Aflatoxin qualitative detection method based on terahertz metamaterial absorber

Abstract

The application discloses a qualitative detection method for aflatoxin based on a terahertz metamaterial absorber, wherein the period P of the terahertz metamaterial absorber is 60 microns, the coating metal is gold, and a silicon substrate is adopted; the terahertz metamaterial absorber comprises a large cross-shaped structure located at a middle position and a plurality of small cross-shaped structures located around the large cross-shaped structure; the length L1 of the large cross-shaped structure is 30 microns, and the width W of the large cross-shaped structure is 2 microns; the small cross-shaped structure comprises a main body part in a cross shape and a frame part arranged outside the main body part, the main body part is perpendicular to the frame part, the length L2 of the main body part is 25 microns, and the length L3 of the frame part is 12 microns; and the distance d between one side of the large cross-shaped structure and the center of the main body part is 15 microns. The application develops a new kind of metamaterial, and realizes the qualitative detection of aflatoxin by using the metamaterial, thereby providing a new way for the qualitative detection of aflatoxin.

Description

Technical Field

[0001] This invention relates to the field of chemical detection technology, and in particular to a qualitative detection method for aflatoxin based on a terahertz metamaterial absorber. Background Technology

[0002] Food safety is crucial to social stability and people's daily lives. Among food safety issues, aflatoxin contamination has become a global problem. Aflatoxins (AFs) are a group of harmful fungal toxins produced by Aspergillus flavus, which can contaminate both water and food. Aflatoxins have been reported to cause hepatocellular carcinoma, Reye's syndrome, and chronic hepatitis, posing serious health risks. There are approximately 20 derivatives, with aflatoxin B1 (AFB1), aflatoxin B2 (AFB2), aflatoxin G1 (AFG1), and aflatoxin G2 (AFG2) being the most common and highly toxic. In 1993, aflatoxins were classified as a Group 1 carcinogen by the International Agency for Research on Cancer (IARC) of the World Health Organization. This highly toxic carcinogen primarily contaminates agricultural products such as peanuts, corn, and grains, causing these foods to mold and spoil.

[0003] Ingesting aflatoxin-contaminated agricultural products and food can cause cellular damage, seriously endangering public health. Many countries have implemented strict regulations to control aflatoxin levels in agricultural products and food to address this issue. For example, US federal law stipulates that the aflatoxin content (the total amount of B1+B2+G1+G2) in human-consumed food and dairy cow feed cannot exceed 15 μg / kg, in human-consumed milk cannot exceed 0.5 μg / kg, and in other animal feed cannot exceed 300 μg / kg. EU countries have even stricter regulations: the limit for aflatoxin B1 in peanuts and nuts and their processed products, and all cereal products and processed products is 2.0 μg / kg; the limit for M1 in raw milk, heat-treated milk, and processed milk products is 0.050 μg / kg; and the limit for M1 in infant food (including infant formula) is 0.025 μg / kg. To prevent aflatoxin contamination of food, relevant departments in China have issued standards and management measures to strictly limit the aflatoxin content in different foods.

[0004] Currently, commonly used physicochemical analysis methods for detecting aflatoxin both domestically and internationally include thin-layer chromatography (TLC), high-performance liquid chromatography (HPLC), enzyme-linked immunosorbent assay (ELISA), and electrochemical methods. While these methods offer high sensitivity for aflatoxin detection, they suffer from drawbacks such as high cost, long detection cycles, poor reproducibility, complex and cumbersome procedures, and potential sample damage. In recent years, many researchers have been exploring spectroscopic detection techniques for qualitative and quantitative analysis of aflatoxin in food, primarily including near-infrared spectroscopy (NIR), fluorescence spectroscopy, Raman spectroscopy, and multispectral techniques. However, while NIR spectroscopy offers advantages such as high efficiency and non-destructive testing, enabling quantitative detection of aflatoxin, it still falls short in terms of accuracy and sensitivity, failing to meet international standards. Fluorescence spectroscopy, while offering fast analysis speed and high precision, is susceptible to numerous interfering factors. While Raman spectroscopy can be enhanced using SERS, the enhancement conditions are extremely demanding, and the selection and preparation of the enhancement substrate are quite challenging. Therefore, finding a rapid, non-destructive, and highly sensitive method for detecting trace aflatoxins in food has always been a major focus of attention in the field of food safety and a hot research topic for scholars both domestically and internationally.

[0005] Terahertz (THz) waves typically refer to electromagnetic waves with frequencies ranging from 0.1 to 10 THz. They possess advantages such as low photon energy, high resolution, and rich optical parameters, providing a theoretical basis for the application of terahertz spectroscopy in biological and chemical molecular detection. In recent years, experts and scholars have successively used terahertz spectroscopy to conduct related research in biological and chemical molecular fields. However, with in-depth research, it has been found that detecting highly toxic and hazardous trace substances using traditional terahertz spectroscopy techniques is quite challenging. This may be because most substances in nature have a relatively weak electromagnetic response to terahertz waves, limiting the detection sensitivity of terahertz spectroscopy. To enhance the detection capability of terahertz spectroscopy for trace components, metamaterial sensors have begun to emerge.

[0006] Metamaterials are artificial composite electromagnetic materials that utilize the periodic arrangement of subwavelength unit structures to achieve a unique response to electromagnetic waves. Metamaterial absorbers, in particular, are devices that effectively absorb incident waves at specific frequencies, significantly attenuating or even eliminating the corresponding transmission and reflection.

[0007] However, there are currently no studies using metamaterials for the qualitative detection of aflatoxin. Summary of the Invention

[0008] In view of the above, the purpose of this invention is to provide a qualitative detection method for aflatoxin based on a terahertz metamaterial absorber, so as to develop a new metamaterial-based method and use the metamaterial to realize the qualitative detection method of aflatoxin, thus providing a new approach for the qualitative detection of aflatoxin.

[0009] This invention provides a qualitative detection method for aflatoxin based on a terahertz metamaterial absorber. The terahertz metamaterial absorber has the following parameters: a period P of 60 μm, a gold coating, and a silicon substrate. The terahertz metamaterial absorber includes a large cross-shaped structure in the center and several smaller cross-shaped structures around the large cross-shaped structure. The length L1 of the large cross-shaped structure is 30 μm, and the width W of the large cross-shaped structure is 2 μm. Each smaller cross-shaped structure includes a cross-shaped main body and an outer frame portion located outside the main body. The main body and the outer frame portion are perpendicular. The length L2 of the main body is 25 μm, and the length L3 of the outer frame portion is 12 μm. The distance d between one side of the large cross-shaped structure and the center of the main body is 15 μm.

[0010] The detection method includes the following steps:

[0011] (1) Prepare multiple aflatoxin B2 standard solutions, aflatoxin G1 standard solutions, and aflatoxin G2 standard solutions of different concentrations, and drip the aflatoxin B2 standard solution onto the corresponding terahertz metamaterial absorber and put it into a drying oven to dry.

[0012] (2) Place the dried terahertz metamaterial absorber into the terahertz system and use the transmission mode to measure and collect the terahertz spectral information of the corresponding sample.

[0013] (3) The original terahertz transmission spectra of aflatoxin B1, G1 and G2 were classified into two categories using a linear discriminant analysis algorithm.

[0014] (4) Feature extraction of terahertz spectral information;

[0015] (5) A qualitative analysis model for the terahertz transmission spectra of aflatoxin B2, G1, and G2 solutions was established using chemometric methods;

[0016] (6) Obtain the terahertz spectral information of the sample containing aflatoxin to be tested and input it into each qualitative analysis model to obtain the type of aflatoxin in the sample to be tested.

[0017] Compared with existing technologies, the qualitative detection method for aflatoxin based on terahertz metamaterial absorbers provided by this invention proposes a new metamaterial-based method and uses this metamaterial to realize the qualitative detection method for aflatoxin, providing a new approach for the qualitative detection of aflatoxin. The qualitative detection of aflatoxin in food using terahertz technology can meet national standards.

[0018] In addition, the qualitative detection method for aflatoxin based on a terahertz metamaterial absorber provided by the present invention also has the following technical features:

[0019] The aforementioned terahertz metamaterial absorber was prepared using the following method:

[0020] The process utilizes 4-inch double-polished silicon wafers and glass sheets. The entire process includes deposition, pretreatment, photoresist coating, pre-baking, exposure, development, photoresist application, hardening, etching, and stripping. First, 5nm chromium is deposited on the silicon substrate, followed by 100nm gold. After metal deposition, pretreatment is performed in an HMDS oven at 120℃ for 10 minutes. Following pretreatment, photoresist AZ6112 is dropped onto the silicon substrate surface and spread using a spin coater at 600 rpm. Then, it is uniformly deposited at 4000 rpm to achieve a thickness of 1-2μm. After spin coating, pre-baking is performed at a temperature and time of 1... 00℃, 120s; After pre-baking, exposure is performed, followed by manual development for 35s using 2.38% 3038 developer. After development, a photoresist stripper is used to remove residual photoresist from the surface. The power of the stripper is set to 200W and the time is 3min. After stripping, hardening is performed, i.e., baking on a hot plate at 110℃ for 120s. Then, IBE etching is performed, i.e., bombarding the sample surface with an ion beam of a certain concentration to achieve the etching purpose. After etching, ultrasonic extraction with acetone is performed for 1 hour. After stripping, the photoresist mask is observed under a microscope to see if it has been completely removed. If it has not been completely removed, it needs to be placed in an acetone solution for ultrasonication until the photoresist mask is completely removed.

[0021] Specifically, step (1) includes:

[0022] Aflatoxin B2 standard solution, aflatoxin G1 standard solution, and aflatoxin G2 standard solution were selected, with a standard solution concentration of 25 μg / ml. Each aflatoxin standard solution was diluted with ultrapure deionized water using a pipette. Eight concentration gradient aflatoxin B2, G1, and G2 solutions were prepared according to the sample ratio table. The prepared standard solutions were placed on a vortex mixer and shaken for 3 minutes to ensure that the aflatoxin standard solutions were fully diluted in deionized water.

[0023] Specifically, step (2) includes:

[0024] The dried terahertz metamaterial absorber was placed in a terahertz time-domain spectrometer. The spectral measurement range of the terahertz time-domain spectrometer was set to 0.1-5.0 THz, and the instrument resolution was set to 7.6 GHz. Before the experiment, the system was preheated using an air compressor with an air dryer, and dry air was continuously filled into the optical cavity. The humidity was kept below 5% and the temperature was kept constant at 25±0.5℃ by monitoring with a hygrometer. During the experiment, after each sample was placed, wait for 2 minutes for the system to stabilize before starting the measurement. Using a pipette, 20 μL of sample solution was pipetted onto the metamaterial sheet in order of increasing sample concentration. Before acquiring the sample spectrum, the metamaterial sheet with the sample was placed in a drying oven for 20 minutes at a temperature of 50℃. Finally, the metamaterial sheet with the sample was placed into the terahertz system and measured using transmission mode. Two points were taken for each concentration sample, and 10 measurements were performed at each point. A total of 160 spectra were obtained for the 8 concentration samples, and a total of 480 terahertz transmission spectra were obtained for the 3 aflatoxin samples.

[0025] Step (4) includes:

[0026] One of the following methods—principal component analysis, competitive adaptive reweighting algorithm, elimination of non-information variables, and continuous projection algorithm—is used to extract features from the original terahertz transmission spectrum.

[0027] The model established in step (5) is one of the following: K-nearest neighbor classification algorithm model, random forest model, and support vector machine model. Attached Figure Description

[0028] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0029] Figure 1 This is a schematic diagram of a terahertz metamaterial absorber.

[0030] Figure 2 This is a scanning electron microscope image of a terahertz metamaterial absorber.

[0031] Figure 3 This is a diagram showing the magnetic field distribution of a terahertz metamaterial absorber.

[0032] Figure 4 Transmission spectra of solutions containing three aflatoxins B2, G1, and G2 on the surface of a metamaterial absorber;

[0033] Figure 5 The LDA binary mixed classification qualitative model for the prediction set samples is a planar classification diagram.

[0034] Figure 6 Confusion matrix diagram of prediction sets for PCA-KNN and CARS-KNN models;

[0035] Figure 7 Plot of out-of-bag (OOB) error rate and prediction set confusion matrix for PCA-RF model;

[0036] Figure 8 The confusion matrix diagram for the CARS-RBF-SVM model. Detailed Implementation

[0037] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be thorough and complete.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0039] Please see Figures 1 to 3 This invention provides a terahertz metamaterial absorber. In this embodiment, the structural parameters of the metamaterial absorber are designed using parameter simulation with FDTD Solutions software. FDTD Solutions is a professional micro-nano photonics simulation and analysis software based on the finite-time difference method to solve the vector Maxwell equation. It can simulate the interaction between electromagnetic waves in the infrared, terahertz, and microwave frequency bands and complex structures with typical subwavelength sizes. It can be used for the design, analysis, and optimization of micro-nano optical materials and micro-nano photonic devices. Considering that metamaterials are composed of periodic structural units, the transmission or reflection properties of a single periodic structural unit of the metamaterial can represent the spectral properties of the entire metamaterial. Therefore, in the simulation process, simulating a single periodic structural unit of the metamaterial can improve the simulation speed.

[0040] Specifically, the terahertz metamaterial absorber has the following parameters: period P is 60 μm, the coating metal is gold, and a silicon substrate is used. In this embodiment, the refractive index of the silicon substrate is 3.335.

[0041] The terahertz metamaterial absorber includes an "X" composite terahertz metamaterial micro / nano structure, specifically a large cross-shaped structure 10 located in the center and several small cross-shaped structures 20 located around the large cross-shaped structure 10. The length L1 of the large cross-shaped structure 10 is 30 μm, and the width W of the large cross-shaped structure 10 is 2 μm. Each small cross-shaped structure 20 includes a cross-shaped main body 21 and an outer frame 22 located outside the main body 21. The main body 21 is perpendicular to the outer frame 22. The length L2 of the main body 21 is 25 μm, and the length L3 of the outer frame 22 is 12 μm. The distance d between one side of the large cross-shaped structure 10 and the center of the main body 21 is 15 μm.

[0042] The fabrication method of the aforementioned terahertz metamaterial absorber includes the following steps: The metamaterial chip is processed using a four-inch double-polished silicon wafer and a glass wafer. The entire processing technology includes deposition, pretreatment, photoresist coating, pre-baking, exposure, development, photoresist application, hardening, etching, and lift-off. First, 5nm of chromium is deposited on the silicon substrate to enhance the adhesion between the gold and the silicon substrate, followed by 100nm of gold deposition. After metal deposition, pretreatment is performed using an HMDS oven at 120℃ for 10 minutes. The main purpose of pretreatment is to increase the adhesion between the photoresist and the silicon substrate. After pretreatment, photoresist AZ6112 is dropped onto the silicon substrate surface and spread using a spin coater at 600 rpm. Then, at 4000 rpm, the AZ6112 is uniformly deposited to a thickness of 1-2 μm. Pre-baking is then performed after spin coating at 100℃ for 120 seconds. After pre-baking, exposure is performed. Following exposure, manual development is carried out for 35 seconds using a 2.38% 3038 developer. After development, a photoresist stripper is used to remove residual photoresist from the surface. The stripper power is set to 200W for 3 minutes. After stripping, hardening is performed by baking on a hot plate at 110℃ for 120 seconds to increase the photoresist mask's resistance to etching, facilitating subsequent etching. IBE etching is then performed, using an ion beam of a specific concentration to bombard the sample surface. After etching, ultrasonic extraction with acetone is performed for 1 hour to remove the photoresist mask. After stripping, the photoresist mask is observed under a microscope to ensure complete removal. If not completely removed, ultrasonic extraction in acetone solution is continued until the photoresist mask is completely removed. Figure 2 This is a scanning electron microscope image of the metamaterial chip after photolithography under an optical microscope.

[0043] Another embodiment of the present invention provides a qualitative detection method for aflatoxin based on a terahertz metamaterial absorber, wherein the terahertz metamaterial absorber is the aforementioned terahertz metamaterial absorber, and the detection method includes the following steps:

[0044] (1) Prepare multiple aflatoxin B2 standard solutions, aflatoxin G1 standard solutions, and aflatoxin G2 standard solutions of different concentrations, and drip the aflatoxin B2 standard solution onto the corresponding terahertz metamaterial absorber and dry them in a drying oven.

[0045] Three aflatoxin standard solutions, B2, G1, and G2, were purchased from Aladdin Reagent Network, with a concentration of 25 μg / ml for each. To better identify trace amounts of aflatoxins B2, G1, and G2, the aflatoxin standard solutions were diluted with ultrapure deionized water (18.2 MΩ·cm, Merck Millipore Ltd., USA) using a pipette. Eight concentration gradients of aflatoxin B2, G1, and G2 solutions were prepared according to the sample ratio table. The prepared sample solutions were vortexed for 3 minutes to ensure sufficient dilution of the aflatoxin B2 standard solution in the deionized water. The concentration ratio table for the three aflatoxin B2, G1, and G2 solutions is shown in Table 1.

[0046] Table 1. Concentration ratios of aflatoxin B2, G1, and G2 solutions (mg / ml)

[0047]

[0048] (2) The dried terahertz metamaterial absorber is placed in the terahertz system and measured using the transmission mode to collect the terahertz spectral information of the corresponding sample.

[0049] In this embodiment, the terahertz transmission spectrum of aflatoxin B2 solution was measured using a TAS7500 terahertz time-domain spectrometer from Advantest, Japan. This system consists of three main parts: a femtosecond laser, a terahertz emitter, and a terahertz detector. The spectral measurement range is 0.1–5.0 THz, and the instrument resolution is 7.6 GHz. To avoid interference from the surrounding environment and to obtain stable sample information, the system was preheated using an air compressor equipped with an air dryer before the experiment. Dry air was continuously filled into the optical cavity, and the humidity was maintained below 5% and the temperature was kept constant at 25 ± 0.5 °C using a hygrometer. During the experiment, after each sample was placed in the system, a two-minute waiting period was allowed until the system stabilized before measurement began. To minimize the influence of concentration residue on the metamaterial sheet, 20 μl of sample solution was pipetted onto the metamaterial sheet in ascending order of sample concentration. Since water strongly absorbs terahertz waves, the metamaterial sheet with the sample was placed in a drying oven at 50°C for 20 minutes before sample spectrum acquisition to ensure the solution on the metamaterial sheet was thoroughly dried. Finally, the metamaterial sheet with the sample was placed in the terahertz system and measured using transmission mode. To reduce the influence of random errors on the experimental results, two points were taken for each concentration sample, and 10 measurements were performed at each point. A total of 160 spectra were obtained from the eight concentration samples, and a total of 480 terahertz transmission spectra were obtained from the three aflatoxin samples.

[0050] (3) The original terahertz transmission spectra of aflatoxin B1, G1 and G2 were classified into two categories using a linear discriminant analysis algorithm.

[0051] First, terahertz time-domain spectra of 24 different concentrations of aflatoxin were collected, and the frequency domain signals were obtained by fast Fourier transform (FFT), which is expressed as formula (1).

[0052]

[0053] Where A(ω) represents the electric field amplitude, The phase difference between the reference signal and the sample signal, E(t), is the terahertz time-domain signal. The refractive index and absorption coefficient of the sample are obtained by formulas (2) and (3).

[0054]

[0055]

[0056] Where n(ω) is the refractive index, α(ω) is the absorption coefficient, ω is the frequency, k(ω) is the extinction coefficient, ρ(ω) is the amplitude ratio function, d is the sample thickness, and c is the speed of light in vacuum.

[0057] Subsequently, the terahertz transmission spectrum of clenbuterol hydrochloride was extracted by comparing the sample spectrum and the reference spectrum.

[0058] T = (A S / A R ) 2 (4)

[0059] Where A S and A R These are the amplitudes of the sample signal and the reference signal, respectively.

[0060] The "X" composite metamaterial combined with THz-TDS technology was used to perform qualitative identification and analysis of trace aflatoxin B2, G1, and G2 solutions. Figure 4 The transmission spectra of solutions containing three aflatoxins, B2, G1, and G2, on the surface of the metamaterial absorber are shown. Considering the significant noise interference in the frequency bands below 0.7 THz and above 3.5 THz, to ensure the accuracy of the measurements, the transmission spectra in the 0.7–3.5 THz frequency band were selected. The 368 spectral data points in this band were used for subsequent modeling.

[0061] in Figure 4 (a) is a magnified view of the transmission spectra of aflatoxin B2, G1, and G2 solutions at 1.2 THz. Figure 4 (b) is a magnified view of the terahertz transmission spectra of aflatoxin B2, G1, and G2 solutions at 2.0 THz. Figure 4 The mid-transmission peaks mainly show the transmission characteristics of the metamaterial itself, and it can be seen that there are transmission peaks of different intensities at around 1.2 THz and 2.0 THz. Figure 4 The red, blue, and green curves represent aflatoxin B2, aflatoxin G1, and aflatoxin G2 solutions, respectively. The straight line, dashed line, and dotted line represent high, medium, and low concentrations, respectively. Figure 4 The transmission peak curves of different types of aflatoxins clearly show significant differences. Aflatoxin G1 solution exhibits the highest transmission peak intensity, followed by aflatoxin B2 solution, while aflatoxin G2 solution shows the lowest intensity. Furthermore, from... Figure 4 As shown in (a), the transmission peak amplitudes of aflatoxins B2, G1, and G2 at around 1.2 THz gradually decrease with increasing solution concentration. Figure 4 (b) It can be seen that the transmission peak amplitudes of aflatoxin B2 and G2 at around 2.0 THz gradually decrease with increasing solution concentration. This provides a theoretical basis for further analysis of the transmission peak characteristics of metamaterials and the response relationship between trace aflatoxin concentration and type.

[0062] In this invention, all terahertz spectra of the collected samples are divided into a calibration set and a prediction set at a ratio of approximately 3:1 using the Kennard-Stone (KS) algorithm for subsequent model building. In this embodiment, the raw terahertz transmission spectra of aflatoxins B1, G1, and G2 are first classified into two categories using the Linear Discriminant Analysis (LDA) algorithm.

[0063] Linear Discriminant Analysis (LDA) is a supervised classification method that linearly transforms an n-dimensional feature vector (or sample) to an m-dimensional space (m < n). This results in samples of the same class being closely spaced, while samples of different classes are far apart, thus improving the separation of samples from different classes. Furthermore, LDA models can effectively predict unknown samples. The goal of the LDA algorithm is to find a transformation matrix that minimizes the within-class variance and maximizes the between-class variance.

[0064]

[0065] Among them, S W Let S be the within-class covariance matrix. B Let be the inter-class covariance matrix.

[0066] This invention establishes a binary mixed classification qualitative identification model for aflatoxin using raw terahertz spectra (LDA). The collected terahertz transmission spectra of three types of aflatoxin are input into the LDA model to establish the binary mixed classification qualitative model. First, the three spectra of aflatoxins B2, G1, and G2 are categorized and numbered: "1", "2", and "3". Then, the KS algorithm is used to select 360 spectra as the calibration set and 120 spectra as the prediction set for the establishment and external validation of the LDA model. Figure 5 The LDA binary mixture classification qualitative model planar classification graph for the prediction set samples, where Figure 5 (a) is a diagram of the B2 and G1 binary mixed classification model. Figure 5 (b) Diagram of the B2 and G2 binary mixed classification model Figure 5 (c) is a diagram of the G1 and G2 binary mixed classification model. From Figure 5 It can be seen that the class boundaries of the planar distribution of aflatoxin G1 and G2 are more obvious than those of the planar distribution of aflatoxin B2, G1 and aflatoxin B2, G2.

[0067] External validation of the established LDA models was performed using 80 prediction set samples. As shown in Table 2, the accuracy of the terahertz spectral LDA binary hybrid classification qualitative models for aflatoxins B2, G1, and G2 shows that there was one false positive for each of the two aflatoxin B2 and G1 models, with a prediction set accuracy of 98.75%. The terahertz spectral LDA binary hybrid classification qualitative models for aflatoxins G1 and G2 showed the highest accuracy, with both the calibration and prediction sets achieving 100% accuracy.

[0068] Table 2. Accuracy of the LDA binary mixed classification qualitative model for terahertz spectra of aflatoxin B2, G1, and G2.

[0069]

[0070] As the number of aflatoxin types increases, the effectiveness of the LDA model will further decrease. To improve the model's accuracy, this invention employs four feature extraction algorithms—PCA, UVE, SPA, and CARS—to extract the main features of the aflatoxin terahertz transmission spectrum, thereby improving model accuracy while reducing computational complexity. Subsequently, the feature-extracted aflatoxin terahertz transmission spectrum, along with the original spectrum, is input into KNN, RF, and SVM models, respectively. The optimal model is then determined through analysis and comparison.

[0071] (3) Extract features from terahertz spectral information.

[0072] Subsequently, to reduce redundant information and simplify the computational load of the model, four algorithms were introduced: Principal Component Analysis (PCA), Competitive Adaptive Reweighting Sampling (CARS), Uninformative Variable Elimination (UVE), and Successive Projection Algorithm (SPA) to extract features from the original terahertz transmission spectrum. Finally, K-Nearest Neighbor (KNN), Random Forest (RF), and Support Vector Machine (SVM) models were established, and the optimal model was selected through analysis and comparison.

[0073] (4) A qualitative analysis model for the terahertz transmission spectra of aflatoxin B2, G1, and G2 solutions was established using chemometric methods.

[0074] This invention establishes the following qualitative identification models: a KNN qualitative identification model based on terahertz spectroscopy combined with feature extraction algorithm for aflatoxin, an RF qualitative identification model based on terahertz spectroscopy combined with feature extraction algorithm for aflatoxin, and an SVM qualitative analysis model based on terahertz spectroscopy combined with feature extraction algorithm for aflatoxin.

[0075] In qualitative discrimination models, prediction accuracy and misclassification rate are often used to evaluate the model. The formula for calculating prediction accuracy is as follows:

[0076]

[0077] Where y i Let y be the number of correctly classified samples predicted by the model, and y be the total number of samples in the prediction set.

[0078] Besides prediction accuracy and misclassification rate, which can be used to evaluate qualitative discriminant models, confusion matrices are also an important evaluation method. Here, to better compare the actual class and the model's predicted class, confusion matrices are used to evaluate the model.

[0079] Establishment of a KNN qualitative identification model for aflatoxin based on terahertz spectroscopy combined with feature extraction algorithm:

[0080] The k-Nearest Neighbor (KNN) classification algorithm inputs the original spectral data and the feature spectral data extracted by four algorithms (PCA, UVE, SPA, and CARS) into the KNN model. The optimal model is obtained by searching for the nearest neighbor number k (k = 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21). Table 3 shows the identification results of the KNN qualitative model based on aflatoxin terahertz spectroscopy combined with feature extraction algorithms. Compared with the KNN model based on the original spectral data, the prediction accuracy of the KNN model after feature extraction by PCA and CARS algorithms is further improved. Both the PCA-KNN and CARS-KNN models have 5 misclassifications in their prediction sets, and the prediction set accuracy reaches 95.83%. Figure 6 The confusion matrices for the prediction sets of PCA-KNN and CARS-KNN models are given, where, Figure 6 (a) is the confusion matrix of the prediction set of the PCA-KNN model; Figure 6 (b) is the confusion matrix of the CARS-KNN model prediction set. It can be seen that the misclassification of the two models is consistent: 1 aflatoxin B2 sample was misclassified as aflatoxin G1, 2 aflatoxin G1 samples were misclassified as aflatoxin B2, and 2 aflatoxin G2 samples were misclassified as aflatoxin B2.

[0081] Table 3. Identification results of aflatoxin using the KNN qualitative model combined with terahertz spectroscopy and feature extraction algorithm.

[0082]

[0083] Establishment of an RF qualitative identification model for aflatoxin based on terahertz spectroscopy combined with feature extraction algorithm:

[0084] Random Forest (RF) is a machine learning method based on decision trees. Here, the original spectral data and the feature spectral data extracted by four algorithms (PCA, UVE, SPA, and CARS) are input into the RF qualitative identification model to qualitatively distinguish aflatoxins B2, G1, and G2. After repeated validation, the number of decision trees (Ntree) was set to 1000, and the number of pre-selected variables (Mtry) was the square root of the number of spectral variables rounded down. The RF model was then constructed, and the prediction accuracy was optimized to some extent. Table 4 shows the identification results of the KNN qualitative model using terahertz spectroscopy combined with feature extraction algorithms for aflatoxins. It can be seen that PCA-RF has the best qualitative identification effect, achieving 115 correct predictions out of 120 predicted samples, with a prediction accuracy of 95.83%.

[0085] Table 4. Identification results of aflatoxin using the RF qualitative model combined with terahertz spectroscopy and feature extraction algorithm.

[0086]

[0087] like Figure 7 (a) shows the relationship between the out-of-bag (OOB) error rate and the number of decision trees. As the number of decisions increases, the OOB error rate first decreases rapidly and then tends to stabilize. The minimum OOB error rate is 0.0611. Figure 7 (b) shows the confusion matrix of the PCA-RF model prediction set, revealing the misclassification distribution: one aflatoxin B2 sample was misclassified as aflatoxin G2, three aflatoxin G1 samples were misclassified as aflatoxin B2, and one aflatoxin G2 sample was misclassified as aflatoxin B2. The overall misclassification rate was 4.17%.

[0088] Establishment of an SVM qualitative analysis model for aflatoxin terahertz spectroscopy combined with feature extraction algorithm:

[0089] SVM qualitative identification models for aflatoxins B2, G1, and G2 were established using The Unscrambler 10.4 software. The SVM type was selected as c-SVC, and two basis functions were compared: radial basis function (RBF) and linear basis function (Linear). Parameters were optimized using a grid search method. The optimal SVM model was found by combining the training model's sum parameter g and penalty parameter c. Table 5 shows the validation results of the SVM model combined with feature extraction methods for the qualitative analysis of the three aflatoxins. The overall prediction accuracy of the SVM model is better than that of the KNN and RF models. The SVM model based on the radial basis function (RBF) is generally better than the SVM model based on the linear basis function (Linear). The optimal model is the RBF-SVM model after selecting feature variables using the CARS algorithm. The calibration set accuracy reached 100%, with 119 correct predictions out of 120 predicted samples, resulting in a prediction accuracy of 99.17%. Its parameter combination is c = 100 and g = 0.015. Figure 8 The CARS-RBF-SVM model confusion matrix shows the misclassification distribution of the prediction set. Only one aflatoxin B2 sample was misclassified as aflatoxin G2, with an overall misclassification rate of only 0.83%.

[0090] Table 5. Validation results of SVM model combined with feature extraction method for three aflatoxins.

[0091]

[0092] (6) Obtain the terahertz spectral information of the sample containing aflatoxin to be tested and input it into each qualitative analysis model to obtain the type of aflatoxin in the sample to be tested.

[0093] Analysis revealed that different types of aflatoxin exhibited transmission peaks of varying intensities at approximately 1.2 THz and 2.0 THz in their terahertz transmission spectra. Comparative analysis of different models showed that the CARS-RBF-SVM model achieved the highest prediction accuracy, with a calibration set precision of 100%. Out of 120 predicted samples, 119 were correctly predicted, resulting in a prediction accuracy of 99.17%. Only one aflatoxin B2 sample was misclassified as aflatoxin G2, leading to an overall misclassification rate of only 0.83%. This study validated the feasibility of using a metamaterial sensor with an "X" composite bimodal structure combined with THz-TDS technology for the qualitative identification of trace aflatoxins B2, G1, and G2 solutions, providing a theoretical basis and a new detection method for the qualitative identification of trace aflatoxins. Furthermore, this research offers important insights for promoting the qualitative detection of other trace substances.

[0094] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A terahertz metamaterial absorber, characterized in that, The terahertz metamaterial absorber has the following parameters: a period P of 60 μm, a gold coating, and a silicon substrate; the terahertz metamaterial absorber includes a large cross-shaped structure in the center and several small cross-shaped structures around the large cross-shaped structure; the length L1 of the large cross-shaped structure is 30 μm, the width W of the large cross-shaped structure is 2 μm, and each small cross-shaped structure includes a cross-shaped main body and an outer frame portion located outside the main body; the main body and the outer frame portion are perpendicular; the length L2 of the main body is 25 μm, and the length L3 of the outer frame portion is 12 μm; the distance d between one side of the large cross-shaped structure and the center of the main body is 15 μm. The terahertz metamaterial absorber was prepared by the following method: The process utilizes 4-inch double-polished silicon wafers and glass sheets. The entire process includes deposition, pretreatment, photoresist coating, pre-baking, exposure, development, coating, hardening, etching, and stripping. First, 5nm of chromium is deposited on the silicon substrate, followed by 100nm of gold. After metal deposition, pretreatment is performed in an HMDS oven at 120℃ for 10 minutes. Following pretreatment, photoresist AZ6112 is dropped onto the silicon substrate surface and spread using a spin coater at 600 rpm. Then, the surface is etched at 4000 rpm. The film is uniformly formed and reaches a thickness of 1-2 μm. After homogenization, pre-baking is performed at 100℃ for 120 seconds. After pre-baking, exposure is performed, followed by manual development for 35 seconds using a 2.38% 3038 developer. After development, a stripper is used to remove residual adhesive from the surface. The stripper power is set to 200W for 3 minutes. After stripping, hardening is performed by baking on a hot plate at 110℃ for 120 seconds. Then, IBE etching is performed, which involves bombarding the sample surface with an ion beam of a certain concentration to achieve the etching purpose. After etching, the photoresist mask is extracted by ultrasonication with acetone for 1 hour. After removal, the photoresist mask is observed under a microscope to see if it has been completely removed. If it has not been completely removed, it needs to be placed in the acetone solution for ultrasonication until the photoresist mask is completely removed.

Images