Meat freshness detection method and system based on Raman spectrum and electronic nose

Abstract

The invention comprises a meat freshness detection method and system based on Raman spectrum and electronic nose. The method comprises the following steps: A, carrying out Raman spectrum acquisition, volatile basic nitrogen determination and electronic nose sampling on a meat sample; b, obtaining a Raman feature vector based on a Raman spectrum acquisition result, obtaining an electronic nose feature vector based on an electronic nose sampling result, splicing the Raman feature vector and the electronic nose feature vector to obtain fused beef sample features, and establishing a partial least square discriminant analysis model based on volatile basic nitrogen determination and the fused beef sample features; c, performing Raman spectrum collection and electronic nose sampling on a meat material to be detected, and obtaining fused beef sample characteristics according to the step B; and judging the freshness of the to-be-detected meat material by utilizing the obtained fused beef sample characteristics and the partial least square discriminant analysis model. Complementary fusion of molecular structure information and volatile gas information is realized, so that the detection efficiency and accuracy can be considered.

Description

Technical Field

[0001] This invention relates to food testing technology, and in particular to a method and system for detecting the freshness of meat based on Raman spectroscopy and an electronic nose. Background Technology

[0002] Beef is an important source of protein in the human diet and occupies a key position in the global food consumption market. Its freshness is directly related to food safety and nutritional value. However, during post-harvest storage and distribution, meat gradually loses its freshness and may even spoil due to the combined effects of microbial growth, enzymatic reactions, and environmental factors. Spoiled meat not only loses its nutritional value, but the volatile basic nitrogen (TVB-N), biogenic amines, aldehydes, ketones, and other harmful substances it produces can cause food poisoning and other food safety incidents, threatening consumer health and causing economic losses to the industry. Therefore, efficient and accurate detection of meat freshness is a crucial technological support for ensuring food safety, regulating market order, and reducing resource waste.

[0003] Currently, methods for detecting meat freshness are mainly divided into traditional and rapid detection methods. Traditional sensory evaluation is highly subjective, while physicochemical testing is complex, time-consuming, and requires sample destruction. Single rapid detection technologies, such as Raman spectroscopy, are a type of scattering spectroscopy. Raman spectroscopy analyzes the scattered spectrum with frequencies different from the incident light to obtain information on molecular vibrations and rotations, and is applied to molecular structure research. Raman spectroscopy can reflect changes in the molecular structure of meat, but it is easily affected by matrix interference. Electronic noses can detect volatile odor components, but their specificity is insufficient. Both methods, when used alone, suffer from limited detection accuracy and weak anti-interference capabilities, making it difficult to meet the needs of rapid and accurate on-site screening. Summary of the Invention

[0004] This invention addresses the technical problems in existing rapid meat detection technologies, where Raman spectroscopy is susceptible to matrix interference, and electronic nose detection lacks specificity and cannot achieve complementary fusion of molecular structure and volatile gas information, resulting in a trade-off between detection efficiency and accuracy. The invention proposes a meat freshness detection method and system based on Raman spectroscopy and an electronic nose, offering the advantages of simple operation, rapid and accurate results, and suitability for on-site integrated meat freshness screening.

[0005] To achieve this goal, the present invention adopts the following technical solution.

[0006] A method for detecting meat freshness based on Raman spectroscopy and an electronic nose, comprising the following steps: A. Raman spectroscopy acquisition, volatile basic nitrogen determination, and electronic nose sampling were performed on meat samples; B. Based on the Raman spectroscopy acquisition results, the Raman feature vector is obtained, and based on the electronic nose sampling results, the electronic nose feature vector is obtained. The Raman feature vector and the electronic nose feature vector are spliced ​​together to obtain the fused beef sample characteristics. Based on the volatile basic nitrogen determination and the fused beef sample characteristics, a partial least squares discriminant analysis model is established. C. For the meat material to be tested, Raman spectroscopy and electronic nose sampling are performed to obtain the characteristics of the fused beef sample according to step B. The freshness of the meat material to be tested is determined by using the obtained characteristics of the fused beef sample and the partial least squares discriminant analysis model.

[0007] In addition, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, the Raman spectroscopy acquisition of meat samples includes: placing the Raman probe at a distance of 5 mm from the surface of the meat sample, using an excitation wavelength of 785 nm, a laser intensity of 100 mW, a power of 80%, and an integration time of 5 s. Meat samples were randomly scanned 10 times at different locations. After obtaining the original spectra, dark current was removed to reduce background noise interference, and the average spectrum of the 10 scans was taken as the Raman spectrum of the meat samples.

[0008] In addition, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, the electronic nose sampling of meat samples includes taking multiple meat samples of the same weight and placing them into sealed sample bottles. The headspace injection temperature of the sealed sample bottles is 60°C, and the samples are heated and shaken for 400 seconds to achieve equilibrium. Using clean and dry air as a carrier, a gas volume of 5000 μL is injected into the electronic nose using an injection needle. The injection is completed in 10 seconds. The temperature of the injection needle is 200°C. After the gas is injected, the electronic nose collects data for 110 seconds.

[0009] In addition, the meat freshness detection method based on Raman spectroscopy and electronic nose in this patent includes a preprocessing step after acquiring Raman spectra of meat samples: the Savitzky-Golay method is used to smooth and reduce noise in the Raman spectra, with a smoothing window size of 5 points, a polynomial fitting order of 2, and baseline correction using an adaptive iterative reweighted penalized least squares method, with 100 iterations and a penalty factor λ=10. 6 The weight update coefficient is 0.5, and finally the effect of concentration difference is eliminated by normalization.

[0010] In addition, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, after sampling the meat sample with electronic nose, the following steps are taken: the baseline drift is corrected by using the moving average method on the collected data, the sliding window length is set to 5 data points, and the maximum response value of the sensor and the time to reach the peak value are extracted as key parameters.

[0011] Furthermore, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, a Raman feature vector is obtained based on the Raman spectroscopy acquisition results, an electronic nose feature vector is obtained based on the electronic nose sampling results, and the fused beef sample features are obtained by splicing the Raman feature vector and the electronic nose feature vector. After preprocessing, five key characteristic peaks were extracted, including 936.19 cm⁻¹ protein backbone C-C stretching peak, 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak, 1319.96 cm⁻¹ amide III band peak, 1449.16 cm⁻¹ lipid and protein CH2 bending vibration peak, and 1653.58 cm⁻¹ amide I band peak. Each characteristic peak corresponds to two parameters: peak intensity and peak shift, forming a total of 10 Wiener Raman eigenvectors, denoted as R = [R1, R2, ..., R]. 10 The characteristic peaks are arranged in sequence. The electronic nose contains N MOS sensors, each corresponding to a maximum response value and peak time, forming a 2N-dimensional electronic nose feature vector, denoted as E=[E1,E2,…,E…]. 2N The sensors are arranged in sequence, where N is an integer greater than 1; The feature vectors of the two are spliced ​​in a fixed order, with the Raman spectroscopy feature vector first and the electronic nose feature vector second. Redundant features are removed from the spliced ​​feature vectors by variance filtering, with the variance threshold set to 0.01. After the redundant features are removed, M effective features are retained, where M is a positive integer less than 2N+10, to obtain the fused beef sample features.

[0012] In addition, the meat freshness detection method based on Raman spectroscopy and electronic nose in this patent includes a partial least squares discriminant analysis model based on the determination of volatile basic nitrogen and the characteristics of the fused beef sample, which includes: Beef freshness grades are determined based on the measured volatile basic nitrogen values, using an integer combined with a unique thermal double code to eliminate numerical order bias; the standard for volatile basic nitrogen content corresponding to the fresh grade is no more than 15mg / 100g, the integer code is 0, and the unique thermal vector is [1,0,0]. The standard for volatile basic nitrogen content corresponding to the near-fresh grade is greater than 15mg / 100g and not more than 25mg / 100g, with an integer code of 1 and a unique heat vector of [0,1,0]. The standard for volatile basic nitrogen content corresponding to the modified grade is more than 25 mg / 100g, with an integer code of 2 and a unique heat vector of [0,0,1]. A partial least squares discriminant analysis model was established using the fused features of the beef samples and the measured volatile basic nitrogen values.

[0013] A meat freshness detection system based on Raman spectroscopy and an electronic nose includes a Raman spectroscopy acquisition unit, a volatile basic nitrogen determination unit, and an electronic nose unit. It also includes a discriminant analysis model unit and a meat freshness discrimination unit. The Raman spectroscopy acquisition unit and the electronic nose unit are used for Raman spectroscopy acquisition and electronic nose sampling of meat samples and meat materials to be tested, respectively; the volatile basic nitrogen determination unit is used for determining the volatile basic nitrogen of meat samples; The discriminant analysis model unit is used to obtain Raman feature vectors based on Raman spectroscopy acquisition results, obtain electronic nose feature vectors based on electronic nose sampling results, splice Raman feature vectors and electronic nose feature vectors to obtain fused beef sample features, and establish a partial least squares discriminant analysis model based on volatile basic nitrogen determination and fused beef sample features. The freshness discrimination unit for the meat to be tested is used to obtain the Raman feature vector based on the acquisition results of the Raman spectroscopy acquisition unit and the electronic nose feature vector based on the sampling results of the meat to be tested by the electronic nose unit. The Raman feature vector and the electronic nose feature vector are spliced ​​to obtain the fused beef sample features. The freshness of the meat to be tested is determined based on the fused beef sample features and the partial least squares discriminant analysis model.

[0014] In addition, the meat freshness detection system based on Raman spectroscopy and electronic nose in this patent also includes a preprocessing unit. The preprocessing unit performs preprocessing operations on the Raman spectrum: the Savitzky-Golay method is used to smooth and reduce noise in the Raman spectrum, with a smoothing window size of 5 points, a polynomial fitting order of 2, and baseline correction is performed using an adaptive iterative reweighted penalized least squares method with 100 iterations and a penalty factor λ=10. 6 The weight update coefficient is 0.5, and it is used to eliminate the influence of concentration differences through normalization.

[0015] In addition, in the meat freshness detection system based on Raman spectroscopy and electronic nose of this patent, the discriminant analysis model unit includes a beef sample feature construction unit. This unit extracts five key characteristic peaks from the preprocessing unit, including a 936.19 cm⁻¹ protein backbone C-C stretching peak, a 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak, a 1319.96 cm⁻¹ amide III band peak, a 1449.16 cm⁻¹ lipid and protein CH2 bending vibration peak, and a 1653.58 cm⁻¹ amide I band peak. Each characteristic peak corresponds to two parameters: peak intensity and peak shift, forming a 10-dimensional Raman eigenvector, denoted as R = [R1, R2, ..., R]. 10 The characteristic peaks are arranged in sequence. And for an electronic nose unit containing N MOS sensors, each sensor corresponds to the maximum response value and peak time, forming a 2N-dimensional electronic nose feature vector, denoted as E=[E1,E2,…,E2N], with the sensors arranged in sequence, where N is an integer greater than 1; The method involves splicing the Raman spectral feature vector first and the electronic nose feature vector last in a fixed order. Redundant features are removed from the spliced ​​feature vector using a variance filtering method with a variance threshold of 0.01. After removing redundant features, M effective features are retained, where M is a positive integer less than 2N+10, resulting in the fused beef sample features.

[0016] This patented method for detecting meat freshness based on Raman spectroscopy and an electronic nose achieves complementary fusion of molecular structure information and volatile gas information, thus balancing detection efficiency and accuracy. It offers the technical advantages of a simple, rapid, accurate, and on-site integrated screening method for meat freshness. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a meat freshness detection method based on Raman spectroscopy and an electronic nose according to a specific embodiment of the present invention.

[0018] Figure 2 The average spectral curve of all Raman spectral data in the meat freshness detection method based on Raman spectroscopy and electronic nose in the specific embodiment of the invention is used to identify 5 characteristic peaks.

[0019] Figure 3 This is the average spectral curve of the preprocessed Raman spectral data in the meat freshness detection method based on Raman spectroscopy and electronic nose according to the specific embodiments of the invention.

[0020] Figure 4 This is a spectral peak shift diagram in the meat freshness detection method based on Raman spectroscopy and electronic nose according to a specific embodiment of the invention.

[0021] Figure 5 This is a comparison chart of peak values ​​from different sensors for the first 50 samples in the meat freshness detection method based on Raman spectroscopy and electronic nose according to a specific embodiment of the invention. Detailed Implementation

[0022] The present invention will now be described in detail with reference to the accompanying drawings.

[0023] The following detailed exemplary embodiments are disclosed. However, the specific structural and functional details disclosed herein are merely for the purpose of describing exemplary embodiments.

[0024] However, it should be understood that the present invention is not limited to the specific exemplary embodiments disclosed, but covers all modifications, equivalents, and substitutions falling within the scope of this disclosure. Throughout the description of the drawings, the same reference numerals denote the same elements.

[0025] Referring to the accompanying drawings, the structures, proportions, sizes, etc., depicted in the drawings are merely for illustrative purposes to aid those skilled in the art in understanding and reading the content disclosed herein. They are not intended to limit the conditions under which the invention can be implemented and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to the size, without affecting the effects and objectives achieved by the invention, should still fall within the scope of the technical content disclosed herein. Furthermore, the positional limitations used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0026] It should also be understood that the term “and / or” as used herein includes any and all combinations of one or more of the related listed items. Furthermore, it should be understood that when a component or unit is referred to as “connected” or “coupled” to another component or unit, it may be directly connected or coupled to the other component or unit, or there may be intermediate components or units. In addition, other words used to describe the relationship between components or units should be understood in the same manner (e.g., “between” versus “directly between,” “adjacent” versus “directly adjacent,” etc.).

[0027] Figure 1 This is a flowchart illustrating a meat freshness detection method based on Raman spectroscopy and an electronic nose according to a specific embodiment of the present invention. As shown in the figure, the specific embodiment of the present invention includes a meat freshness detection method based on Raman spectroscopy and an electronic nose, which includes the following steps: A. Raman spectroscopy acquisition, volatile basic nitrogen determination, and electronic nose sampling were performed on meat samples; B. Based on the Raman spectroscopy acquisition results, the Raman feature vector is obtained, and based on the electronic nose sampling results, the electronic nose feature vector is obtained. The Raman feature vector and the electronic nose feature vector are spliced ​​together to obtain the fused beef sample characteristics. Based on the volatile basic nitrogen determination and the fused beef sample characteristics, a partial least squares discriminant analysis model is established. C. For the meat material to be tested, Raman spectroscopy and electronic nose sampling are performed to obtain the characteristics of the fused beef sample according to step B. The freshness of the meat material to be tested is determined by using the obtained characteristics of the fused beef sample and the partial least squares discriminant analysis model.

[0028] In addition, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, the Raman spectroscopy acquisition of meat samples includes: placing the Raman probe at a distance of 5 mm from the surface of the meat sample, using an excitation wavelength of 785 nm, a laser intensity of 100 mW, a power of 80%, and an integration time of 5 s. Meat samples were randomly scanned 10 times at different locations. After obtaining the original spectra, dark current was removed to reduce background noise interference, and the average spectrum of the 10 scans was taken as the Raman spectrum of the meat samples.

[0029] In addition, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, the electronic nose sampling of meat samples includes taking multiple meat samples of the same weight and placing them into sealed sample bottles. The headspace injection temperature of the sealed sample bottles is 60°C, and the samples are heated and shaken for 400 seconds to achieve equilibrium. Using clean and dry air as a carrier, a gas volume of 5000 μL is injected into the electronic nose using an injection needle. The injection is completed in 10 seconds. The temperature of the injection needle is 200°C. After the gas is injected, the electronic nose collects data for 110 seconds.

[0030] In addition, the meat freshness detection method based on Raman spectroscopy and electronic nose in this patent includes a preprocessing step after acquiring Raman spectra of meat samples: the Savitzky-Golay method is used to smooth and reduce noise in the Raman spectra, with a smoothing window size of 5 points, a polynomial fitting order of 2, and baseline correction using an adaptive iterative reweighted penalized least squares method, with 100 iterations and a penalty factor λ=10. 6 The weight update coefficient is 0.5, and finally the effect of concentration difference is eliminated by normalization.

[0031] The 5-point window fitting range is moderate, avoiding insufficient noise reduction and blurred characteristic peaks caused by excessively small or large windows. A second-order polynomial is used to adapt to the smooth variation trend of the Raman spectrum, resulting in small fitting errors and no additional distortion. Savitzky-Golay smoothing filters out random spectral noise (mainly at the beginning and end), accurately reflecting the intensity and shift information of five key characteristic peaks, including the 936.19 cm⁻¹ protein backbone C-spanning peak and the 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak.

[0032] Figure 3 The figure shows the average spectral curve of the preprocessed Raman spectral data in the meat freshness detection method based on Raman spectroscopy and electronic nose according to a specific embodiment of the invention. As shown in the figure, the preprocessed curve is significantly smoother.

[0033] The iteration count was 100 to ensure sufficient convergence of baseline correction. Baseline correction is used to eliminate interference from fluorescence background, instrument noise, etc., making the characteristic peaks clearly separated from the baseline and improving peak signal identification. A penalty factor λ=10 was used. 6 This is to balance baseline smoothness and signal retention, ensuring the baseline closely matches the original spectral background without distorting characteristic peak shapes and guaranteeing the accuracy of peak intensity and peak shift parameters. A weight update coefficient of 0.5 smooths the baseline correction process and improves data consistency.

[0034] Normalization eliminates intensity interference caused by differences in sample concentration and fluctuations in sampling volume, unifying the data scale. It ensures that the spectra of meat samples from different batches and concentrations can be directly compared, avoiding misjudgment of characteristic peak intensity changes (such as lipid oxidation peak intensity) due to concentration differences, and improving the generalization ability and discrimination accuracy of the PLS-DA model.

[0035] The core principle of the Savitzky-Golay smoothing method lies in its creation of a locally approximating smoothing window. Data within this smoothing window is treated as a whole and fitted using a polynomial. The size of this smoothing window can be adjusted as needed; typically, choosing a suitable smoothing window can better eliminate noise while preserving important trends and periodicity.

[0036] Its workflow is as follows: (1) Sliding window processing: Moving a fixed-size window (usually an odd number of points) across the data sequence; (2) Local polynomial fitting: Performing polynomial least squares fitting on the data points within each window; (3) Center point replacement: Replacing the original data points with the value of the fitted polynomial at the center point of the window to generate a smooth sequence. In addition, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, after sampling the meat sample with electronic nose, the following steps are taken: the baseline drift is corrected by using the moving average method on the collected data, the sliding window length is set to 5 data points, and the maximum response value of the sensor and the time to reach the peak value are extracted as key parameters.

[0037] Furthermore, in the meat freshness detection method based on Raman spectroscopy and electronic nose of this patent, a Raman feature vector is obtained based on the Raman spectroscopy acquisition results, an electronic nose feature vector is obtained based on the electronic nose sampling results, and the fused beef sample features are obtained by splicing the Raman feature vector and the electronic nose feature vector. After preprocessing, five key characteristic peaks were extracted, including 936.19 cm⁻¹ protein backbone C-C stretching peak, 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak, 1319.96 cm⁻¹ amide III band peak, 1449.16 cm⁻¹ lipid and protein CH2 bending vibration peak, and 1653.58 cm⁻¹ amide I band peak. Each characteristic peak corresponds to two parameters: peak intensity and peak shift, forming a total of 10 Wiener Raman eigenvectors, denoted as R = [R1, R2, ..., R]. 10 The characteristic peaks are arranged in sequence. These five characteristic peaks correspond to structural changes in proteins and lipids during meat spoilage, exhibiting strong signal specificity and comprehensively reflecting the characteristics of decreased freshness. The 1653.58 cm⁻¹ amide I peak is primarily related to changes in protein secondary structure; α-helices transform into β-sheets, and the number of double bonds in unsaturated fatty acids decreases, leading to a decline in the tenderness and texture of beef. The 1449.16 cm⁻¹ lipid and protein CH2 bending vibration peak reflects the impact of lipid oxidation on beef freshness; a decrease in CH2 groups leads to an imbalance in the total lipid and protein ratio, resulting in off-flavors. The 1319.96 cm⁻¹ amide III peak represents an increase in protein β-sheet structure and changes in CH3 / CN bonds caused by microbial decomposition, also contributing to decreased beef freshness. The 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak reflects the reduction in protein content caused by microbial decomposition, lowering the nutritional value of beef and indicating changes in its freshness. The 936.19 cm⁻¹ C-C stretching peak of the protein backbone reflects the breakage of C-C bonds in the protein backbone, resulting in a decrease in glycogen content, which reduces the water retention capacity of beef and leads to juice loss. Changes in these characteristic peaks, whether direct or indirect, ultimately contribute to a decline in the freshness of the beef. Figure 2 The average spectral curve of all Raman spectral data in the meat freshness detection method based on Raman spectroscopy and electronic nose in the specific embodiment of the invention is used to identify five characteristic peaks. As shown in the figure, these five characteristic peaks are clearly representative. Furthermore, Figure 4 This is a spectral peak shift diagram in the meat freshness detection method based on Raman spectroscopy and electronic nose according to a specific embodiment of the invention.

[0038] The electronic nose contains N MOS sensors, each corresponding to a maximum response value and peak time, forming a 2N-dimensional electronic nose feature vector, denoted as E=[E1,E2,…,E…]. 2N The sensors are arranged in sequence, where N is an integer greater than 1; The feature vectors of the two are spliced ​​in a fixed order, with the Raman spectroscopy feature vector first and the electronic nose feature vector second. Redundant features are removed from the spliced ​​feature vectors by variance filtering, with the variance threshold set to 0.01. After the redundant features are removed, M effective features are retained, where M is a positive integer less than 2N+10, to obtain the fused beef sample features.

[0039] In addition, the meat freshness detection method based on Raman spectroscopy and electronic nose in this patent includes a partial least squares discriminant analysis model based on the determination of volatile basic nitrogen and the characteristics of the fused beef sample, which includes: Beef freshness grades are determined based on the measured volatile basic nitrogen values, using an integer combined with a unique thermal double code to eliminate numerical order bias; the standard for volatile basic nitrogen content corresponding to the fresh grade is no more than 15mg / 100g, the integer code is 0, and the unique thermal vector is [1,0,0]. The standard for volatile basic nitrogen content corresponding to the near-fresh grade is greater than 15mg / 100g and not more than 25mg / 100g, with an integer code of 1 and a unique heat vector of [0,1,0]. The standard for volatile basic nitrogen content corresponding to the modified grade is more than 25 mg / 100g, with an integer code of 2 and a unique heat vector of [0,0,1]. One-hot vectors can convert each freshness level into an independent binary vector, with no magnitude relation, completely avoiding the model's misunderstanding of the partial order of the levels. Integer encoding corresponds to the three freshness levels, facilitating manual calibration during model training.

[0040] A partial least squares discriminant analysis model was established using the fused features of the beef samples and the measured volatile basic nitrogen values.

[0041] The input features are M-dimensional effective feature vectors obtained by fusing Raman spectroscopy and electronic nose data and removing redundancy. Each sample corresponds to one M-dimensional vector, forming a feature matrix X (number of samples × M dimensions).

[0042] Standardize the feature matrix X (e.g., center or normalize) so that the mean of each feature is 0 and the variance is 1, thus eliminating the interference of scale differences between different features on the model.

[0043] The optimal number of principal components was determined using 10-fold cross-validation, and the model was optimized.

[0044] The output is labeled with the freshness level corresponding to the volatile basic nitrogen (TVB-N) value.

[0045] The dataset is divided into a training set and a prediction set in a 7:3 ratio.

[0046] Accuracy, precision, recall, and F1 score are used as model evaluation metrics.

[0047] A meat freshness detection system based on Raman spectroscopy and an electronic nose includes a Raman spectroscopy acquisition unit, a volatile basic nitrogen determination unit, and an electronic nose unit. It also includes a discriminant analysis model unit and a meat freshness discrimination unit. The Raman spectroscopy acquisition unit and the electronic nose unit are used for Raman spectroscopy acquisition and electronic nose sampling of meat samples and meat materials to be tested, respectively; the volatile basic nitrogen determination unit is used for determining the volatile basic nitrogen of meat samples; The discriminant analysis model unit is used to obtain Raman feature vectors based on Raman spectroscopy acquisition results, obtain electronic nose feature vectors based on electronic nose sampling results, splice Raman feature vectors and electronic nose feature vectors to obtain fused beef sample features, and establish a partial least squares discriminant analysis model based on volatile basic nitrogen determination and fused beef sample features. The freshness discrimination unit for the meat to be tested is used to obtain the Raman feature vector based on the acquisition results of the Raman spectroscopy acquisition unit and the electronic nose feature vector based on the sampling results of the meat to be tested by the electronic nose unit. The Raman feature vector and the electronic nose feature vector are spliced ​​to obtain the fused beef sample features. The freshness of the meat to be tested is determined based on the fused beef sample features and the partial least squares discriminant analysis model.

[0048] In addition, the meat freshness detection system based on Raman spectroscopy and electronic nose in this patent also includes a preprocessing unit. The preprocessing unit performs preprocessing operations on the Raman spectrum: the Savitzky-Golay method is used to smooth and reduce noise in the Raman spectrum, with a smoothing window size of 5 points, a polynomial fitting order of 2, and baseline correction is performed using an adaptive iterative reweighted penalized least squares method with 100 iterations and a penalty factor λ=10. 6 The weight update coefficient is 0.5, and it is used to eliminate the influence of concentration differences through normalization.

[0049] In addition, in the meat freshness detection system based on Raman spectroscopy and electronic nose of this patent, the discriminant analysis model unit includes a beef sample feature construction unit. This unit extracts five key characteristic peaks from the preprocessing unit, including a 936.19 cm⁻¹ protein backbone C-C stretching peak, a 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak, a 1319.96 cm⁻¹ amide III band peak, a 1449.16 cm⁻¹ lipid and protein CH2 bending vibration peak, and a 1653.58 cm⁻¹ amide I band peak. Each characteristic peak corresponds to two parameters: peak intensity and peak shift, forming a 10-dimensional Raman eigenvector, denoted as R = [R1, R2, ..., R]. 10 The characteristic peaks are arranged in sequence. And for an electronic nose unit containing N MOS sensors, each sensor corresponds to the maximum response value and peak time, forming a 2N-dimensional electronic nose feature vector, denoted as E=[E1,E2,…,E2N], with the sensors arranged in sequence, where N is an integer greater than 1; The method involves splicing the Raman spectral feature vector first and the electronic nose feature vector last in a fixed order. Redundant features are removed from the spliced ​​feature vector using a variance filtering method with a variance threshold of 0.01. After removing redundant features, M effective features are retained, where M is a positive integer less than 2N+10, resulting in the fused beef sample features.

[0050] The following two specific examples will be used to illustrate the specific implementation of the present invention in detail. Example 1

[0051] Step 1: The beef is degreased and detendonized, then minced using a meat grinder and mixed thoroughly. Any remaining tendons and membranes are manually removed, and the beef is portioned into 10g (±0.001g) samples. Each sample is placed in a petri dish, the surface smoothed, sealed with a sealing film, and stored in a 4℃ constant temperature incubator. Samples are taken daily, with 6 samples taken each time for Raman spectroscopy, electronic nose sampling, and freshness index determination, for 14 consecutive days. 70% of the samples are used as the calibration set, and 30% as the prediction set.

[0052] Step 2: The Raman probe is positioned 5 mm from the surface of the beef sample. The excitation wavelength is 785 nm, the laser intensity is 100 mW, the power is 80%, and the integration time is 5 s. The sample is randomly scanned 10 times at different locations. After acquiring the raw spectrum, the dark current is subtracted using software to reduce background noise interference. The average spectrum of the 10 scans is taken to represent the spectral information of the sample.

[0053] Step 3: After the Raman spectroscopy acquisition is completed, the volatile basic nitrogen (TVB-N) value is determined. For example, the fully automated Kjeldahl method specified in the national standard GB5009.228-2016 "National Food Safety Standard - Determination of Volatile Basic Nitrogen in Food" can be used for measurement.

[0054] Step 4: After completing the above steps, perform electronic nose sampling. Accurately weigh 2g (±0.001g) from each of the 6 samples and place them into sealed sample vials. The headspace temperature of the sample vials is 60℃. Heat and shake for 400s to equilibrate. Using clean, dry air as a carrier, inject 5000ul of gas into the electronic nose using a syringe needle. The injection is completed in 10 seconds. The temperature of the syringe needle is 200℃. After the gas is injected, the electronic nose collects data for 110 seconds.

[0055] Step 5: The Savitzky-Golay algorithm is used to smooth and denoise the Raman spectrum. The smoothing window size is set to 5 points, the polynomial fitting order is 2, and baseline correction is performed using the adaptive iterative reweighted penalized least squares method (airPLS). The number of iterations is set to 100, and the penalty factor λ=10. 6 The weighting update coefficient was set to 0.5, and the effect of concentration differences was eliminated through normalization. The electronic nose signal was corrected for baseline drift using a moving average method, with a sliding window length of 5 data points. The maximum response value of the sensor and the time to reach the peak value were extracted as key parameters.

[0056] Step 6: After preprocessing, extract 5 key characteristic peaks (936.19 cm⁻¹ protein backbone C-C stretching peak, 1002.72 cm⁻¹ phenylalanine benzene ring respiration peak, 1319.96 cm⁻¹ amide III band peak, 1449.16 cm⁻¹ lipid and protein CH2 bending vibration peak, 1653.58 cm⁻¹ amide I band peak). Each characteristic peak corresponds to two parameters: peak intensity and peak shift, forming a total of 5 × 2 = 10 Wiraman eigenvectors, denoted as R = [R1, R2, ..., R 10 The characteristic peaks are arranged sequentially. The electronic nose contains 10 MOS sensors, each corresponding to the maximum response value and peak time, forming a 10×2=20-dimensional electronic nose feature vector, denoted as E=[E1,E2,…,E…]. 20 The same principle applies to each sensor. A fixed sequence is used to stitch the Raman spectral features first, followed by the electronic nose features. Redundant features are then removed using a variance filtering method with a variance threshold set to 0.01. After removing redundant features, 25-30 effective features are retained.

[0057] Step 7: Classify beef freshness grades according to the measured TVB-N values, using integer and unique thermal dual coding to eliminate numerical order bias. The standard for volatile basic nitrogen (TVB-N) content corresponding to the fresh grade is no more than 15 mg / 100g, with an integer code of 0 and a unique thermal vector of [1,0,0]. The standard for TVB-N content corresponding to the slightly fresh grade is greater than 15 mg / 100g but no more than 25 mg / 100g, with an integer code of 1 and a unique thermal vector of [0,1,0]. The standard for TVB-N content corresponding to the spoiled grade is greater than 25 mg / 100g, with an integer code of 2 and a unique thermal vector of [0,0,1]. A partial least squares discriminant analysis (PLS-DA) model is established using the fused characteristics of the beef samples and the measured TVB-N values. Figure 5 This is a comparison chart of peak values ​​from different sensors for the first 50 samples in the meat freshness detection method based on Raman spectroscopy and electronic nose according to a specific embodiment of the invention. Example 2

[0058] Beef that has passed quarantine and inspection was purchased from the market. A small portion of the beef was processed, including degreasing and removing tendons. After being minced and mixed evenly, any remaining tendons were manually removed before being packaged into 10g (±0.001g) samples. Raman spectroscopy was performed, TVB-N values ​​were determined, and electronic nose sampling was conducted. During Raman spectroscopy, the distance between the Raman probe and the surface of the beef sample was 5 mm, the excitation wavelength was 785nm, the laser intensity was 100mW, the power was 80%, and the integration time was 5s. The sample was randomly scanned 10 times at different locations. After the original spectrum was acquired, the software was used to subtract dark current to reduce background noise interference, and the average spectrum of the 10 scans was taken to represent the spectral information of the sample. The determination of volatile basic nitrogen (TVB-N) values ​​can be referenced, for example, the national standard GB5009.228-2016 "National Food Safety Standard - Determination of Volatile Basic Nitrogen in Food". During electronic nose sampling, 2g (±0.001g) of each sample was accurately weighed and placed into sealed sample vials. The headspace temperature of the sample vials was 60℃, and the samples were heated and shaken for 400s to achieve equilibration. Using clean, dry air as a carrier, 5000ul of gas was injected into the electronic nose using a syringe needle. The injection was completed in 10 seconds. The temperature of the syringe needle was 200℃. After the gas was injected, the electronic nose collected data for 110 seconds.

[0059] After preprocessing and feature stitching of the collected data, the Raman spectroscopy preprocessing method included Savitzky-Golay curve smoothing, adaptive iterative reweighted penalized least squares (airPLS) baseline correction, and normalization to eliminate the influence of concentration differences. The electronic nose data preprocessing method used a moving average to correct baseline drift. Raman spectroscopy features were stitched together in a fixed order, first the electronic nose features, then the pre-trained classification model. The model was then input into the pre-trained classification model for classification, ultimately outputting a clear classification level of fresh, semi-fresh, or spoiled. The classification results were verified by comparison with measured volatile basic nitrogen (TVB-N).

[0060] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as mentioned above, it should be understood that the present invention is not limited to the forms disclosed in this specification and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described in this specification through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for detecting freshness of meat based on Raman spectroscopy and electronic nose, characterized by, The method comprises the following steps: A. Raman spectrum acquisition, determination of volatile basic nitrogen and electronic nose sampling are performed on the meat sample; B. Raman characteristic vectors are obtained based on the Raman spectrum acquisition results, electronic nose characteristic vectors are obtained based on the electronic nose sampling results, the Raman characteristic vectors and the electronic nose characteristic vectors are spliced to obtain the fused beef sample characteristics, and a partial least squares discriminant analysis model is established based on the determination of volatile basic nitrogen and the fused beef sample characteristics; C. Raman spectrum acquisition and electronic nose sampling are performed on the meat sample to be detected, and the fused beef sample characteristics are obtained according to step B; the freshness of the meat sample to be detected is determined by using the obtained fused beef sample characteristics and the partial least squares discriminant analysis model.

2. The method for detecting freshness of meat based on Raman spectroscopy and electronic nose according to claim 1, characterized by, The Raman spectrum acquisition of the meat sample comprises the following steps: the distance between the Raman probe and the surface of the meat sample is 5 mm, the excitation wavelength is 785 nm, the laser intensity is 100 mW, the power is 80%, and the integration time is 5 s; The original spectrum is obtained by randomly scanning the meat sample 10 times at different positions, and the background noise interference is removed after the dark current is removed, and the average spectrum of 10 times is taken as the Raman spectrum of the meat sample.

3. The freshness detection method of meat based on Raman spectroscopy and electronic nose according to claim 1, characterized in that, The electronic nose sampling of the meat sample comprises the following steps: a plurality of meat samples with the same weight are taken and are respectively placed in sealed sample bottles, the headspace sampling temperature of the sealed sample bottles is 60 DEG C, the bottles are balanced by heating and oscillation for 400 seconds, clean and dry air is used as the carrier, a 5000 ul gas is injected into the electronic nose by using a syringe, the injection is completed in 10 seconds, the temperature of the syringe is 200 DEG C, and data collection of the electronic nose is performed for 110 seconds after the gas is injected.

4. The freshness detection method of meat based on Raman spectroscopy and electronic nose according to claim 1, characterized in that, After Raman spectrum collection of meat samples, pretreatment steps of Raman spectrum are included: Savitzky-Golay method is used for smoothing and denoising of Raman spectrum, the smoothing window size is set to 5 points, the polynomial fitting order is 2, the baseline correction is carried out by using adaptive iteratively reweighted penalized least squares method, the iteration number is set to 100 times, the penalty factor λ = 10 6 , the weight update coefficient is 0.5, and finally the influence of concentration difference is eliminated by normalization processing.

5. The method of claim 4, wherein the Raman spectroscopy and electronic nose based freshness detection method of meat is characterized by, The electronic nose sampling of the meat sample comprises the following steps: the collected data is corrected by using a moving average method to correct the baseline drift, the sliding window length is set to 5 data points, and the maximum response value of the sensor and the time to reach the peak value are taken as the key parameters.

6. The method of claim 5, wherein the Raman spectroscopy and electronic nose based freshness detection method of meat is characterized by, The Raman characteristic vectors are obtained based on the Raman spectrum acquisition results, the electronic nose characteristic vectors are obtained based on the electronic nose sampling results, and the Raman characteristic vectors and the electronic nose characteristic vectors are spliced to obtain the fused beef sample characteristics, which comprises the following steps: Five key feature peaks were extracted after pretreatment, including 936.19 cm⁻¹ protein main chain C-C stretching peak, 1002.72 cm⁻¹ phenylalanine benzene ring breathing peak, 1319.96 cm⁻¹ amide III band peak, 1449.16 cm⁻¹ lipid, protein CH2 bending vibration peak, 1653.58 cm⁻¹ amide I band peak, each feature peak corresponds to peak intensity and peak displacement 2 parameters, a total of 10-dimensional Raman feature vectors, recorded as R = [R1, R2, …, R 10 ] in turn. ] in turn. The electronic nose contains N MOS sensors, each of which corresponds to a maximum response value and a peak time, and together forms a 2N-dimensional electronic nose feature vector, denoted as E = [E1, E2, …, E 2N ], each sensor is arranged in sequence, wherein N is an integer greater than 1; The Raman spectrum characteristic vectors are spliced in front of the electronic nose characteristic vectors in a fixed order, redundant features are removed from the spliced characteristic vectors by using a variance filtering method, the variance threshold is set to 0.01, M-dimensional effective features are retained after the redundant features are removed, M is a positive integer less than 2N+10, and the fused beef sample characteristics are obtained.

7. The method of claim 1, wherein the method is characterized by, The partial least squares discriminant analysis model is established based on the determination of volatile basic nitrogen and the fused beef sample characteristics, which comprises the following steps: The freshness grades of beef are divided according to the determined values of volatile basic nitrogen, and integer and single-hot double coding are used to eliminate numerical bias sequence misleading; the standard of fresh grade is that the content of volatile basic nitrogen is not more than 15 mg / 100 g, the integer coding is 0, and the single-hot vector is [1, 0, 0]; The standard of sub-fresh grade is that the content of volatile basic nitrogen is greater than 15 mg / 100 g and not more than 25 mg / 100 g, the integer coding is 1, and the single-hot vector is [0, 1, 0]. The volatile base nitrogen content corresponding to the deteriorated grade is more than 25 mg / 100 g, the integer code is 2, and the one-hot vector is [0, 0, 1]; The characteristic-fused beef sample characteristics and the measured volatile base nitrogen values are used to establish a partial least squares discriminant analysis model.

8. A meat freshness detection system based on Raman spectrum and electronic nose, comprising a Raman spectrum acquisition unit, a volatile base nitrogen determination unit and an electronic nose unit, further comprising a discriminant analysis model unit and a to-be-detected meat freshness discriminant unit, wherein, The Raman spectrum acquisition unit and the electronic nose unit are respectively used for Raman spectrum acquisition and electronic nose sampling on meat samples and to-be-detected meat materials; The volatile base nitrogen determination unit is used for volatile base nitrogen determination on meat samples; The discriminant analysis model unit is used for obtaining a Raman feature vector based on the Raman spectrum acquisition result, obtaining an electronic nose feature vector based on the electronic nose sampling result, splicing the Raman feature vector and the electronic nose feature vector to obtain a fused beef sample feature, and establishing a partial least squares discriminant analysis model based on the volatile base nitrogen determination and the fused beef sample feature; The to-be-detected meat freshness discriminant unit is used for obtaining a Raman feature vector based on the acquisition result of the Raman spectrum acquisition unit on the to-be-detected meat material, obtaining an electronic nose feature vector based on the sampling result of the electronic nose unit on the to-be-detected meat material, splicing the Raman feature vector and the electronic nose feature vector to obtain a fused beef sample feature, and discriminating the freshness of the to-be-detected meat material according to the fused beef sample feature and the partial least squares discriminant analysis model.

9. The freshness detection system of meat based on Raman spectroscopy and electronic nose as claimed in claim 8, wherein, The pre-processing unit is further configured to perform pre-processing operations on the Raman spectrum, including smoothing and denoising the Raman spectrum by using a Savitzky-Golay method, setting a smoothing window size to 5 points, and setting a polynomial fitting order to 2; baseline correction by using an adaptive iteratively reweighted penalized least squares method, setting an iteration number to 100 times, and setting a penalty factor λ to 10; and eliminating the influence of concentration difference by normalization processing, and setting a weight update coefficient to 0.

5. 6 , the weight update coefficient is 0.5, and the concentration difference is eliminated by normalization processing.

10. The meat freshness detection system based on Raman spectroscopy and electronic nose according to claim 9, characterized in that, The discriminant analysis model unit comprises a beef sample feature fusion construction unit, which is configured to extract five key feature peaks after the pretreatment unit is pretreated, including a 936.19 cm-1 protein main chain C-C stretching peak, a 1002.72 cm-1 phenylalanine benzene ring breathing peak, a 1319.96 cm-1 amide III band peak, a 1449.16 cm-1 lipid and protein CH2 bending vibration peak and a 1653.58 cm-1 amide I band peak, each feature peak corresponding to two parameters of peak intensity and peak displacement, and forming a 10-dimensional Raman feature vector, denoted as R = [R1, R2, …, R 10 ] in sequence. and for each electronic nose unit containing N MOS sensors, each sensor corresponding to a maximum response value and a peak time, forming a 2N-dimensional electronic nose feature vector, denoted as E = [E1, E2, …, EN, E1, E2, …, EN], each sensor being arranged in sequence, wherein N is an integer greater than 1. 2N ], each sensor being arranged in sequence, wherein N is an integer greater than 1. and the two feature vectors are spliced in a fixed order with the Raman spectrum feature vector in front and the electronic nose feature vector behind, redundant features are removed from the spliced feature vector by variance filtering method, the variance threshold is set to 0.01, after the redundant feature removal, M-dimensional effective features are reserved, M is a positive integer less than 2N+10, and a fused beef sample feature is obtained.

Images