Machine learning assisted sers substrate based method for identifying ophiopogon japonicus

Abstract

This invention relates to the field of Ophiopogon japonicus identification technology, and in particular to a method for identifying Ophiopogon japonicus based on a machine learning-assisted SERS substrate. The method includes preparing an AuNBPs@ZIF-8@AuNPs composite SERS substrate; mixing and incubating the Ophiopogon japonicus extract to be tested with the AuNBPs@ZIF-8@AuNPs composite SERS substrate at the optimal detection concentration to form a mixed liquid; collecting the SERS spectral data of the mixed liquid; and inputting the SERS spectral data into a machine learning model to output the identification result. This invention prepares a high-purity AuNBPs@ZIF-8 solution, then uses NaBH4 to in-situ reduce HAuCl4 adsorbed on the ZIF-8 surface, ultimately constructing an AuNBPs@ZIF-8@AuNPs composite SERS substrate. Subsequently, Ophiopogon japonicus extract is mixed with this SERS material and incubated, and sufficient spectral data is collected using a portable Raman spectrometer. The spectral data is then imported into a pre-built machine learning model for training and optimization. Finally, an intelligent identification system is used to identify the origin and authenticity of Ophiopogon japonicus, enabling rapid and accurate identification of Ophiopogon japonicus from different origins and its adulterants.

Description

Technical Field

[0001] This invention relates to the field of Ophiopogon japonicus identification technology, and in particular to a method for identifying Ophiopogon japonicus based on machine learning-assisted SERS. Background Technology

[0002] As a highly representative Yin-nourishing Chinese medicinal herb in traditional Chinese medicine, the correlation between its medicinal value and active ingredient system has become a focus of industry research. Modern pharmacological studies have confirmed that the polysaccharides, amino acids, sterols, and ophiopogon saponins abundant in ophiopogon not only form the material basis for its traditional effects of nourishing Yin and moistening the lungs, and calming the mind, but also demonstrate diverse pharmacological activities such as antioxidation, immunomodulation, anti-inflammation, and potential anti-tumor activity in modern research. However, significant variety differences and quality irregularities exist in the ophiopogon market, posing a serious challenge to the stable realization of its medicinal value. Currently, the ophiopogon circulating in the market is mainly from Sichuan and Zhejiang, while... Due to core differences in growth cycle, intrinsic active ingredients (such as flavonoids, polysaccharides, saponins), and trace element content, these varieties exhibit vastly different quality, clinical efficacy, and market value, placing higher demands on "precise selection." Existing identification techniques for Ophiopogon japonicus have significant limitations: First, traditional sensory identification methods heavily rely on personal experience, are highly subjective, have low accuracy, and are difficult to achieve objective and reproducible identification; second, while sophisticated instrumental analysis methods such as high-performance liquid chromatography (HPLC) offer high accuracy, they suffer from drawbacks such as complex operation, time consumption, high cost, and the need for a professional laboratory environment, failing to meet the immediate needs of rapid on-site testing.

[0003] Surface-enhanced Raman spectroscopy (SERS), as a rapid, highly sensitive, and non-destructive fingerprint analysis technique, provides a new pathway for the rapid identification of traditional Chinese medicinal materials. However, traditional SERS substrates have limitations in sensitivity and stability, restricting their detection performance. Meanwhile, although machine learning (ML) technology can effectively process complex spectral data, the development and application of traditional ML algorithms rely on specialized programming software, making them complex to operate and posing a high technical barrier for the main users of traditional Chinese medicinal material identification, resulting in difficulties in technology implementation.

[0004] Therefore, there is an urgent need in this field for a rapid on-site identification technology for Ophiopogon japonicus that integrates high specificity, high sensitivity, ease of operation, and low usage threshold.

[0005] Chinese Patent Publication No. CN120275534A discloses a method for quality detection of Ophiopogon japonicus, a method for establishing an HPLC-ELSD fingerprint of Ophiopogon japonicus, and a method for identifying Ophiopogon japonicus with the core removed. The method employs high-performance liquid chromatography (HPLC) with the following chromatographic conditions: Kromasil-100-5-C18 column (4.6 mm × 250 mm, 5 μm); column temperature 35℃, drift tube temperature 100℃, gas flow rate 3.0 L / min; injection volume 20 μL; mobile phase acetonitrile (A)-water (B) solution, gradient elution: 0–20 min, 20%–38% A; 20–30 min, 38%–55% A; 30–50 min, 55%–70% A; flow rate 1.0 mL / min; equilibration 10 min. Therefore, existing Ophiopogon japonicus identification techniques lack a method that combines speed, accuracy, and on-site detection. Summary of the Invention

[0006] Therefore, this invention provides a method for identifying Ophiopogon japonicus based on a machine learning-assisted SERS substrate, in order to overcome the problem of low accuracy in the identification of Ophiopogon japonicus due to insufficient sensitivity and stability of the SERS substrate in the prior art.

[0007] To achieve the above objectives, this invention provides a method for identifying the traditional Chinese medicine Ophiopogon japonicus based on a machine learning-assisted SERS basis, comprising:

[0008] Step S1: Prepare AuNBPs@ZIF-8@AuNPs composite SERS substrate;

[0009] Step S2: Prepare the Ophiopogon japonicus extract to be tested, and mix and incubate the Ophiopogon japonicus extract to be tested with the AuNBPs@ZIF-8@AuNPs composite SERS substrate at the optimal detection concentration to form a mixed liquid;

[0010] Step S3: Use a Raman spectrometer to acquire SERS spectral data of the mixed liquid;

[0011] The SERS spectral data includes text datasets and image datasets;

[0012] Step S4: Input the SERS spectral data into a pre-trained machine learning model to output the identification result through the identification system.

[0013] Further, in step S1, the preparation of the AuNBPs@ZIF-8@AuNPs composite SERS substrate includes:

[0014] Step S101: Synthesize AuNBPs@ZIF-8 solution by chemical reduction method;

[0015] Step S102: Mix the HAuCl4 solution with the AuNBPs@ZIF-8 solution at the optimal mixing time to form a mixed solution;

[0016] Step S103: Add NaBH4 solution of optimal concentration to the mixed solution for in-situ reduction and stir under ice bath conditions to obtain AuNBPs@ZIF-8@AuNPs composite SERS substrate.

[0017] Furthermore, in step S4, the machine learning model is a KNN model or a CNN model;

[0018] When the input SERS spectral data is a text dataset, the KNN model is invoked for recognition.

[0019] When the input SERS spectral data is an image dataset, the CNN model is invoked for recognition.

[0020] Furthermore, the parameters of the KNN model are: n_neighbors=5, weights='uniform', algorithm='auto';

[0021] The CNN model is a 2D-CNN model containing three convolutional layers, three max pooling layers, one Dropout layer, and three fully connected layers.

[0022] Further, in step S4, the training process of the machine learning model includes:

[0023] SERS spectral data from the roots of Sichuan Ophiopogon japonicus, Zhejiang Ophiopogon japonicus, mountain Ophiopogon japonicus, and Lophatherum gracile were collected to construct a dataset;

[0024] The dataset is divided into a training set and a test set;

[0025] The training set is input into the machine learning algorithm for training, and the model performance is verified through the test set to obtain a trained machine learning model.

[0026] The SERS spectral data were acquired using a portable Raman spectrometer with an excitation source of 785 nm and a spectral measurement range of 200 cm⁻¹. -1 Up to 2000cm -1 The power is 100mW and the acquisition time is 3000ms.

[0027] Furthermore, the optimal mixing time is 6 hours, and the optimal concentration is 0.1 g / L;

[0028] In step S2, the incubation time is 20 minutes, the optimal detection concentration is 0.1 nM, and the volume ratio of the Ophiopogon japonicus extract to the AuNBPs@ZIF-8@AuNPs composite SERS substrate is 9:1.

[0029] Furthermore, the identification result includes the category of the Ophiopogon japonicus extract to be tested, which includes, but is not limited to, Sichuan Ophiopogon japonicus, Zhejiang Ophiopogon japonicus, mountain Ophiopogon japonicus, or Lophatherum gracile root.

[0030] Furthermore, the identification system includes:

[0031] The front-end interaction module is used for users to upload SERS spectral data files, the file format of which includes text format or image format;

[0032] The backend server module is connected to the frontend interaction module and is used to receive the SERS spectral data file and call the corresponding machine learning model to identify the SERS spectral data file.

[0033] The result return module is used to return the authentication result output by the backend server module to the frontend interaction module and display it to the user.

[0034] Furthermore, the enhancement factor of the AuNBPs@ZIF-8@AuNPs composite SERS substrate is 2.4 × 10⁻⁶. 7 Uniformity RSD < 5%, reproducibility RSD < 5%.

[0035] Furthermore, in the AuNBPs@ZIF-8@AuNPs composite SERS substrate, the average particle size of ZIF-8 is 320 nm, and the average particle size of AuNPs generated in situ on its surface is 14 nm.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows: In this embodiment, a high-purity AuNBPs@ZIF-8 solution is first prepared by chemical reduction, and then HAuCl4 adsorbed on the ZIF-8 surface is reduced in situ using NaBH4 to finally construct an AuNBPs@ZIF-8@AuNPs composite SERS substrate; subsequently, the Ophiopogon japonicus extract is mixed with the SERS material and incubated, and sufficient spectral data is collected using a portable Raman spectrometer; the spectral data is imported into a pre-built machine learning model to complete training and optimization; finally, the Ophiopogon japonicus is identified by an intelligent identification system to distinguish between authentic and counterfeit products. The authentic identification identifies Sichuan Ophiopogon japonicus or Zhejiang Ophiopogon japonicus, and the counterfeit identification identifies mountain Ophiopogon japonicus or Lophatherum gracile root; thus achieving rapid and accurate identification of Ophiopogon japonicus from different producing areas and its adulterants. Attached Figure Description

[0037] Figure 1This is a schematic diagram illustrating the preparation of the AuNBPs@ZIF-8@AuNPs SERS substrate and its rapid detection of Ophiopogon japonicus using machine learning, as described in an embodiment of the present invention.

[0038] Figure 2 The ultraviolet absorption spectra of AuNBPs, AuNBPs@ZIF-8, and AuNBPs@ZIF-8@AuNPs are shown in the embodiments of the present invention.

[0039] Figure 3 This is a schematic diagram illustrating the characterization of AuNBPs@ZIF-8@AuNPs using TEM and SEM according to an embodiment of the present invention. Figure 3 In the figure, A is a TEM image of AuNBPs in an embodiment of the present invention; Figure 3 B in the figure is a TEM image of AuNBPs@ZIF-8 in the embodiment of the present invention; Figure 3 C in the figure represents the TEM image of AuNBPs@ZIF-8@AuNPs in the embodiment of the present invention; Figure 3 In the figure, D is the SEM image of AuNBPs@ZIF-8@AuNPs in the embodiment of the present invention;

[0040] Figure 4 This is an energy dispersive X-ray spectroscopy (EDX) elemental mapping diagram of AuNBPs@ZIF-8@AuNPs according to an embodiment of the present invention;

[0041] Figure 5 This is a histogram showing the particle size distribution of ZIF-8 and AuNPs in AuNBPs@ZIF-8@AuNPs in an embodiment of the present invention.

[0042] Figure 6 This is a schematic diagram illustrating the Raman signal intensity testing of 4-NTP after incubation with a solution of AuNBPs@ZIF-8@AuNPs in an embodiment of the present invention; wherein, Figure 6 In the figure, A represents the Raman spectroscopy of HAuCl4 and AuNBPs@ZIF-8 at different mixing times in the embodiments of the present invention; Figure 6 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, which serves as the characteristic peak, is represented by error bars indicating ±SD (n=3).

[0043] Figure 7 These are TEM images and UV absorption spectra of HAuCl4 and AuNBPs@ZIF-8 at different mixing times from 1 h to 18 h, as described in this embodiment of the invention. Figure 7 The AE in the figure is a TEM image of HAuCl4 and AuNBPs@ZIF-8 at different mixing times from 1h to 18h in the embodiments of the present invention. Figure 7F in the figure represents the UV absorption spectra of HAuCl4 and AuNBPs@ZIF-8 from different mixing times from 1h to 18h in the embodiments of the present invention.

[0044] Figure 8 This is a SERS spectrum and signal intensity histogram of NaBH4 concentration optimized according to an embodiment of the present invention, wherein... Figure 8 In this paper, A represents the SERS spectrum of NaBH4 concentration optimization in an embodiment of the present invention. Figure 8 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, which serves as the characteristic peak, is represented by error bars indicating ±SD (n=3).

[0045] Figure 9 The images shown are TEM images and UV absorption spectra of AuNBPs@ZIF-8@AuNPs prepared with different concentrations of NaBH4 in this embodiment of the invention. Figure 9 AC in the figure represents TEM images of AuNBPs@ZIF-8@AuNPs prepared with different concentrations of NaBH4 in the embodiments of the present invention. Figure 9 In this embodiment, D represents the ultraviolet absorption spectra of AuNBPs@ZIF-8@AuNPs prepared with different concentrations of NaBH4 in this invention.

[0046] Figure 10 This is a SERS spectrum and signal intensity histogram of HAuCl4 volume optimized according to an embodiment of the present invention, wherein... Figure 10 In this paper, A represents the SERS spectrum of HAuCl4 volume optimized according to an embodiment of the present invention; Figure 10 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, which serves as the characteristic peak, is represented by error bars indicating ±SD (n=3).

[0047] Figure 11 The images shown are TEM images and UV absorption spectra of different volumes of HAuCl4, ranging from 20 μL to 140 μL, in the embodiments of the present invention. Figure 11 The AE images in the figures are TEM images of different volumes of HAuCl4, from 20 μL to 140 μL, in the embodiments of the present invention. Figure 11 In this paper, F represents the ultraviolet absorption spectra of different volumes of HAuCl4, from 20 μL to 140 μL, in the embodiments of the present invention.

[0048] Figure 12 This is a SERS spectrum and signal intensity histogram of AuNBPs@ZIF-8@AuNPs solution concentration optimization according to an embodiment of the present invention. Figure 12 In this paper, A represents the SERS spectrum of the AuNBPs@ZIF-8@AuNPs solution concentration optimization in this embodiment of the invention; Figure 12In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, which serves as the characteristic peak, is represented by error bars indicating ±SD (n=3).

[0049] Figure 13 This is a SERS spectrum and signal intensity histogram of the AuNBPs@ZIF-8@AuNPs solution with optimized incubation time for signal molecules according to an embodiment of the present invention. Figure 13 In this paper, A represents the SERS spectrum of the AuNBPs@ZIF-8@AuNPs solution with optimized incubation time for signal molecules in an embodiment of the present invention. Figure 13 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, which serves as the characteristic peak, is represented by error bars indicating ±SD (n=3).

[0050] Figure 14 This is a performance comparison chart of the SERS substrate according to an embodiment of the present invention;

[0051] Figure 15 For the embodiments of the present invention, 20 points (A, B) were randomly selected on the AuNBPs@ZIF-8@AuNPs substrate and 10 self-assembled substrates (C, D) prepared in different batches were measured. -6 Line graph and bar chart of M's 4-NTP SERS signal;

[0052] Figure 16 The images and SERS spectra of Ophiopogon japonicus from different origins and its adulterants are shown in the embodiments of the present invention. Figure 16 A1-A4 in the figures are photographs of Ophiopogon japonicus from different origins and its adulterated medicinal materials in the embodiments of the present invention; Figure 16 B in the figure represents the SERS spectra of Ophiopogon japonicus from different origins and its adulterated medicinal materials in the embodiments of the present invention;

[0053] Figure 17 This is a diagram of the 2D-CNN architecture according to an embodiment of the present invention;

[0054] Figure 18 The following are confusion matrix diagrams and ROC curves for several traditional machine learning models according to embodiments of the present invention. Figure 18 In the diagram, AG represents the confusion matrix of several traditional machine learning models in the embodiments of this invention. Figure 18 H represents the ROC curves of several traditional machine learning models in the embodiments of the present invention;

[0055] Figure 19 This is a graph showing the accuracy and loss variation during the training process of the CNN model in an embodiment of the present invention. Figure 19 In the figure, A represents the accuracy change during the training process of the CNN model in this embodiment of the invention; Figure 19B in the figure represents the loss variation diagram during the training process of the CNN model in this embodiment of the invention;

[0056] Figure 20 The diagram shows the confusion matrix and ROC curve of the CNN model in this embodiment of the invention. Figure 20 In the diagram, A represents the confusion matrix of the CNN model in this embodiment of the invention. Figure 20 In the figure, B represents the ROC curve of the CNN model in the embodiment of the present invention;

[0057] Figure 21 This is a flowchart illustrating the identification process of Ophiopogon japonicus according to an embodiment of the present invention;

[0058] Figure 22 This is a diagram of the interface for recognizing Ophiopogon japonicus according to an embodiment of the present invention, wherein, Figure 22 In this context, A represents the "Home" page interface of the smart mini-program in this embodiment of the invention; Figure 22 In this context, BC refers to the "Result" page of the smart mini-program in this embodiment of the invention. Detailed Implementation

[0059] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0060] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0061] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate the direction or positional relationship, are based on the direction or positional relationship shown in the drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0062] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0063] Please see Figure 1The diagram shown is a flowchart illustrating the identification method for the traditional Chinese medicine Ophiopogon japonicus based on a machine learning-assisted SERS basis according to an embodiment of the present invention. The present invention provides a method for identifying the traditional Chinese medicine Ophiopogon japonicus based on a machine learning-assisted SERS basis, comprising:

[0064] Step S1: Prepare AuNBPs@ZIF-8@AuNPs composite SERS substrate;

[0065] Step S2: Prepare the Ophiopogon japonicus extract to be tested, and mix and incubate the Ophiopogon japonicus extract to be tested with the AuNBPs@ZIF-8@AuNPs composite SERS substrate at the optimal detection concentration to form a mixed liquid;

[0066] Step S3: Use a Raman spectrometer to acquire SERS spectral data of the mixed liquid;

[0067] The SERS spectral data includes text datasets and image datasets;

[0068] Step S4: Input the SERS spectral data into a pre-trained machine learning model to output the identification result;

[0069] The identification results include the category of the Ophiopogon japonicus extract to be tested, which includes Ophiopogon japonicus from Sichuan, Ophiopogon japonicus from Zhejiang, Ophiopogon japonicus from the mountains, or Lophatherum gracile root.

[0070] In this embodiment, a high-purity AuNBPs@ZIF-8 solution was first prepared by chemical reduction. Then, HAuCl4 adsorbed on the ZIF-8 surface was reduced in situ using NaBH4 to finally construct an AuNBPs@ZIF-8@AuNPs composite SERS substrate. Subsequently, the Ophiopogon japonicus extract was mixed with the SERS material and incubated. Sufficient spectral data were collected using a portable Raman spectrometer. The spectral data were then imported into a pre-built machine learning model for training and optimization. Finally, an intelligent identification system was used to identify the origin and authenticity of Ophiopogon japonicus. The origin identification identified Sichuan Ophiopogon japonicus or Zhejiang Ophiopogon japonicus, while the authenticity identification identified mountain Ophiopogon japonicus or Lophatherum gracile root. This enabled rapid and accurate identification of Ophiopogon japonicus from different origins and its adulterants.

[0071] This embodiment creatively introduces metal-organic framework (MOF) materials. ZIF-8, a zeolite imidazole ester framework material, is a typical MOF material with a tunable pore structure (pore size and distribution can be precisely controlled by adjusting synthesis conditions), ultra-high specific surface area, excellent chemical stability, and abundant multifunctional active sites. Combining it with metal nanoparticles exhibiting localized surface plasmon resonance (LSPR) effects not only enables efficient enrichment and screening of target active ingredients in Ophiopogon japonicus extract but also significantly improves the detection sensitivity and long-term stability of the SERS substrate, providing a basis for... The acquisition of stronger and more precise SERS characteristic signals provides a material basis. Therefore, by utilizing the high specific surface area and molecular sieve effect of the ZIF-8 metal-organic framework, efficient and selective enrichment of characteristic small molecule metabolites in the complex extract of Ophiopogon japonicus is achieved. Simultaneously, through a precisely controlled in-situ reduction process, uniformly distributed and uniformly sized (approximately 14 nm) AuNPs are generated on the ZIF-8 surface, forming strong electromagnetic coupling with the core AuNBPs, resulting in order-of-magnitude signal amplification. In particular, through the above synergistic design, the SERS substrate obtained in this embodiment reaches 2.4 × 10⁻⁶. 7 The enhancement factor is much higher than that of conventional gold nanorods or simply modified SERS substrates. In addition, the substrate exhibits excellent uniformity (RSD < 5%) and reproducibility (RSD < 5%), solving the industry pain points of large signal fluctuations and significant batch-to-batch differences in traditional SERS substrates. This lays a solid foundation for the rapid, stable, and quantitative identification of Chinese medicinal materials. Moreover, these performance indicators are achieved through the specific structural design and precise process parameters (such as 6h mixing time and 0.1g / L NaBH4 concentration) in this embodiment.

[0072] In this embodiment, a dual-model adaptation architecture of KNN model and 2D-CNN model is adopted to be suitable for different data formats (text-type spectral data and image-type spectral maps) in actual application scenarios. That is, the optimal model is called according to the input data format to ensure the accurate adaptation of the identification method, improve the practicality and accuracy of Ophiopogon japonicus identification, and thus realize the rapid, accurate and stable identification of the authenticity and genuineness of Ophiopogon japonicus.

[0073] For text datasets, the KNN algorithm with a comprehensive recognition rate of 98% was selected by comparing various traditional machine learning algorithms. For image datasets, a multi-layer 2D-CNN model was designed. By increasing the network depth (three convolutional layers and fully connected layers) and combining techniques such as Dropout, the model's ability to extract subtle features and generalize in spectral images was effectively improved, ultimately increasing the recognition accuracy from the initial 0.87 to 0.98. This achieves efficient processing of multi-source spectral data.

[0074] Specifically, in step S1, the preparation of the AuNBPs@ZIF-8@AuNPs composite SERS substrate includes:

[0075] Step S101: Synthesize AuNBPs@ZIF-8 solution by chemical reduction method;

[0076] Step S102: Mix the HAuCl4 solution with the AuNBPs@ZIF-8 solution at the optimal mixing time to form a mixed solution;

[0077] Step S103: Add a 0.1 g / L NaBH4 solution to the mixed solution for in-situ reduction, and stir for 1 hour under ice bath conditions to obtain AuNBPs@ZIF-8@AuNPs composite material. Add the composite material to silicon wafers, PDMS, PMMA, glass slides, etc. at the optimal concentration to obtain AuNBPs@ZIF-8@AuNPs composite SERS substrate. That is, after adding the composite material solution to substrates such as silicon wafers, PDMS, PMMA, and glass slides and drying them, the SERS substrate is obtained. The analyte is added to this substrate and then detected by Raman spectroscopy.

[0078] In this embodiment, the preparation process of AuNBPs@ZIF-8 is as follows: referring to and improving upon the method of Wang Jing et al., 18.0 mL of dimethylimidazole (1.32 M) aqueous solution was placed in a 100.0 mL round-bottom flask, and 2.7 mL of CTAB aqueous solution (1.0 mM) was added. The mixture was stirred at 150 rpm for 10 min. Then, 16.0 mL of ZnNO3 solution (24.0 mM), 16.0 mL of AuNBPs aqueous solution (3.0 nM), and 1.6 mL of CTAB aqueous solution (5.0 mM) were added sequentially. After stirring for another 5 min, the mixture was allowed to stand overnight at room temperature, centrifuged at 3000 rpm for 10 min, and the precipitate was dispersed in methanol and stored at room temperature.

[0079] The preparation process of AuNBPs@ZIF-8@AuNPs is as follows: 1.0 mL of AuNBPs@ZIF-8 solution was placed in a 2.0 mL centrifuge tube, and 100.0 μL of HAuCl4·4H2O solution was added. The mixture was stirred on a rotary mixer for 6 h, then transferred to a 2.0 mL round-bottom flask. 0.1 g / L NaBH4 solution was rapidly added at 1000 rpm, and the mixture was stirred for 1 h in an ice bath. The mixture was then centrifuged at 3000 rpm for 10 min using a refrigerated centrifuge. The precipitate was dispersed in methanol and stored at room temperature for later use. The precipitate contained chloroauric acid (HAuCl4·4H2O, 99.9%), trisodium citrate (99%), silver nitrate (AgNO3, 99.8%), hydrochloric acid (HCl, 12 M), and nitric acid. Zinc (ZnNO3, 99%) was purchased from Sinopharm Chemical Reagent Co., Ltd.; hexadecyltrimethylammonium bromide (CTAB) and hexadecyltrimethylammonium chloride (CTAC, 98%) were purchased from Sangon Biotech (Shanghai) Co., Ltd.; sodium borohydride (NaBH4, 98%) was purchased from Aladdin Reagent Co., Ltd.; ascorbic acid (Vc, 99%) was purchased from Sigma-Aldrich (USA) Reagent Co., Ltd.; the refrigerated centrifuge was Eppendorf, model 5418R, equipped with 18 x 1.5 / 2.0 mL airtight centrifuge tube rotors.

[0080] Specifically, in step S2, the preparation of the Ophiopogon japonicus extract to be tested includes:

[0081] Step S201: Dry and pulverize the raw material to be tested;

[0082] Step S202: Add water at a material-to-liquid ratio of 1g:5mL and extract in a 50℃ water bath for 2.5 hours;

[0083] In step S203, the extract cooled to room temperature is filtered through a 0.45 μm polyethersulfone membrane, and the filtrate is brought to a final volume of 2 mL to obtain the Ophiopogon japonicus extract for SERS detection.

[0084] In this embodiment, the raw material to be tested is Ophiopogon japonicus and its adulterants. The process of preparing the Ophiopogon japonicus extract to be tested is as follows: all Ophiopogon japonicus and its adulterants are dried overnight in an oven at 50°C. The dried sample is then pulverized using a small Chinese herbal medicine pulverizer. The powder is passed through a 200-mesh Chinese herbal medicine filter screen and then separately prepared. 1g of powder is placed in a 10mL round-bottom flask, 5mL of water is added, and the mixture is heated in a 50°C water bath for 2.5h. After the extract cools to room temperature, it is passed through a 0.45μm polyethersulfone (PES) membrane. The final filtrate is brought to a final volume of 2mL and stored in a refrigerator at 4°C for later use.

[0085] Example 1: Preparation and characterization of SERS substrate;

[0086] See Figure 2As shown, this is the ultraviolet absorption spectrum of AuNBPs, AuNBPs@ZIF-8, and AuNBPs@ZIF-8@AuNPs in this embodiment of the invention; in this embodiment, the three synthesized materials were characterized by a UV-Vis spectrophotometer, as shown below. Figure 2 As shown, AuNBPs at 511 cm -1 There is a transverse absorption peak at 776 cm⁻¹. -1 There is a longitudinal absorption peak at the surface of AuNBPs. After ZIF-8 is deposited on the surface of AuNBPs, the transverse plasmon resonance peak of AuNBPs shows a slight shift, while the longitudinal plasmon resonance peak shows a significant red shift. Compared with water (1.333), ZIF-8 has a higher refractive index (1.352), which leads to plasmon shift. After AuNPs are generated on the surface of ZIF-8, the refractive index changes again, the ultraviolet peak shows a slight blue shift, and the peak width becomes wider.

[0087] See Figure 3 As shown, Figure 3 In the diagram, A represents the TEM image of AuNBPs from an embodiment of the present invention. Figure 3 B in the figure is a TEM image of AuNBPs@ZIF-8 in an embodiment of the present invention. Figure 3 C in the figure represents the TEM image of AuNBPs@ZIF-8@AuNPs in the embodiment of the present invention. Figure 3 D in the figure represents the SEM image of AuNBPs@ZIF-8@AuNPs in the embodiment of the present invention. The AuNBPs@ZIF-8@AuNPs were further characterized by TEM and SEM. It can be clearly seen that AuNPs of relatively uniform size and distribution were generated on the surface of ZIF-8. The TEM image shows that AuNBPs are encapsulated in ZIF-8, and the SEM image shows the three-dimensional structure of ZIF-8.

[0088] See Figure 4 As shown, it is an energy dispersive X-ray energy dispersive spectroscopy (EDX) elemental mapping diagram of AuNBPs@ZIF-8@AuNPs in the embodiment of the present invention; Figure 4 In the image, A represents the energy dispersive X-ray spectroscopy (EDX) elemental mapping of AuNBPs@ZIF-8@AuNPs. The upper left corner is a bright-field image showing the overall morphology of the material. The upper right corner shows the remaining images, which are mappings of Zn and Au elements. The blue / yellow areas correspond to the enrichment regions of Zn and Au elements, indicating the co-distribution of Zn and Au within the nanoparticles, further illustrating the formation of the core-shell structure of the material. Figure 4B in the diagram represents the EDS plot, which quantitatively shows the atomic percentages of Au, Zu, C, N, and O. Zn (3.2%) and Au (1.55%) are the main metallic elements in the material. Through energy-dispersive X-ray spectroscopy (EDX) elemental mapping characterization, the core-shell structures of Au, ZIF-8, and Au can be observed, such as... Figure 4 As shown, elemental analysis by EDX revealed the presence of Au, Zu, C, N, and O elements on the AuNBPs@ZIF-8@AuNPs substrate, indicating the successful synthesis of AuNBPs@ZIF-8@AuNPs nanomaterials.

[0089] See Figure 5 As shown, this is a histogram of the particle size distribution of ZIF-8 and AuNPs in AuNBPs@ZIF-8@AuNPs according to an embodiment of the present invention; in this embodiment, the particle sizes of ZIF-8 and AuNPs in AuNBPs@ZIF-8@AuNPs were statistically analyzed to obtain the particle size distribution histogram, as shown. Figure 5 As shown, the average particle size of ZIF-8 is around 320 nm, while the particle size of AuNPs is approximately 14 nm.

[0090] Example 2: Condition optimization of SERS substrate;

[0091] In this embodiment, the conditions for the SERS substrate include the optimal mixing time for mixing the HAuCl4 solution with the AuNBPs@ZIF-8 solution, the optimal concentration of the NaBH4 reducing agent for in-situ reduction, the optimal volume of the HAuCl4 precursor, and the optimal detection concentration and optimal incubation time of the AuNBPs@ZIF-8@AuNPs substrate solution during mixed incubation.

[0092] In this embodiment, the optimal mixing time is 6 hours, and the analysis process is as follows:

[0093] See Figure 6 A and Figure 6 As shown in B, Figure 6 In the figure, A represents the Raman spectroscopy of HAuCl4 and AuNBPs@ZIF-8 at different mixing times in the embodiments of the present invention. Figure 6 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram is used as a characteristic peak, with error bars representing ±SD (n=3). NaBH4 is used to in-situ reduce HAuCl4, which is fully adsorbed on the ZIF-8 surface, to generate AuNPs. The mixing time of HAuCl4 and AuNBPs@ZIF-8 is optimized to ensure sufficient adsorption of HAuCl4 onto the ZIF-8 surface. Figure 6 A and Figure 6 As shown in B, through 10 -6The performance of the SERS substrate was evaluated using M 4-NTP. As the mixing time increased, the SERS Raman signal intensity of 4-NTP gradually increased. After the mixing time reached 6 h, further increasing the mixing time did not significantly change the SERS signal intensity, indicating that the HAuCl4 adsorbed on the ZIF-8 surface did not reach saturation before 6 h. It reached saturation after the adsorption time reached 6 h. Further increasing the mixing time did not significantly change the amount of HAuCl4 adsorbed.

[0094] Continue reading Figure 7 As shown, Figure 7 The images shown in Figure AE are TEM images of HAuCl4 and AuNBPs@ZIF-8 mixed for different times from 1 h to 18 h in the embodiments of the present invention. Figure 7 In this embodiment, F represents the UV absorption spectra of HAuCl4 and AuNBPs@ZIF-8 at different mixing times from 1 h to 18 h in this invention embodiment; through... Figure 7 TEM images of AE and Figure 7 The ultraviolet absorption spectrum of F in the figure clearly demonstrates this point. Figure 7 A to Figure 7 The mixing times for E in the sample were 1 h, 3 h, 6 h, 10 h, and 18 h. In the early stages, when the mixing time for HAuCl4 and AuNBPs@ZIF-8 was relatively short, only a small portion of the added NaBH4 reacted with the HAuCl4 adsorbed on the ZIF-8 surface. The remaining NaBH4 directly damaged the ZIF-8. This can be seen from the TEM images at mixing times of 1 h and 3 h, which show varying degrees of damage to the ZIF-8. With further increases in mixing time, the ZIF-8 showed a tendency to remain intact, and the number of AuNPs on the surface remained relatively stable. Figure 7 The F in the figure shows that the UV spectral data for mixing times of 1h and 3h are similar, and the UV data for the other mixing times are similar. Therefore, the optimal mixing time is 6 hours.

[0095] In this embodiment, the optimal concentration of NaBH4 reducing agent is 0.1 g / L, and the analysis process is as follows:

[0096] See Figure 8 As shown, Figure 8 In the figure, A represents the SERS spectrum of NaBH4 concentration optimization in an embodiment of the present invention. Figure 8 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, representing the characteristic peaks, is shown with error bars indicating ±SD (n=3). The size of surface-generated AuNPs is crucial for SERS signal enhancement; the size of AuNPs can be controlled by adjusting the concentration of NaBH4, such as... Figure 8As shown, the SERS signal intensity was weak when the NaBH4 concentration was 0.01 g / L, and the signal intensity increased significantly when the NaBH4 concentration increased to 0.1 g / L, until the signal intensity weakened again when the NaBH4 concentration was 1 g / L.

[0097] Continue reading Figure 9 As shown, Figure 9 In the diagram, AC represents TEM images of AuNBPs@ZIF-8@AuNPs prepared with different concentrations of NaBH4 in the embodiments of this invention. Figure 9 In the figure, D represents the ultraviolet absorption spectra of AuNBPs@ZIF-8@AuNPs prepared with different concentrations of NaBH4 in the embodiments of the present invention; the reasons for the above-mentioned changes in signal intensity were observed by TEM and ultraviolet characterization. Figure 9 The concentrations of NaBH4 corresponding to AC in the samples were 0.01 g / L, 0.1 g / L, and 1 g / L, respectively. As the NaBH4 concentration increased, the size of AuNPs on the ZIF-8 surface also increased. However, when the NaBH4 concentration reached 1 g / L, damage to the ZIF-8 shell was clearly observed. This indicates that the rapid injection of high-concentration NaBH4, with its strong reducing properties, damaged the ZIF-8 shell, resulting in a weakening of the SERS signal intensity. Figure 9 The UV absorption spectrum of D in the image also shows that NaBH4 concentrations of 0.1 g / L and 1 g / L are significantly lower at 800 cm⁻¹ compared to 0.01 g / L. -1 The UV absorption peak at 0.01 g / L exhibited a blue shift, indicating that the AuNPs obtained from the reduction of NaBH4 by 0.01 g / L were smaller compared to other concentrations. Furthermore, it can be observed that the 1 g / L concentration showed a blue shift at 500 cm⁻¹. -1 The ultraviolet absorption peak at the point shifted significantly upward, indicating that the ZIF-8 shell was damaged. Therefore, the optimal concentration of NaBH4 reducing agent is 0.1 g / L.

[0098] In this embodiment, the optimal volume of the HAuCl4 precursor was 100.0 μL, and the analysis process is as follows:

[0099] See Figure 10 As shown, Figure 10 In this diagram, A represents the SERS spectrum of HAuCl4 volume optimized according to an embodiment of the present invention. Figure 10 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, representing the characteristic peaks, is shown with error bars indicating ±SD (n=3). The number of AuNPs generated in situ on the AuNBPs@ZIF-8 surface has a significant impact on the SERS signal. The number of AuNPs on the AuNBPs@ZIF-8 surface was optimized by controlling the volume of HAuCl4 and its corresponding change in NaBH4 volume. Figure 10It can be seen that as the volume of HAuCl4 increases, the SERS signal intensity gradually increases until the volume of HAuCl4 reaches 100 μL. Further increasing the volume of HAuCl4 will result in a decrease in the SERS signal intensity.

[0100] Continue reading Figure 11 As shown, Figure 11 The images shown in AE are TEM images of different volumes of HAuCl4, ranging from 20 μL to 140 μL, from the embodiments of the present invention. Figure 11 The volumes of HAuCl4 corresponding to each AE graph in the figure are 20 μL, 60 μL, 80 μL, 100 μL, and 140 μL, respectively. Figure 11 F in the figure represents the UV absorption spectra of different volumes of HAuCl4 from 20 μL to 140 μL in the embodiments of the present invention; the reasons for the above phenomena were characterized by TEM and UV. Figure 11 As shown, the study found that the number of AuNPs generated on the AuNBPs@ZIF-8 surface increased with the increase of HAuCl4 volume. However, when the volume reached 140 μL, a large number of HAuCl4 nucleated on its own in the solution, resulting in a decrease in the number of AuNPs on the AuNBPs@ZIF-8 surface. This led to a decrease in the SERS signal, and no significant change was observed in the UV spectrum. This indicates that the size of the AuNPs is small and cannot be well characterized by UV. Therefore, the optimal volume of the HAuCl4 precursor is 100.0 μL.

[0101] In this embodiment, the optimal detection concentration of the AuNBPs@ZIF-8@AuNPs substrate solution is 0.1 nM, and the analysis process is as follows:

[0102] See Figure 12 As shown, Figure 12 In this paper, A represents the SERS spectrum of the AuNBPs@ZIF-8@AuNPs solution concentration optimized in this embodiment of the invention. Figure 12 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram, representing the characteristic peaks, is shown with error bars indicating ±SD (n=3). The concentration of the AuNBPs@ZIF-8@AuNPs solution was optimized. Different concentrations resulted in variations in the spacing between AuNBPs@ZIF-8@AuNPs in the solution. Specifically, when the solution concentration was too low, the spacing between nanoparticles was too large, making it difficult to form strong "hot spots," resulting in a weak Raman signal. When the solution concentration was too high, nanoparticles stacked, partially covering the "hot spots," further weakening the Raman signal. Figure 12As shown, the Raman signal intensity of 4-NTPs changes with the concentration of the AuNBPs@ZIF-8@AuNPs solution. Initially, when the concentration of the AuNBPs@ZIF-8@AuNPs solution is low, the SERS signal becomes stronger with increasing concentration. When the concentration reaches a critical value of 0.1 nM, further increases in the concentration of the AuNBPs@ZIF-8@AuNPs solution lead to a decrease in SERS. This indicates that after 0.1 nM, some AuNBPs@ZIF-8@AuNPs in the solution exhibit partial aggregation, resulting in the coverage of some "hot spots" and a decrease in signal intensity. Therefore, the optimal detection concentration of the AuNBPs@ZIF-8@AuNPs substrate solution is 0.1 nM.

[0103] In this embodiment, the optimal incubation time is 20 minutes, and the analysis process is as follows:

[0104] See Figure 13 As shown, Figure 13 In this paper, A represents the SERS spectrum of the AuNBPs@ZIF-8@AuNPs solution after optimization of the incubation time with the signal molecule in this embodiment of the invention. Figure 13 In this context, B represents an embodiment of the invention with a height of 1380cm. -1 The SERS signal intensity histogram is presented as a characteristic peak, with error bars representing ±SD (n=3). After incubating a solution of the signal molecule 4-NTP with AuNBPs@ZIF-8@AuNPs for a period of time, the Raman signal intensity of 4-NTP was measured. Figure 13 As shown, the SERS signal intensity of 4-NTPs increases with the increase of incubation time. After 20 minutes, the signal intensity no longer changes. This indicates that the signal molecules were not completely adsorbed onto AuNBPs@ZIF-8@AuNPs at the beginning. As the incubation time increases, the amount of 4-NTPs adsorbed on AuNBPs@ZIF-8@AuNPs reaches saturation. After that, further increasing the incubation time does not significantly change the amount of adsorbed signal molecules, nor does it significantly change the SERS signal intensity. Therefore, the optimal incubation time is 20 minutes.

[0105] Example 3, Performance analysis of SERS substrate;

[0106] In this embodiment, the enhancement factor of the SERS substrate is 2.4 × 10⁻⁶. 7 The analysis process is as follows:

[0107] See Figure 14 As shown, this is a performance comparison chart of the SERS substrate according to an embodiment of the present invention; to explore the enhancement performance of the AuNBPs@ZIF-8@AuNPs substrate, it was compared with AuNBPs and AuNBPs@ZIF-8. 10 -6SERS spectra of 4-NTPs on three different substrates: AuNBPs, AuNBPs@ZIF-8, and AuNBPs@ZIF-8@AuNPs, as shown in the figure. Figure 14 As shown, both AuNBPs and AuNBPs@ZIF-8 exhibited SERS enhancement of the 4-NTP signal. However, compared to AuNBPs@ZIF-8@AuNPs, the SERS signal was significantly weaker. While AuNBPs@ZIF-8 had a weaker SERS signal than AuNBPs, it demonstrated better enrichment and screening of small molecules. AuNBPs@ZIF-8@AuNPs combined the advantages of both, possessing excellent SERS enhancement performance while also exhibiting the high selectivity and high enrichment capacity of ZIF-8. The calculated EF value was 2.4 × 10⁷, demonstrating that the AuNBPs@ZIF-8@AuNPs substrate is more effective than using AuNBPs or AuNBPs@ZIF-8 alone.

[0108] In this embodiment, the homogeneity RSD of the AuNBPs@ZIF-8@AuNPs SERS substrate is <5%, and the reproducibility RSD is <5%. The analysis process is as follows:

[0109] See Figure 15 As shown, this embodiment of the invention measures 10 points (A, B) randomly selected on an AuNBPs@ZIF-8@AuNPs substrate and 10 self-assembled substrates (C, D) prepared in different batches. -6 Line and bar charts of the 4-NTP SERS signal of M; the uniformity and reproducibility of the SERS substrate are important indicators for evaluating the performance of the SERS substrate, such as... Figure 15 As shown, 20 points were randomly selected on the same batch of substrates to detect 1×10⁻⁶ samples. -6 The 4-NTP signal of M had an average signal intensity of 25545 and an RSD of only 4.3%, indicating that the substrate has good homogeneity. Ten batches of SERS substrates were selected to detect the signal molecules, and the average signal intensity was 25254 with an RSD of 4.1%, indicating that it has good reproducibility.

[0110] Furthermore, currently reported methods for identifying Ophiopogon japonicus, such as HPLC and UPLC / Q-TOF MS, generally require 30-1 hour of detection time, while the detection method described in this embodiment takes less than 20 minutes. The pretreatment of currently reported methods such as HPLC and UPLC / Q-TOF MS is more complicated, while the detection method described in this embodiment only requires mixing and incubation. Moreover, HPLC is a large and expensive laboratory instrument, while this embodiment uses a portable Raman spectrometer, which is suitable for on-site detection. The method of this embodiment achieves an accuracy of 98% in identifying Ophiopogon japonicus from Sichuan and Zhejiang, which is comparable to the 99% accuracy of the HPLC method.

[0111] Example 4: Analysis of Raman fingerprint spectrum of Ophiopogon japonicus;

[0112] See Figure 16 As shown, Figure 16 A1-A4 in the figures are photographs of Ophiopogon japonicus from different origins and its adulterants, as described in the embodiments of the present invention. Figure 16 B in the figure represents the SERS spectra of Ophiopogon japonicus and its adulterants from different origins in the embodiments of the present invention. By collecting different batches of Ophiopogon japonicus and its adulterants from different origins, it was found that it is difficult to accurately determine the category by visual inspection alone, without sufficient experience and comparison. Figure 16 As shown in A1-A4, further identification was performed by collecting Raman spectra of the four Chinese medicinal herbs; for example... Figure 16 As shown in B, some differences were found between Zhejiang Ophiopogon japonicus and Sichuan Ophiopogon japonicus in some weak Raman peaks. Zhejiang Ophiopogon japonicus had one more peak at 518 cm⁻¹ than Sichuan Ophiopogon japonicus. -1 The peak at that location, along with the Sichuan Ophiopogon japonicus at 732cm. -1 and 760cm -1 Two more peaks appeared, with the light bamboo leaf peak at 690cm. -1 and 734cm -1 There are two unique peaks, and the mountain lilyturf is 518cm high. -1 732cm -1 and 760cm -1 There is no peak.

[0113] Example 5: Selection of ML Algorithm and Model Training;

[0114] In this embodiment, the process of acquiring SERS spectral data of the mixed liquid using a Raman spectrometer is as follows: the incubated mixed liquid is centrifuged to remove unadsorbed Ophiopogon japonicus extract, resuspended in water, and data is acquired using a portable Raman spectrometer with an excitation source of 785 nm and a spectral measurement range of 200 cm⁻¹. -1 -2000cm -1The laser power used was 100mW, and the single acquisition time was 3000ms. To improve the generalization ability of the model and maximize the diversity and randomness of the sampling, three people randomly sampled the samples. Approximately 24 points were randomly sampled from each batch of Ophiopogon japonicus. Adulterants were divided into Ophiopogon japonicus from Hubei, Ophiopogon japonicus from Shandong, and Lophatherum gracile; each adulterant was also divided into 10 batches, with approximately 12 points randomly sampled from each batch. A total of 816 data points were obtained. 20% of all data was randomly selected as the test set, and the remainder as the training set. This dataset was named the "Text Dataset." The entire spectral range was selected as all feature values, and no other complicated preprocessing was performed. All Raman spectral data were plotted into images using the `plot` function in Python, serving as the dataset for the CNN model. This dataset was named the "Image Dataset."

[0115] Specifically, in step S4, the machine learning model is a KNN model or a CNN model;

[0116] When the input SERS spectral data is a text dataset, the KNN model is invoked for recognition.

[0117] When the input SERS spectral data is an image dataset, the CNN model is invoked for recognition.

[0118] In this embodiment, the text dataset is in text format and the image dataset is in image format; the text dataset is a numerical sequence in .csv format; and the image dataset is a spectral image generated using the Python plot function.

[0119] Specifically, the parameters of the KNN model are: n_neighbors=5, weights='uniform', algorithm='auto';

[0120] The convolutional neural network model is a 2D-CNN model containing three convolutional layers, three max pooling layers, one Dropout layer, and three fully connected layers.

[0121] Specifically, all models in this embodiment use the open-source Python-based machine learning toolkit Sklearn, and the specific parameter values ​​of the traditional machine learning models and CNN models used are shown in Table 1:

[0122]

[0123] Data augmentation was performed on the image dataset by randomly selecting 20% ​​of the data for rotation. Then, a suitable CNN model was built, using a 2D-CNN model. The specific model structure is as follows... Figure 17As shown, this model consists of three convolutional layers, three max-pooling layers, one Dropout layer, and three fully connected layers. The three convolutional layers can extract more complex features from the input image, while the use of Dropout prevents overfitting and enhances the model's generalization ability. The max-pooling layer reduces the spatial size of features, decreasing computation while preserving important features. However, this model is only suitable for visual tasks and has limited processing capabilities for text, audio, and other similar data. It is best suited for processing data with local spatial correlations.

[0124] Specifically, in this embodiment, the selection process for the KNN model is as follows: Machine learning models LR, SVM, KNN, DT, RF, NB, and ANN were used to train Zhejiang Ophiopogon japonicus, Sichuan Ophiopogon japonicus, and two adulterants, Ophiopogon japonicus var. sarcodactylis and Lophatherum gracile. The performance of the trained models was evaluated using the accuracy on the validation set and the accuracy, precision, recall, and F1 score on the test set, as shown in Table 2. Table 2 compares the performance of different classifiers on the test set. It can be seen that the LR, SVM, KNN, RF, and NN models all achieved over 90% accuracy in all metrics. The DT and NB models performed poorly in predicting the Ophiopogon japonicus and its adulterants. The validation and test accuracy show that the LR and KNN models had the highest correct classification rate. The precision shows that these two models had the best ability to distinguish negative samples, reaching 98%. The model with the best recall was the KNN model, also reaching 98%, indicating good ability to identify positive samples and a high recall rate. The closer the F1 score is to 1, the better the overall model performance. The table shows that the F1 score... The model with the highest score is still the KNN model. In summary, the KNN model is the model with the best overall performance.

[0125]

[0126] See Figure 18 As shown, Figure 18 In the diagram, AG represents the confusion matrix of several traditional machine learning models in the embodiments of this invention. Figure 18 H in the diagram represents the ROC curves of several traditional machine learning models in this invention embodiment. Further evaluation of the models is conducted using the ROC curves and confusion matrix. The confusion matrix shows that the KNN model misidentified three data points, the LR model misclassified four data points, the SVM and RF models had 8 and 9 misclassifications respectively, and other models had more than 10 misclassifications. The closer the ROC curve is to the upper left corner, the better the learner's performance. Figure 18It is evident from the H-values ​​that the ROC curve of the SVM model completely envelops the ROC curves of the DT, NN, RF, and NB models, while the ROC curves of LR and KNN completely envelop the ROC curve of the SVM model. Furthermore, by comparing the AUC of LR and KNN, it is found that the KNN model performs better. Overall, the KNN model has the best performance in classifying Ophiopogon japonicus and its adulterants.

[0127] Specifically, in this embodiment, the selection and performance evaluation process of the convolutional neural network model is as follows: all randomly collected image files were used as the dataset to train the constructed 2D-CNN model. After evaluation by various metrics, it was found that the recall rate and F1 score were low, only 0.83 (Table 3), indicating that the model's ability to distinguish between positive and negative samples was different, and the recognition rate of negative samples was low. By examining the dataset, it was found that the number of samples of different categories was not uniform. Data augmentation was used to make the sample size more balanced, thereby increasing the model's adaptability. The model performance before and after data augmentation was evaluated, and it was found that the recall rate increased significantly, that is, the recognition accuracy of negative samples was significantly improved, as shown in Table 3.

[0128]

[0129] After data augmentation, the data in the dataset was good and the model was in a relatively balanced state, but none of the metrics reached 0.9, indicating that the model's performance needed further improvement. In this embodiment, the complexity of the CNN model was increased by increasing the network depth, from one convolutional layer and fully connected layer to three. As the depth increased, the model's learning ability also improved, and the performance became better and better. Accuracy, recall, precision, and F1-score all reached above 0.98, and the loss was also in the order of 10⁻². As the depth continued to increase, in this embodiment, the number of convolutional layers and fully connected layers was increased to 4 and 5, respectively. It was found that the metrics showed a downward trend, indicating that the model began to overfit. That is, the model's learning ability was too strong, causing new samples that were slightly different from the learned samples to be classified as other categories, reducing the generalization ability and thus causing a decrease in accuracy when identifying new samples. As shown in Table 4,

[0130]

[0131] Based on the optimal model structure, the number of training epochs was optimized. Too many training epochs increase training time and lead to overfitting, while too few training epochs prevent the model from converging. As shown in Table 5, the model's performance steadily improves with increasing training epochs, reaching its maximum at 24 epochs. Further increasing the number of training epochs resulted in increased loss and a declining trend in model performance. Therefore, 24 epochs were ultimately selected as the optimal number of training epochs.

[0132]

[0133] See Figure 19 As shown, Figure 19 In the figure, A represents the accuracy change during the training process of the CNN model in this embodiment of the invention. Figure 19 B in the figure represents the loss change graph during the training process of the CNN model in this embodiment of the invention. According to the accuracy and loss changes during the training process of the CNN model, it can be found that the accuracy gradually increases and the loss gradually decreases. As the training rounds reach 24, both gradually stabilize, indicating that the model has reached a convergent state and there is no overfitting.

[0134] See Figure 20 As shown, Figure 20 In the diagram, A represents the confusion matrix of the CNN model in this embodiment of the invention. Figure 20 B in the figure represents the ROC curve of the CNN model in this embodiment of the invention. The CNN model is further evaluated using the ROC curve and the confusion matrix. It can be seen that the ROC curve is infinitely close to the upper left corner, indicating that the CNN model performs well. The confusion matrix shows that the CNN model failed to distinguish between Sichuan Ophiopogon japonicus and Zhejiang Ophiopogon japonicus in five data points. In the process of identifying mountain Ophiopogon japonicus, the Raman spectra of seven mountain Ophiopogon japonicus were identified as other adulterants. The identification error rate is low, which proves the feasibility of the CNN model for Ophiopogon japonicus identification.

[0135] Example 6: Building a Smart Mini Program;

[0136] See Figure 21The diagram shows a flowchart of the Ophiopogon japonicus identification process according to an embodiment of the present invention. In this embodiment, the Ophiopogon japonicus identification method is applied to an online Ophiopogon japonicus identification system. This system loads image or text files via a front-end mini-program, transmits them to a back-end server, where a trained machine learning model performs the identification. Finally, the identification result is sent back to the front-end page. First, the front-end mini-program judges the loaded file. If the file is loaded via the camera icon 1 or the "Open the Image" button 2, it is cached in the system. If the file is transmitted via the "Open the file" button 3, no caching is performed. Then, the loaded file is uploaded to the back-end server via an interface. The back-end server first uses an algorithm to identify the received file format. If it is an image file, a trained CNN model is called for identification; if it is a text file, a pre-trained KNN model is called for identification. After identification, the result is sent back to the front-end mini-program. The mini-program judges the returned result and displays different descriptions of the medicinal materials on the mini-program interface based on the result. It also checks if there are cached files in the system. If they exist, it indicates that an image file was loaded, and the cached image needs to be displayed on the "result" page. After "returning to the homepage," the cached file is cleared. If no cached file exists, it means a text file is being loaded. The pre-stored image of *Ophiopogon japonicus* will be displayed on the "result" page. A complete flowchart of the *Ophiopogon japonicus* recognition process is shown below. Figure 21 As shown.

[0137] This embodiment uses Flask for front-end and back-end data transfer. Flask is a lightweight web application framework written in Python. It is simple to use, flexible, and powerful, suitable for rapid development of web applications and APIs. It has the advantages of being easy to learn, lightweight and flexible, highly extensible, having a built-in development server, and quick to deploy. Its ease of learning is reflected in Flask's concise and clear API, which is easy to understand and use. It provides a simple routing system, making it easy to define URLs and corresponding processing functions. Its lightweight and flexible nature stems from Flask's low dependency ratio and relatively small core library, allowing for customization according to project needs. It does not force the use of specific databases or tables. Developers can choose between a standalone processing or template engine based on their preferences. Flask boasts a rich library of extensions that can easily integrate various functionalities, such as database connectivity, form validation, and authentication. These extensions can be selected based on project needs, and developers can also write their own. Flask includes a simple development server for convenient and rapid development and debugging. During the development phase, this server can be used for real-time debugging without the need to configure and deploy other servers. Due to Flask's lightweight and flexible nature, it can be easily deployed to various server environments. It can be deployed on traditional web servers or using WSGI containers for rapid deployment.

[0138] In summary, Flask is a simple, flexible, and powerful web application framework, particularly suitable for developing small to medium-sized projects; such as rapid prototyping and building complex web applications. The specific workflow of Flask framework is as follows: Figure 21 As shown.

[0139] See Figure 22 As shown, Figure 22 In this context, A represents the "Home" interface of the smart mini-program in this embodiment of the invention. Figure 22 In this context, BC refers to the "Result" page of the smart mini-program in this embodiment of the invention. Figure 22 B in the equation is based on the analysis of the obtained SERS spectral data. Figure 22 The C in the figure is based on the analysis of the obtained SERS spectral images;

[0140] The designed interface for recognizing Ophiopogon japonicus is as follows: Figure 22 As shown, it is divided into a "Home" interface and a "Result" interface. The "Home" interface is as follows: Figure 22As shown in A, the page has three functions: the camera icon in the center, which, when clicked, uses the API to call the WeChat camera function to directly take a picture of the acquired Raman spectrum image, uploads the image to the backend server, and determines whether to use a CNN model to recognize the image format or a traditional machine learning model to predict and recognize the .csv format file based on the uploaded file format. Clicking "Select from Image" on the left allows you to select all images in your phone's album, and clicking "Select from File" on the right allows you to select files using WeChat Transfer Assistant or your phone's file manager. The selected files are also uploaded to the backend server for recognition. After the images are sent to the backend server, the backend uses a pre-trained ML model to predict the data and then sends the prediction results back to the "Result" interface of the mini-program. The mini-program interface will then redirect accordingly. Figure 22 As shown in B, the uploaded Raman spectrum image will be displayed at the top of the page, and the system's recognition results and some information about the medicinal material will be displayed at the bottom of the page. At the very bottom of the page is the "Back" button 4. Clicking it will return you to the "Home" interface to select a new file for recognition again. If you upload a file for recognition, you will be redirected to... Figure 22 Page C in the image shows an image of Ophiopogon japonicus at the top and the KNN model's recognition results and some information about the identified medicinal herb at the bottom. Similarly, a "Back" button is located at the very bottom of the page, which returns to the "Home" screen.

[0141] In this embodiment, multiple different files were imported into the system for identification, and the system identification did not encounter any problems, indicating that the smart mini-program built in this experiment is fully functional.

[0142] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for identifying the Chinese medicinal herb Ophiopogon japonicus based on machine learning-assisted SERS, characterized in that, include: Step S1: Prepare AuNBPs@ZIF-8@AuNPs composite SERS substrate; Step S2: Prepare the Ophiopogon japonicus extract to be tested, and mix and incubate the Ophiopogon japonicus extract to be tested with 0.1 nM composite material at a mixing volume ratio of 9:1 for 20 minutes to form a mixed liquid; Step S3: Use a Raman spectrometer to acquire SERS spectral data of the mixed liquid; The SERS spectral data includes text datasets and image datasets; Step S4: Input the SERS spectral data into a pre-trained machine learning model to output the identification result through the identification system; In step S1, the preparation of the AuNBPs@ZIF-8@AuNPs composite SERS substrate includes: Step S101: Synthesize AuNBPs@ZIF-8 solution by chemical reduction method; Step S102: Mix the HAuCl4 solution with the AuNBPs@ZIF-8 solution over 6 hours to form a mixed solution; Step S103: Add 0.1 g / L NaBH4 solution to the mixed solution for in-situ reduction and stir under ice bath conditions to obtain AuNBPs@ZIF-8@AuNPs composite material. Add the composite material dropwise to a silicon wafer, PDMS, PMMA and glass slide at a concentration of 0.1 nM to obtain AuNBPs@ZIF-8@AuNPs composite SERS substrate.

2. The method for identifying Ophiopogon japonicus based on machine learning-assisted SERS as described in claim 1, characterized in that, In step S4, the machine learning model is a KNN model or a CNN model; When the input SERS spectral data is a text dataset, the KNN model is invoked for recognition. When the input SERS spectral data is an image dataset, the CNN model is invoked for recognition.

3. The method for identifying Ophiopogon japonicus based on machine learning-assisted SERS as described in claim 2, characterized in that, The parameters of the KNN model are: n_neighbors=5, weights='uniform', algorithm='auto'; The CNN model is a 2D-CNN model containing three convolutional layers, three max pooling layers, one Dropout layer, and three fully connected layers.

4. The method for identifying Ophiopogon japonicus based on machine learning-assisted SERS as described in claim 1, characterized in that, In step S4, the training process of the machine learning model includes: SERS spectral data from the roots of Sichuan Ophiopogon japonicus, Zhejiang Ophiopogon japonicus, mountain Ophiopogon japonicus, and Lophatherum gracile were collected to construct a dataset; The dataset is divided into a training set and a test set; The training set is input into the machine learning algorithm for training, and the model performance is verified through the test set to obtain a trained machine learning model. The SERS spectral data were acquired using a portable Raman spectrometer with an excitation source of 785 nm and a spectral measurement range of 200 cm⁻¹. -1 Up to 2000cm -1 The power is 100mW and the acquisition time is 3000ms.

5. The method for identifying Ophiopogon japonicus based on machine learning-assisted SERS as described in claim 1, characterized in that, The identification results include the category of the Ophiopogon japonicus extract to be tested, which includes, but is not limited to, Sichuan Ophiopogon japonicus, Zhejiang Ophiopogon japonicus, mountain Ophiopogon japonicus, or Lophatherum gracile root.

6. The method for identifying Ophiopogon japonicus based on machine learning-assisted SERS as described in claim 1, characterized in that, The identification system includes: The front-end interaction module is used for users to upload SERS spectral data files, the file format of which includes text format or image format; The backend server module is connected to the frontend interaction module and is used to receive the SERS spectral data file and call the corresponding machine learning model to identify the SERS spectral data file. The result return module is used to return the authentication result output by the backend server module to the frontend interaction module and display it to the user.

7. The method for identifying Ophiopogon japonicus based on machine learning-assisted SERS as described in claim 6, characterized in that, In the AuNBPs@ZIF-8@AuNPs composite SERS substrate, the average particle size of ZIF-8 is 320 nm, and the average particle size of AuNPs generated in situ on its surface is 14 nm.

Images