Rice aging detection method based on Raman spectrum characteristic spectrum peak

Abstract

The invention discloses a Raman spectrum characteristic spectrum peak-based rice aging detection method, and relates to the field of rice aging detection, and the method comprises the following steps: constructing a rice sample set; acquiring original Raman spectrum data in a preset wave number range; performing zero-phase smooth filtering and continuous wavelet transform to obtain optimized Raman spectrum data; extracting characteristic spectrum peaks of a plurality of preset wave numbers and a three-dimensional characteristic vector of each characteristic spectrum peak; generating an aging degree grading rule set and an anti-aging characteristic judgment threshold value; processing a to-be-detected sample through S2-S4 to obtain a three-dimensional feature vector, comparing the three-dimensional feature vector with the aging degree grading rule set and the anti-aging characteristic judgment threshold value, and outputting a corresponding aging detection result. According to the method, a dual data set is constructed, a three-dimensional characteristic spectrum peak vector is extracted, and noise suppression and a self-adaptive clustering algorithm are combined, so that quantitative classification of the rice aging degree and synchronous identification of anti-aging varieties are realized, and the detection precision and efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of rice aging detection, and more particularly to a rice aging detection method based on Raman spectroscopy characteristic peaks. Background Art

[0002] As a staple food crop, rice quality and safety are crucial to consumer health and food security. During long-term storage, changes in internal chemical composition lead to quality deterioration (aging), seriously affecting rice's taste, nutritional value, and market acceptance.

[0003] Currently, rice aging detection methods based on changes in chemical composition can provide accurate data, but they have significant limitations. For example, the fatty acid value determination method requires complex experimental operations and is costly; the viscosity determination method is limited by the influence of specific varieties and environmental conditions and is difficult to be widely applied; although volatile organic compound detection has high sensitivity, it requires strict equipment requirements and is complicated to operate. In addition, it usually requires professional technicians to operate, which increases the difficulty of application and the risk of error. More importantly, chemical detection methods are often time-consuming and cannot meet the needs of rapid detection, which limits their practical application.

[0004] In addition to chemical detection methods, biological detection methods judge the degree of aging by evaluating biological characteristics of rice such as enzyme activity and gene expression during the aging process. However, such methods also face many challenges: they usually involve long-term artificial accelerated aging experiments or complex molecular biology techniques, such as RNA sequencing and proteomics analysis, which are not only time-consuming and labor-intensive, but also have strict requirements on sample quality. Secondly, changes in biological indicators are often affected by multiple factors, and the interpretation of results is relatively complicated, which can easily lead to misjudgment. Moreover, due to significant differences between different varieties, the standardization and promotion of biological detection methods face great difficulties, and it is difficult to form a unified application standard.

[0005] Although instrumental detection methods such as electronic noses, electronic tongues, and near-infrared spectrometers offer new avenues for rice staling detection, they also have drawbacks. For example, sample pretreatment is cumbersome and prone to introducing errors; the equipment is expensive and requires high maintenance; and some techniques rely on complex algorithms, resulting in a steep learning curve for non-experts. Furthermore, some instrumental detection methods have a limited scope of application and cannot fully cover all types of rice and their storage conditions, limiting their widespread application.

[0006] Therefore, how to design a rice aging detection method based on Raman spectral characteristic peaks to achieve rapid and accurate assessment of the degree of rice aging, thereby effectively addressing the limitations of existing methods and promoting the development of rice quality detection technology is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0007] In view of this, the present invention provides a rice aging detection method based on Raman spectral characteristic peaks. By constructing a multidimensional data set, optimizing spectral feature extraction and adaptive classification algorithms, it can achieve a comprehensive and accurate assessment of the aging degree of Chinese medicinal materials, effectively identify anti-aging varieties, and improve the accuracy and intelligence level of detection.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A method for detecting rice aging based on Raman spectroscopy characteristic peaks comprises the following steps:

[0010] S1. Constructing a rice sample set; the rice sample set includes sample subsets of different grades stored for different times and sample subsets of different varieties stored for the same time;

[0011] S2. Based on the rice sample set, collecting raw Raman spectral data within a preset wavenumber range under controlled environmental conditions;

[0012] S3, performing zero-phase smoothing filtering and continuous wavelet transform on the original Raman spectrum data to obtain optimized Raman spectrum data;

[0013] S4, extracting a plurality of characteristic peaks of preset wavenumbers from the optimized Raman spectrum data, and obtaining a three-dimensional feature vector of each characteristic peak;

[0014] S5. Based on the three-dimensional feature vectors of different sample subsets, generate a set of aging degree classification rules and a threshold value for determining anti-aging characteristics;

[0015] S6. Process the sample to be tested through S2-S4 to obtain a three-dimensional feature vector, compare it with the aging degree classification rule set and the anti-aging characteristic judgment threshold, and output the corresponding aging detection result.

[0016] Preferably, in S1, the sample subsets graded by different storage times are divided into three levels according to the storage time; among them, 1-6 months is light aging, 7-12 months is moderate aging, and 13-18 months is heavy aging.

[0017] Preferably, in said S2, the controlled environmental conditions are temperature 20±2°C, relative humidity 52±5%; the preset wave number range is 400-3400cm -1 .

[0018] Preferably, in S3, the zero-phase smoothing filtering includes:

[0019] Forward filtering y1=lfilter(b, a, x) and reverse filtering y2=reverse(lfilter(b, a, reverse(y1))); where lfilter represents one-way recursive filtering, reverse represents the reverse operation, a and b represent filter coefficients, and x represents the input spectrum vector.

[0020] Preferably, in S3, the continuous wavelet transform uses Ricker wavelet basis function:

[0021]

[0022] Where σ represents the scale parameter and t represents the time variable.

[0023] Preferably, in said S4, the plurality of preset wave numbers are 443 cm -1 , 865cm -1 、1260cm -1 、1339cm -1 、1382cm -1 、1461cm -1 、2910cm -1 ;

[0024] The three-dimensional eigenvector is defined as v = [Δλ, H, W] T ; Wherein, Δλ represents the peak position offset, H represents the peak height, and W represents the peak width.

[0025] Preferably, in S5, generating a set of aging degree grading rules includes:

[0026] The three-dimensional feature vector v1 of the graded sample subsets with different storage times is clustered using a Gaussian mixture model, and the multi-component normal distribution is fitted using the expectation maximization algorithm.

[0027] Dynamically determine the optimal number of clusters k based on the Bayesian Information Criterion BIC;

[0028] Calculate the mean vector μ for each cluster i and the covariance matrix Σ i ;

[0029] Define the classification rules. When When When the temperature is 100℃, it is judged to be slightly aged; among them, It represents the critical value of the chi-square distribution with 3 degrees of freedom under probability p.

[0030] Preferably, determining the optimal number of clusters k includes:

[0031] The model with the smallest BIC value is preferred, and when the BIC difference of multiple models is less than 2, the model with the smallest number of clusters k is selected.

[0032] Preferably, in S5, generating the anti-aging property determination threshold comprises:

[0033] The three-dimensional feature vector v2 of the sample subsets of different varieties with the same storage time is clustered with outliers robustly, and the minimum covariance determinant algorithm MCD is used to estimate the mean vector μ of the sample set r and the covariance matrix Σ r ;

[0034] Compute the robust Mahalanobis distance for all samples:

[0035]

[0036] The 95% quantile Q of the robust Mahalanobis distance 0.95 As the initial threshold, the final threshold θ was obtained by optimizing it on an independent validation set through receiver operating characteristic curve analysis;

[0037] Define the discrimination threshold of anti-aging varieties as D R >θ.

[0038] Preferably, in the minimum covariance determinant algorithm MCD, the subsample size is set to 75% of the total sample size, and the number of iterations is ≥50 times; when the sample size is greater than 100, an approximate algorithm based on Fast-MCD is used to solve the problem.

[0039] It can be seen from the above technical solution that compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0040] 1. This method constructs a dual dataset of graded samples with different storage times and variety difference samples with the same storage time, combines it with multiple preset three-dimensional vectors of characteristic spectral peaks, and simultaneously quantifies the degree of aging and identifies anti-aging varieties. This overcomes the limitations of single-dimensional detection of traditional methods and significantly improves the comprehensiveness and accuracy of the assessment.

[0041] 2. A preprocessing combination of zero-phase smoothing filtering and Ricker wavelet continuous transform is used to effectively eliminate environmental noise, fluorescence interference, and baseline drift, ensuring the extraction of pure spectral features under complex conditions and laying a high signal-to-noise ratio data foundation for subsequent analysis.

[0042] 3. Based on Gaussian mixture clustering, the classification threshold is dynamically determined and the robust outlier detection algorithm is used to generate the anti-aging judgment threshold, which can adaptively optimize the classification boundary and avoid reliance on manual experience. The Bayesian criterion and receiver curve verification are combined to ensure the generalization ability and reliability of the rule set in multi-sample scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0044] Figure 1 A flow chart of a method for detecting rice aging based on Raman spectroscopy characteristic peaks provided by an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of the process of outputting corresponding aging detection results in a specific application scenario provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] like Figure 1 As shown, this embodiment provides a method for detecting rice aging based on Raman spectroscopy characteristic peaks, comprising the following steps:

[0048] S1. Constructing a rice sample set; the rice sample set includes sample subsets of different grades stored for different times and sample subsets of different varieties stored for the same time;

[0049] S2. Based on the rice sample set, collecting raw Raman spectral data within a preset wavenumber range under controlled environmental conditions;

[0050] S3, performing zero-phase smoothing filtering and continuous wavelet transform on the original Raman spectrum data to obtain optimized Raman spectrum data;

[0051] S4, extracting a plurality of characteristic peaks of preset wavenumbers from the optimized Raman spectrum data, and obtaining a three-dimensional feature vector of each characteristic peak;

[0052] S5. Based on the three-dimensional feature vectors of different sample subsets, generate a set of aging degree classification rules and a threshold value for determining anti-aging characteristics;

[0053] S6. Process the sample to be tested through S2-S4 to obtain a three-dimensional feature vector, compare it with the aging degree classification rule set and the anti-aging characteristic judgment threshold, and output the corresponding aging detection result.

[0054] This method achieves quantitative evaluation of aging degree and simultaneous identification of anti-aging varieties by constructing a dual data set and combining it with three-dimensional characteristic spectral peak vectors, breaking through the limitations of single-dimensional detection of traditional methods; at the same time, zero-phase smoothing filtering and Ricker wavelet transform are used to effectively remove noise interference and improve the signal-to-noise ratio of spectral feature extraction; Gaussian mixture clustering and robust outlier detection algorithm are combined to adaptively determine the classification threshold, and the Bayesian criterion and ROC curve verification are introduced, which significantly enhances the generalization ability and judgment reliability of the model.

[0055] The following further describes each step in the above method in detail;

[0056] In this embodiment, S1, a rice sample set is constructed; the rice sample set includes sample subsets of different grades stored for different times and sample subsets of different varieties stored for the same time;

[0057] Specifically, representative varieties were collected from a major rice-producing area and divided into three levels according to storage time, including:

[0058] The mildly aged group (1-6 months) simulated the state of new grain to short-term storage, with a total of 120 samples covering 37 varieties; the moderately aged group (7-12 months) reflected the changes in quality during mid-term storage, with a total of 122 samples, and the varieties were consistent with the mild group; the severely aged group (13-18 months) represented the state of long-term storage deterioration, with a total of 117 samples, and the varieties were the same as above.

[0059] A separate storage period (e.g., 12 months) was set up, containing 3 kernels from each of the 37 varieties (a total of 111 kernels). By comparing the aging degree of different varieties within the same storage period, naturally resistant varieties were selected.

[0060] In this embodiment S2, based on the rice sample set, raw Raman spectral data within a preset wavenumber range is collected under controlled environmental conditions;

[0061] Specifically, rice samples to be tested are placed in an environmentally controlled chamber or a constant temperature and humidity laboratory, maintaining a stable temperature of 20±2°C and a relative humidity of 52±5%. This strict control minimizes the effects of fluctuations in ambient temperature and humidity on rice molecular vibrations, Raman scattering signal intensity, and peak shape, ensuring comparability of spectral data collected at different times and from different batches.

[0062] Using an Advantage 532 or other similar Raman spectrometer, after shelling and removing surface fluorescent impurities, place the sample on the sample stage. Adjust the laser spot to precisely focus on the endosperm area of ​​the rice kernel. Set the spectrometer parameters: excitation wavelength 532nm, integration time 4 seconds, resolution approximately 10cm -1 Scan and record the sample at 400cm -1to 3400cm -1 Raw Raman spectra within a range of wavenumbers. This range covers the key molecular vibrational fingerprints of rice's main components (starch, protein, and lipids). The raw spectral data is saved in a specific format.

[0063] In this embodiment, S3, the original Raman spectrum data is subjected to zero-phase smoothing filtering and continuous wavelet transform to obtain optimized Raman spectrum data;

[0064] Zero-phase smoothing filtering aims to eliminate high-frequency random noise (such as electronic noise and photon counting noise) in the spectral signal while avoiding peak position distortion caused by phase shift. It involves forward filtering y1 = lfilter(b, a, x) and reverse filtering y2 = reverse(lfilter(b, a, reverse(y1))). Here, lfilter represents a one-way recursive filter, reverse denotes the reverse operation, a and b represent the filter coefficients, and x represents the input spectral vector. The filter coefficients b (numerator) and a (denominator) are selected based on the noise characteristics and are commonly selected from Savitzky-Golay or Butterworth low-pass filters. The zero-phase design ensures that the peak position of the smoothed spectrum is not shifted.

[0065] Furthermore, the continuous wavelet transform uses the Ricker wavelet basis function to remove the slowly changing fluorescence background that is prevalent in Raman spectra, making the characteristic peaks clearer and more prominent, facilitating subsequent quantitative analysis:

[0066]

[0067] Where σ represents the scale parameter and t represents the time variable.

[0068] Specifically, the smoothed spectral data y, the wavelet function, and a scale parameter widths array are input. The wavelet transform can effectively separate the high-frequency Raman peak signal and the low-frequency baseline signal. The baseline signal is extracted by analyzing the wavelet coefficients or reconstructing specific scale components. The baseline is then subtracted from the original smoothed spectrum to obtain the optimized Raman spectral data after baseline correction. The selection of the scale parameter widths requires empirical debugging to best match the width of the baseline change.

[0069] In this embodiment, S4, a plurality of characteristic peaks of preset wave numbers are extracted from the optimized Raman spectrum data, and a three-dimensional feature vector of each characteristic peak is obtained;

[0070] Among them, multiple preset wave numbers are 443cm -1 , 865cm -1 、1260cm -1 、1339cm -1、1382cm -1 、1461cm -1 、2910cm -1 ;

[0071] In this embodiment, seven characteristic peak positions sensitive to rice aging are predetermined: 443 cm -1 (Pyran ring skeleton vibration), 865 cm -1 (CH deformation and CO ring vibration), 1260 cm -1 (CN stretching vibration and in-plane CH deformation), 1339 cm -1 (CC stretching vibration and CH deformation), 1382 cm -1 (CC stretching vibration, CH2 and NH2 rocking vibration), 1461 cm -1 (CH in-plane bending, CH2 symmetric, CH3 antisymmetric vibration), 2910 cm -1 (Antisymmetric CH2 stretching and NH2 stretching vibrations).

[0072] The three-dimensional eigenvector is defined as v = [Δλ, H, W] T ; Wherein, Δλ represents the peak position offset, H represents the peak height, and W represents the peak width.

[0073] Specifically, peak position offset: calculate the difference between the actual central wavenumber position of the peak and the preset standard position. Aging may cause perturbations in the molecular structure, causing slight shifts in the characteristic peak; peak height: directly read the spectral intensity value at the peak apex. Changes in peak height reflect changes in the concentration of corresponding chemical bonds or functional groups; peak width: usually calculate the full width at half maximum (FWHM) of the peak, that is, measure the width of the peak at half the peak height. Changes in peak width can reflect changes in the order or crystallinity of the molecular environment.

[0074] Furthermore, the data of these seven characteristic spectral peaks can be arranged in sequence to form a 21-dimensional feature vector, which is used to characterize the aging state fingerprint of the rice sample.

[0075] In this embodiment, S5, based on the three-dimensional feature vectors of different sample subsets, a set of aging degree classification rules and a threshold value for determining anti-aging properties are generated;

[0076] The generation of the aging degree grading rule set includes:

[0077] The three-dimensional feature vector v1 of the graded sample subsets with different storage times is clustered using a Gaussian mixture model, and the multi-component normal distribution is fitted using the expectation maximization algorithm.

[0078] Dynamically determine the optimal number of clusters k based on the Bayesian Information Criterion BIC;

[0079] Calculate the mean vector μ for each clusteri and the covariance matrix Σ i ;

[0080] Define the classification rules. When When When the temperature is 100℃, it is judged to be slightly aged; among them, It represents the critical value of the chi-square distribution with 3 degrees of freedom under probability p.

[0081] Furthermore, determining the optimal number of clusters k includes:

[0082] The model with the smallest BIC value is preferred, and when the BIC difference of multiple models is less than 2, the model with the smallest number of clusters k is selected.

[0083] Through Gaussian mixture model clustering and Bayesian Information Criterion (BIC) dynamic optimization, it adaptively determines the optimal number of rice aging classifications (k=3), avoiding bias in manually pre-set classifications. It also combines the Mahalanobis distance of multidimensional feature vectors with the chi-square distribution critical value to construct statistically rigorous classification boundaries, significantly improving the accuracy of distinguishing between light, medium, and heavy aging states. This method breaks through the limitations of traditional single-threshold discrimination and maintains classification reliability even under complex sample distributions.

[0084] Furthermore, generating the anti-aging characteristic determination threshold includes:

[0085] The three-dimensional feature vector v2 of the sample subsets of different varieties with the same storage time is clustered with outliers robustly, and the minimum covariance determinant algorithm MCD is used to estimate the mean vector μ of the sample set r and the covariance matrix Σ r ;

[0086] Compute the robust Mahalanobis distance for all samples:

[0087]

[0088] The 95% quantile Q of the robust Mahalanobis distance 0.95 As the initial threshold, the final threshold θ was obtained by optimizing it on an independent validation set through receiver operating characteristic curve analysis;

[0089] Define the discrimination threshold of anti-aging varieties as D R >θ.

[0090] Furthermore, in the minimum covariance determinant algorithm MCD, the subsample size is set to 75% of the total sample size, and the number of iterations is ≥50 times; when the sample size is greater than 100, an approximate algorithm based on Fast-MCD is used to solve the problem.

[0091] It uses the minimum covariance determinant (MCD) algorithm to robustly estimate the core distribution parameters of the variety set, effectively resisting the interference of outliers; taking the 95% quantile of the robust Mahalanobis distance as the benchmark, the final threshold is obtained through ROC curve optimization, realizing sensitive screening of aging-resistant varieties and providing high-reliability criteria for breeding screening.

[0092] In this embodiment, S6, the sample to be tested is processed through S2-S4 to obtain a three-dimensional feature vector, which is compared with the aging degree classification rule set and the anti-aging characteristic judgment threshold, and the corresponding aging detection result is output.

[0093] like Figure 2 As shown, in the grain storage quality monitoring and rice breeding screening scenario, the implementation process of step S6 includes the following specific steps:

[0094] 1) Preprocessing and feature extraction of rice samples to be tested;

[0095] In the grain depot inspection or breeding laboratory scenario, take the rice sample to be tested (such as rice grains sampled from storage or new variety grains), perform environmental stabilization treatment according to the S2 standard process, and use a Raman spectrometer to collect 400-3400cm -1 The original spectrum within the wavenumber range is obtained. Then, zero-phase smoothing filtering and Ricker wavelet transform are performed in S3 to remove noise and fluorescence background, obtaining an optimized spectrum. Finally, S4 is used to extract the three-dimensional vectors (peak displacement, peak height, and peak width) of the seven preset characteristic peaks and merge them into a 21-dimensional feature vector.

[0096] 2) Grading and determination of aging degree;

[0097] The 21-dimensional feature vector of the sample to be tested is input into the pre-trained aging grading rule set. The system automatically calculates the statistical distance between the vector and the cluster center of each aging grade (light / medium / heavy), and determines the classification based on the preset distance threshold range:

[0098] If it falls within the statistical boundary of the lightly aged cluster, the output aging level is: Level 1 (light); if it crosses the distribution domain of the moderately aged cluster, the output aging level is: Level 2 (moderate); if it exceeds the moderate cluster threshold and enters the heavy aged interval, the output aging level is: Level 3 (heavy).

[0099] 3) Anti-aging property screening;

[0100] The characteristic vector of the sample to be tested is simultaneously compared with the anti-aging judgment threshold, and the robust distance of the vector relative to the distribution center of the mainstream varieties is calculated; if the distance value exceeds the preset threshold θ, it is judged to have anti-aging characteristics, otherwise it is marked as a regular variety.

[0101] 4) Comprehensive result generation and feedback;

[0102] The system integrates the grading and resistance determination results to generate a structured report. For example, the test conclusion is: aging status: Level 2 (moderate aging, equivalent to 7-12 months of storage) and anti-aging characteristics: positive. It also gives a recommendation that the sample is suitable for blending with medium- and long-term reserve grains or as a candidate for the anti-aging variety resource library. The results can be displayed in real time on the quality inspection terminal or connected to the warehouse management system to trigger automatic warehouse separation instructions.

[0103] The rice aging detection method based on Raman spectroscopy in this embodiment integrates dual data set construction, three-dimensional spectral peak feature quantification and adaptive threshold decision-making mechanism, and realizes accurate grading of aging degree and simultaneous identification of anti-aging varieties. Noise interference is effectively suppressed through zero-phase filtering and Ricker wavelet transform, and the model generalization ability is significantly improved by combining Bayesian criterion and ROC optimization, breaking through the limitations of traditional single-dimensional detection. In grain storage quality monitoring and rice breeding screening, two-dimensional conclusions can be quickly output to provide data-driven decision support for dynamic storage regulation, germplasm resource optimization and processing technology adaptation, with high precision, strong robustness and applicability for field deployment.

[0104] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. References to the same or similar parts between the various embodiments are sufficient. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For relevant parts, refer to the method description.

[0105] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A rice aging detection method based on Raman spectroscopy characteristic peaks, characterized in that: The following steps are involved: S1. Constructing a rice sample set; the rice sample set includes sample subsets of different grades stored for different times and sample subsets of different varieties stored for the same time; S2. Based on the rice sample set, collecting raw Raman spectral data within a preset wavenumber range under controlled environmental conditions; S3, performing zero-phase smoothing filtering and continuous wavelet transform on the original Raman spectrum data to obtain optimized Raman spectrum data; S4, extracting a plurality of characteristic peaks of preset wavenumbers from the optimized Raman spectrum data, and obtaining a three-dimensional feature vector of each characteristic peak; S5. Based on the three-dimensional feature vectors of different sample subsets, generate a set of aging degree classification rules and a threshold value for determining anti-aging characteristics; S6. Process the sample to be tested through S2-S4 to obtain a three-dimensional feature vector, compare it with the aging degree classification rule set and the anti-aging characteristic judgment threshold, and output the corresponding aging detection result.

2. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 1, wherein: In S1, the sample subsets graded by different storage times are divided into three levels according to the storage time; among them, 1-6 months is light aging, 7-12 months is medium aging, and 13-18 months is heavy aging.

3. The rice aging detection method based on Raman spectroscopy characteristic peaks according to claim 1, characterized in that: In S2, the controlled environmental conditions are temperature 20±2°C, relative humidity 52±5%; the preset wave number range is 400-3400cm -1 .

4. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 1, wherein: In S3, the zero-phase smoothing filter includes: Forward filtering y1=lfilter(b, a, x) and reverse filtering y2=reverse(lfilter(b, a, reverse(y1))); where lfilter represents one-way recursive filtering, reverse represents the reverse operation, a and b represent filter coefficients, and x represents the input spectrum vector.

5. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 1, wherein: In S3, the continuous wavelet transform uses the Ricker wavelet basis function: Where σ represents the scale parameter and t represents the time variable.

6. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 1, wherein: In the S4, multiple preset wave numbers are 443cm -1 , 865cm -1 、1260cm -1 、1339cm -1 、1382cm -1 、1461cm -1 、2910cm -1 ; The three-dimensional eigenvector is defined as v = [Δλ, H, W] T ; Wherein, Δλ represents the peak position offset, H represents the peak height, and W represents the peak width.

7. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 1, wherein: In said S5, generating a set of aging degree classification rules includes: The three-dimensional feature vector v1 of the graded sample subsets with different storage times is clustered using a Gaussian mixture model, and the multi-component normal distribution is fitted using the expectation maximization algorithm. Dynamically determine the optimal number of clusters k based on the Bayesian Information Criterion BIC; Calculate the mean vector μ for each cluster i and the covariance matrix Σ i ; Define the classification rules. When When When the temperature is 100℃, it is judged to be slightly aged; among them, It represents the critical value of the chi-square distribution with 3 degrees of freedom under probability p.

8. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 7, wherein: Determining the optimal number of clusters k includes: The model with the smallest BIC value is preferred, and when the BIC difference of multiple models is less than 2, the model with the smallest number of clusters k is selected.

9. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 1, wherein: In S5, generating the anti-aging property determination threshold includes: The three-dimensional feature vector v2 of the sample subsets of different varieties with the same storage time is clustered with outliers robustly, and the minimum covariance determinant algorithm MCD is used to estimate the mean vector μ of the sample set r and the covariance matrix Σ r ; Compute the robust Mahalanobis distance for all samples: The 95% quantile Q of the robust Mahalanobis distance 0.95 As the initial threshold, the final threshold θ was obtained by optimizing it on an independent validation set through receiver operating characteristic curve analysis; Define the discrimination threshold of anti-aging varieties as D R >θ.

10. The method for detecting rice aging based on Raman spectroscopy characteristic peaks according to claim 9, characterized in that: In the minimum covariance determinant algorithm MCD, the subsample size is set to 75% of the total sample size, and the number of iterations is ≥50 times; when the sample size is greater than 100, an approximate algorithm based on Fast-MCD is used to solve the problem.

Images