A water body drug residue Raman spectrum recognition method, device and system based on a neural network algorithm

Abstract

The application relates to a water body drug residue Raman spectrum recognition method, device and system based on a neural network algorithm, and relates to the technical field of Raman spectrum analysis. The method comprises the following steps: through Raman spectrum technology, a Raman spectrum detection substrate is used to perform Raman spectrum detection analysis on a to-be-detected target object solution containing at least one target object, Raman spectrum data is obtained, then a neural network algorithm model is used to perform target object type analysis on the Raman spectrum data, key characteristics of the target object are obtained, and then, based on the key characteristics, category differentiation is performed in combination with preset key characteristic samples, and a target object detection recognition result is obtained. The application detects a metal enhanced substrate by using a Raman enhanced spectrum technology, and identifies the target object in an actual water sample in combination with model analysis, improves detection sensitivity, and thus solves the problem that mixed target objects in a water environment cannot be recognized in the prior art, and provides a stable and reliable scheme for water body target object recognition.

Description

Technical Field

[0001] This application relates to the field of Raman spectroscopy analysis technology, and in particular to a method, device and system for Raman spectroscopy identification of drug residues in water based on neural network algorithm. Background Technology

[0002] Raman spectroscopy is a non-destructive and rapid detection technique that can obtain the fingerprint information of target molecules and has the characteristic of specific identification. It has been applied in fields such as chemical analysis and biomedical detection.

[0003] Existing target object detection methods based on Raman spectroscopy are mostly focused on the quantitative detection of known substances. In terms of Raman spectroscopy data processing, existing methods lack sufficient flexibility when processing spectral data of complex samples and mixtures, which can easily lead to misjudgments and make it difficult to identify multiple pollutants in water. Summary of the Invention

[0004] This application provides a method, apparatus, and system for Raman spectroscopy identification of drug residues in water based on a neural network algorithm. On one hand, this application utilizes surface-enhanced Raman spectroscopy (SERS) technology, leveraging the plasmon resonance enhancement effect of noble metal nanomaterials to significantly enhance the Raman signal of target molecules adsorbed on the surface, thereby improving the stability of the detection substrate and thus increasing its detection sensitivity. On the other hand, this application uses a neural network algorithm model to extract features and classify categories from the detected Raman spectra to achieve the identification of unknown substances. For cases involving multiple mixtures, a neural network algorithm model capable of identifying multiple target substances is used for the detection of mixtures in the environment, achieving the identification of multiple pollutants in aquaculture water.

[0005] In a first aspect, this application provides a method for Raman spectroscopy identification of drug residues in water based on a neural network algorithm, including:

[0006] Raman spectroscopy is used to perform Raman spectroscopy detection and analysis on the solution of the target analyte using a prepared Raman spectroscopy-enhanced detection substrate, thereby obtaining Raman spectral data. The Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified.

[0007] The target species are analyzed by using a neural network algorithm model to obtain the key characteristics of the target species from the Raman spectral data;

[0008] Based on the key features, and combined with preset key feature samples, category distinction is performed to obtain the target object detection and recognition results;

[0009] The target solution to be detected is an aqueous solution of the target substance detected by a metal substrate, wherein the metal substrate is an aluminum foil SERS substrate, and the target solution to be detected contains at least one target substance selected from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green.

[0010] Optionally, Raman spectroscopy is used to perform Raman spectroscopy analysis on the target analyte solution using the prepared Raman spectroscopy-enhanced detection substrate to obtain Raman spectral data, including:

[0011] Raman spectroscopy is used to acquire raw Raman spectral data from the solution of the target analyte using a Raman spectroscopy-enhanced detection substrate. The raw Raman spectral data includes single-target Raman spectra or mixed-target Raman spectra.

[0012] Preprocessing is performed on the original Raman spectral data to obtain Raman spectral data;

[0013] The aluminum foil SERS substrate in the Raman spectroscopy enhanced detection substrate is obtained by a Raman enhanced detection substrate preparation method, which involves reacting a metal foil with a silver nitrate solution containing hydrofluoric acid to form a nano-silver structure on the metal foil.

[0014] Optionally, raw Raman spectroscopy data can be acquired from the analyte solution using Raman spectroscopy-enhanced detection substrate, including:

[0015] Under the target temperature conditions, the solution of the target analyte is detected by Raman spectroscopy using a Raman spectroscopy-enhanced detection substrate to obtain the Raman spectral characteristic peak information at the corresponding solution concentration of the target analyte, which is used as the original Raman spectral data.

[0016] The Raman spectral characteristic peak information includes the Raman spectral characteristic peak intensity and / or peak area intensity.

[0017] Optionally, a neural network algorithm model can be used to analyze the target species in the Raman spectral data to obtain the key features of the target species to be detected and identified, including:

[0018] The Raman spectral data is input into a neural network algorithm model for target classification to obtain a first classification result, which may include single target classification results or multi-target classification results.

[0019] Based on the first classification result, an interpretability analysis is performed using the neural network algorithm model, and key features of at least one target object corresponding to the identified Raman spectral data are output.

[0020] Optionally, for the first classification result, an interpretability analysis is performed using the neural network algorithm model, including:

[0021] Based on the first classification result, target neurons for identification and analysis are selected from the neural network algorithm model;

[0022] Key features are obtained by performing feature analysis on the Raman spectral data using the target neuron.

[0023] Optionally, before performing target species analysis on the Raman spectral data using a neural network algorithm model, the method further includes:

[0024] Using Raman spectroscopy, the sample solution is detected and analyzed using a prepared Raman spectroscopy-enhanced substrate to obtain a Raman spectroscopy sample set, which includes single-target spectral samples and mixed-target spectral samples.

[0025] The Raman spectrum sample set is sorted and divided using a preset random function to obtain a training set, a validation set, and a test set.

[0026] Based on the compositional characteristics of the Raman spectrum sample set, a multilayer perceptron algorithm model based on neural networks is established.

[0027] The multilayer perceptron algorithm model is trained using the training set, and validated using the validation set during the training process. The model is then tested using the test set until the multilayer perceptron algorithm model training is completed, resulting in a performance evaluation result of the model and a trained neural network algorithm model.

[0028] Based on the performance evaluation results, the SHAP parsing method is used to analyze the model and obtain key feature samples of each category of target objects;

[0029] The sample solution contains at least one concentration of the target substance solution sample, and the key characteristic samples are concentrated in the 447-450 cm⁻¹ region. -1 1104-1114cm -1 and 571cm -1 Within the Raman shift range, and the key feature sample matches the characteristic peak of the target object, it is used to distinguish the spectra of different categories of target objects.

[0030] Optionally, based on the compositional characteristics of the Raman spectral sample set, a multilayer perceptron algorithm model based on a neural network is established, including:

[0031] Spectral feature number analysis was performed on the Raman spectrum sample set to obtain the spectral feature number;

[0032] A multilayer perceptron algorithm model based on neural networks is established, and the number of input layer neurons of the multilayer perceptron algorithm model is set to correspond to the number of spectral features, and the output layer neurons of the multilayer perceptron algorithm model are set to correspond to the target object recognition category.

[0033] Optionally, the sample solution is detected and analyzed using Raman spectroscopy on a prepared Raman spectroscopy-enhanced substrate to obtain a Raman spectroscopy sample set, including:

[0034] Under room temperature conditions, Raman spectroscopy is used to collect Raman spectral data of sample solutions of different concentrations in the sample using Raman spectroscopy enhancement substrate. Working curves of the relationship between the target analyte and the Raman spectral characteristic peak intensity or peak area intensity at different concentrations are obtained as Raman spectral sample sets, and a Raman enhancement detection method for the target analyte is established.

[0035] In the substrate preparation method of the Raman spectroscopy-enhanced detection substrate, the reaction solution is prepared by reacting hydrofluoric acid of a preset concentration and silver nitrate solution of at least one concentration according to the detection requirements of the target analyte.

[0036] Secondly, this application provides a water-based drug residue Raman spectroscopy identification device based on a neural network algorithm, comprising:

[0037] The Raman spectroscopy detection and analysis module is used to perform Raman spectroscopy detection and analysis on the solution of the target analyte to be detected through the prepared Raman spectroscopy enhancement substrate, and to obtain Raman spectral data, wherein the Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified;

[0038] The analysis module is used to perform target species analysis on the Raman spectral data using a neural network algorithm model to obtain the key characteristics of the target species;

[0039] The category differentiation module is used to differentiate categories based on the key features and in combination with preset key feature samples to obtain target object detection and identification results; wherein, the metal substrate solution is an aluminum foil SERS substrate, and the target object aqueous solution contains at least one target object to be detected and identified from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue and malachite green.

[0040] Thirdly, this application provides a water-based drug residue Raman spectroscopy identification system based on a neural network algorithm, comprising:

[0041] A Raman spectroscopy detection device is used to perform Raman spectroscopy detection and analysis on a solution of a target analyte prepared on a Raman spectroscopy-enhanced substrate to obtain Raman spectral data, wherein the Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified.

[0042] The detection and analysis equipment is connected to the Raman spectroscopy detection equipment and is used to analyze the target species in the Raman spectral data through a neural network algorithm model to obtain the key features of the target species; based on the key features, and combined with preset key feature samples, the categories are distinguished to obtain the target species detection and identification results; wherein, the metal substrate solution is an aluminum foil SERS substrate, and the target species aqueous solution contains at least one target species to be detected and identified, namely levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green.

[0043] In summary, this application utilizes Raman spectroscopy to perform Raman spectral analysis on solutions containing single or multiple target objects using a Raman-enhanced substrate. This yields Raman spectral data, which is then analyzed using a neural network algorithm to identify the target object types. Key characteristics of the target objects are identified, and based on these key characteristics and pre-defined key feature samples, category differentiation is performed to obtain the target object detection and identification results. This application utilizes Raman-enhanced spectroscopy to detect on a metal-enhanced substrate and combines this with model analysis to identify target objects in actual water samples, improving detection sensitivity. This solves the problem of existing technologies struggling to identify mixed target objects in aquatic environments, providing a stable and reliable solution for identifying target objects in water bodies. Attached Figure Description

[0044] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0045] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A schematic flowchart illustrating a method for identifying drug residues in water using Raman spectroscopy based on a neural network algorithm, provided in this application embodiment;

[0047] Figure 2 This is a flowchart illustrating the steps of an optional embodiment of the present application for a method for identifying drug residues in water using Raman spectroscopy based on a neural network algorithm.

[0048] Figure 3 This application provides an optional example of the Raman spectra of a metal substrate and a target on different metal substrates;

[0049] Figure 4 This is a schematic diagram of a multilayer perceptron model structure provided as an optional example in this application;

[0050] Figure 5 This is a schematic diagram illustrating the accuracy and loss rate of a multilayer perceptron model on the training and validation sets, provided as an optional example of this application.

[0051] Figure 6 This is a schematic diagram of the SHAP value analysis of the influence of the Raman displacement characteristics of the target object on the model output, provided as an optional example of this application;

[0052] Figure 7 A structural block diagram of a water drug residue Raman spectroscopy identification device based on a neural network algorithm provided in this application embodiment;

[0053] Figure 8 This is a schematic diagram of the structure of a water drug residue Raman spectroscopy identification system based on a neural network algorithm provided in an embodiment of this application. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0055] To facilitate understanding of the embodiments of this application, further explanations and descriptions will be provided below in conjunction with the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application.

[0056] Figure 1 This is a schematic flowchart illustrating a method for identifying drug residues in water using Raman spectroscopy based on a neural network algorithm, provided in an embodiment of this application. Figure 1 As shown in the embodiments of this application, the method for identifying drug residues in water based on neural network algorithms using Raman spectroscopy may specifically include the following steps:

[0057] Step 110: Using Raman spectroscopy, the prepared Raman spectroscopy-enhanced detection substrate is used to perform Raman spectroscopy detection and analysis on the solution of the target analyte to obtain Raman spectral data.

[0058] Wherein, the Raman spectral data includes the Raman spectrum of at least one target to be detected and identified, the target solution to be detected is an aqueous solution of the target detected by a metal substrate, the metal substrate is an aluminum foil SERS substrate, and the target solution to be detected includes at least one target to be detected and identified from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue and malachite green.

[0059] Specifically, the target analyte is typically aquatic drug residues. The aqueous solution of the target analyte usually contains at least one target analyte, such as a single target analyte (i.e., only one type of drug residue to be detected and identified), or a mixture of multiple target analytes (e.g., containing at least two types of drug residues to be detected and identified). This embodiment does not limit the types of target analytes contained in the aqueous solution of the target analyte. In this embodiment, the metal substrate solution is used as the Raman spectroscopy-enhanced detection substrate (or Raman spectroscopy-enhanced substrate). It can be prepared by various methods, mainly by reacting a metal with a silver nitrate solution containing hydrofluoric acid. For example, the aluminum foil SERS substrate (also called the aluminum foil SERS detection substrate) is mainly prepared by reacting aluminum foil with a silver nitrate solution containing hydrofluoric acid. The aluminum foil SERS detection substrate can be used for the detection of drug residues in water. This embodiment uses this substrate to establish a Raman-enhanced detection method for multiple drug residues in water to achieve quantitative analysis and for the identification and detection of unknown target analytes.

[0060] It should be noted that, in this embodiment, the Raman spectroscopy-enhancing substrate can be prepared not only with aluminum foil, but also with either copper or zinc sheets, combined with a silver nitrate solution containing hydrofluoric acid. This embodiment selects inexpensive metal materials and prepares the SERS substrate through chemical substitution self-assembly. The chemical substitution method rapidly generates uniform silver nanostructures with Raman enhancement effects on the metal foil, thus achieving a rapid and low-cost method for preparing the detection substrate, and providing an effective way to address the instability issues of existing detection substrates.

[0061] In practical implementation, this embodiment can utilize a Raman spectrometer or other Raman spectroscopy detection equipment to detect and analyze the Raman spectrum of the solution. This primarily involves using a prepared Raman-enhanced substrate to detect and analyze the Raman spectrum of the solution containing the target analyte. By detecting the Raman signal of the target analyte, its Raman spectrum is acquired and used as the Raman spectrum of the target analyte.

[0062] In practical applications, since the Raman signal of the target object is enhanced on the aluminum foil SERS substrate, the Raman spectrum acquired during detection and analysis using Raman spectroscopy-enhanced detection is essentially the Raman spectrum of the target object obtained on the aluminum foil SERS substrate. For cases containing only one target object, the acquired Raman spectrum is the spectrum of that target object obtained on the aluminum foil SERS substrate. For cases containing mixed target objects, Raman spectroscopy can acquire the Raman spectra corresponding to multiple target objects on the aluminum foil SERS substrate.

[0063] In related technologies, methylene blue, malachite green, and various antibiotics are commonly used disinfectants in aquaculture. However, these substances all possess certain toxicity and potential hazards. Residual drugs in water can enter the food chain through aquatic products, thus harming human health. Simultaneously, drug residues in water can enter the environment through wastewater discharge, causing ecological damage. Currently, the main methods for detecting these target substances in water are high-performance liquid chromatography (HPLC) and liquid chromatography-mass spectrometry (LC / MS). However, these instruments are complex to operate and costly, generally requiring complex sample pretreatment processes, thus limiting their application. Furthermore, existing laboratory detection methods are mostly for the quantitative detection of known substances, while screening for unknown substances places higher demands on instruments and personnel. In addition, there are schemes using traditional Raman spectroscopy for target substance detection.

[0064] Compared to existing target detection technologies, this embodiment utilizes surface-enhanced Raman spectroscopy (SERS), a technique that leverages the plasmon resonance enhancement effect of metal nanomaterials to significantly enhance the Raman signal of adsorbed target molecules, thereby improving detection sensitivity. This embodiment uses SERS to achieve quantitative analysis of known substances, and combined with data processing algorithms and recognition models, it can identify unknown substances.

[0065] It should be noted that in this application, Raman spectroscopy is a non-destructive and rapid detection technique that can obtain the fingerprint information of target molecules and has the characteristic of specific identification. It has been applied in fields such as chemical analysis and biomedical detection.

[0066] Optionally, the above-mentioned method of using Raman spectroscopy to perform Raman spectroscopy detection and analysis on the solution of the target analyte using the prepared Raman spectroscopy-enhanced detection substrate to obtain Raman spectral data may include the following sub-steps:

[0067] Sub-step 1101: Raw Raman spectral data are collected from the solution of the target analyte by using Raman spectroscopy to enhance the detection substrate.

[0068] The raw Raman spectral data includes single-target Raman spectra or mixed-target Raman spectra.

[0069] Sub-step 1102: Preprocess the original Raman spectral data to obtain Raman spectral data.

[0070] The aluminum foil SERS substrate in the Raman spectroscopy enhanced detection substrate is obtained by a Raman enhanced detection substrate preparation method, which involves reacting a metal foil with a silver nitrate solution containing hydrofluoric acid to form a nano-silver structure on the metal foil.

[0071] A unified description is provided for sub-steps 1101 and 1102:

[0072] In this specific implementation, after acquiring the raw Raman spectral data from the solution of the target analyte using the Raman spectroscopy-enhanced substrate, the raw Raman spectral data can be preprocessed in a series of ways. The preprocessing process includes, but is not limited to, shift clipping, Airpls baseline processing, Whittaker smoothing filtering, and normalization methods to obtain the Raman spectral data.

[0073] In practical implementation, this embodiment can collect Raman spectra under certain temperature conditions, such as room temperature / normal temperature conditions.

[0074] In one optional embodiment, this embodiment uses Raman spectroscopy technology and a Raman spectroscopy-enhanced detection substrate to collect raw Raman spectral data from the solution of the target analyte. This may include: under target temperature conditions, detecting the solution of the target analyte using Raman spectroscopy detection method and a Raman spectroscopy-enhanced detection substrate to obtain Raman spectral characteristic peak information at the corresponding solution concentration of the target analyte, which is used as the raw Raman spectral data; wherein the Raman spectral characteristic peak information includes the Raman spectral characteristic peak intensity and / or peak area intensity.

[0075] Preferably, the target temperature is room temperature / normal temperature, which corresponds to 25 degrees Celsius, or 25℃.

[0076] For example, in this embodiment, the Raman spectrum of the target object can be acquired at 25°C using a Raman detection system equipped with a 785nm laser. The Raman laser power is 500mW, and the Raman signal is acquired twice, with each acquisition lasting 5 seconds, to obtain the Raman spectrum corresponding to the target object.

[0077] Step 120: Analyze the target species in the Raman spectral data using a neural network algorithm model to obtain the key characteristics of the target species.

[0078] In this embodiment, the key features mainly refer to the characteristics of the Raman spectrum of the target object. The key features of a single target object and the key features of a mixture of target objects may differ. The key features are primarily concentrated in the 447-450 cm⁻¹ region. -1 1104-1114cm -1 and 571cm -1 Within the Raman displacement range.

[0079] In this implementation, a neural network algorithm model can be pre-trained, and then Raman spectral data can be input into the neural network algorithm model. The model can then identify and extract features from the Raman spectral data. This embodiment can also utilize the SHAP model interpretation tool to perform interpretability analysis on the model, and use SHAP to identify key features of different types of spectra from the Raman spectral data.

[0080] Optionally, this embodiment uses a neural network algorithm model to analyze the target species in the Raman spectral data and obtain the key characteristics of the target species, which may include the following sub-steps:

[0081] Sub-step 1201: Input the Raman spectral data into the neural network algorithm model to classify the target object and obtain the first classification result.

[0082] The first classification result includes single-object classification result or multi-object classification result;

[0083] Sub-step 1202: Based on the first classification result, perform an interpretability analysis using the neural network algorithm model, and output the key features of at least one target object corresponding to the identified Raman spectral data.

[0084] A unified description is provided for sub-steps 1201 and 1202:

[0085] In its implementation, the neural network algorithm model can identify the number of target species contained in the target solution from the input Raman spectral data. This number of species serves as the classification result, distinguishing whether the target aqueous solution contains a single target species or a mixture of target species. The classification result for multiple target species can be further subdivided into classification results for two mixed target species and classification results for more than two mixed target species; this embodiment does not impose any limitations on this. By identifying the number of target species, the model can perform further analytical analysis based on the number of species to obtain the key characteristics of the target species.

[0086] Optionally, in this embodiment, for the first classification result, an interpretability analysis is performed using the neural network algorithm model. Specifically, this may include: selecting target neurons from the neural network algorithm model for identification and analysis based on the first classification result; and performing feature analysis on the Raman spectral data using the target neurons to obtain key features.

[0087] In existing technologies, the Hit Quality Index (HQI) similarity method is used to assess the similarity of spectral data in Raman spectroscopy data processing. However, the HQI method lacks sufficient flexibility and robustness when processing spectral data of complex samples and mixtures, leading to misclassification and thus limiting its application in Raman spectroscopy identification. Environments (such as aquaculture) often contain mixtures of multiple antibiotics; detecting these antibiotic mixtures requires training a model capable of recognizing multiple target substances for detecting mixtures in the environment.

[0088] To address the aforementioned issues, the neural network algorithm model in this embodiment can include multiple output layer neurons, with a certain correspondence between the output layer neurons and the target objects. Preferably, the output layer can have 7 neurons corresponding to 7 categories. These 7 categories of neurons are mainly divided into neurons for single target object recognition and neurons for mixed target object recognition (such as neurons for recognizing two mixed target objects, and neurons for recognizing more than two mixed target objects). In this embodiment, corresponding neurons can be selected for recognition and analysis for different categories of target objects to distinguish the spectra of different categories of target objects.

[0089] For example, in the above neural network algorithm model, the neurons of the 7 categories can be divided as follows: 3 neurons correspond to single target object recognition and classification, 3 neurons correspond to recognition and classification in the case of two mixed target objects, and 1 neuron corresponds to recognition and classification in the case of three or more mixed target objects.

[0090] Step 130: Based on the key features, classify the target object by combining the preset key feature samples to obtain the target object detection and recognition result.

[0091] In this embodiment, the key feature samples are established in advance by analyzing the sample solution of the target analyte. The key feature samples are mainly concentrated in the 447-450 cm⁻¹ region. -1 1104-1114cm -1 and 571cm -1 Within the Raman shift range. This embodiment utilizes key features and key feature samples for comparative analysis to determine the target object detection and identification results. In this embodiment, each feature sample in the key feature samples matches the characteristic peaks of the target object and plays a crucial role in distinguishing the spectra of different categories of target objects.

[0092] As can be seen, this application embodiment utilizes Raman spectroscopy technology to perform Raman spectral analysis on a solution containing one or more target objects using a Raman spectroscopy-enhanced detection substrate, obtaining Raman spectral data. Then, a neural network algorithm model is used to analyze the Raman spectral data to determine the target object types, obtaining key features of the target objects. Based on these key features, and combined with preset key feature samples, category differentiation is performed to obtain the target object detection and identification results. On one hand, this application utilizes Raman-enhanced spectroscopy technology to detect metal-enhanced substrates, solving the problem of difficulty in identifying multiple pollutants in water due to the complexity of current SERS substrate fabrication. On the other hand, this embodiment proposes a SERS identification method based on a deep learning algorithm. Based on the acquired Raman spectra, the algorithm model analyzes and identifies target objects in actual water samples, achieving the identification of multiple pollutants in water while further improving detection sensitivity. This solves the problem of existing technologies' difficulty in identifying mixed target objects in the aquatic environment, providing a stable and reliable solution for the identification of target objects in water.

[0093] Reference Figure 2 This document illustrates a flowchart of a method for identifying drug residues in water using Raman spectroscopy based on a neural network algorithm, according to an optional embodiment of this application. Specifically, before analyzing the target species using the Raman spectral data through the neural network algorithm model, the method for identifying drug residues in water using Raman spectroscopy based on a neural network algorithm model can first train the neural network algorithm model and establish key feature samples for different target species. The model training stage may specifically include the following steps:

[0094] Step 210: Using Raman spectroscopy, the prepared Raman spectroscopy-enhanced substrate is used to detect and analyze the sample solution to obtain a Raman spectroscopy sample set.

[0095] The sample solution contains at least one concentration of the target analyte solution sample, and the Raman spectral sample set contains single target analyte spectral samples and mixed target analyte spectral samples.

[0096] In this embodiment, the enhanced detection substrate is composed of an aluminum foil SERS substrate. The scheme in this embodiment is a Raman detection method for drug residues in water based on the detection substrate. The sample solution mainly contains single-target solutions and mixed-target solutions of different concentrations, such as solutions with concentrations of 1×10⁻⁶. -4 ~1×10 -8Multiple substrate solutions were prepared by placing Raman-enhanced detection substrates in drug solutions of varying concentrations (mol / L) to form substrate solution samples. The target analytes for Raman detection of drug residues in water primarily include, but are not limited to, one or more of the following: levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green.

[0097] Compared to existing methods for detecting unknowns based on metal substrates, this embodiment uses a Raman spectroscopy-enhanced metal substrate. Addressing the difficulties in preparing existing metal substrates, the substrate preparation method proposed in this embodiment is simple to operate, fast to prepare, and scalable. It solves the problems of complex and time-consuming preparation processes in existing detection substrates, and also addresses the difficulty of current machine learning models in identifying mixed target substances in aquatic environments. The specific method is as follows:

[0098] The preparation method for the aluminum foil SERS detection substrate includes the following steps: Immersing an aluminum foil of a certain size in a silver nitrate solution containing hydrofluoric acid for 2–20 seconds (s). Preferably, the silver nitrate concentration is 0.005–0.4 mol / L, and the hydrofluoric acid concentration is 0.1 mol / L. After the aluminum foil turns a metallic yellowish-brown color, it is removed, rinsed with deionized water to remove residual hydrofluoric acid, and then air-dried to complete the preparation of the aluminum foil SERS substrate.

[0099] In related technologies, existing techniques for detecting unknown substances also utilize silver nanoparticles. Compared to this method, this embodiment uses hydrofluoric acid to remove the oxide film on the metal foil and generates a uniform silver nanostructure with Raman enhancement through a displacement reaction between the metal element and silver nitrate. This substrate preparation method is low-cost and quick. The metal-enhanced substrate preparation method provided in this embodiment creates fixed "hot spots," resulting in better reproducibility of Raman-enhanced detection. Spiking recovery experiments were conducted on actual water samples with different concentrations of fleroxacin, pefloxacin, and levofloxacin. The results show that the recovery rate of this substrate in water samples is between 97.8% and 110.4%, with an RSD of <6.73%.

[0100] In practical applications, this embodiment uses fleroxacin, levofloxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green as typical target compounds for the detection of drug residues in water using aluminum foil SERS detection substrates. Different concentrations of target compound solutions are combined with aluminum foil SERS detection substrates to form solution samples for drug residue detection in water.

[0101] For example, using fleroxacin, levofloxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green as typical target compounds, concentrations of 1×10⁻⁶ were prepared respectively. -4 ~1×10 -8 A mol / L target analyte solution was added dropwise to the detection substrate. After the solution evaporated completely, it was detected using a Raman spectrometer at room temperature. The Raman spectrometer used had a laser wavelength of 785 nm and a laser power of 50–500 mW. The sample acquisition time was 3–5 seconds. Working curves showing the intensity or area of ​​characteristic Raman peaks at different concentrations of the target analyte were obtained using the instrument, thus establishing a Raman-enhanced detection method for the target analyte. For the above target analyte, the limit of detection of this method was 10.62–23 μg / L, with a linear range of four orders of magnitude.

[0102] In practical applications, the widespread use of SERS technology is limited by the stability and reliability issues of the substrate. The SERS substrate is the core of SERS technology, and a good SERS substrate should have the advantages of simple preparation method, short preparation time, low cost, good reproducibility and high sensitivity.

[0103] The following explains the preparation process of the relevant substrate solution samples involved in the rapid detection of target analytes in water using an aluminum foil SERS substrate:

[0104] ① For the rapid detection of methylene blue residues in water using aluminum foil SERS substrates, the sample solution preparation stage can be achieved, for example, through the following:

[0105] First, the Raman-enhanced detection substrate was prepared as follows: 40 mL of 5 mmol / L silver nitrate solution was prepared, and 80 μL of hydrofluoric acid solution was added and stirred thoroughly. The cut aluminum foil was immersed in the solution for 10 seconds until its color changed from silver to yellowish-brown. The foil was then removed, rinsed with deionized water to remove residual hydrofluoric acid and silver nitrate, and dried for later use. A concentration of 1 × 10⁻⁶ was prepared. -8 ~1×10 -4 A standard solution of methylene blue was prepared at mol / L (3.199 μg / L to 31.99 mg / L). During the detection process, the aluminum foil SERS substrate was immersed in the target solution of different concentrations for 10 minutes. Raman spectra of methylene blue were acquired at room temperature (25℃) using a Raman detection system equipped with a 785 nm laser. The Raman laser power was 500 mW, and the Raman signal was acquired twice, with each acquisition lasting 5 seconds. The characteristic peak was observed at 446 cm⁻¹. -1 For example, its Raman signal shows a significant enhancement (e.g. Figure 3(As shown). Aluminum foil SERS substrates were obtained using silver nitrate solutions with concentrations of 0.005 mol / L, 0.01 mol / L, 0.02 mol / L, 0.03 mol / L, and 0.04 mol / L, respectively. The obtained substrates were used to detect 1×10⁻⁶... -4 A substrate prepared with 0.02 mol / L silver nitrate solution and mol / L methylene blue solution showed the best enhancement effect for methylene blue, with the characteristic peak of methylene blue at 446 cm⁻¹. -1 The enhancement factor is 4.2 × 10⁻⁶. 5 ,like Figure 3 As shown. Among them, Figure 3 In this context, MB represents methylene blue, AlF represents aluminum foil substrate, and AlF SERS represents aluminum foil SERS substrate.

[0106] ②Preparing an aluminum foil SERS substrate and detecting fleroxacin in water can, for example, be achieved through the following steps during the sample solution preparation stage:

[0107] Referring to the experimental conditions mentioned above ①, the silver nitrate concentration was 40 mmol / L, and the reaction time between silver nitrate and aluminum foil was 2-10 s to obtain the aluminum foil SERS substrate. The substrate was then used for detection at 1×10⁻⁶ ppm. -4 The substrate contained mol / L fleroxacin and obtained Raman spectra. The recoveries of fleroxacin on this substrate ranged from 91.74% to 116.34%, and the RSDs ranged from 3.36% to 4.48%.

[0108] ③ For rapid identification of data related to sulfadiazine, methylene blue, malachite green, and their mixtures, the following can be used as an example during the sample solution preparation stage:

[0109] Using the experimental conditions mentioned in ① above, this method was employed to establish Raman spectral datasets for sulfadiazine, methylene blue, and malachite green, and their mixtures. The procedure was as follows: aluminum foil was placed in a solution containing 0.02 mol / L silver nitrate and 0.1 mol / L hydrofluoric acid for 10 seconds. After the aluminum foil changed color from silvery-white to yellowish-brown, it was removed, rinsed with deionized water to remove the hydrofluoric acid, and air-dried to prepare the aluminum foil SERS substrate. The concentrations of sulfadiazine, methylene blue, and malachite green were prepared to be 1 × 10⁻⁶. -4 ~1×10 -8 Aluminum foil SERS substrates were immersed in single and mixed target analyte solutions at mol / L.

[0110] Therefore, this embodiment demonstrates the rapid generation of uniform silver nanostructures with Raman enhancement on metal foils via a chemical substitution method. This substrate preparation method is simple and achieves high sensitivity, high reproducibility, low cost, and large-scale fabrication.

[0111] After the substrate solution sample is prepared in this embodiment, it can be detected by a Raman spectrometer at room temperature to establish a Raman spectrum dataset of the sample. The dataset is then processed using shift clipping, Airpls baseline processing, Whittaker smoothing filtering, and normalization methods to obtain a Raman spectrum sample set.

[0112] In one optional embodiment, this embodiment uses Raman spectroscopy to detect and analyze sample solutions using a prepared Raman spectroscopy-enhancing substrate to obtain a Raman spectroscopy sample set. This includes: at room temperature, using Raman spectroscopy, acquiring Raman spectral data of sample solutions at different concentrations using the Raman spectroscopy-enhancing substrate, obtaining working curves showing the relationship between the target analyte and the intensity or area of ​​its Raman spectral characteristic peaks at different concentrations, which serve as the Raman spectroscopy sample set, and establishing a Raman-enhanced detection method for the target analyte; wherein, in the substrate preparation method of the Raman spectroscopy-enhanced detection substrate, the reaction solution is prepared by reacting hydrofluoric acid of a preset concentration and at least one concentration of silver nitrate solution according to the detection requirements of the target analyte.

[0113] In a specific implementation, this embodiment can use a Raman spectrometer to detect at room temperature and obtain working curves showing the relationship between the intensity of the characteristic peaks or the peak area intensity of the target substance at different concentrations and its Raman spectrum, thereby establishing a Raman-enhanced detection method for the target substance.

[0114] In practical applications, the method for identifying drug residues in water in this embodiment specifically includes the following steps: Raman spectra of single and mixed target substances such as sulfadiazine, methylene blue, and malachite green are acquired using Raman detection to establish a dataset. Then, the acquired Raman spectra are processed with background subtraction, shift clipping, Airpls baseline processing, Whittaker smoothing filtering, and normalization. Preferably, the Raman shift is clipped to 400-1750 cm⁻¹. -1 The Airpls baseline processing parameters are set to: lamdba=100000, order=2, and the Whittaker smoothing filter parameters are set to lamdba=100000, order=2.

[0115] Step 220: Sort and divide the Raman spectrum sample set using a preset random function to obtain a training set, a validation set, and a test set.

[0116] This embodiment can sort and divide the collected Raman spectrum sample set, treating it as a dataset. Preferably, the division can be as follows: 80% training set and 20% test set, with 20% of the training set selected as a validation set to verify the model's accuracy and loss rate. Before training the model, the data can be normalized, and the data order can be randomly shuffled using the `np.random.permutation` function, followed by one-hot encoding to convert the class labels. This allows the divided data to be subsequently fed into the multilayer perceptron algorithm to train the model.

[0117] Step 230: Based on the compositional characteristics of the Raman spectrum sample set, establish a multilayer perceptron algorithm model based on neural networks.

[0118] In practical implementation, addressing the shortcomings of current SERS substrate fabrication, such as complexity and difficulty in identifying multiple pollutants in water, this embodiment establishes a multilayer perceptron algorithm model based on the Tensorflow framework, also known as a multilayer perceptron neural network model. This model has one input layer, three hidden layers, and one output layer. The Raman shift after the above-mentioned pruning is 400-1750 cm⁻¹. -1 The interpolation gap is 1cm. -1 The model has 1351 features, which are used as the input layer's 1351 neurons. The number of hidden layers and neurons is determined by a grid search procedure performed on the training and validation sets; there are a total of 3 layers, with 768, 384, and 48 neurons per layer, respectively. ReLU is chosen as the activation function for all hidden layers. The output layer contains 7 neurons corresponding to the 7 classification categories, and SOFTMAX is chosen as the activation function. An example model structure is as follows: Figure 4 As shown.

[0119] Optionally, this embodiment establishes a multilayer perceptron algorithm model based on neural networks based on the compositional characteristics of the Raman spectrum sample set. This may include: performing spectral feature number analysis on the Raman spectrum sample set to obtain spectral feature numbers; establishing a multilayer perceptron algorithm model based on neural networks, setting the number of input layer neurons of the multilayer perceptron algorithm model to correspond to the number of spectral features, and setting the output layer neurons of the multilayer perceptron algorithm model to correspond to the target object recognition category.

[0120] Step 240: Train the multilayer perceptron algorithm model using the training set, validate the model using the validation set during the training process, and test the model using the test set until the multilayer perceptron algorithm model training is completed, thereby obtaining the model performance evaluation results and the trained neural network algorithm model.

[0121] In this specific implementation, the multilayer perceptron algorithm model is trained using a training set, validated using a validation set, and tested using a test set. During model training, the parameters of the multilayer perceptron algorithm model can be optimized. Specifically, parameter optimization includes iterative optimization using a stochastic gradient descent (SGD) optimizer. The SGD parameters are set as follows: categorical_crossentropy is selected as the loss function; the learning rate is 0.01; the batch size is 20; and the number of iterations is 200. Iteration continues until the loss function no longer decreases, thus obtaining the parameter-optimized multilayer perceptron neural network algorithm model, and model training is complete.

[0122] In related technologies, existing methods for identifying unknown substances also utilize pollutant identification algorithms, such as the existing technology (taking patent application CN116678864A as an example), which uses the PCA-SVM algorithm to identify different types of quinolone antibiotics. However, this technology is not suitable for complex environmental systems with pollutants. PCA-SVM cannot distinguish the Raman spectra of mixtures of multiple antibiotics, therefore this data processing method is not suitable for use under complex conditions. The solution in this application establishes a multilayer perceptron model for identifying the Raman spectra of mixed target substances of different concentrations. This model achieves an accuracy of 97.8% in identifying different components, a 5-fold cross-validation of 97.4%, and an F1 score of 98.2%, indicating that the model not only has high accuracy but also good robustness. In addition, this embodiment also uses the SHAP interpretation model to evaluate the rationality of the selected features. The aluminum foil SERS substrate proposed in this embodiment, combined with this model, can detect and identify target substances in actual water samples, showing good application prospects in real water samples.

[0123] For example, during the iteration process, the accuracy and loss rate of the training set and validation set in each iteration can be referenced. Figure 5 As shown, the trained deep learning neural network model can quickly and accurately identify single and mixed targets of sulfadiazine, methylene blue, and malachite green in water. The model accuracy is 97.8%, the 5-fold cross-validation is 97.4%, and the F1 score is 98.2%, proving that the model can accurately identify complex samples.

[0124] Step 250: The prediction results are analyzed using a preset analysis method to obtain key feature samples of each category of target objects.

[0125] The key feature samples are concentrated in the 447-450cm range. -1 1104-1114cm -1 and 571cm -1Within the Raman shift range, and the key feature sample matches the characteristic peak of the target object, it is used to distinguish the spectra of different categories of target objects.

[0126] In this specific implementation, the Kernel-Explainer method within the SHAP model explanation tool was used to interpret the explanatory variables and feature selections of the model, thus evaluating its rationality. For example... Figure 6 As shown in the SHAP summary diagram, the model accurately identified the key features of seven target categories, with the key features mainly concentrated in the 447-450 cm⁻¹ range. -1 1104-1114cm -1 and 571cm -1 Within the Raman shift range, these key features match the characteristic peaks of the target analytes, which is of great significance for distinguishing the spectra of different types of target analytes and plays a crucial role in this distinction, indicating that the selected features of the model are reasonable. Finally, the model was used to detect and identify antibiotics spiked in actual water samples with different matrices, achieving an accuracy of 100%.

[0127] As an example, firstly, taking the preparation of aluminum foil SERS substrate and the rapid detection and identification of malachite green solution in aquaculture water as an example, referring to the experimental conditions in ① above, 1×10 -5 Malachite green was applied at a concentration of mol / L. An aluminum foil SERS substrate was immersed in actual aquaculture water for 10 minutes. After removal and allowing the solution to evaporate, the Raman signal on the aluminum foil SERS substrate was detected. The collected Raman spectral data were preprocessed and then input into a multilayer perceptron model for identification. This model accurately identified the target substance in the water as malachite green based on the collected Raman spectral data.

[0128] Taking the preparation of aluminum foil SERS substrate and the rapid detection and identification of mixed solutions of methylene blue, malachite green and sulfadiazine at different concentrations in aquaculture water as an example: referring to the experimental conditions in ① above, 1×10 - 5 mol / L malachite green, 1×10 -5 mol / L sulfadiazine and 1×10 -6 A mol / L solution of methylene blue was used. An aluminum foil SERS substrate was then immersed in the mixed solution for 30 minutes. After removal and evaporation of the solution, the Raman signal on the aluminum foil SERS substrate was detected. The acquired Raman spectral data were preprocessed and input into a multilayer perceptron model for identification. This model accurately identified the target substances in the water as malachite green, sulfadiazine, and methylene blue based on the acquired Raman spectral data.

[0129] Therefore, this embodiment proposes a Raman spectroscopy identification method for drug residues in aquatic water based on a neural network algorithm, namely, a Raman spectroscopy identification method for drug residues in aquaculture water based on a multilayer perceptron deep learning algorithm. The detection substrate is obtained by treating a metal foil with a silver nitrate solution containing hydrofluoric acid. Based on this substrate, a Raman-enhanced detection method for multiple drug residues in water is established to achieve quantitative analysis. The identification of unknown substances is based on the above detection method, collecting model data and establishing a dataset, using a multilayer perceptron neural network model, and training and optimizing the parameter model through the dataset.

[0130] It should be noted that the identification algorithm in this embodiment uses a deep learning multilayer perceptron algorithm to model the Raman spectra of single and mixed targets of sulfadiazine, methylene blue and malachite green. The resulting model, combined with the detection substrate, can be used for rapid detection and identification of drug residues in the aquatic environment.

[0131] Furthermore, the method provided in this embodiment can detect sulfadiazine, methylene blue, and malachite green in water. The aluminum foil SERS substrate can be rapidly prepared using a chemical replacement method, reducing operating costs and shortening preparation time. By training the model and creating recognition software, pollutants and their types can be identified quickly and accurately.

[0132] In summary, this embodiment, in terms of model training and construction of key sample features, firstly uses Raman spectroscopy to analyze solution samples using a prepared Raman spectroscopy-enhanced detection substrate, obtaining Raman spectral sample sets of single and mixed targets. Then, the Raman spectral sample sets are sorted and divided to obtain training, validation, and test sets. Based on the compositional characteristics of the Raman spectral sample sets, a multilayer perceptron algorithm model based on a neural network is established. The model is trained using the training set, validated using the validation set, and tested using the test set, resulting in a trained neural network algorithm model. Finally, a preset analytical method is used to analyze the model, obtaining key feature samples of each category of target. Therefore, this patent employs a multilayer perceptron deep learning model to identify mixed pollutants and uses SHAP to analyze the deep learning model, providing a new technical solution for high-sensitivity, high-reproducibility detection of surface-enhanced Raman spectroscopy and rapid and accurate identification of drug residues in water using deep learning models. After effective training, the model can identify single and mixed targets such as sulfadiazine, methylene blue, and malachite green in water, thus accurately identifying complex samples. Furthermore, using the SHAP model interpretation tool, it can quickly and accurately identify pollutants and their types.

[0133] Furthermore, in this embodiment, a multilayer perceptron algorithm is used to train the recognition model to obtain a trained neural network recognition model. The recognition model can then be applied in practice. The model automatically captures the key features of the Raman spectral data of the target object to be detected and identified, and then combines the key features for recognition and classification.

[0134] It should be noted that there are many options for the solution ratio. The concentration and ratio of each solution in the above embodiments are only examples. This application does not limit the ratio of each target solution and the metal substrate solution.

[0135] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should know that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps may be performed in other orders or simultaneously.

[0136] like Figure 7 As shown in the figure, this application embodiment also provides a water drug residue Raman spectroscopy identification device 700 based on a neural network algorithm, including:

[0137] The Raman spectroscopy detection and analysis module 710 is used to perform Raman spectroscopy detection and analysis on the solution of the target analyte by means of a prepared Raman spectroscopy-enhanced detection substrate to obtain Raman spectral data, wherein the Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified.

[0138] The analysis module 720 is used to perform target species analysis on the Raman spectral data through a neural network algorithm model to obtain the key characteristics of the target species;

[0139] The category differentiation module 730 is used to differentiate categories based on the key features and in combination with preset key feature samples to obtain target object detection and identification results; wherein, the target object solution to be detected is a target object aqueous solution detected by a metal substrate, the metal substrate is an aluminum foil SERS substrate, and the target object solution to be detected contains at least one target object to be detected and identified from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue and malachite green.

[0140] Optional, the Raman spectroscopy detection and analysis module 710 includes:

[0141] The acquisition submodule is used to acquire raw Raman spectral data from the solution of the target analyte using Raman spectroscopy-enhanced detection substrate. The raw Raman spectral data includes single-target Raman spectra or mixed-target Raman spectra.

[0142] The preprocessing submodule is used to preprocess the original Raman spectral data to obtain Raman spectral data; wherein, the aluminum foil SERS substrate in the Raman spectroscopy enhancement detection substrate is obtained by a Raman enhancement detection substrate preparation method, which utilizes the reaction of metal foil with silver nitrate solution containing hydrofluoric acid to form a nano-silver structure on the metal foil.

[0143] Optionally, the acquisition submodule is specifically used to detect the solution of the target analyte under target temperature conditions using Raman spectroscopy detection method and Raman spectroscopy-enhanced detection substrate, to obtain Raman spectral characteristic peak information at the corresponding solution concentration of the target analyte, which is used as the original Raman spectral data; wherein, the Raman spectral characteristic peak information includes Raman spectral characteristic peak intensity and / or peak area intensity.

[0144] Optional, the analysis module 720 includes:

[0145] The classification submodule is used to input the Raman spectral data into a neural network algorithm model to classify target objects and obtain a first classification result, which includes a single target object classification result or a multi-target object classification result.

[0146] The parsing submodule is used to perform parsability analysis on the first classification result using the neural network algorithm model, and output the key features of at least one target object corresponding to the identified Raman spectral data.

[0147] Optionally, the parsing submodule is specifically used to select target neurons for identification and analysis from the neural network algorithm model based on the first classification result; and to perform feature analysis on the Raman spectral data through the target neurons to obtain key features.

[0148] Optionally, the water drug residue Raman spectroscopy identification device 700 based on neural network algorithm also includes:

[0149] The sample detection and analysis module is used to detect and analyze sample solutions using Raman spectroscopy technology and a prepared Raman spectroscopy-enhanced substrate to obtain a Raman spectroscopy sample set, which includes single-target spectral samples and mixed-target spectral samples.

[0150] The sorting and partitioning module is used to sort and partition the Raman spectrum sample set using a preset random function to obtain a training set, a validation set, and a test set.

[0151] The model building module is used to establish a multilayer perceptron algorithm model based on neural networks based on the composition characteristics of the Raman spectrum sample set.

[0152] The model training and verification module is used to train the multilayer perceptron algorithm model using the training set, verify the model using the verification set during the training process, and test the model using the test set until the multilayer perceptron algorithm model training is completed, thereby obtaining the performance evaluation results of the model and the trained neural network algorithm model.

[0153] The parsing module is used to analyze the model based on the performance evaluation results using the SHAP parsing method to obtain key feature samples of each category of target objects.

[0154] The key feature samples are concentrated in the 447-450cm range. -1 1104-1114cm -1 and 571cm -1 Within the Raman shift range, and the key feature sample matches the characteristic peak of the target object, it is used to distinguish the spectra of different categories of target objects;

[0155] Optionally, the model building module is specifically used to perform spectral feature number analysis based on the Raman spectral sample set to obtain spectral feature numbers; establish a multilayer perceptron algorithm model based on a neural network, and set the number of input layer neurons of the multilayer perceptron algorithm model to correspond to the number of spectral feature numbers, and set the output layer neurons of the multilayer perceptron algorithm model to correspond to the target object recognition category;

[0156] Optionally, the sample detection and analysis module is specifically used to acquire Raman spectral data of sample solutions of different concentrations in the sample using Raman spectroscopy at room temperature and a Raman spectroscopy-enhanced substrate. This yields a working curve showing the relationship between the target analyte and the intensity or area of ​​its characteristic Raman spectral peaks at different concentrations, serving as a Raman spectral sample set. A Raman-enhanced detection method for the target analyte is then established. The substrate preparation method for the Raman spectroscopy-enhanced detection substrate involves reacting a pre-concentrated hydrofluoric acid solution with at least one concentration of silver nitrate solution according to the detection requirements of the target analyte.

[0157] It should be noted that the water drug residue Raman spectroscopy identification device based on neural network algorithm provided in the various embodiments of this patent can execute the water drug residue Raman spectroscopy identification method based on neural network algorithm provided in any embodiment of this application, and has the corresponding functions and beneficial effects of the method.

[0158] In specific implementations, such as Figure 8 As shown in the figure, this embodiment also provides a water drug residue detection and identification system for implementing a water drug residue Raman spectroscopy identification method based on a neural network algorithm. The system includes:

[0159] Raman spectroscopy detection device 81 is used to perform Raman spectroscopy detection and analysis on a solution of a target analyte using a prepared Raman spectroscopy-enhanced detection substrate to obtain Raman spectral data, wherein the Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified.

[0160] The detection and analysis device 82 is connected to the Raman spectroscopy detection device and is used to analyze the target species in the Raman spectral data through a neural network algorithm model to obtain the key features of the target species; based on the key features, it is combined with preset key feature samples to classify the species and obtain the target species detection and identification results; wherein, the target species solution to be detected is an aqueous solution of the target species detected by a metal substrate, the metal substrate is an aluminum foil SERS substrate, and the target species solution to be detected contains at least one target species to be detected and identified from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue and malachite green.

[0161] Optional, Raman spectroscopy detection equipment 81, including:

[0162] The first Raman spectrometer is used to acquire raw Raman spectral data from a solution of a target analyte by using Raman spectroscopy to enhance the detection substrate. The raw Raman spectral data includes single-target Raman spectra or mixed-target Raman spectra.

[0163] A spectral data processor is used to preprocess the original Raman spectral data to obtain Raman spectral data; wherein the aluminum foil SERS substrate in the Raman spectral enhancement detection substrate is obtained by a Raman enhancement detection substrate preparation method, which utilizes the reaction of a metal foil with a silver nitrate solution containing hydrofluoric acid to form a nano-silver structure on the metal foil.

[0164] Optionally, the first Raman spectrometer is specifically used to detect the solution of the target analyte under target temperature conditions by Raman spectroscopy detection method using a Raman spectroscopy-enhanced detection substrate, to obtain Raman spectral characteristic peak information at the corresponding solution concentration of the target analyte, which is used as the original Raman spectral data; wherein, the Raman spectral characteristic peak information includes Raman spectral characteristic peak intensity and / or peak area intensity.

[0165] Optional, the detection and analysis equipment 82 includes:

[0166] A classification device is used to input the Raman spectral data into a neural network algorithm model to classify target objects and obtain a first classification result, wherein the first classification result includes a single target object classification result or a multi-target object classification result;

[0167] The analysis device is used to perform an analyzeability analysis on the first classification result using the neural network algorithm model, and output the key features of at least one target object corresponding to the identified Raman spectral data.

[0168] Optional, sorting equipment includes:

[0169] A classifier is used to input the Raman spectral data into a neural network algorithm model to classify target objects and obtain a first classification result, which includes a single target object classification result or a multi-target object classification result.

[0170] A parser is used to perform parsability analysis on the first classification result using the neural network algorithm model, and output the key features of at least one target object corresponding to the identified Raman spectral data.

[0171] Optionally, a classifier is specifically used to select target neurons for identification and analysis from the neural network algorithm model based on the first classification result; and to perform feature analysis on the Raman spectral data through the target neurons to obtain key features.

[0172] Optionally, the water drug residue detection and identification system also includes:

[0173] The second Raman spectrometer is used to detect and analyze sample solutions using Raman spectroscopy technology and a prepared Raman spectroscopy-enhancing substrate to obtain a Raman spectroscopy sample set, which includes single-target spectral samples and mixed-target spectral samples.

[0174] The model building and training device is connected to the second Raman spectrometer and is used to establish a multilayer perceptron algorithm model based on the composition characteristics of the Raman spectrum sample set; the multilayer perceptron algorithm model is trained using the training set, and validated using the validation set during the model training process; the model is tested using the test set until the multilayer perceptron algorithm model training is completed, and the performance evaluation results of the model and the trained neural network algorithm model are obtained.

[0175] The analysis device, connected to the model building and training device, is used to analyze the model using the SHAP analysis method based on the performance evaluation results, obtaining key feature samples of each category of target objects; wherein, the sample solution contains target object solution samples of at least one concentration, and the key feature samples are concentrated in the 447-450 cm⁻¹ region. -1 1104-1114cm -1 and 571cm -1 Within the Raman shift range, and the key feature sample matches the characteristic peak of the target object, it is used to distinguish the spectra of different categories of target objects.

[0176] Optional model building and training equipment includes:

[0177] The feature number analyzer, connected to the second Raman spectrometer, is used to perform spectral feature number analysis based on the Raman spectrum sample set to obtain spectral feature numbers;

[0178] A model builder is used to establish a multilayer perceptron algorithm model based on a neural network, and to set the number of input layer neurons of the multilayer perceptron algorithm model to correspond to the number of spectral features, and to set the output layer neurons of the multilayer perceptron algorithm model to correspond to the target object recognition category.

[0179] Optionally, a second Raman spectrometer is specifically used to acquire Raman spectral data of sample solutions of different concentrations in a sample at room temperature using Raman spectroscopy and a Raman spectroscopy-enhanced substrate. This yields a working curve showing the relationship between the intensity of the characteristic Raman peaks or the peak area intensity of the target analyte at different concentrations, which serves as a Raman spectral sample set. A Raman-enhanced detection method for the target analyte is then established. The substrate preparation method for the Raman spectroscopy-enhanced detection substrate involves reacting a predetermined concentration of hydrofluoric acid and at least one concentration of silver nitrate solution according to the detection requirements of the target analyte.

[0180] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the water drug residue Raman spectroscopy identification method based on a neural network algorithm provided in any of the foregoing method embodiments.

[0181] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0182] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for Raman spectroscopy identification of drug residues in water based on a neural network algorithm, characterized in that, include: Raman spectroscopy is used to perform Raman spectroscopy detection and analysis on the solution of the target analyte using a prepared Raman spectroscopy-enhanced detection substrate, thereby obtaining Raman spectral data. The Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified. The target species are analyzed by using a neural network algorithm model to obtain the key characteristics of the target species from the Raman spectral data; Based on the key features, and combined with preset key feature samples, category distinction is performed to obtain the target object detection and recognition results; The target analyte solution is an aqueous solution of the target analyte detected by a metal substrate, wherein the metal substrate is an aluminum foil SERS substrate, and the target analyte solution contains at least one target analyte selected from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green. The Raman spectral data is analyzed for target species using a neural network algorithm model to obtain key features of the target species to be detected and identified. This includes: inputting the Raman spectral data into the neural network algorithm model for target species classification to obtain a first classification result, which may include single-target species classification results or multi-target species classification results; performing an interpretability analysis on the first classification result using the neural network algorithm model and outputting key features of at least one target species corresponding to the identified Raman spectral data. Based on the first classification result, an interpretability analysis is performed using the neural network algorithm model, including: selecting target neurons from the neural network algorithm model for identification and analysis based on the first classification result; performing feature analysis on the Raman spectral data using the target neurons to obtain key features; the neural network model includes neurons of 7 categories, with 3 neurons corresponding to single target object identification and classification, 3 neurons corresponding to mixed identification and classification of two target objects, and 1 neuron corresponding to mixed identification and classification of three or more target objects; The aluminum foil SERS substrate in the Raman spectroscopy enhanced detection substrate is obtained by a Raman enhanced detection substrate preparation method, which involves reacting a metal foil with a silver nitrate solution containing hydrofluoric acid to form a silver nanostructure on the metal foil.

2. The method according to claim 1, characterized in that, Raman spectroscopy was used to analyze the solution of the target analyte using a prepared Raman-enhanced detection substrate, yielding Raman spectral data, including: Raman spectroscopy is used to acquire raw Raman spectral data from the solution of the target analyte using a Raman spectroscopy-enhanced detection substrate. The raw Raman spectral data includes single-target Raman spectra or mixed-target Raman spectra. Preprocessing is performed on the original Raman spectral data to obtain Raman spectral data.

3. The method according to claim 2, characterized in that, Raw Raman spectroscopy data are acquired from the analyte solution using Raman spectroscopy-enhanced detection substrates, including: Under the target temperature conditions, the solution of the target analyte is detected by Raman spectroscopy using a Raman spectroscopy-enhanced detection substrate to obtain the Raman spectral characteristic peak information at the corresponding solution concentration of the target analyte, which is used as the original Raman spectral data. The Raman spectral characteristic peak information includes the Raman spectral characteristic peak intensity and / or peak area intensity.

4. The method according to any one of claims 1-3, characterized in that, Before performing target species analysis on the Raman spectral data using a neural network algorithm model, the following steps are also included: Using Raman spectroscopy, the sample solution is detected and analyzed using a prepared Raman spectroscopy-enhanced substrate to obtain a Raman spectroscopy sample set, which includes single-target spectral samples and mixed-target spectral samples. The Raman spectrum sample set is sorted and divided using a preset random function to obtain a training set, a validation set, and a test set. Based on the compositional characteristics of the Raman spectrum sample set, a multilayer perceptron algorithm model based on neural networks is established. The multilayer perceptron algorithm model is trained using the training set, and validated using the validation set during the training process. The model is then tested using the test set until the multilayer perceptron algorithm model training is completed, resulting in a performance evaluation result of the model and a trained neural network algorithm model. Based on the performance evaluation results, the SHAP parsing method is used to analyze the model and obtain key feature samples of each category of target objects; The sample solution contains at least one concentration of the target substance solution sample, and the key feature samples are concentrated in... , and Within the Raman shift range, and the key feature sample matches the characteristic peak of the target object, it is used to distinguish the spectra of different categories of target objects.

5. The method according to claim 4, characterized in that, Based on the compositional characteristics of the Raman spectral sample set, a multilayer perceptron algorithm model based on a neural network is established, including: Spectral feature number analysis was performed on the Raman spectrum sample set to obtain the spectral feature number; A multilayer perceptron algorithm model based on neural networks is established, and the number of input layer neurons of the multilayer perceptron algorithm model is set to correspond to the number of spectral features, and the output layer neurons of the multilayer perceptron algorithm model are set to correspond to the target object recognition category.

6. The method according to claim 4, characterized in that, Using Raman spectroscopy, the sample solution was analyzed using a prepared Raman spectroscopy-enhanced substrate to obtain a Raman spectroscopy sample set, including: Under room temperature conditions, Raman spectroscopy is used to collect Raman spectral data of sample solutions of different concentrations in the sample using Raman spectroscopy enhancement substrate. Working curves of the relationship between the target analyte and the Raman spectral characteristic peak intensity or peak area intensity at different concentrations are obtained as Raman spectral sample sets, and a Raman enhancement detection method for the target analyte is established. In the substrate preparation method of the Raman spectroscopy-enhanced detection substrate, the reaction solution is prepared by reacting hydrofluoric acid of a preset concentration and silver nitrate solution of at least one concentration according to the detection requirements of the target analyte.

7. A Raman spectroscopy identification device for drug residues in water based on a neural network algorithm, characterized in that, include: The Raman spectroscopy detection and analysis module is used to perform Raman spectroscopy detection and analysis on the solution of the target analyte using a prepared Raman spectroscopy-enhanced detection substrate, and to obtain Raman spectral data, wherein the Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified. The analysis module is used to perform target species analysis on the Raman spectral data using a neural network algorithm model to obtain the key characteristics of the target species; The category differentiation module is used to differentiate categories based on the key features and in combination with preset key feature samples to obtain target object detection and identification results. The target object solution to be tested is an aqueous solution of the target object detected via a metal substrate, wherein the metal substrate is an aluminum foil SERS substrate, and the target object solution to be tested contains at least one target object to be detected and identified, selected from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green. A neural network algorithm model is used to analyze the Raman spectral data to obtain the key features of the target object to be detected and identified, including: inputting the Raman spectral data into the neural network algorithm model for target object classification to obtain a first classification result, which may include single-target object classification results or multi-target object classification results; and performing resolvability analysis on the first classification result using the neural network algorithm model. The system analyzes and outputs key features of at least one target corresponding to the identified Raman spectral data; for the first classification result, it performs an interpretability analysis through the neural network algorithm model, including: selecting target neurons for identification analysis from the neural network algorithm model for the first classification result; performing feature analysis on the Raman spectral data through the target neurons to obtain key features; the neural network model includes neurons of 7 categories, with 3 neurons corresponding to single target identification classification, 3 neurons corresponding to mixed identification classification of two target objects, and 1 neuron corresponding to mixed identification classification of three or more target objects; the aluminum foil SERS substrate in the Raman spectroscopy enhancement detection substrate is obtained by a Raman enhancement detection substrate preparation method, which utilizes the reaction of metal foil with silver nitrate solution containing hydrofluoric acid to form a nano-silver structure on the metal foil.

8. A Raman spectroscopy identification system for drug residues in water based on a neural network algorithm, characterized in that, The system includes: A Raman spectroscopy detection device is used to perform Raman spectroscopy detection and analysis on a solution of a target analyte using a prepared Raman spectroscopy-enhanced detection substrate, thereby obtaining Raman spectral data, wherein the Raman spectral data includes the Raman spectrum of at least one target analyte to be detected and identified. A detection and analysis device, connected to a Raman spectroscopy detection device, is used to analyze the Raman spectral data for target species using a neural network algorithm model to obtain key features of the target species; based on the key features, and combined with preset key feature samples, category differentiation is performed to obtain target species detection and identification results; wherein, the target species solution to be tested is an aqueous solution of the target species detected by a metal substrate, the metal substrate being an aluminum foil SERS substrate, and the target species solution to be tested contains at least one target species to be detected and identified from levofloxacin, fleroxacin, pefloxacin, sulfadiazine, methylene blue, and malachite green; the analysis of the Raman spectral data for target species using a neural network algorithm model to obtain key features of the target species to be detected and identified includes: inputting the Raman spectral data into the neural network algorithm model for target species classification to obtain a first classification result, the first classification result including a single target species classification result or a multi-target species classification result; for The first classification result is analyzed for resolvability using the neural network algorithm model, and key features of at least one target corresponding to the identified Raman spectral data are output. The resolvability analysis of the first classification result using the neural network algorithm model includes: selecting target neurons from the neural network algorithm model for identification analysis; performing feature analysis on the Raman spectral data using the target neurons to obtain key features; the neural network model includes neurons of 7 categories, with 3 neurons corresponding to single target identification classification, 3 neurons corresponding to mixed identification classification of two target objects, and 1 neuron corresponding to mixed identification classification of three or more target objects; the aluminum foil SERS substrate in the Raman spectroscopy enhancement detection substrate is obtained by a Raman enhancement detection substrate preparation method, utilizing the reaction of a metal foil with a silver nitrate solution containing hydrofluoric acid to form a nano-silver structure on the metal foil.

Images