CN-122016978-A - Electronic tongue analysis method based on multi-frequency pulse voltammetric time spectrum fingerprint

CN122016978ACN 122016978 ACN122016978 ACN 122016978ACN-122016978-A

Abstract

The invention discloses an electronic tongue analysis method based on multi-frequency pulse voltammetry frequency spectrum fingerprints, which relates to the field of electrochemical analysis and comprises the steps of initializing an electronic tongue system, carrying out joint judgment through constant temperature stability, angle in place and electrode stability criteria, outputting a segmented multi-frequency pulse voltammetry excitation signal to a working electrode, obtaining a complete current response sequence through a FIFO buffer period pulling mechanism, sequentially gating a plurality of working electrode channels to obtain the current response sequence of each channel, constructing a two-dimensional Mel frequency spectrum fingerprint, obtaining a qualitative identification result and an index quantitative prediction result of a sample based on a time spectrum feature-attention fusion model and a time spectrum feature-regression model, generating a contribution degree heat map by combining a gradient weighting type activation mapping strategy, and outputting an interpretability analysis result. The accuracy and the result reliability of the electronic tongue in complex sample detection are improved through closed-loop stable scheduling, hardware channel multiplexing calibration and timely spectral fingerprint deep learning analysis.

Inventors

TANG XIUYING
HE MENG
YU SHI
XU JIAQI
ZHANG BIN

Assignees

中国农业大学

Dates

Publication Date: 20260512
Application Date: 20260305

Claims (10)

1. An electronic tongue analysis method based on multi-frequency pulse voltammetry frequency spectrum fingerprint is characterized by comprising the following steps: S1, initializing an electronic tongue system, establishing communication with lower hardware and completing parameter loading and channel mapping; S2, carrying out joint judgment through a constant temperature stability criterion, an angle in-place stability criterion and an electrode stability criterion, and opening a signal acquisition window when all stability criteria are met simultaneously; S3, outputting a segmented multi-frequency pulse volt-ampere excitation signal to a working electrode in the signal acquisition window; S4, collecting response current signals of the working electrode in a continuous collection mode, and obtaining a complete current response sequence through a FIFO buffer memory periodic pulling mechanism; S5, sequentially gating a plurality of working electrode channels, repeatedly executing steps S3 to S4 on each channel, and setting a waiting period after channel switching to obtain a current response sequence of each channel; s6, constructing a corresponding two-dimensional Mel spectrum fingerprint based on the current response sequence of each channel; S7, inputting the two-dimensional Mel frequency spectrum fingerprint characteristics into a pre-trained time spectrum characteristic-attention fusion model and a time spectrum characteristic-regression model to respectively obtain a qualitative recognition result and an index quantitative prediction result of the sample; and S8, combining a gradient weighting type activation mapping strategy, generating a corresponding contribution degree heat map for the qualitative identification result and the index quantitative prediction result, and outputting an interpretability analysis result.
2. The method for analyzing an electronic tongue based on multi-frequency pulse voltammetry spectrum fingerprints according to claim 1, wherein in S2, the constant temperature stability criterion is a current temperature T (T) and a target temperature The difference of (2) satisfies And last for a preset period of time , Is a preset temperature tolerance threshold; The angle in-place stability criterion is that after the movement mechanism enters an in-place state, the preset damping waiting time is continued ; The electrode stability criterion is the baseline drift of the current response during a pre-sampling period before excitation begins Variance of And the root mean square noise RMS are smaller than the corresponding preset threshold value.
3. The method according to claim 1, wherein in S3, the segmented multi-frequency pulse voltammetric excitation signal is composed of a set of discrete reference potential sequences By generating the reference potential sequence Setting different repetition times or time steps in different frequency bands to form segmented excitation waveforms at least comprising a low frequency band, a medium frequency band and a high frequency band.
4. The method for electronic tongue analysis based on multi-frequency pulse voltammetry frequency spectrum fingerprint according to claim 1, wherein in S4, the FIFO buffer period pulling mechanism comprises: s41, after starting the continuous acquisition mode, periodically inquiring the readable sample number in the FIFO buffer memory ; S42, according to the total number of the current required samples And single read upper limit Determining the current reading quantity ; S43, read The sample data are additionally stored in a response sequence buffer corresponding to the current channel; S44, before stopping the continuous acquisition operation, the steps S41 to S43 are circularly executed until all the residual data in the FIFO buffer memory are read.
5. The method for analyzing an electronic tongue based on multi-frequency pulse voltammetry spectrum fingerprint according to claim 1, wherein in S5, a plurality of working electrode channels are sequentially gated, and in particular, three-bit address signals (a, B, C) are output to an analog multiplexer through an upper computer; the analog multiplexer inputs channels at a plurality of working electrodes according to the address signals (A, B, C) One of the channels is conducted with a common end COM, and the common end COM is connected to the input end of the constant potential measuring circuit.
6. The method for electronic tongue analysis based on multi-frequency pulse voltammetric frequency spectrum fingerprinting according to claim 1, wherein S6 comprises: Framing and windowing are carried out on the current response sequence x [ n ] to obtain a kth frame windowing signal ; Windowing signal for each frame Performing short-time Fourier transform to obtain a time-frequency power spectrum P (k, f), wherein f is a linear frequency; through a set of triangular Mel filter banks Mapping the time-frequency power spectrum P (k, f) to a Mel frequency domain to obtain a Mel frequency spectrum Wherein m is a mel-band index; for the mel spectrum Carrying out logarithmic compression and normalization processing to obtain final two-dimensional Mel frequency spectrum fingerprint characteristics Wherein Delta is an anti-zero constant, which is a normalization function.
7. The method for electronic tongue analysis based on multi-frequency pulsed voltammetric spectrum fingerprinting according to claim 1, wherein in S7, the time-spectral feature-attention fusion model comprises: The multi-scale feature extraction module is used for extracting features under different receptive fields from the input two-dimensional Mel spectrum fingerprint; the channel attention module is used for learning and recalibrating importance weights of different characteristic channels; the spatial attention module is used for learning and recalibrating importance weights of different spatial positions in the feature map; And the classification head is used for outputting the class probability distribution of the sample.
8. The method for electronic tongue analysis based on multi-frequency pulse voltammetric spectrum fingerprint according to claim 1, wherein in S7, the time-spectrum feature-regression model comprises: The convolution feature extraction trunk is used for extracting a primary feature map from an input two-dimensional Mel spectrum fingerprint; The global dependency modeling module is used for modeling the long-range dependency relationship of the primary feature map through a self-attention mechanism based on a transducer encoder structure; and the regression prediction head is used for outputting the continuous predicted value of the target index.
9. The method for electronic tongue analysis based on multi-frequency pulse voltammetry spectral fingerprinting according to claim 1, wherein in S8, the gradient weighted class activation mapping strategy comprises: For the target class c or regression task, calculating the feature map output by the last convolution layer of the network Gradient weights of (2) : Wherein H and W are the spatial height and width of the feature map, The score or the output value of the model for the target c is obtained, k is a feature map index, and i and j are spatial position indexes; Weighting the gradient And corresponding feature map Weighted summation is carried out, and a heat map is obtained through activation of a ReLU function : The heat map is processed Upsampling to the same size as the input fingerprint feature F (k, m) generates a visualized contribution heat map.
10. An electronic tongue system for implementing the method of any one of claims 1 to 9, comprising: An electrode array module including a plurality of working electrodes, a reference electrode and a counter electrode; The constant potential excitation and measurement module is used for generating and outputting the segmented multi-frequency pulse volt-ampere excitation signal to the electrode, measuring the response current of the working electrode in real time and converting the response current into a voltage signal; The channel selection and multiplexing module comprises an analog multiplexer and a driving circuit thereof, and is used for switching and gating among a plurality of working electrode channels; The data acquisition and control module comprises a path of digital-to-analog converter for outputting excitation waveforms, a path of analog-to-digital converter for acquiring current response signals and a plurality of paths of digital input-output interfaces for controlling peripheral equipment; the movement and temperature control execution module is used for controlling the lifting and rotating movement of the sample platform and maintaining the temperature of the solution constant and providing an in-place state and temperature feedback signal; and the upper computer software module is used for executing all the processes of system initialization, stability criterion judgment, acquisition window control, excitation signal output, data acquisition, fingerprint construction, model reasoning, heat map generation and result storage.

Description

Electronic tongue analysis method based on multi-frequency pulse voltammetric time spectrum fingerprint Technical Field The invention relates to the field of electrochemical analysis, in particular to an electronic tongue analysis method based on multi-frequency pulse voltammetry frequency spectrum fingerprints. Background In the fields of quality control, flavor evaluation and safety monitoring of the food industry, an electronic tongue is a rapid analysis technology based on a sensor array, and qualitative distinction and quantitative prediction of physicochemical indexes and sensory attributes of liquid samples are realized by simulating a human taste perception mechanism. However, the measurement process of the existing electronic tongue system is seriously dependent on preset fixed time sequence or experience judgment of operators, the output of an excitation signal, the starting of data acquisition and the switching among multiple channels are often based on simple delay control, the actual state of the system cannot be sensed and responded in real time, the acquisition is fast and rapid when the temperature is not balanced, the mechanical movement is not completely static or an electrode interface is not stable, random errors are introduced, and the reliability of a measurement result is difficult to ensure when a food system with complex components is handled. Furthermore, in order to realize parallel detection of the multi-electrode array, in the prior art, an independent front end measurement circuit is often configured for each working electrode, or a manual switching mode is adopted, and inherent differences of hardware parameters, such as operational amplifier gain, zero drift, wiring distribution parameters and the like, among all channels are difficult to be completely consistent. In addition, in the signal processing and analysis level, the traditional method generally directly utilizes time domain feature points or simple transformation features of a current-time response curve, and fails to fully mine rich kinetic information contained in multi-band pulse voltammetric excitation, meanwhile, a conventional machine learning model has limited modeling capability on deep association and global context dependency between features, and the decision process lacks visual physical explanation, so that a trusted evidence chain is difficult to form in a quality control scene requiring high credibility. Therefore, how to design an electronic tongue analysis method based on multi-frequency pulse voltammetry frequency spectrum fingerprint, which can realize intelligent closed-loop scheduling, ensure multi-channel consistent measurement and have stronger interpretable analysis capability is a problem to be solved by the technicians in the field. Disclosure of Invention In view of the above, the invention provides an electronic tongue analysis method based on multi-frequency pulse voltammetry frequency spectrum fingerprint, which aims to solve the problems of poor acquisition consistency, insufficient multi-channel data comparability, limited feature expression capability and the like of the traditional electronic tongue under a complex food system, and improves the capability of stably and reliably identifying and predicting the quality of complex samples by integrating closed-loop scheduling control, calibratable multiplexing measurement hardware and a deep learning analysis model based on time spectrum fingerprint. In order to achieve the above purpose, the present invention adopts the following technical scheme: In a first aspect, the present invention provides an electronic tongue analysis method based on multi-frequency pulse voltammetric frequency spectrum fingerprint, comprising the steps of: S1, initializing an electronic tongue system, establishing communication with lower hardware and completing parameter loading and channel mapping; S2, carrying out joint judgment through a constant temperature stability criterion, an angle in-place stability criterion and an electrode stability criterion, and opening a signal acquisition window when all stability criteria are met simultaneously; S3, outputting a segmented multi-frequency pulse volt-ampere excitation signal to a working electrode in the signal acquisition window; S4, collecting response current signals of the working electrode in a continuous collection mode, and obtaining a complete current response sequence through a FIFO buffer memory periodic pulling mechanism; S5, sequentially gating a plurality of working electrode channels, repeatedly executing steps S3 to S4 on each channel, and setting a waiting period after channel switching to obtain a current response sequence of each channel; s6, constructing a corresponding two-dimensional Mel spectrum fingerprint based on the current response sequence of each channel; S7, inputting the two-dimensional Mel frequency spectrum fingerprint characteristics into a pre-trained time spectrum characteris