A terahertz metamaterial enhanced spectral detection system

Abstract

The application belongs to the technical field of terahertz detection, and discloses a terahertz metamaterial enhanced spectral detection system, which comprises the following steps: constructing a metamaterial enhancement factor tensor matrix according to the interaction between a terahertz wave and a metamaterial; based on the metamaterial enhancement factor tensor matrix, optimizing local field enhancement and surface plasmon resonance of a target sample to obtain a characteristic enhancement spectroscopy mapping function; based on the characteristic enhancement spectroscopy mapping function, constructing a multi-dimensional detection characteristic space; based on the multi-dimensional detection characteristic space, forming a spectral signal enhancement topology network; based on the spectral signal enhancement topology network, generating an enhanced spectral feature discriminant function; and based on the enhanced spectral feature discriminant function, obtaining a high-precision terahertz spectral detection result; the application can mine the internal correlation between spectral features, enhance the comprehensive characterization of sample information, and thus realize high-precision terahertz spectral detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of terahertz detection, and more particularly, to a terahertz metamaterial enhanced spectral detection system. BACKGROUND

[0002] With the rapid development of terahertz technology, terahertz waves (frequency range of about 0.1 THz to 10 THz) are widely used in non-destructive testing, quality control, safety detection, material characterization and other fields due to their strong penetration, non-ionization, non-invasive and other advantages. At the same time, metamaterials, as a new type of artificial material, have the ability to precisely control electromagnetic response, especially in the terahertz waveband, showing effective adjustment of wave absorption, transmission and reflection performance, significantly improving the flexibility and efficiency of the terahertz detection system. However, the current terahertz detection technology still faces some challenges, including unclear interaction mechanism between terahertz waves and materials, low efficiency of spectral data processing, insufficient sensitivity and anti-interference ability of the system, and high requirements for sample morphology, etc. Therefore, an intelligent terahertz metamaterial spectral detection system is urgently needed to solve the limitations of existing terahertz detection technology.

[0003] The patent with publication number CN119574497A discloses a detection method for terahertz trace molecule fingerprint spectrum; including: collecting the time domain signal of the sample under the condition that the sample layer and the substrate layer composite sample are placed in the terahertz wave radiation area for radiation; the sample layer is obtained by coating the analyte on the substrate layer formed based on the selected substrate material; based on fast Fourier transform, the time domain signal of the sample is converted into the frequency domain signal of the sample; based on the frequency domain signal of the sample and the reference frequency domain signal obtained by testing the air, the transmission spectrum is obtained, and the absorption fingerprint signal of the analyte is extracted from the transmission spectrum; based on the absorption fingerprint signals of analytes with different contents, a standard equation between the content of the analyte and the fingerprint signal intensity is established. The detection method has low cost, short detection time and simple operation, and can realize ultra-wide spectrum detection of the analyte.

[0004] However, the above-mentioned technology realizes terahertz spectral detection, but relies on traditional Fourier transform and filtering algorithm to process spectral signals, which is difficult to fully meet the extraction needs of non-linear and non-stationary signal characteristics in complex samples, resulting in limited detection accuracy. Moreover, the analysis of spectral data is still based on a linear model to establish a standard equation, lacking in-depth mining of potential correlations and complex structures between spectral features. In addition, the above-mentioned technology has limited processing in terms of signal-to-noise ratio improvement and anti-interference ability, which is difficult to meet the actual application needs in higher precision and more complex environments.

[0005] In view of this, the present application proposes a terahertz metamaterial enhanced spectral detection system to solve the above-mentioned problems. SUMMARY

[0006] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a terahertz metamaterial-enhanced spectral detection system, comprising:

[0007] The matrix construction module is used to construct a microscopic electromagnetic response model based on the interaction between terahertz waves and metamaterials, perform resonance characteristic analysis on the microscopic electromagnetic response model, and construct the metamaterial enhancement factor tensor matrix.

[0008] The function mapping module is used to perform local field enhancement and surface plasmon resonance optimization on target samples based on the metamaterial enhancement factor tensor matrix, and obtain the feature enhancement spectroscopic mapping function.

[0009] The spatial construction module is used to perform nonlinear mode decomposition on the feature-enhanced spectral mapping function using the Hilbert-Huang transform, extract the intrinsic spectral component set, and construct a multi-dimensional detection feature space.

[0010] The topology forming module is used to perform topological invariant calculations and quantum correlation analysis on the multi-dimensional detection feature space to form a spectral signal enhancement topology network.

[0011] The function generation module is used to perform multi-scale entropy metric evaluation and signal-to-noise ratio adaptive optimization based on the spectral signal enhancement topology network, and generate enhanced spectral feature discrimination functions.

[0012] The intelligent detection module combines variational Bayesian inference and sensitivity analysis methods to perform uncertainty quantification and sensitivity analysis on the enhanced spectral feature discrimination function, thereby obtaining high-precision terahertz spectral detection results.

[0013] Furthermore, the method for constructing the metamaterial enhancement factor tensor matrix includes:

[0014] Preset simulation parameters are used to simulate the interaction between terahertz waves and metamaterials using electromagnetic simulation software, obtaining electromagnetic field distribution data, transmission coefficient data, and reflection coefficient data. Based on the simulation parameters, electromagnetic field distribution data, transmission coefficient data, and reflection coefficient data, a corresponding microscopic electromagnetic response model is constructed.

[0015] Based on the transmission coefficient and reflection coefficient data, obtain the transmission coefficient and reflection coefficient corresponding to each frequency; construct transmission coefficient curves and reflection coefficient curves respectively based on the transmission coefficient and reflection coefficient corresponding to each frequency; mark the frequencies in the transmission coefficient curve that correspond to the minimum transmission coefficient as candidate resonant frequencies; mark the frequencies in the reflection coefficient curve that correspond to the maximum reflection coefficient as candidate resonant frequencies; square the mode length of the transmission coefficient and the mode length of the reflection coefficient corresponding to each frequency in turn, and then add them together to obtain the energy sum corresponding to each frequency; subtract the energy sum corresponding to each frequency from the sum to obtain the absorptivity corresponding to each frequency; construct an absorptivity curve based on the absorptivity corresponding to each frequency; mark the frequencies in the absorptivity curve that correspond to the maximum absorptivity as candidate resonant frequencies.

[0016] Electromagnetic field distribution data corresponding to each candidate resonant frequency is acquired, and the local enhancement factor corresponding to each candidate resonant frequency is calculated. The local enhancement factor includes the local electric field enhancement factor and the local magnetic field enhancement factor. Each local electric field enhancement factor is compared with a preset factor threshold. Candidate resonant frequencies with a local electric field enhancement factor greater than or equal to the factor threshold are marked as resonant frequencies, while candidate resonant frequencies with a local electric field enhancement factor less than the factor threshold are not marked. Based on the local enhancement factor corresponding to each resonant frequency, an enhancement factor tensor matrix corresponding to each resonant frequency is constructed. The enhancement factor tensor matrices corresponding to each resonant frequency are merged to construct a metamaterial enhancement factor tensor matrix.

[0017] Furthermore, the method for obtaining the feature-enhanced spectroscopic mapping function includes:

[0018] The basic optical parameters of the target sample are obtained, including the complex permittivity, Raman scattering cross section, molecular dipole moment, and surface molecular density. The incident electric field vector corresponding to each resonant frequency is obtained from the simulation parameters. Each incident electric field vector is multiplied by the corresponding enhancement factor tensor matrix to obtain the sample local field corresponding to each resonant frequency. The square of the modulus corresponding to each sample local field is divided by the square of the modulus corresponding to the incident electric field vector to obtain the local field enhancement factor for each resonant frequency. The resonant frequency with the largest local electric field enhancement factor in the metamaterial enhancement factor tensor matrix is ​​marked as the dominant frequency. The sample characteristic frequency is obtained from the complex permittivity. Based on the dominant frequency and the sample characteristic frequency, the surface plasmon resonance frequency is calculated.

[0019] The incident frequency is obtained from the simulation parameters, and the Raman shift is obtained from the Raman scattering cross section. The scattering frequency is obtained by subtracting the Raman shift from the incident frequency. Based on the local field enhancement factor of each resonant frequency, the local field enhancement factor corresponding to the scattering frequency and the incident frequency is calculated using an interpolation algorithm. The radiative attenuation rate and the non-radiative attenuation rate are obtained. The total attenuation rate is obtained by adding the radiative attenuation rate to the non-radiative attenuation rate. The quantum efficiency factor is obtained by dividing the radiative attenuation rate by the total attenuation rate. The characteristic enhancement spectral mapping function is obtained by multiplying the quantum efficiency factor, the local field enhancement factor of the scattering frequency, the local field enhancement factor of the incident frequency, and the Raman scattering cross section in sequence.

[0020] Furthermore, the method for calculating the surface plasmon resonance frequency includes:

[0021] The molecular square is obtained by squaring the mode length corresponding to the molecular dipole moment; the sample square is obtained by squaring the mode length corresponding to the local field of the dominant frequency; the molecular square, sample square, and surface molecular density are multiplied sequentially, divided by the reduced Planck constant, and then the square root is taken to obtain the coupling strength; the loss factor is obtained by fitting the Lorentz model at the dominant frequency based on the complex permittivity; the loss factor is multiplied by the imaginary unit, added to the dominant frequency, and then subtracted from the sample characteristic frequency to obtain the complex detuning factor; the frequency shift is obtained by squaring the coupling strength and dividing by the complex detuning factor; the surface plasmon resonance frequency is obtained by adding the frequency shift to the dominant frequency.

[0022] Furthermore, the method for extracting the intrinsic spectral component set includes:

[0023] Decompose the feature-enhanced spectroscopy mapping function into Each intrinsic spectral component has an intrinsic spectral component, and a Hilbert transform is performed on each intrinsic spectral component to obtain its amplitude spectrum and instantaneous frequency spectrum. The input is a positive integer; the signal-to-noise ratio (SNR) and contribution of each intrinsic spectral component are calculated sequentially and compared with the corresponding component threshold in the preset threshold set; if both the SNR and contribution are greater than or equal to the corresponding component threshold, the corresponding intrinsic spectral component is marked as a superior spectral component; the threshold set includes the component thresholds corresponding to the SNR and contribution; if there is an input with an SNR or contribution less than the corresponding component threshold, the corresponding intrinsic spectral component is not marked; based on all superior spectral components, an intrinsic spectral component set is constructed.

[0024] The method for constructing a multi-dimensional detection feature space includes:

[0025] The intrinsic spectral components are concentrated, and the amplitude spectrum and instantaneous frequency spectrum corresponding to each superior spectral component are regarded as a set of components, with each set of components corresponding to a superior spectral component. Each set of components is input into a trained feature extraction model to extract spectral feature data. The feature extraction model is a deep neural network model. Based on the spectral feature data corresponding to each superior spectral component, a multi-dimensional detection feature space is constructed.

[0026] Furthermore, the decomposition of the feature-enhanced spectral mapping function into... The steps for identifying each intrinsic spectral component include:

[0027] Step S101: Mark the feature-enhanced spectral mapping function as a mapping function, and obtain all the maximum and minimum points in the mapping function;

[0028] Step S102: Construct the upper envelope using cubic spline interpolation based on all maximum points;

[0029] Step S103: Construct the lower envelope using cubic spline interpolation based on all local minima;

[0030] Step S104: Take the average value of the upper and lower envelopes to obtain the mean curve;

[0031] Step S105: Subtract the mean curve from the mapping function to obtain the candidate spectral components, and determine whether to retain the candidate spectral components; if not retained, use the candidate spectral components as the mapping function and return to step S101; if retained, subtract the candidate spectral components from the mapping function to obtain the residual components, and proceed to step S106.

[0032] Step S106: Determine whether the residual component is a monotonic function; if not, use the residual component as the mapping function and return to step S101; if yes, retain the remaining component. Each candidate spectral component is used as an intrinsic spectral component;

[0033] In step S105, the method for determining whether to retain candidate spectral components includes:

[0034] The total number of maxima and minima in the candidate spectral components is counted and marked as the number of extreme points; the number of zero-crossing points in the candidate spectral components is counted and marked as the number of zeros, where a zero-crossing point is the intersection of the candidate spectral component with the zero axis; the number of extreme points is subtracted from the number of zeros to obtain the difference; if the difference is greater than 1, the candidate spectral component is not retained; if the difference is less than or equal to 1, the candidate spectral component is retained.

[0035] In step S106, the method for determining whether the residual component is a monotonic function includes:

[0036] The total number of maximum and minimum points in the residual components is counted and marked as the residual quantity. The residual quantity is compared with a preset monotonic threshold. If the residual quantity is greater than or equal to the monotonic threshold, the residual component is a monotonic function; if the residual quantity is less than the monotonic threshold, the residual component is not a monotonic function.

[0037] Furthermore, the method for forming the spectral signal enhancement topology network includes:

[0038] In the multi-dimensional detection feature space, the spectral feature data corresponding to each superior spectral component are normalized. A dimensionality reduction algorithm is used to project the multi-dimensional detection feature space onto a three-dimensional topological space, where feature points in the three-dimensional topological space correspond one-to-one with the superior spectral components of the multi-dimensional detection feature space. A clustering algorithm is then used to cluster the feature points in the three-dimensional topological space to obtain... Sub-topological spaces The integer is greater than 1; in each sub-topological space, three adjacent feature points form a triangular unit, and the Berry curvature of each triangular unit is calculated; the Berry curvatures of all corresponding triangular units in each sub-topological space are summed sequentially, and then divided by . The first Chern number of each sub-topological space is obtained; the Delaunay triangulation algorithm is used to construct the triangulation network of each sub-topological space, and the Euler characteristic of each sub-topological space is calculated based on the triangulation network; the Euler characteristic and the first Chern number are used as the topological invariants of each sub-topological space.

[0039] Each superior spectral component is mapped to a quantum state, and the quantum correlation degree between any two quantum states is calculated. A correlation threshold is preset, which is set by those skilled in the art based on the actual situation. Each quantum correlation degree is compared with the preset correlation threshold. If the quantum correlation degree is greater than or equal to the correlation threshold, the corresponding two quantum states are considered as a quantum set. If the quantum correlation degree is less than the correlation threshold, the corresponding two quantum states are not considered as a quantum set. Each superior spectral component is treated as a node, and the topological invariants and normalized spectral feature data of each node are used as the corresponding node attributes. Correlation edges are established between the nodes corresponding to each quantum set, and the weight of the correlation edge is the quantum correlation degree of the corresponding quantum set. Based on all nodes, node attributes, and correlation edges, a spectral signal enhancement topology network is formed.

[0040] Furthermore, the method for calculating the Berry curvature is as follows: three feature points in the triangular unit are randomly labeled as the first point, the second point, and the third point; the cross product of the second point and the third point is calculated to obtain the cross product point; the dot product of the cross product point and the first point is calculated to obtain the dot product value; the modulus of the first point, the second point, and the third point is calculated in sequence and multiplied in sequence to obtain the total modulus; the dot product value is divided by the total modulus, and then the Berry curvature is calculated using the arctangent function.

[0041] The Euler characteristic number is calculated as follows: the number of vertices, edges, and faces in the triangulation network are counted sequentially. The number of vertices is the number of feature points, the number of edges is the number of edges between all feature points, and the number of faces is the number of triangular units. The Euler characteristic number is obtained by subtracting the number of edges from the number of vertices and adding the number of faces.

[0042] The method for calculating the quantum correlation is as follows: calculate the inner product between two quantum states and label it as similarity; calculate the inner product of each quantum state itself and label it as norm; multiply the two norms and then take the square root to obtain the norm product; divide the similarity by the norm product to obtain the quantum correlation.

[0043] Furthermore, the method for generating the enhanced spectral feature discrimination function includes:

[0044] Based on the node attributes of each node in the spectral signal enhancement topology network, an attribute sequence is constructed; a scale value is set, and based on the scale value, the attribute sequence is coarsened to... Granulated sequences, The value is an integer greater than 1; calculate the multivariate sample entropy of each granulated sequence in turn, multiply it by the weight coefficients in the preset first weight set, and then add them in turn to obtain the multiscale entropy value; calculate the signal-to-noise ratio of each node, multiply the multiscale entropy, the signal-to-noise ratio of each node and the topological invariant by the weight coefficients in the preset second weight set, and then add them in turn to obtain the node weight of each node.

[0045] Each associated edge is multiplied by the average of the weights of the two corresponding nodes to obtain the adjustment edge. A preset edge threshold is set, and each adjustment edge is compared with the edge threshold. Adjustment edges with values ​​greater than or equal to the edge threshold are retained in the spectral signal enhancement topology network, while those with values ​​less than the edge threshold are deleted from the spectral signal enhancement topology network. The spectral signal enhancement topology network is mapped to the feature space to construct a feature vector, and the feature vector is regularized to obtain a regularized processing vector. The feature vector includes all node attributes, multi-scale entropy values, signal-to-noise ratio, and adjustment edges in the spectral signal enhancement topology network. Based on the regularized processing vector, a feature judgment model is constructed using the random forest algorithm, and the constructed feature judgment model is transformed into an enhanced spectral feature discrimination function.

[0046] Furthermore, the method for obtaining high-precision terahertz spectral detection results includes:

[0047] Variational Bayesian inference is used to quantify the uncertainty of the enhanced spectral feature discriminant function and obtain confidence intervals. Sensitivity analysis is then performed on the enhanced spectral feature discriminant function to obtain the first-order sensitivity index and the total sensitivity index. Based on the confidence intervals, the first-order sensitivity index, and the total sensitivity index, an optimization algorithm is used to optimize the model hyperparameters of the feature judgment model. The optimization objective is to minimize the confidence interval width while minimizing the model's prediction error. Based on the optimized feature judgment model, the optimized enhanced spectral feature discriminant function is obtained and marked as the optimized discriminant function. High-precision terahertz spectral detection results are obtained based on the output of the optimized discriminant function.

[0048] The technical effects and advantages of the terahertz metamaterial-enhanced spectral detection system of this invention are as follows:

[0049] By constructing a metamaterial enhancement factor tensor matrix, the interaction between terahertz waves and metamaterials is modeled and analyzed, achieving precise optimization of local field enhancement and surface plasmon resonance in target samples, significantly enhancing the detection sensitivity of spectral signals. Hilbert-Huang transform is used for nonlinear mode decomposition to extract intrinsic spectral components representing different physical characteristics from the feature-enhanced spectral functions, constructing a multi-dimensional detection feature space. This effectively extracts nonlinear and non-stationary feature signals from complex samples, improving the ability to describe complex structures in spectral signals. Based on a spectral signal enhancement topology network, the intrinsic correlations between spectral features are explored through topological invariant analysis and quantum correlation calculation, enhancing the comprehensive characterization of sample information. Advanced technologies such as multi-scale entropy measurement, adaptive signal-to-noise ratio, and variational Bayesian inference are integrated to specifically enhance the robustness and anti-interference ability of spectral signals, generating highly robust and adaptive enhanced spectral feature discrimination functions, achieving high-precision terahertz spectral detection. This effectively meets the practical application needs in complex environments and high-precision scenarios, contributing to the widespread application of terahertz technology in various fields. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of a terahertz metamaterial-enhanced spectral detection system according to Embodiment 1 of the present invention;

[0051] Figure 2 This is a flowchart of a terahertz metamaterial-enhanced spectral detection system according to Embodiment 1 of the present invention. Detailed Implementation

[0052] 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.

[0053] Example 1

[0054] Please see Figure 1 and Figure 2 As shown in this embodiment, a terahertz metamaterial-enhanced spectral detection system includes a matrix construction module, a function mapping module, a space construction module, a topology forming module, a function generation module, and an intelligent detection module. The modules are connected via wired and / or wireless means to achieve data transmission between the modules.

[0055] The matrix construction module is used to construct a microscopic electromagnetic response model based on the interaction between terahertz waves and metamaterials, perform resonance characteristic analysis on the microscopic electromagnetic response model, and construct the metamaterial enhancement factor tensor matrix.

[0056] Methods for constructing the metamaterial enhancement factor tensor matrix include:

[0057] Pre-set simulation parameters are provided, which are pre-defined by those skilled in the art based on actual research needs. These parameters include, but are not limited to, material parameters (such as dielectric constant, permeability, and conductivity), excitation source parameters (such as incident wave type, direction, and frequency), and boundary condition parameters (such as periodic boundaries, PML, and metallic boundaries). Based on these pre-set simulation parameters, electromagnetic simulation software (such as CST Microwave Studio, COMSOL Multiphysics, and Lumerical FDTD) is used to simulate the interaction between terahertz waves and metamaterials (i.e., the electromagnetic field distribution and response characteristics inside and around the metamaterial structure under terahertz wave electromagnetic excitation), obtaining electromagnetic field distribution data, transmission coefficient data, and reflection coefficient data. Based on the simulation parameters, electromagnetic field distribution data, transmission coefficient data, and reflection coefficient data, a corresponding microscopic electromagnetic response model is constructed to describe the local electromagnetic response behavior of metamaterial units under terahertz wave excitation.

[0058] Based on the transmission coefficient and reflection coefficient data, obtain the transmission coefficient and reflection coefficient corresponding to each frequency; construct transmission coefficient curves and reflection coefficient curves respectively based on the transmission coefficient and reflection coefficient corresponding to each frequency; mark the frequencies in the transmission coefficient curve that correspond to the minimum transmission coefficient as candidate resonant frequencies; mark the frequencies in the reflection coefficient curve that correspond to the maximum reflection coefficient as candidate resonant frequencies; square the mode length of the transmission coefficient and the mode length of the reflection coefficient corresponding to each frequency in turn, and then add them together to obtain the energy sum corresponding to each frequency; subtract the energy sum corresponding to each frequency from the sum to obtain the absorptivity corresponding to each frequency; construct an absorptivity curve based on the absorptivity corresponding to each frequency; mark the frequencies in the absorptivity curve that correspond to the maximum absorptivity as candidate resonant frequencies.

[0059] Electromagnetic field distribution data corresponding to each candidate resonant frequency is acquired, and the local enhancement factor corresponding to each candidate resonant frequency is calculated. The local enhancement factor includes the local electric field enhancement factor and the local magnetic field enhancement factor. It should be noted that the calculation methods for the local electric field enhancement factor and the local magnetic field enhancement factor are existing technologies, and the specific calculation process will not be elaborated here. Each local electric field enhancement factor is compared with a preset factor threshold. Candidate resonant frequencies with a local electric field enhancement factor greater than or equal to the factor threshold are marked as resonant frequencies, while candidate resonant frequencies with a local electric field enhancement factor less than the factor threshold are not marked. The factor threshold is preset by those skilled in the art according to the actual situation. Based on the local enhancement factor corresponding to each resonant frequency, an enhancement factor tensor matrix corresponding to each resonant frequency is constructed. The enhancement factor tensor matrices corresponding to each resonant frequency are merged to construct a metamaterial enhancement factor tensor matrix.

[0060] It should be noted that, since both the incident field direction and the local field response direction include... , and There are three directions, therefore each enhancement factor tensor matrix contains nine local electric field enhancement factors and nine local magnetic field enhancement factors, i.e. Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor, Directional incident Directional local enhancement factor.

[0061] The function mapping module is used to perform local field enhancement and surface plasmon resonance optimization on target samples based on the metamaterial enhancement factor tensor matrix, and obtain the feature enhancement spectroscopic mapping function.

[0062] The target sample is the sample for terahertz metamaterial spectroscopy detection;

[0063] Methods for obtaining feature-enhanced spectroscopic mapping functions include:

[0064] The basic optical parameters of the target sample are obtained, including the complex permittivity, Raman scattering cross section, molecular dipole moment, and surface molecular density. The complex permittivity describes the response characteristics of the target sample under electromagnetic fields of different frequencies, the Raman scattering cross section describes the ability of molecules to produce Raman scattering under excitation at a specific frequency, the molecular dipole moment describes the asymmetry of charge distribution in the molecule, and the surface molecular density describes the number of target sample molecules per unit area of ​​the metamaterial. The basic optical parameters are obtained by those skilled in the art through experimental measurement or by consulting scientific literature.

[0065] The incident electric field vector corresponding to each resonant frequency is obtained from the simulation parameters. Each incident electric field vector is multiplied by the corresponding enhancement factor tensor matrix to obtain the sample local field corresponding to each resonant frequency. The square of the corresponding modulus of each sample local field is divided by the square of the corresponding modulus of the incident electric field vector to obtain the local field enhancement factor for each resonant frequency. The resonant frequency with the largest local electric field enhancement factor in the metamaterial enhancement factor tensor matrix is ​​marked as the dominant frequency. The sample characteristic frequency is obtained from the complex permittivity function. Based on the dominant frequency and the sample characteristic frequency, the surface plasmon resonance frequency is calculated to achieve surface plasmon resonance optimization of the target sample.

[0066] The incident frequency is obtained from the simulation parameters, and the Raman shift is obtained from the Raman scattering cross section. The scattering frequency is obtained by subtracting the Raman shift from the incident frequency. Based on the local field enhancement factor of each resonant frequency, the local field enhancement factor corresponding to the scattering frequency and the incident frequency is calculated by interpolation algorithms (such as Hermite interpolation, Lorentz fitting interpolation, etc.). The radiative attenuation rate and the non-radiative attenuation rate are obtained. The total attenuation rate is obtained by adding the radiative attenuation rate to the non-radiative attenuation rate. The quantum efficiency factor is obtained by dividing the radiative attenuation rate by the total attenuation rate. The characteristic enhancement spectroscopic mapping function is obtained by multiplying the quantum efficiency factor, the local field enhancement factor of the scattering frequency, the local field enhancement factor of the incident frequency, and the Raman scattering cross section in sequence. The radiative attenuation rate and the non-radiative attenuation rate are obtained by those skilled in the art through electromagnetic simulation of the metamaterial using the finite element method or FDTD method.

[0067] Methods for calculating surface plasmon resonance frequencies include:

[0068] The molecular square is obtained by squaring the mode length corresponding to the molecular dipole moment; the sample square is obtained by squaring the mode length corresponding to the dominant frequency in the sample local field; the molecular square, sample square, and surface molecular density are multiplied sequentially, divided by the reduced Planck constant, and then the square root is taken to obtain the coupling strength; the value of the reduced Planck constant is approximately... Based on the complex permittivity, the loss factor is obtained by fitting the Lorentz model at the dominant frequency. The Lorentz model is an existing technology, and the specific fitting process will not be elaborated here. The loss factor is multiplied by the imaginary unit, added to the dominant frequency, and then the sample characteristic frequency is subtracted to obtain the complex detuning factor. The coupling strength is squared and then divided by the complex detuning factor to obtain the frequency shift. The dominant frequency is added to the frequency shift to obtain the surface plasmon resonance frequency.

[0069] The spatial construction module is used to perform nonlinear mode decomposition on the feature-enhanced spectral mapping function using the Hilbert-Huang transform, extract the intrinsic spectral component set, and construct a multi-dimensional detection feature space.

[0070] Methods for extracting intrinsic spectral component sets include:

[0071] Decompose the feature-enhanced spectroscopy mapping function into Each intrinsic spectral component has an intrinsic spectral component, and a Hilbert transform is performed on each intrinsic spectral component to obtain its amplitude spectrum and instantaneous frequency spectrum. The integer is greater than 0. It should be noted that the Hilbert transform is a prior art technique, and the specific process will not be elaborated here. The signal-to-noise ratio and contribution of each intrinsic spectral component are calculated sequentially and compared with the corresponding component threshold in the preset threshold set. If the signal-to-noise ratio and contribution are both greater than or equal to the corresponding component threshold, the corresponding intrinsic spectral component is marked as a superior spectral component. The threshold set includes the component thresholds corresponding to the signal-to-noise ratio and contribution. The threshold set is preset by those skilled in the art according to the actual situation. If there is a signal-to-noise ratio or contribution less than the corresponding component threshold, the corresponding intrinsic spectral component is not marked. Based on all the superior spectral components, an intrinsic spectral component set is constructed. Each superior spectral component corresponds to a typical response mode, such as a sharp plasmon enhancement peak, a characteristic vibration mode of the sample, or the influence of low-frequency resonance coupling.

[0072] Methods for calculating the signal-to-noise ratio and contribution of each intrinsic spectral component include:

[0073] Low-pass filtering is used to separate the signal and noise components in each intrinsic spectral component. Low-pass filtering is an existing technology, and the specific method will not be described in detail here. The mean of the signal component corresponding to each intrinsic spectral component is calculated and marked as the signal mean. The standard deviation of the noise component corresponding to each intrinsic spectral component is calculated and marked as the noise standard deviation. The signal mean of each intrinsic spectral component is divided by the corresponding noise standard deviation to obtain the signal-to-noise ratio of each intrinsic spectral component.

[0074] Perform a Hilbert transform on the feature-enhanced spectral mapping function to obtain the amplitude spectrum, which is then labeled as the overall amplitude spectrum. Integrate the amplitude spectrum of each intrinsic spectral component and the overall amplitude spectrum in the frequency domain to obtain the intrinsic amplitude and overall amplitude, respectively. Divide the intrinsic amplitude of each intrinsic spectral component by the overall amplitude to obtain the contribution of each intrinsic spectral component.

[0075] Decompose the feature-enhanced spectroscopy mapping function into The steps for identifying each intrinsic spectral component include:

[0076] Step S101: Mark the feature-enhanced spectral mapping function as a mapping function, and obtain all the maximum and minimum points in the mapping function;

[0077] Step S102: Construct the upper envelope using cubic spline interpolation based on all maximum points; cubic spline interpolation is an existing technology, and the specific process will not be described in detail here.

[0078] Step S103: Construct the lower envelope using cubic spline interpolation based on all local minima;

[0079] Step S104: Take the average value of the upper and lower envelopes to obtain the mean curve;

[0080] Step S105: Subtract the mean curve from the mapping function to obtain the candidate spectral components, and determine whether to retain the candidate spectral components; if not retained, use the candidate spectral components as the mapping function and return to step S101; if retained, subtract the candidate spectral components from the mapping function to obtain the residual components, and proceed to step S106.

[0081] Step S106: Determine whether the residual component is a monotonic function; if not, use the residual component as the mapping function and return to step S101; if yes, retain the remaining component. Each candidate spectral component is taken as an intrinsic spectral component.

[0082] In step S105 above, the method for determining whether to retain candidate spectral components includes:

[0083] The total number of maxima and minima in the candidate spectral components is counted and marked as the number of extreme points; the number of zero-crossing points in the candidate spectral components is counted and marked as the number of zeros, where a zero-crossing point is the intersection of the candidate spectral component with the zero axis; the number of extreme points is subtracted from the number of zeros to obtain the difference; if the difference is greater than 1, the candidate spectral component is not retained; if the difference is less than or equal to 1, the candidate spectral component is retained.

[0084] In step S106 above, the method for determining whether the residual component is a monotonic function includes:

[0085] The total number of maximum and minimum points in the residual components is counted and marked as the residual quantity. The residual quantity is compared with a preset monotonic threshold. If the residual quantity is greater than or equal to the monotonic threshold, the residual component is a monotonic function; if the residual quantity is less than the monotonic threshold, the residual component is not a monotonic function. The monotonic threshold is preset by those skilled in the art based on the actual situation.

[0086] Methods for constructing multi-dimensional detection feature spaces include:

[0087] The intrinsic spectral components are concentrated, and the amplitude spectrum and instantaneous frequency spectrum corresponding to each superior spectral component are regarded as a set of components, with each set of components corresponding one-to-one with the superior spectral components. Each set of components is input into a trained feature extraction model to extract spectral feature data. The spectral feature data includes peak position, peak width, peak intensity, energy integral, etc. The feature extraction model is a deep neural network model, which is a current technology, and the specific training process will not be elaborated here. Based on the spectral feature data corresponding to each superior spectral component, a multi-dimensional detection feature space is constructed.

[0088] The topology forming module is used to perform topological invariant calculations and quantum correlation analysis on the multi-dimensional detection feature space to form a spectral signal enhancement topology network.

[0089] Methods for forming spectral signal enhancement topologies include:

[0090] In the multi-dimensional detection feature space, the spectral feature data corresponding to each superior spectral component are normalized (e.g., min-max normalization, Z-score standardization, etc.) to eliminate the influence of dimensions and ensure that features of different dimensions have a uniform scale. Dimensionality reduction algorithms (e.g., principal component analysis, t-distributed random neighborhood embedding) are used to project the multi-dimensional detection feature space onto a three-dimensional topological space, where feature points in the three-dimensional topological space correspond one-to-one with superior spectral components of the multi-dimensional detection feature space. Clustering algorithms (e.g., K-means clustering, DBSCAN, etc.) are then used to cluster the feature points in the three-dimensional topological space to obtain... Sub-topological spaces The integer is greater than 1; in each sub-topological space, three adjacent feature points form a triangular unit, and the Berry curvature of each triangular unit is calculated; the Berry curvatures of all corresponding triangular units in each sub-topological space are summed sequentially, and then divided by . The first Chern number of each sub-topological space is obtained; the Delaunay triangulation algorithm is used to construct the triangulation network of each sub-topological space, and the Euler characteristic of each sub-topological space is calculated based on the triangulation network; the Euler characteristic and the first Chern number are used as the topological invariants of each sub-topological space.

[0091] Each superior spectral component is mapped to a quantum state, and the quantum correlation degree between every two quantum states is calculated. A correlation threshold is preset, which is set by those skilled in the art based on the actual situation. Each quantum correlation degree is compared with the preset correlation threshold. If the quantum correlation degree is greater than or equal to the correlation threshold, the corresponding two quantum states are considered as a quantum set. If the quantum correlation degree is less than the correlation threshold, the corresponding two quantum states are not considered as a quantum set. Each superior spectral component is treated as a node, and the topological invariants and normalized spectral feature data of each node are used as the corresponding node attributes. Correlation edges are established between the nodes corresponding to each quantum set, and the weight of the correlation edge is the quantum correlation degree of the corresponding quantum set. Based on all nodes, node attributes, and correlation edges, a spectral signal enhancement topology network is formed. It should be noted that the Delaunay triangulation algorithm and the method of mapping spectral components to quantum states are existing technologies, and the specific process will not be elaborated here.

[0092] The method for calculating Berry curvature is as follows: three feature points in the triangular unit are randomly labeled as the first point, the second point, and the third point; the cross product of the second point and the third point is calculated to obtain the cross product point; the dot product of the cross product point and the first point is calculated to obtain the dot product value; the modulus of the first point, the second point, and the third point is calculated in sequence and multiplied in sequence to obtain the total modulus; the dot product value is divided by the total modulus, and then the Berry curvature is calculated using the arctangent function.

[0093] The Euler characteristic is calculated as follows: count the number of vertices, edges, and faces in the triangulation network in sequence. The number of vertices is the number of feature points, the number of edges is the number of edges between all feature points, and the number of faces is the number of triangular units. Subtract the number of edges from the number of vertices and add the number of faces to obtain the Euler characteristic.

[0094] The method for calculating quantum correlation is as follows: calculate the inner product between two quantum states and label it as similarity; calculate the inner product of each quantum state itself and label it as norm; multiply the two norms and then take the square root to obtain the norm product; divide the similarity by the norm product to obtain the quantum correlation.

[0095] It should be noted that the calculation methods for cross product, dot product, and inner product are all existing technologies, and the specific calculation process will not be elaborated on here.

[0096] The function generation module is used to perform multi-scale entropy metric evaluation and signal-to-noise ratio adaptive optimization based on the spectral signal enhancement topology network, and generate enhanced spectral feature discrimination functions.

[0097] Methods for generating enhanced spectral feature discriminant functions include:

[0098] Based on the node attributes of each node in the spectral signal enhancement topology network, an attribute sequence is constructed, where each item in the attribute sequence corresponds to a set of node attributes. A scale value is set, which is pre-defined by those skilled in the art based on the actual situation. Based on the scale value, the attribute sequence is coarsened to... Granulated sequences, The value is an integer greater than 1. The multivariate sample entropy of each granulated sequence is calculated sequentially, multiplied by the weight coefficients in the preset first weight set, and then summed sequentially to obtain the multiscale entropy value. It should be noted that the calculation method for the multivariate sample entropy is existing technology, and the specific calculation process will not be elaborated upon here. The signal-to-noise ratio (SNR) of each node is calculated by multiplying the multiscale entropy, the SNR of each node, and the topological invariant by the weight coefficients in the preset second weight set, and then summing sequentially to obtain the node weight of each node. It should be understood that the calculation process for the SNR of each node is consistent with the calculation process for the SNR of the intrinsic spectral components described above. Both the first and second weight sets are preset by those skilled in the art based on actual conditions.

[0099] Each associated edge is multiplied by the average of the node weights of the corresponding two nodes to obtain the adjustment edge; a preset edge threshold is set, which is pre-set by those skilled in the art according to the actual situation; each adjustment edge is compared with the edge threshold, and adjustment edges with values ​​greater than or equal to the edge threshold are retained in the spectral signal enhancement topology network, while adjustment edges with values ​​less than the edge threshold are deleted from the spectral signal enhancement topology network; the spectral signal enhancement topology network is mapped to the feature space to construct feature vectors, and the feature vectors are regularized to obtain regularized processing vectors; the feature vectors include all node attributes, multi-scale entropy values, signal-to-noise ratio, and adjustment edges in the spectral signal enhancement topology network, and the regularization processing is a prior art, the specific process of which will not be elaborated here; based on the regularized processing vectors, a feature judgment model is constructed using the random forest algorithm, and the constructed feature judgment model is transformed into an enhanced spectral feature discrimination function; the random forest algorithm is a prior art, and the specific process of which will not be elaborated here.

[0100] The output of the enhanced spectral feature discrimination function is set by those skilled in the art according to the specific application scenario, including classification tasks (such as identifying different drug components, different types of plastics, distinguishing organic compounds, etc.), regression tasks (such as predicting the concentration of a compound in a target sample, estimating the purity percentage of a target sample, measuring specific physical parameters of a target sample, etc.), anomaly detection (such as detecting food contamination, identifying drug adulteration, etc.).

[0101] The intelligent detection module combines variational Bayesian inference and sensitivity analysis methods to perform uncertainty quantification and sensitivity analysis on the enhanced spectral feature discrimination function, thereby obtaining high-precision terahertz spectral detection results.

[0102] Methods for obtaining high-precision terahertz spectral detection results include:

[0103] Variational Bayesian inference is used to quantify the uncertainty of the enhanced spectral feature discriminant function (ESD) and obtain confidence intervals. A confidence interval is a statistical concept representing the range of uncertainty in the model's prediction results. More specifically, a confidence interval is an interval estimate, indicating that the true value has a certain probability (usually 95%) of falling within this interval. Variational Bayesian inference is an existing technique, and its specific process will not be elaborated upon here. Sensitivity analysis methods (such as the Sobol method and variance decomposition) are used to perform sensitivity analysis on the ESD, obtaining the first-order sensitivity index and the total sensitivity index. The first-order sensitivity index represents the contribution of a single variable to the output variance of the ESD under independent action. The total sensitivity index represents the overall contribution of a single variable and its interactions with all other variables to the output variance of the ESD. The output variance is the statistical variance of the output results produced by the ESD, measuring the dispersion of the output results.

[0104] Based on confidence intervals, first-order sensitivity indices, and total sensitivity indices, optimization algorithms (such as simulated annealing, genetic algorithms, and clonal selection algorithms) are used to optimize the hyperparameters of the feature judgment model (such as the number of trees, maximum depth, and minimum number of leaf node samples). The optimization objective is to minimize the confidence interval width while minimizing the model's prediction error. Based on the optimized feature judgment model, an optimized enhanced spectral feature discrimination function is obtained and marked as the optimized discrimination function. Based on the output of the optimized discrimination function, high-precision terahertz spectral detection results are obtained.

[0105] This embodiment constructs a metamaterial enhancement factor tensor matrix to model and analyze the interaction between terahertz waves and metamaterials, achieving precise optimization of local field enhancement and surface plasmon resonance in target samples, significantly enhancing the detection sensitivity of spectral signals. It employs Hilbert-Huang transform for nonlinear mode decomposition, extracting intrinsic spectral components representing different physical characteristics from the feature-enhanced spectral functions, constructing a multi-dimensional detection feature space. This effectively extracts nonlinear and non-stationary feature signals from complex samples, improving the ability to describe complex structures in spectral signals. Based on the spectral signal enhancement topology network, it mines the intrinsic correlations between spectral features through topological invariant analysis and quantum correlation calculation, enhancing the comprehensive characterization of sample information. It integrates advanced technologies such as multi-scale entropy measurement, adaptive signal-to-noise ratio, and variational Bayesian inference to specifically enhance the robustness and anti-interference ability of spectral signals, generating highly robust and adaptive enhanced spectral feature discrimination functions, achieving high-precision terahertz spectral detection. This effectively meets the practical application needs in complex environments and high-precision scenarios, contributing to the widespread application of terahertz technology in various fields.

[0106] Example 2

[0107] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable code that, when executed by the one or more processors, can perform a terahertz metamaterial-enhanced spectroscopy detection system as described above.

[0108] The method or system according to the embodiments of this application can also be implemented using the architecture of the electronic device shown in this application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as a ROM or hard disk, may store a terahertz metamaterial-enhanced spectroscopy detection system provided in this application. Furthermore, the electronic device may also include a user interface. Of course, the architecture shown in this application is merely exemplary; when implementing different devices, one or more components in the electronic device shown in this application may be omitted according to actual needs.

[0109] Example 3

[0110] One embodiment of this application discloses a computer-readable storage medium. The computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, a terahertz metamaterial-enhanced spectroscopy detection system according to an embodiment of this application, as described with reference to the above figures, can be executed. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0111] Furthermore, according to embodiments of this application, the processes described in the above-referenced flowcharts can be implemented as computer software programs. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as a terahertz metamaterial-enhanced spectroscopy detection system. When this computer program is executed by a central processing unit (CPU), it performs the functions defined in the method of this application.

[0112] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0113] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0114] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0115] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0116] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0117] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0118] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0119] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A terahertz metamaterial-enhanced spectroscopic detection system, characterized in that, include: The matrix construction module is used to construct a microscopic electromagnetic response model based on the interaction between terahertz waves and metamaterials, perform resonance characteristic analysis on the microscopic electromagnetic response model, and construct the metamaterial enhancement factor tensor matrix. The function mapping module is used to perform local field enhancement and surface plasmon resonance optimization on target samples based on the metamaterial enhancement factor tensor matrix, and obtain the feature enhancement spectroscopic mapping function. The spatial construction module is used to perform nonlinear mode decomposition on the feature-enhanced spectral mapping function using the Hilbert-Huang transform, extract the intrinsic spectral component set, and construct a multi-dimensional detection feature space. The method for extracting the intrinsic spectral component set includes: Decompose the feature-enhanced spectroscopy mapping function into Each intrinsic spectral component has an intrinsic spectral component, and a Hilbert transform is performed on each intrinsic spectral component to obtain its amplitude spectrum and instantaneous frequency spectrum. The input is a positive integer; the signal-to-noise ratio (SNR) and contribution of each intrinsic spectral component are calculated sequentially and compared with the corresponding component threshold in the preset threshold set; if both the SNR and contribution are greater than or equal to the corresponding component threshold, the corresponding intrinsic spectral component is marked as a superior spectral component; the threshold set includes the component thresholds corresponding to the SNR and contribution; if there is an input with an SNR or contribution less than the corresponding component threshold, the corresponding intrinsic spectral component is not marked; based on all superior spectral components, an intrinsic spectral component set is constructed. The method for constructing a multi-dimensional detection feature space includes: The intrinsic spectral components are concentrated, and the amplitude spectrum and instantaneous frequency spectrum corresponding to each superior spectral component are regarded as a set of components, with each set of components corresponding to a superior spectral component. Each set of components is input into a trained feature extraction model to extract spectral feature data. The feature extraction model is a deep neural network model. Based on the spectral feature data corresponding to each superior spectral component, a multi-dimensional detection feature space is constructed. The topology forming module is used to perform topological invariant calculations and quantum correlation analysis on the multi-dimensional detection feature space to form a spectral signal enhancement topology network. The function generation module is used to perform multi-scale entropy metric evaluation and signal-to-noise ratio adaptive optimization based on the spectral signal enhancement topology network, and generate enhanced spectral feature discrimination functions. The intelligent detection module combines variational Bayesian inference and sensitivity analysis methods to perform uncertainty quantification and sensitivity analysis on the enhanced spectral feature discrimination function, thereby obtaining high-precision terahertz spectral detection results.

2. The terahertz metamaterial-enhanced spectral detection system according to claim 1, characterized in that, The method for constructing the metamaterial enhancement factor tensor matrix includes: Preset simulation parameters are used to simulate the interaction between terahertz waves and metamaterials using electromagnetic simulation software, obtaining electromagnetic field distribution data, transmission coefficient data, and reflection coefficient data. Based on the simulation parameters, electromagnetic field distribution data, transmission coefficient data, and reflection coefficient data, a corresponding microscopic electromagnetic response model is constructed. Based on the transmission coefficient and reflection coefficient data, obtain the transmission coefficient and reflection coefficient corresponding to each frequency; construct transmission coefficient curves and reflection coefficient curves respectively based on the transmission coefficient and reflection coefficient corresponding to each frequency; mark the frequencies in the transmission coefficient curve that correspond to the minimum transmission coefficient as candidate resonant frequencies; mark the frequencies in the reflection coefficient curve that correspond to the maximum reflection coefficient as candidate resonant frequencies; square the mode length of the transmission coefficient and the mode length of the reflection coefficient corresponding to each frequency in turn, and then add them together to obtain the energy sum corresponding to each frequency; subtract the energy sum corresponding to each frequency from the sum to obtain the absorptivity corresponding to each frequency; construct an absorptivity curve based on the absorptivity corresponding to each frequency; mark the frequencies in the absorptivity curve that correspond to the maximum absorptivity as candidate resonant frequencies. Electromagnetic field distribution data corresponding to each candidate resonant frequency is acquired, and the local enhancement factor corresponding to each candidate resonant frequency is calculated. The local enhancement factor includes the local electric field enhancement factor and the local magnetic field enhancement factor. Each local electric field enhancement factor is compared with a preset factor threshold. Candidate resonant frequencies with a local electric field enhancement factor greater than or equal to the factor threshold are marked as resonant frequencies, while candidate resonant frequencies with a local electric field enhancement factor less than the factor threshold are not marked. Based on the local enhancement factor corresponding to each resonant frequency, an enhancement factor tensor matrix corresponding to each resonant frequency is constructed. The enhancement factor tensor matrices corresponding to each resonant frequency are merged to construct a metamaterial enhancement factor tensor matrix.

3. The terahertz metamaterial-enhanced spectroscopic detection system according to claim 2, characterized in that, The method for obtaining the feature-enhanced spectroscopic mapping function includes: The basic optical parameters of the target sample are obtained, including the complex permittivity, Raman scattering cross section, molecular dipole moment, and surface molecular density. The incident electric field vector corresponding to each resonant frequency is obtained from the simulation parameters. Each incident electric field vector is multiplied by the corresponding enhancement factor tensor matrix to obtain the sample local field corresponding to each resonant frequency. The square of the modulus corresponding to each sample local field is divided by the square of the modulus corresponding to the incident electric field vector to obtain the local field enhancement factor for each resonant frequency. The resonant frequency with the largest local electric field enhancement factor in the metamaterial enhancement factor tensor matrix is ​​marked as the dominant frequency. The sample characteristic frequency is obtained from the complex permittivity. Based on the dominant frequency and the sample characteristic frequency, the surface plasmon resonance frequency is calculated. The incident frequency is obtained from the simulation parameters, and the Raman shift is obtained from the Raman scattering cross section. The scattering frequency is obtained by subtracting the Raman shift from the incident frequency. Based on the local field enhancement factor of each resonant frequency, the local field enhancement factor corresponding to the scattering frequency and the incident frequency is calculated using an interpolation algorithm. The radiative attenuation rate and the non-radiative attenuation rate are obtained. The total attenuation rate is obtained by adding the radiative attenuation rate to the non-radiative attenuation rate. The quantum efficiency factor is obtained by dividing the radiative attenuation rate by the total attenuation rate. The characteristic enhancement spectral mapping function is obtained by multiplying the quantum efficiency factor, the local field enhancement factor of the scattering frequency, the local field enhancement factor of the incident frequency, and the Raman scattering cross section in sequence.

4. The terahertz metamaterial-enhanced spectroscopic detection system according to claim 3, characterized in that, The method for calculating the surface plasmon resonance frequency includes: The molecular square is obtained by squaring the mode length corresponding to the molecular dipole moment; the sample square is obtained by squaring the mode length corresponding to the local field of the dominant frequency; the molecular square, sample square, and surface molecular density are multiplied sequentially, divided by the reduced Planck constant, and then the square root is taken to obtain the coupling strength; the loss factor is obtained by fitting the Lorentz model at the dominant frequency based on the complex permittivity; the loss factor is multiplied by the imaginary unit, added to the dominant frequency, and then subtracted from the sample characteristic frequency to obtain the complex detuning factor; the frequency shift is obtained by squaring the coupling strength and dividing by the complex detuning factor; the surface plasmon resonance frequency is obtained by adding the frequency shift to the dominant frequency.

5. The terahertz metamaterial-enhanced spectroscopic detection system according to claim 4, characterized in that, The decomposition of the feature-enhanced spectroscopy mapping function into The steps for identifying each intrinsic spectral component include: Step S101: Mark the feature-enhanced spectral mapping function as a mapping function, and obtain all the maximum and minimum points in the mapping function; Step S102: Construct the upper envelope using cubic spline interpolation based on all maximum points; Step S103: Construct the lower envelope using cubic spline interpolation based on all local minima; Step S104: Take the average value of the upper and lower envelopes to obtain the mean curve; Step S105: Subtract the mean curve from the mapping function to obtain the candidate spectral components, and determine whether to retain the candidate spectral components; if not retained, use the candidate spectral components as the mapping function and return to step S101; if retained, subtract the candidate spectral components from the mapping function to obtain the residual components, and proceed to step S106. Step S106: Determine whether the residual component is a monotonic function; if not, use the residual component as the mapping function and return to step S101; if yes, retain the remaining component. Each candidate spectral component is used as an intrinsic spectral component; In step S105, the method for determining whether to retain candidate spectral components includes: The total number of maxima and minima in the candidate spectral components is counted and marked as the number of extreme points; the number of zero-crossing points in the candidate spectral components is counted and marked as the number of zeros, where a zero-crossing point is the intersection of the candidate spectral component with the zero axis; the number of extreme points is subtracted from the number of zeros to obtain the difference; if the difference is greater than 1, the candidate spectral component is not retained; if the difference is less than or equal to 1, the candidate spectral component is retained. In step S106, the method for determining whether the residual component is a monotonic function includes: The total number of maximum and minimum points in the residual components is counted and marked as the residual quantity. The residual quantity is compared with a preset monotonic threshold. If the residual quantity is greater than or equal to the monotonic threshold, the residual component is a monotonic function; if the residual quantity is less than the monotonic threshold, the residual component is not a monotonic function.

6. The terahertz metamaterial-enhanced spectral detection system according to claim 5, characterized in that, The method for forming a spectral signal enhancement topology network includes: In the multi-dimensional detection feature space, the spectral feature data corresponding to each superior spectral component are normalized. A dimensionality reduction algorithm is used to project the multi-dimensional detection feature space onto a three-dimensional topological space, where feature points in the three-dimensional topological space correspond one-to-one with the superior spectral components of the multi-dimensional detection feature space. A clustering algorithm is then used to cluster the feature points in the three-dimensional topological space to obtain... Sub-topological spaces The integer is greater than 1; in each sub-topological space, three adjacent feature points form a triangular unit, and the Berry curvature of each triangular unit is calculated; the Berry curvatures of all corresponding triangular units in each sub-topological space are summed sequentially, and then divided by . The first Chern number of each sub-topological space is obtained; the Delaunay triangulation algorithm is used to construct the triangulation network of each sub-topological space, and the Euler characteristic of each sub-topological space is calculated based on the triangulation network; the Euler characteristic and the first Chern number are used as the topological invariants of each sub-topological space. Each superior spectral component is mapped to a quantum state, and the quantum correlation degree between any two quantum states is calculated. A correlation threshold is preset, which is set by those skilled in the art based on the actual situation. Each quantum correlation degree is compared with the preset correlation threshold. If the quantum correlation degree is greater than or equal to the correlation threshold, the corresponding two quantum states are considered as a quantum set. If the quantum correlation degree is less than the correlation threshold, the corresponding two quantum states are not considered as a quantum set. Each superior spectral component is treated as a node, and the topological invariants and normalized spectral feature data of each node are used as the corresponding node attributes. Correlation edges are established between the nodes corresponding to each quantum set, and the weight of the correlation edge is the quantum correlation degree of the corresponding quantum set. Based on all nodes, node attributes, and correlation edges, a spectral signal enhancement topology network is formed.

7. The terahertz metamaterial-enhanced spectroscopic detection system according to claim 6, characterized in that, The method for calculating the Berry curvature is as follows: three feature points in the triangular unit are randomly labeled as the first point, the second point, and the third point in sequence; the cross product of the second point and the third point is calculated to obtain the cross product point; Calculate the dot product of the cross product point and the first point to obtain the dot product value; calculate the modulus of the first, second and third points in sequence, and multiply them in sequence to obtain the total modulus; divide the dot product value by the total modulus, and then calculate the Bailey curvature using the arctangent function; The Euler characteristic number is calculated as follows: the number of vertices, edges, and faces in the triangulation network are counted sequentially. The number of vertices is the number of feature points, the number of edges is the number of edges between all feature points, and the number of faces is the number of triangular units. The Euler characteristic number is obtained by subtracting the number of edges from the number of vertices and adding the number of faces. The method for calculating the quantum correlation is as follows: calculate the inner product between two quantum states and label it as similarity; calculate the inner product of each quantum state itself and label it as norm; multiply the two norms and then take the square root to obtain the norm product; divide the similarity by the norm product to obtain the quantum correlation.

8. The terahertz metamaterial-enhanced spectroscopic detection system according to claim 7, characterized in that, The method for generating the enhanced spectral feature discriminant function includes: Based on the node attributes of each node in the spectral signal enhancement topology network, an attribute sequence is constructed; a scale value is set, and based on the scale value, the attribute sequence is coarsened to... Granulated sequences, The value is an integer greater than 1; calculate the multivariate sample entropy of each granulated sequence in turn, multiply it by the weight coefficients in the preset first weight set, and then add them in turn to obtain the multiscale entropy value; calculate the signal-to-noise ratio of each node, multiply the multiscale entropy, the signal-to-noise ratio of each node and the topological invariant by the weight coefficients in the preset second weight set, and then add them in turn to obtain the node weight of each node. Each associated edge is multiplied by the average of the weights of the two corresponding nodes to obtain the adjustment edge. A preset edge threshold is set, and each adjustment edge is compared with the edge threshold. Adjustment edges with values ​​greater than or equal to the edge threshold are retained in the spectral signal enhancement topology network, while those with values ​​less than the edge threshold are deleted from the spectral signal enhancement topology network. The spectral signal enhancement topology network is mapped to the feature space to construct a feature vector, and the feature vector is regularized to obtain a regularized processing vector. The feature vector includes all node attributes, multi-scale entropy values, signal-to-noise ratio, and adjustment edges in the spectral signal enhancement topology network. Based on the regularized processing vector, a feature judgment model is constructed using the random forest algorithm, and the constructed feature judgment model is transformed into an enhanced spectral feature discrimination function.

9. A terahertz metamaterial-enhanced spectroscopic detection system according to claim 8, characterized in that, The method for obtaining high-precision terahertz spectral detection results includes: Variational Bayesian inference is used to quantify the uncertainty of the enhanced spectral feature discriminant function and obtain confidence intervals. Sensitivity analysis is then performed on the enhanced spectral feature discriminant function to obtain the first-order sensitivity index and the total sensitivity index. Based on the confidence intervals, the first-order sensitivity index, and the total sensitivity index, an optimization algorithm is used to optimize the model hyperparameters of the feature judgment model. The optimization objective is to minimize the confidence interval width while minimizing the model's prediction error. Based on the optimized feature judgment model, the optimized enhanced spectral feature discriminant function is obtained and marked as the optimized discriminant function. High-precision terahertz spectral detection results are obtained based on the output of the optimized discriminant function.

Images