nonlinear interaction mapping on the response signal matrix by a nonlinear two-frequency cross spectrum compression algorithm to obtain a one-dimensional nonlinear compressed current response sequence; Based on the one-dimensional nonlinear compressed current response sequence, a variable window integral singular response identification mechanism is introduced, and a singular response strength is calculated; S2. Based on the singular response strength, a singular response strength sequence is constructed, a singular response occurrence point is determined, and a singular response weighting coefficient is obtained; A local window tensor is constructed, and the singular response weighting coefficient is combined to obtain a singular weighted fusion tensor; an inversion mechanism based on current response gradient and time delay mapping is introduced, and asynchronous reconstruction processing is performed on the singular weighted fusion tensor to obtain a characteristic representation of the flavone target object after asynchronous reconstruction processing; and the flavone compound is intelligently analyzed based on the characteristic representation of the flavone target object after asynchronous reconstruction processing. 2.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 1, characterized in that, The S1 specifically comprises: In the implementation process of the nonlinear two-frequency cross spectrum compression algorithm, based on the response signal matrix in time-frequency dimension, a current derivative cooperative term is constructed, the consistency and cooperativity of the flavone compound reaction trend under different excitation frequencies are quantified, the frequency interaction weight is calculated in combination with the frequency domain distance between the excitation frequencies, and a compression kernel matrix is constructed. 3.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 2, characterized in that, The S1 specifically comprises: Based on the compression kernel matrix, the response signal matrix in time-frequency dimension is weighted in the frequency dimension to generate a frequency spectrum compression representation, and a one-dimensional nonlinear compressed current response sequence is obtained. 4.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 1, characterized in that, The S1 specifically comprises: In the implementation process of the variable window integral singular response identification mechanism, based on the one-dimensional nonlinear compressed current response sequence, a local stationarity coefficient and a normalized periodic sine mapping term are constructed, and a steady-state jump response score of a time segment is calculated to obtain a singular response strength. 5.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 1, characterized in that, The S2 specifically comprises: The singular response strength sequence is processed to obtain a maximum value point set, and based on the maximum value point set, in combination with a preset threshold, a singular response occurrence point is determined. 6.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 5, characterized in that, The S2 specifically comprises: After the singular response strength corresponding to the singular response occurrence point is normalized, a singular response weighting coefficient is obtained; based on the singular response weighting coefficient, a local window tensor is weighted and fused to obtain a singular weighted fusion tensor; and the local window tensor is obtained by feature extraction on the one-dimensional nonlinear compressed current response corresponding to the singular response occurrence point. 7.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 1, characterized in that, The S2 specifically comprises: In the implementation process of the inversion mechanism based on current response gradient and time delay mapping, based on the singular weighted fusion tensor, a time diffusion scale factor is introduced, a gradient penalty term is combined, and sliding integral reconstruction is performed on each time step to obtain an asynchronous response window vector of the flavone target object. 8.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 7, characterized in that, The S2 specifically comprises: Based on the asynchronous response window vector of the flavone target object, in combination with an attention mechanism, an association weight between the flavone target objects is obtained; based on the association weight between the flavone target objects, the asynchronous response window vector of the flavone target object is weighted and accumulated to obtain a characteristic representation of the flavone target object after asynchronous reconstruction processing. 9.The intelligent analysis method for electrochemical recognition of flavonoids according to claim 8, characterized in that, The S2 specifically comprises: The feature representation tensors under the unified semantic space are generated by merging the feature representations of all flavonoid target objects after asynchronous reconstruction processing, and the feature representation tensors are input into an inference network, and the classification probability of each flavonoid target object is output, so that intelligent analysis of flavonoid compounds is realized.

Images