A non-destructive detection method for biological amine of Spanish mackerel based on Raman spectroscopy

Abstract

The application discloses a nondestructive detection method for biological amine of Spanish mackerel based on Raman spectrum technology, and belongs to the technical field of food nondestructive detection. ‑1 The method first cuts fresh Spanish mackerel into pieces and refrigerates the pieces for different lengths of time to prepare different freshness samples, collects Raman spectra of the samples in the range of 400-2000 cm ‑1 under 532nm laser light, simultaneously determines the histamine, putrescine and cadaverine content in the samples by HPLC, carries out principal component analysis on the spectrum baseline correction and normalization, realizes accurate classification of the freshness of Spanish mackerel by combining LDA, SVM and RF machine learning models, the accuracy of the SVM model reaches 100%, the spectrum characteristic variables are screened by CARS, PLSR and SVR regression models are established, and accurate quantification of the three core biological amine is realized, and the prediction correlation coefficients are all higher than 0.9. The application is nondestructive, rapid and convenient to operate, improves the early warning sensitivity of Spanish mackerel corruption, can effectively avoid the risk of excessive histamine, and also provides technical reference for biological amine detection of similar marine products.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing technology for food, specifically relating to a non-destructive testing method for biogenic amines in mackerel based on Raman spectroscopy. Background Technology

[0002] Fish is an important component of aquatic products, containing essential amino acids, fat-soluble vitamins, micronutrients, and highly unsaturated fatty acids. However, due to its high nutrient and water content, fish is highly susceptible to spoilage during storage, leading to a sharp decline in its sensory quality and the potential for the growth of harmful microorganisms that directly endanger consumer health and food safety. Biogenic amines are a crucial indicator of freshness; excessive intake can be harmful to humans. Therefore, detecting biogenic amine content helps in the early warning of fish spoilage. Currently, the determination of biogenic amine content in food mainly relies on various traditional analytical methods, including high-performance liquid chromatography (HPLC), thin-layer chromatography (TLC), capillary electrophoresis, and liquid chromatography-mass spectrometry (LC-MS). While these techniques provide accurate results, they are cumbersome to operate, time-consuming, and involve destructive sample processing steps. Therefore, non-destructive and rapid detection techniques have become an important research direction in the field of freshness assessment.

[0003] Raman spectroscopy, as an advanced non-destructive testing technique, requires no destructive treatment of the sample, can acquire chemical information from the sample in a short time, and is less affected by moisture interference, making it particularly advantageous for analyzing fish products. Publication number CN118758892B proposes a non-destructive and rapid detection method for the texture of fish flesh at different freshness levels, using spectroscopic technology to solve the "destructive" problem of texture detection, eliminating the need for sample destructive treatment. However, during the spoilage process of mackerel, the accumulation rate of histamine and other biogenic amines caused by microbial action far exceeds the textural changes caused by protein denaturation, making texture indicators insufficiently sensitive for early warning of mackerel spoilage.

[0004] Biogenic amines are recognized markers of microbial metabolism and freshness loss during fish storage, and their accumulation is directly associated with a decline in fish freshness. Important biogenic amines of concern include histamine, cadaverine, and putrescine. Histamine is mainly found in blue-skinned, red-fleshed marine fish, and excessive histamine levels can lead to food poisoning and other risks. Mackerel, as a typical fish with a potential high histamine risk, was assessed using Raman spectroscopy combined with targeted biogenic amine analysis. This approach aims to advance fish quality assessment from simple classification to a deeper understanding of quality deterioration mechanisms. Summary of the Invention

[0005] Technical issues Existing methods for detecting biogenic amines in mackerel mainly rely on traditional methods such as high-performance liquid chromatography (HPLC), which are cumbersome to operate, time-consuming to detect, and require destructive sample processing. Existing non-destructive spectroscopic detection technologies only target the texture of fish flesh. Since the accumulation of biogenic amines in the early stages of mackerel spoilage is much faster than the changes in texture, texture indicators are not sensitive enough to provide early warning of spoilage, thus failing to achieve rapid, non-destructive, and accurate detection of biogenic amines in mackerel.

[0006] Technical content To address the aforementioned technical problems, this invention provides a non-destructive detection method for mackerel biogenic amines based on Raman spectroscopy. This method aims to solve the technical challenges of existing mackerel biogenic amine detection methods that rely on traditional analytical methods, are cumbersome and time-consuming, require destructive processing, and are limited to fish flesh texture and lack sensitivity for early warning of mackerel spoilage.

[0007] This invention is implemented as follows: a non-destructive detection method for biogenic amines in mackerel based on Raman spectroscopy, comprising the following steps: S1: Cut fresh mackerel into pieces, seal them, refrigerate and let them stand to obtain mackerel meat samples of different freshness levels; S2: Raman spectroscopy was performed on mackerel meat samples of different freshness levels from S1, using a 532nm laser with a spectral range of 400-2000 cm⁻¹. -1 Spectra were collected at randomly selected test points on each fish sample. S3: The biogenic amines in the fish meat samples that underwent spectral acquisition in S2 were determined by high performance liquid chromatography. After extraction with perchloric acid, defatting with n-hexane, and derivatization with dansyl chloride, the samples were injected into a high performance liquid chromatography system equipped with an ultraviolet detector and a Polaris C18-A column to analyze the content of histamine, putrescine, and cadaverine in the fish meat. S4: The raw Raman spectra collected in S2 are preprocessed with baseline correction and area normalization to eliminate fluorescence background, sample differences and numerical deviations caused by the instrument. Then, principal component analysis is performed on the preprocessed spectral data to preliminarily determine the freshness grade of the mackerel meat. S5: The preprocessed Raman spectral feature data is randomly divided into training set:test set = 4:1, and input into the SVM machine learning model for training and validation. The accuracy and stability of the model in classifying fish meat of different freshness is evaluated by 5-fold cross-validation, so as to achieve accurate classification of mackerel freshness. S6: The Raman spectroscopy dataset is divided into a training set:prediction set ratio of 3:2. First, competitive adaptive resampling is used to select the variables with the most predictive value from the spectral data. Then, partial least squares regression and support vector regression quantitative prediction models are established for histamine, putrescine, and cadaverine, respectively. Five-fold cross-validation is used to avoid overfitting or underfitting of the model. The model performance is evaluated by calibration / prediction root mean square error, calibration / prediction correlation coefficient, and residual prediction bias, so as to achieve accurate quantitative detection of biogenic amines in mackerel.

[0008] Preferably, when preparing mackerel samples of different freshness in step S1, the refrigeration and static time is 0-5 days.

[0009] Preferably, in step S3, the mobile phase for high-performance liquid chromatography (HPLC) consists of water and acetonitrile. The initial acetonitrile ratio is set to 60%, increasing to 75% within 15 minutes, 85% within 22 minutes, and 90% within 25 minutes. After maintaining the 90% ratio for 3 minutes, it decreases back to 60%. The detection wavelength is set to 254 nm. The biogenic amine content is expressed as milligrams of biogenic amine per kilogram of fresh sample and is quantified by comparing the retention time of pure standards and using a calibration curve.

[0010] Preferably, the area normalization preprocessing in step S4 is performed on the integral area of ​​specific characteristic peaks in the spectrum, which effectively eliminates interference from other peaks or background and improves the effectiveness of spectral data; principal component analysis, by observing the degree of aggregation and separation of fish meat samples of different freshness in the scatter plot, intuitively reveals the spectral differences and intrinsic relationships between samples.

[0011] Preferably, in step S5, during the machine learning model training process, accuracy and F1 score are used as evaluation indicators of the model's classification ability to accurately determine the freshness of mackerel and solve the problems of weak Raman signals in early spoilage and difficulty in distinguishing samples with similar storage dates.

[0012] Preferably, PLSR and SVR models are selected in the construction of the regression quantitative prediction model in step S6.

[0013] Preferably, in step S6, during the feature variable screening and regression model construction process, the model parameters are optimized for the three core biogenic amines—histamine, putrescine, and cadaverine—so that the model prediction correlation coefficients all exceed 0.9, ensuring the accuracy of quantitative detection of biogenic amines and achieving effective early warning of mackerel spoilage.

[0014] Beneficial effects 1. This invention enables non-destructive and rapid detection of biogenic amines in mackerel without damaging the sample. Detection can be completed simply by acquiring chemical information through Raman spectroscopy, greatly simplifying the pretreatment process and shortening the detection time. Furthermore, Raman scattering is less affected by moisture, is well-suited to the characteristics of mackerel samples, is easy to operate, and can meet the needs of rapid on-site detection, providing an efficient technical means for rapid evaluation of mackerel freshness and food safety.

[0015] 2. This invention couples Raman spectroscopy with machine learning. Principal component analysis is used to initially determine the freshness of mackerel, and then an SVM model is used to achieve accurate classification, solving the problems of weak early spoilage signals and difficulty in distinguishing samples with similar storage periods. At the same time, for histamine, putrescine, and cadaverine, PLSR and SVR quantitative prediction models are selected. After feature variable screening and optimization, the prediction correlation coefficients all exceed 0.9, breaking through the limitation of existing spectroscopic technologies that only detect texture. This enables accurate quantification of core biogenic amines, significantly improving the sensitivity of early warning of mackerel spoilage and effectively avoiding food safety risks caused by excessive histamine.

[0016] 3. This invention combines qualitative classification of freshness with quantitative detection of biogenic amines, upgrading the quality of mackerel from simple classification to precise assessment of the degree of deterioration; the model undergoes multi-fold cross-validation to avoid overfitting or underfitting, resulting in high repeatability and reliability; and the method can be extended to the detection of biogenic amines in similar blue-skinned, red-fleshed marine fish, providing a referable technical solution for the quality and safety testing of aquatic products. Attached Figure Description

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 Raman spectra of mackerel refrigerated for different days; Figure 2 Main component analysis of fish meat at different freshness levels; Figure 3 : A machine learning-based fish freshness classification model; (A) LDA model confusion matrix; (B) LDA score map; (C) RF model confusion matrix; (D) SVM model confusion matrix; Figure 4 Changes in biogenic amine content in mackerel during refrigerated storage; Figure 5 Results of PLSR prediction model for (A) histamine, (B) putrescine, and (C) cadaverine content based on Raman spectroscopy; Figure 6 Results of SVR prediction model based on Raman spectroscopy for the contents of (A) histamine, (B) putrescine, and (C) cadaverine. Detailed Implementation

[0018] The technical solution of the present invention will be described in detail below with reference to the embodiments.

[0019] The model used in the following examples: For LDA, SVM, and RF models, see the article "Black Tea Fingerprint Recognition: A Novel Workflow for Geographic Origin Identification When Spectroscopy Meets Machine Learning" (DOI: 10.1016 / j.foodchem.2023.138029). The PLSR and SVR regression prediction models can be found in the article "Rapid and non-destructive detection of pork freshness by visible-near-infrared spectroscopy based on convolutional neural network hybrid model" (DOI: 10.1016 / j.jfca.2025.107199).

[0020] Example 1 This example involves collecting Raman spectra of fish meat at different freshness levels and performing Raman spectroscopy analysis, including the following steps: (1) Preparation of fish meat samples with different freshness: Fresh mackerel were deheaded and gutted, and the fish meat was cut into uniformly sized pieces, resulting in 50 pieces of mackerel meat, which were then placed in resealable bags. To prepare fish meat samples with different freshness, the fish meat was divided into 5 groups and refrigerated for 0-4 days. (2) Raman spectroscopy acquisition: A 532 nm laser light source was selected, with a wavelength range of 400 to 2000 cm⁻¹. -1 The single acquisition integration time was 10 s, with 6 cumulative scans, 50% power, a confocal pinhole of 400 mm, and a grating of 600 gr / mm. During the acquisition process, five locations were randomly measured on each fish fillet, resulting in the acquisition of 250 Raman spectra of fish meat at different freshness levels. (3) Spectral preprocessing: To reduce data redundancy and improve modeling accuracy, the original Raman spectra were denoised and baseline corrected to eliminate fluorescence background interference. Subsequently, normalization was performed to eliminate differences between samples or numerical deviations caused by measuring instruments.

[0021] Figure 1 The normalized Raman average spectra of mackerel refrigerated for different days are shown. The spectra are displayed at 720, 752, 826, 850, 934, 1002, 1128, 1155, 1207, 1446, 1515, and 1654 cm⁻¹. -1 Multiple characteristic peaks were observed, which can be attributed to the stretching effect of proteins, fatty acids, and nucleic acid molecules in fish muscle. With prolonged storage time, the intensity of some Raman peaks decreased, from 1551-1603 cm⁻¹. -1 Changes in band shape within the spectral range are related to the degradation of proteins, fatty acids, and nucleic acids, as well as alterations in protein secondary structure during fish decay. Additionally, at 1155 cm... -1 and 1515 cm -1 The Raman intensity at a certain point is attributed to the color of the fish flesh itself. The changes in Raman spectral intensity and morphology provide a theoretical basis for using Raman spectroscopy to predict the content of targeted biogenic amines.

[0022] Example 2 This example combines non-targeted Raman spectroscopy with multivariate statistical analysis to classify fish meat freshness. The preparation of fish meat samples with different freshness levels and the acquisition of fish meat Raman spectra are the same as in Example 1.

[0023] (1) Freshness and type identification of fish meat: Principal component analysis was performed on the normalized Raman spectra. By analyzing the degree of aggregation and separation of fish meat samples with different freshness in the scatter plot, the freshness status of the fish meat was preliminarily determined, and the spectral differences and intrinsic relationships between different samples were intuitively revealed. The results are as follows: Figure 2 As shown.

[0024] (2) Classification model based on Raman spectroscopy and machine learning: Based on the K-fold cross-validation method, the analyzed fish samples were randomly divided into training and test sets at a ratio of 4:1 to establish a dataset for fish freshness. This dataset was then input into LDA, SVM, and RF models respectively. The classification ability of each model was judged by evaluating the accuracy (ACC) and F1 score. The results are shown in Table 1, and the detailed confusion matrix results are as follows: Figure 3 As shown.

[0025] ACC (%) ×100% Recall (%) = ×100% Precision (%) = ×100% F1-score=2×

[0026] Where TP = true positive, TN = true negative, FP = false positive, and FN = false negative.

[0027] Preliminary exploratory analysis using PCA visualized the distribution trend of the samples over time. The first two principal components explained a total of 26.16% of the total variance, with PC1 explaining 19.4% and PC2 explaining 6.76%. Principal component analysis can roughly distinguish fish samples of different freshness levels, but some samples with similar storage dates overlapped in the principal component score plots. This suggests that the Raman signals generated in the early stages may be too weak to achieve clear differentiation.

[0028] The accuracy rates of machine learning models in classifying fish of different freshness levels were as follows: SVM model 100%, LDA model 98%, and RF model 96%. All three models were able to accurately predict the type of fish freshness with few misclassifications. Compared to unsupervised analysis, machine learning has a significant advantage in classification accuracy.

[0029] Table 1. Classification results of fish meat of different freshness based on different machine learning models

[0030] Example 3 This example measures the content of biogenic amines in fish meat of different freshness levels. At the same time, CARS features are extracted from Raman spectra, and PLSR (partial least squares regression) and SVR (support vector machine) regression prediction models for targeting biogenic amines in mackerel are established. The results are shown in Table 2.

[0031] (1) Determination of biogenic amines: Biogenic amines in mackerel meat stored for different durations (0-4 days) were determined by high performance liquid chromatography (LC-16, Shimadzu Corporation, Kyoto, Japan). During the detection process, the fish meat samples were first extracted with 0.4 mol / L perchloric acid, followed by defatting with n-hexane. The extract was then derivatized with dansyl chloride and injected into the LC-16 system (Shimadzu) for analysis. This system was equipped with a UV detector (254 nm) and a Polaris C18-A column (Agilent ZORBAX Eclipse Plus C18, 4.6 × 250 mm, 5 µm). The mobile phase consisted of water (A) and acetonitrile (B), with the initial B phase ratio set at 60%. This ratio increased to 75% within 15 minutes, 85% within 22 minutes, and 90% within 25 minutes. After maintaining a 90% concentration for 3 minutes, the concentration was gradually reduced to 60%. The detection wavelength was set to 254 nm, and a UV detector was used for detection. Then, the samples detected by chromatography of biogenic amines were matched one-to-one with the samples acquired by Raman spectroscopy.

[0032] (2) Construction and Evaluation of the Prediction Model: To reduce random errors and build a more reliable model, the Raman spectral samples were divided into a calibration set and a prediction set in a 3:2 ratio. The Competitive Adaptive Reweighted Sampling (CARS) variable selection algorithm was used to screen characteristic variables in the spectra. Subsequently, PLSR and SVR were used to construct a regression model to predict the target biogenic amine content of mackerel. The model's predictive performance was evaluated using the root mean square error of calibration / prediction (RMSEC / RMSEP), the calibration / prediction correlation coefficient (Rc / Rp), and the residual prediction bias (RPD). The model was then evaluated and validated.

[0033] Table 2 Results of PLSR and SVR models in predicting biogenic amines

[0034] To investigate the correlation between biogenic amine accumulation and Raman spectral changes in mackerel, biogenic amines in the corresponding samples were measured. Histamine, putrescine, and cadaverine were the main biogenic amines during the cold storage of fish, and their contents all increased with increasing storage time. Histamine showed the largest increase in content, reaching 393.89±62.86 mg / kg, exceeding the limit for histamine content in my country (400 mg / kg).

[0035] The content of targeted biogenic amines was predicted using PLSR and SVR regression models. The PLSR model, constructed based on key variables, showed good predictive performance for cadaverine and putrescine content, with prediction correlation coefficients (Rp) exceeding 0.9. In contrast, the SVR model was more accurate than PLSR for histamine content data, but its root mean square error was relatively high due to the larger sample concentration range.

[0036] The results of Examples 1, 2, and 3 demonstrate that this invention integrates Raman spectroscopy with machine learning to assess fish freshness and explore its correlation with biogenic amines. Unsupervised principal component analysis was used to classify fish samples, effectively assessing freshness levels. Compared to traditional models, the machine learning model exhibits superior classification performance and reveals the correlation between biogenic amine concentration and fish spoilage, highlighting the importance of these indicators in assessing seafood integrity. This demonstrates that the method of this invention has good beneficial effects and significant inventiveness, showing great promise for practical applications in assessing fish meat quality and safety.

[0037] The above are merely some preferred embodiments of the present invention, and the present invention is not limited to the contents of these embodiments. For those skilled in the art, various changes and modifications can be made within the scope of the present invention's technical solutions, and any such changes and modifications are within the protection scope of the present invention.

Claims

1. A non-destructive method for detecting biogenic amines in mackerel based on Raman spectroscopy, characterized in that, Includes the following steps: S1: Cut fresh mackerel into pieces, seal them, refrigerate and let them stand to obtain mackerel meat samples of different freshness levels; S2: Raman spectroscopy was performed on mackerel meat samples of different freshness levels from S1, using a 532nm laser with a spectral range of 400-2000 cm⁻¹. -1 Spectra were collected at randomly selected test points on each fish sample. S3: The biogenic amines in the fish meat samples that underwent spectral acquisition in S2 were determined by high performance liquid chromatography. After extraction with perchloric acid, defatting with n-hexane, and derivatization with dansyl chloride, the samples were injected into a high performance liquid chromatography system equipped with an ultraviolet detector and a Polaris C18-A column to analyze the content of histamine, putrescine, and cadaverine in the fish meat. S4: The raw Raman spectra collected in S2 are preprocessed with baseline correction and area normalization to eliminate fluorescence background, sample differences and numerical deviations caused by the instrument. Then, principal component analysis is performed on the preprocessed spectral data to preliminarily determine the freshness grade of the mackerel meat. S5: The preprocessed Raman spectral feature data is randomly divided into training set:test set = 4:1, and input into the SVM machine learning model for training and validation. The accuracy and stability of the model in classifying fish meat of different freshness is evaluated by 5-fold cross-validation, so as to achieve accurate classification of mackerel freshness. S6: The Raman spectroscopy dataset is divided into a training set:prediction set ratio of 3:

2. First, competitive adaptive resampling is used to select the variables with the most predictive value from the spectral data. Then, partial least squares regression and support vector regression quantitative prediction models are established for histamine, putrescine, and cadaverine, respectively. Five-fold cross-validation is used to avoid overfitting or underfitting of the model. The model performance is evaluated by calibration / prediction root mean square error, calibration / prediction correlation coefficient, and residual prediction bias, so as to achieve accurate quantitative detection of biogenic amines in mackerel.

2. The non-destructive testing method for mackerel biogenic amines according to claim 1, characterized in that, When preparing mackerel samples of different freshness in step S1, the refrigeration and static time is 0-5 days.

3. The non-destructive testing method for mackerel biogenic amines according to claim 1, characterized in that, In step S3, the mobile phase for high-performance liquid chromatography (HPLC) consists of water and acetonitrile. The initial acetonitrile ratio is set at 60%, increasing to 75% within 15 minutes, 85% within 22 minutes, and 90% within 25 minutes. After maintaining the 90% ratio for 3 minutes, it drops back to 60%. The detection wavelength is set at 254 nm. The biogenic amine content is expressed as milligrams of biogenic amine per kilogram of fresh sample and is quantified by comparing the retention time of pure standards and using a calibration curve.

4. The non-destructive testing method for mackerel biogenic amines according to claim 1, characterized in that, In step S4, area normalization preprocessing is performed on the integral area of ​​specific characteristic peaks in the spectrum, effectively eliminating interference from other peaks or background and improving the effectiveness of spectral data. Principal component analysis, by observing the degree of aggregation and separation of fish meat samples of different freshness in the scatter plot, intuitively reveals the spectral differences and intrinsic relationships between samples.

5. The non-destructive testing method for mackerel biogenic amines according to claim 1, characterized in that, In step S5, during the machine learning model training process, accuracy and F1 score are used as evaluation indicators of the model's classification ability to accurately determine the freshness of mackerel.

6. The non-destructive testing method for mackerel biogenic amines according to claim 1, characterized in that, In step S6, PLSR and SVR models were selected for constructing the regression quantitative prediction model.

7. The non-destructive testing method for mackerel biogenic amines according to claim 1, characterized in that, In step S6, during the feature variable screening and regression model construction process, the model parameters were optimized for the three core bioamines: histamine, putrescine, and cadaverine, so that the model prediction correlation coefficients all exceeded 0.9, ensuring the accuracy of quantitative detection of bioamines.

Images