Aquatic product drug residue detection method and system based on Raman spectrum

Abstract

The invention relates to an aquatic product drug residue detection method and system based on Raman spectrum. The method comprises the following steps: acquiring Raman spectrum data generated after an aquatic product sample is irradiated by laser; preprocessing the Raman spectrum data, and extracting characteristic peak data; comparing the characteristic peak data with a pre-established drug residue standard spectrum library to obtain a drug residue concentration value corresponding to the characteristic peak data; constructing a quantitative detection model by adopting a machine learning algorithm based on the characteristic peak data and the corresponding drug residue concentration values; and based on the quantitative detection model, predicting the drug residue concentration of the aquatic product. The method can be widely applied to the field of food safety, and an efficient and reliable solution is provided for detection of drug residues in aquatic products.

Description

Technical Field

[0001] The present invention relates to the technical field, and particularly to a method and system for detecting drug residues in aquatic products based on Raman spectroscopy. Background Art

[0002] In the method and system for detecting drug residues in aquatic products based on Raman spectroscopy, spectral acquisition is one of the key steps. In actual operation, the adjustment of the power and wavelength of the laser has a direct impact on the quality of the spectral signal. However, there are differences in the physical properties of different aquatic product samples, such as the color, texture, moisture content, etc. of the samples. These factors will cause different degrees of scattering and absorption when the laser interacts with the sample, thereby affecting the stability of the spectral signal. For example, a dark sample may absorb more laser energy, resulting in a decrease in the intensity of the spectral signal; while a high-moisture sample may cause laser scattering, making noise appear in the spectrogram. In addition, the surface roughness of the sample will also affect the focusing effect of the laser, further affecting the accuracy of spectral acquisition.

[0003] In the link of spectral map comparison, the standard spectral maps of known drug residues are usually obtained under ideal conditions, while the spectral maps of actual samples may be affected by various interference factors, such as impurities in the samples, background noise, etc. These interferences may cause changes in the position, intensity, and shape of the characteristic peaks, thereby increasing the difficulty of comparison. Especially when the drug residue concentration is low, the characteristic peaks may be submerged by noise and it is difficult to accurately identify.

[0004] In the link of spectral map comparison, the standard spectral maps of known drug residues are usually obtained under ideal conditions, while the spectral maps of actual samples may be affected by various interference factors, such as impurities in the samples, background noise, etc. These interferences may cause changes in the position, intensity, and shape of the characteristic peaks, thereby increasing the difficulty of comparison. Especially when the drug residue concentration is low, the characteristic peaks may be submerged by noise and it is difficult to accurately identify.

[0005] Furthermore, when establishing a prediction model for quantitative detection, the accuracy of the model depends on a large amount of high-quality spectral data. However, the spectral data of actual samples often has great variability, such as differences between different batches of samples, changes in acquisition conditions, etc. These variabilities may cause overfitting or underfitting phenomena during the model training process, affecting the generalization ability of the model. In addition, there are many types of drug residues, and the spectral characteristics of each drug are different. How to effectively extract and utilize these characteristics is also a technical problem that needs to be solved during the model establishment process. Summary of the Invention

[0006] The object of the present invention is to propose a method for detecting drug residues in aquatic products based on Raman spectroscopy to solve the problems existing in the above-mentioned prior art.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for detecting drug residues in aquatic products based on Raman spectroscopy, comprising:

[0009] Obtaining Raman spectral data generated after the aquatic product sample is irradiated by laser;

[0010] Preprocessing the Raman spectral data and extracting characteristic peak data;

[0011] Comparing the characteristic peak data with a pre-established drug residue standard spectral library to obtain the drug residue concentration value corresponding to the characteristic peak data;

[0012] Based on the characteristic peak data and the corresponding drug residue concentration value, using a machine learning algorithm to construct a quantitative detection model;

[0013] Based on the quantitative detection model, predicting the drug residue concentration of aquatic products.

[0014] Optionally, preprocessing the Raman spectral data includes:

[0015] Using wavelet transform to remove high-frequency noise in the Raman spectral data, retaining low-frequency characteristic peak information, obtaining a denoised spectral map, and if the noise residue is higher than a preset threshold, adjusting the wavelet transform parameters and denoising again.

[0016] Optionally, extracting characteristic peak data includes:

[0017] Using a preset value as a deviation judgment condition to extract the characteristic peak position value, characteristic peak and intensity value from the denoised spectral map;

[0018] According to the difference between the characteristic peak position value and the standard spectral map position value, judging whether the deviation value exceeds the preset value; if the deviation value exceeds the preset value, using a correction algorithm to adjust the characteristic peak position value to obtain a corrected characteristic peak position value;

[0019] According to the corrected characteristic peak position value, recalculating the characteristic peak intensity value to obtain a corrected characteristic peak intensity value;

[0020] Using the corrected characteristic peak position value and intensity value to update the characteristic peak information in the denoised spectral map.

[0021] Optionally, comparing the characteristic peak data with a pre-established drug residue standard spectral library to obtain the drug residue concentration value corresponding to the characteristic peak data includes:

[0022] Judging the matching degree between the characteristic peak data and the drug residue standard spectral library, and if the matching degree is lower than the preset threshold, re-extracting the characteristic peak data;

[0023] According to the re-extracted characteristic peak information, a correction algorithm is used to adjust the characteristic peak position value and intensity value, and the corrected characteristic peak data is obtained. The corrected characteristic peak data is compared with the drug residue standard spectral library to determine whether the matching degree meets the preset threshold. If it meets, the final spectral comparison result is obtained;

[0024] Based on the final spectral comparison result, the drug residue concentration value corresponding to the characteristic peak data is obtained.

[0025] Optionally, a machine learning algorithm is used to construct a quantitative detection model, including:

[0026] According to the characteristic peak data and the corresponding drug residue concentration value, a regression model is constructed; wherein, the regression model adopts a support vector regression model with a radial basis kernel function;

[0027] Judge whether the fitting degree of the regression model meets the preset threshold. If the fitting degree is lower than the preset threshold, adjust the regression model parameters; through the adjusted regression model parameters, reconstruct the regression model; judge whether the fitting degree of the reconstructed regression model meets the preset threshold. If it meets, determine the quantitative detection model.

[0028] A detection system for drug residues in aquatic products based on Raman spectroscopy, the system includes: a data acquisition module, a feature extraction module, a comparison module, a model construction module, and a prediction module;

[0029] The data acquisition module is used to obtain Raman spectral data generated after the aquatic product sample is irradiated by laser;

[0030] The comparison module is used to preprocess the Raman spectral data and extract characteristic peak data;

[0031] The feature extraction module is used to compare the characteristic peak data with a pre-established drug residue standard spectral library to obtain the drug residue concentration value corresponding to the characteristic peak data;

[0032] The model construction module is used to construct a quantitative detection model based on the characteristic peak data and the corresponding drug residue concentration value by using a machine learning algorithm;

[0033] The prediction module is used to predict the drug residue concentration of aquatic products based on the quantitative detection model.

[0034] Optionally, the feature extraction module extracts characteristic peak data including:

[0035] Using a preset value as the deviation judgment condition, extract the characteristic peak position value and intensity value from the denoised spectrogram;

[0036] Based on the difference between the characteristic peak position value and the standard spectrogram position value, determine whether the deviation value exceeds the preset value; if the deviation value exceeds the preset value, use a correction algorithm to adjust the characteristic peak position value to obtain the corrected characteristic peak position value;

[0037] According to the corrected characteristic peak position value, recalculate the characteristic peak intensity value to obtain the corrected characteristic peak intensity value;

[0038] Use the corrected characteristic peak position value and intensity value to update the characteristic peak information in the denoised spectrogram.

[0039] Optionally, the model construction module uses a machine learning algorithm to construct a quantitative detection model, including:

[0040] According to the characteristic peak data and the corresponding drug residue concentration value, construct a regression model; wherein, the regression model uses a support vector regression model with a radial basis kernel function;

[0041] Judge whether the fitting degree of the regression model meets the preset threshold. If the fitting degree is lower than the preset threshold, adjust the regression model parameters; with the adjusted regression model parameters, reconstruct the regression model; judge whether the fitting degree of the reconstructed regression model meets the preset threshold. If it meets, determine the quantitative detection model.

[0042] The beneficial effects of the present invention are:

[0043] The present invention discloses a method for detecting drug residues in aquatic products based on Raman spectroscopy. The method first optimizes the laser parameters according to the sample characteristics to obtain a stable spectral signal. Subsequently, wavelet transform is used to remove noise, and the characteristic peak data is extracted and corrected. The processed data is compared with the standard spectral library to construct a regression model of drug residue concentration and characteristic peak intensity. Finally, cross-validation is used to evaluate the generalization ability of the model and predict the concentration of actual samples. The present invention realizes the rapid and accurate detection of drug residues by optimizing steps such as laser parameter optimization, signal processing, feature extraction, and model construction, improving the detection efficiency and accuracy. This method can be widely applied in fields such as food safety and environmental monitoring, providing an efficient and reliable solution for drug residue detection. Description of the Drawings

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1Schematic diagram of a method for detecting drug residues in aquatic products based on Raman spectroscopy according to an embodiment of the present invention. Detailed implementation manners

[0046] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0048] As Figure 1 shown, this embodiment proposes a method for detecting drug residues in aquatic products based on Raman spectroscopy, including:

[0049] Obtaining Raman spectral data generated after a laser irradiates an aquatic product sample;

[0050] Preprocessing the Raman spectral data and extracting characteristic peak data;

[0051] Comparing the characteristic peak data with a pre-established drug residue standard spectral library to obtain the drug residue concentration value corresponding to the characteristic peak data, including:

[0052] Based on the characteristic peak data and the corresponding drug residue concentration values, using a machine learning algorithm to construct a quantitative detection model;

[0053] Based on the quantitative detection model, predicting the drug residue concentration of the aquatic product.

[0054] Specifically, in this embodiment, obtaining the Raman spectral data generated after a laser irradiates an aquatic product sample includes:

[0055] Adjusting the laser power and wavelength according to the color and texture of the sample. If the sample color is darker, the laser power is reduced; if the sample texture is rough, the laser wavelength is increased to obtain optimized laser parameters;

[0056] Irradiating the sample with the optimized laser parameters to obtain an original spectral signal. If the signal intensity is lower than a preset threshold, the laser power and wavelength are readjusted until the signal intensity meets the requirements to obtain a stable spectral signal.

[0057] Furthermore, preprocessing the Raman spectral data includes:

[0058] Wavelet transform is used to remove the high-frequency noise in the Raman spectrum data and retain the low-frequency characteristic peak information to obtain the denoised spectrogram. If the noise residue is higher than the preset threshold, the wavelet transform parameters are adjusted and denoising is performed again.

[0059] Specifically, the signal preprocessing process in this embodiment is as follows: A preset value is used as the noise judgment condition to obtain the noise amount in the denoised spectrogram. According to the comparison result between the noise amount and the preset value, it is determined whether it is necessary to adjust the wavelet transform parameters. If the noise amount is higher than the preset value, the wavelet transform parameters are reset to separate the high-frequency band and low-frequency band information. The adjusted wavelet transform parameters are used to process the spectral signal again to obtain a new denoised spectrogram. It is judged whether the noise amount in the new denoised spectrogram meets the requirements of the preset value. If the noise amount is still higher than the preset value, the process of adjusting the wavelet transform parameters and processing the spectral signal is repeated. Finally, a denoised spectrogram with a noise amount lower than the preset value is obtained, and the low-frequency band characteristic peak information is retained.

[0060] Further, extracting the characteristic peak data includes:

[0061] Using a preset value as the deviation judgment condition, extracting the characteristic peak position value and intensity value from the denoised spectrogram;

[0062] According to the difference between the characteristic peak position value and the position value of the standard spectrogram, it is judged whether the deviation value exceeds the preset value; if the deviation value exceeds the preset value, a correction algorithm is used to adjust the characteristic peak position value to obtain the corrected characteristic peak position value;

[0063] According to the corrected characteristic peak position value, the characteristic peak intensity value is recalculated to obtain the corrected characteristic peak intensity value;

[0064] Using the corrected characteristic peak position value and intensity value to update the characteristic peak information in the denoised spectrogram.

[0065] Specifically, in this embodiment, a preset value is used as the deviation judgment condition to extract the characteristic peak position value and intensity value from the denoised spectrogram. According to the difference between the characteristic peak position value and the position value of the standard spectrogram, it is judged whether the deviation value exceeds the preset value. If the deviation value exceeds the preset value, a correction algorithm is used to adjust the characteristic peak position value to obtain the corrected characteristic peak position value. According to the corrected characteristic peak position value, the characteristic peak intensity value is recalculated to obtain the corrected characteristic peak intensity value. Using the corrected characteristic peak position value and intensity value to update the characteristic peak information in the denoised spectrogram. According to the updated characteristic peak information, the separation effect between the high-frequency band and the low-frequency band in the spectrogram is judged. If the separation effect between the high-frequency band and the low-frequency band does not meet the preset requirements, the denoising algorithm parameters are adjusted and the spectral signal is processed again.

[0066] Further, comparing the characteristic peak data with a pre-established drug residue standard spectrogram library to obtain the drug residue concentration value corresponding to the characteristic peak data includes:

[0067] Judge the matching degree between the characteristic peak data and the drug residue standard spectral library. If the matching degree is lower than the preset threshold, re-extract the characteristic peak data;

[0068] According to the re-extracted characteristic peak information, adopt a correction algorithm to adjust the characteristic peak position value and intensity value, and obtain the corrected characteristic peak data. Compare the corrected characteristic peak data with the drug residue standard spectral library to judge whether the matching degree meets the preset threshold. If it meets, obtain the final spectral comparison result;

[0069] Based on the final spectral comparison result, obtain the drug residue concentration value corresponding to the characteristic peak data.

[0070] Specifically, in this embodiment, a preset threshold is adopted to judge the matching degree between the corrected characteristic peak data and the drug residue standard spectral library. If the matching degree is lower than the preset threshold, re-extract the characteristic peak information. According to the re-extracted characteristic peak information, adopt a correction algorithm to adjust the characteristic peak position value and intensity value, and obtain the corrected characteristic peak data. Compare the corrected characteristic peak data with the drug residue standard spectral library to judge whether the matching degree meets the preset threshold. If it meets, obtain the final spectral comparison result. If the matching degree still does not meet the preset threshold, adjust the spectral signal processing parameters and re-perform characteristic peak extraction and correction. Through the adjusted parameters, re-extract the characteristic peak information, adopt a correction algorithm to adjust the characteristic peak position value and intensity value, and obtain the new corrected characteristic peak data. Compare the new corrected characteristic peak data with the drug residue standard spectral library to judge whether the matching degree meets the preset threshold. If it meets, obtain the final spectral comparison result. If the matching degree still does not meet the preset threshold, use a machine learning algorithm to classify the spectral data to determine the best matching result.

[0071] The characteristic peak matching degree is a key indicator for evaluating the reliability of spectral analysis results. In drug residue detection, the characteristic peaks in the standard spectral library are usually used as a benchmark, and the preset threshold is set at 95%. When the matching degree between the characteristic peaks of the test sample and the standard library is lower than this threshold, the extraction of characteristic peaks needs to be carried out again. Taking the detection of a certain pesticide residue as an example, the main characteristic peaks in the standard spectrum are located at 1500 wavenumbers and 1800 wavenumbers. If the deviation exceeds plus or minus 5 wavenumbers in actual detection, it is considered that the matching degree is insufficient. The choice of correction algorithm directly affects the adjustment effect of characteristic peaks. Commonly used correction methods include peak shape fitting and peak position calibration. Taking the detection of a certain veterinary drug residue as an example, the shape of the characteristic peak can be optimized by Gaussian fitting, which can effectively eliminate the interference caused by background noise. When the characteristic peak is distorted, the polynomial fitting method is used for peak shape correction to improve the peak position accuracy. Adjusting the spectral signal processing parameters is an important means to improve the matching degree. Taking the detection of a certain antibiotic residue as an example, by adjusting the baseline correction parameters and the smoothing window size, the extraction effect of characteristic peaks can be effectively improved. When the signal-to-noise ratio is large, increasing the smoothing times can improve the spectral quality, but it should be noted that excessive smoothing may cause peak shape distortion. Machine learning algorithms play an important role in spectral matching. Taking the support vector machine as an example, by establishing a multi-dimensional feature vector of the characteristic peak position and intensity, automatic classification of spectral diagrams can be achieved. In practical applications, the principal component analysis method is used for dimensionality reduction to extract the most representative features, which can significantly improve the classification accuracy. Taking the detection of a certain hormone drug residue as an example, the random forest algorithm is used to classify the characteristic peak data, and the accuracy rate can reach 98% in the training set and remain above 95% in the test set. The quality of characteristic peak data directly affects the final matching result. Taking the detection of a certain antibacterial drug residue as an example, by setting the peak intensity threshold and peak width limit, the interference of false peaks can be effectively excluded. When the signal-to-noise ratio of the characteristic peak is lower than 10, it is necessary to increase the sampling times or adjust the instrument parameters to improve the data quality. By establishing a complete data quality control system, the reliability and accuracy of spectral analysis results can be ensured.

[0072] Furthermore, adopting machine learning algorithms, constructing a quantitative detection model includes:

[0073] According to the characteristic peak data and the corresponding drug residue concentration values, a regression model is constructed; among them, the regression model adopts a support vector regression model with a radial basis kernel function;

[0074] Judge whether the fitting degree of the regression model meets the preset threshold. If the fitting degree is lower than the preset threshold, adjust the regression model parameters; through the adjusted regression model parameters, reconstruct the regression model; judge whether the fitting degree of the reconstructed regression model meets the preset threshold. If it meets, determine the quantitative detection model.

[0075] Specifically, in this embodiment, spectral comparison results are used to obtain characteristic peak intensity values and drug residue concentration values. Based on the characteristic peak intensity values and drug residue concentration values, a regression model is constructed. It is judged whether the goodness of fit of the regression model meets a preset threshold. If the goodness of fit is lower than the preset threshold, the regression model parameters are adjusted. With the adjusted regression model parameters, the regression model is reconstructed. It is judged whether the goodness of fit of the reconstructed regression model meets the preset threshold. If it meets, a quantitative detection model is determined. Based on the quantitative detection model, the drug residue concentration detection result is obtained. A machine learning algorithm is used to optimize the regression model parameters and improve the accuracy of the quantitative detection model.

[0076] The relationship between the characteristic peak intensity value and the drug residue concentration value in the spectral comparison result reflects the basis of quantitative detection. The characteristic peak intensity of Raman spectroscopy shows a non-linear change law with the change of concentration. This non-linear relationship needs to be fitted by a suitable regression model. Taking cefixime as an example, when the concentration is less than 0.1 mg / L, the characteristic peak intensity and the concentration are basically linearly related. When the concentration is greater than 1 mg / L, the growth of the characteristic peak intensity slows down, showing a significant non-linear characteristic. In this case, a cubic polynomial regression model can be used for fitting, and the requirement for the goodness of fit of the model is that the coefficient of determination reaches above 0.98. If the goodness of fit does not meet the requirement, the regression model parameters need to be adjusted. Taking the detection of cefixime as an example, by introducing cross terms and high-order terms, the fitting ability of the model is improved. When the goodness of fit of the cubic polynomial model is insufficient, a quartic term can be added, and the cross term of concentration and characteristic peak wave number is introduced to enhance the model's ability to describe non-linear changes. The establishment of a quantitative detection model requires verifying its stability and reliability. Taking the detection of sulfonamide drug residues as an example, the leave-one-out method is used to divide the data set into a training set and a validation set. The training set is used to construct the regression model, and the validation set is used to evaluate the model performance. The prediction error of the model on the validation set should be controlled within ±5%. The application of machine learning algorithms can significantly improve the accuracy of the quantitative detection model. Taking support vector regression as an example, the data is mapped to a high-dimensional space through a kernel function to better describe the non-linear relationship. For the detection of cephalosporin drug residues, a support vector regression model with a radial basis kernel function can control the prediction error within ±3%. The actual application scenario needs to be considered during the model optimization process. Taking the detection of penicillin residues in dairy products as an example, considering the influence of matrix effects, a matrix correction term is introduced into the regression model. By adding standard products with known concentrations, a matrix effect correction curve is established to improve the accuracy of the model in actual samples. This method can control the relative standard deviation of the detection results within 5%.

[0077] Specifically, this embodiment also uses a cross-validation method to evaluate the generalization ability of the quantitative detection model. If the model error is higher than the preset threshold, the regression model is reconstructed to obtain a generalization model with the error meeting the requirements.

[0078] Spectral data is adopted to obtain the characteristic peak intensity values and drug residue concentration values. An initial regression model is constructed based on the characteristic peak intensity values and drug residue concentration values. The cross-validation method is used to obtain the error value of the initial regression model. If the error value is higher than the preset threshold, the regression model parameters are adjusted. A regression model is reconstructed according to the adjusted regression model parameters. The cross-validation method is used to obtain the error value of the reconstructed regression model. If the error value is lower than the preset threshold, a generalization model is determined.

[0079] The drug residue concentration of the actual sample is predicted through the generalization model. If the deviation between the prediction result and the standard value exceeds the preset range, the model parameters are readjusted to obtain a prediction result that meets the accuracy requirements.

[0080] This embodiment also proposes a detection system for aquatic product drug residues based on Raman spectroscopy. The system includes: a data acquisition module, a feature extraction module, a comparison module, a model construction module, and a prediction module;

[0081] The data acquisition module is used to obtain the Raman spectroscopy data generated after the aquatic product sample is irradiated by laser;

[0082] The comparison module is used to preprocess the Raman spectroscopy data and extract the characteristic peak data;

[0083] The feature extraction module is used to compare the characteristic peak data with a pre-established drug residue standard spectrum library to obtain the drug residue concentration value corresponding to the characteristic peak data;

[0084] The model construction module is used to construct a quantitative detection model based on the characteristic peak data and the corresponding drug residue concentration values by using a machine learning algorithm;

[0085] The prediction module is used to predict the drug residue concentration of the aquatic product based on the quantitative detection model.

[0086] Furthermore, the feature extraction module extracts the characteristic peak data including:

[0087] Using a preset value as the deviation judgment condition, the characteristic peak position values and intensity values are extracted from the denoised spectrogram;

[0088] According to the difference between the characteristic peak position value and the standard spectrogram position value, it is judged whether the deviation value exceeds the preset value; if the deviation value exceeds the preset value, the characteristic peak position value is adjusted by using a correction algorithm to obtain the corrected characteristic peak position value;

[0089] According to the corrected characteristic peak position value, the characteristic peak intensity value is recalculated to obtain the corrected characteristic peak intensity value;

[0090] The characteristic peak information in the denoised spectrogram is updated by using the corrected characteristic peak position value and intensity value.

[0091] Furthermore, the model construction module uses machine learning algorithms to construct a quantitative detection model, including:

[0092] Construct a regression model based on the characteristic peak data and the corresponding drug residue concentration values; among them, the regression model uses a support vector regression model with a radial basis kernel function.

[0093] Judge whether the fitting degree of the regression model meets the preset threshold. If the fitting degree is lower than the preset threshold, adjust the regression model parameters; use the adjusted regression model parameters to reconstruct the regression model; judge whether the fitting degree of the reconstructed regression model meets the preset threshold. If it meets, determine the quantitative detection model.

[0094] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for detecting drug residues in aquatic products based on Raman spectroscopy, characterized in that: include: Obtaining Raman spectrum data generated by aquatic product samples after being irradiated by laser; Preprocessing the Raman spectrum data and extracting characteristic peak data; Comparing the characteristic peak data with a pre-established drug residue standard spectral library to obtain a drug residue concentration value corresponding to the characteristic peak data; Based on the characteristic peak data and the corresponding drug residue concentration value, a quantitative detection model is constructed using a machine learning algorithm; Based on the quantitative detection model, the drug residue concentration in aquatic products is predicted.

2. The method for detecting drug residues in aquatic products based on Raman spectroscopy according to claim 1, characterized in that: Preprocessing the Raman spectrum data includes: The high-frequency noise in the Raman spectrum is removed by wavelet transform, and the low-frequency characteristic peak information is retained to obtain a denoised spectrum. If the residual noise is higher than a preset threshold, the wavelet transform parameters are adjusted and denoised again.

3. The method for detecting drug residues in aquatic products based on Raman spectroscopy according to claim 2, characterized in that: Extracting characteristic peak data includes: Using the preset value as the deviation judgment condition, the characteristic peak position value, characteristic peak and intensity value are extracted from the denoised spectrum; According to the difference between the characteristic peak position value and the standard spectrum position value, it is judged whether the deviation value exceeds the preset value; if the deviation value exceeds the preset value, the characteristic peak position value is adjusted by using a correction algorithm to obtain a corrected characteristic peak position value; Recalculate the characteristic peak intensity value according to the corrected characteristic peak position value to obtain the corrected characteristic peak intensity value; The corrected characteristic peak position and intensity values ​​are used to update the characteristic peak information in the denoised spectrum.

4. The method for detecting drug residues in aquatic products based on Raman spectroscopy according to claim 1, characterized in that: Comparing the characteristic peak data with a pre-established drug residue standard spectral library to obtain a drug residue concentration value corresponding to the characteristic peak data; Determine the matching degree between the characteristic peak data and the drug residue standard spectral library. If the matching degree is lower than a preset threshold, re-extract the characteristic peak data. According to the re-extracted characteristic peak information, a correction algorithm is used to adjust the characteristic peak position value and intensity value to obtain corrected characteristic peak data; Compare the corrected characteristic peak data with the drug residue standard spectral library to determine whether the matching degree meets the preset threshold. If so, obtain the final spectral comparison result; Based on the final spectral comparison result, the drug residue concentration value corresponding to the characteristic peak data is obtained.

5. The method for detecting drug residues in aquatic products based on Raman spectroscopy according to claim 1, characterized in that: Using machine learning algorithms, quantitative detection models are constructed including: A regression model is constructed according to the characteristic peak data and the corresponding drug residue concentration value; wherein the regression model adopts a support vector regression model of a radial basis kernel function; Determine whether the fit of the regression model meets the preset threshold. If the fit is lower than the preset threshold, adjust the regression model parameters. Reconstruct the regression model through the adjusted regression model parameters. Determine whether the fit of the reconstructed regression model meets the preset threshold. If so, determine the quantitative detection model.

6. A drug residue detection system for aquatic products based on Raman spectroscopy, characterized in that: Used to implement the method for detecting drug residues in aquatic products based on Raman spectroscopy as claimed in any one of claims 1 to 5, the system comprises: a data acquisition module, a feature extraction module, a comparison module, a model building module, and a prediction module; The data acquisition module is used to obtain Raman spectrum data generated by the aquatic product sample after being irradiated by laser; The comparison module is used to pre-process the Raman spectrum data and extract characteristic peak data; The feature extraction module is used to compare the characteristic peak data with a pre-established drug residue standard spectrum library to obtain the drug residue concentration value corresponding to the characteristic peak data; The model building module is used to build a quantitative detection model based on the characteristic peak data and the corresponding drug residue concentration value using a machine learning algorithm; The prediction module is used to predict the drug residue concentration in aquatic products based on the quantitative detection model.

7. The aquatic product drug residue detection system based on Raman spectroscopy according to claim 6 is characterized in that: The feature extraction module extracts characteristic peak data including: Using the preset value as the deviation judgment condition, the characteristic peak position value, characteristic peak and intensity value are extracted from the denoised spectrum; According to the difference between the characteristic peak position value and the standard spectrum position value, it is judged whether the deviation value exceeds the preset value; if the deviation value exceeds the preset value, the characteristic peak position value is adjusted by using a correction algorithm to obtain a corrected characteristic peak position value; Recalculate the characteristic peak intensity value according to the corrected characteristic peak position value to obtain the corrected characteristic peak intensity value; The corrected characteristic peak position and intensity values ​​are used to update the characteristic peak information in the denoised spectrum.

8. The aquatic product drug residue detection system based on Raman spectroscopy according to claim 6 is characterized in that: The model building module adopts a machine learning algorithm to build a quantitative detection model including: A regression model is constructed according to the characteristic peak data and the corresponding drug residue concentration value; wherein the regression model adopts a support vector regression model of a radial basis kernel function; Determine whether the fit of the regression model meets the preset threshold. If the fit is lower than the preset threshold, adjust the regression model parameters. Reconstruct the regression model through the adjusted regression model parameters. Determine whether the fit of the reconstructed regression model meets the preset threshold. If so, determine the quantitative detection model.

Images