Application of detecting zanthoxylum oil endogenous precipitate based on visible / near infrared spectrum

Abstract

The invention belongs to the technical field of food detection, and relates to an application of detecting zanthoxylum oil endogenous precipitate based on visible / near infrared spectrum. The method comprises the following steps: firstly, collecting spectral data of zanthoxylum oil samples with and without endogenous precipitates; carrying out pretreatment to obtain pretreated spectral data; further extracting characteristic variables, taking the extracted characteristic variables as input variables, taking whether the zanthoxylum oil has the endogenous precipitates as output variables, establishing a near infrared spectrum detection model and a near infrared spectrum prediction model for whether the zanthoxylum oil has the endogenous precipitates, and obtaining a final near infrared analysis model through cross validation; therefore, whether the zanthoxylum oil sample has the endogenous precipitate or not is judged. The NIR technology is applied to zanthoxylum oil endogenous precipitate detection for the first time, the detection efficiency is high (the detection time is shorter than or equal to 30 seconds), the result accuracy is high (the accuracy rate reaches 97%), and the technical blank of rapid and accurate detection of zanthoxylum oil endogenous precipitate is filled; and the method is simple to operate, free of complex pretreatment and wide in application prospect.

Description

Technical Field

[0001] This invention belongs to the field of food detection technology, specifically, it relates to the use of visible / near-infrared spectroscopy to detect endogenous precipitation in Sichuan pepper oil. Background Technology

[0002] Sichuan pepper oil effectively retains the unique numbing flavor and aroma of Sichuan peppercorns and is widely used in Chinese cooking. During the production process, if natural macromolecules such as pectin, polysaccharides, and proteins in the raw materials are not completely removed during pretreatment and pressing, they may remain. During subsequent storage or distribution, these residues can crystallize and polymerize under the influence of external factors such as temperature fluctuations and light, forming insoluble microparticles that aggregate and precipitate. Simultaneously, during the pressing and refining stage, incomplete removal of the colloidal system formed by water-soluble macromolecules such as pectin and polysaccharides and trace impurities in the oil, coupled with insufficient filtration precision, allows residual trace colloids to easily combine with free fatty acids and pigments to form flocculants. Furthermore, metal ions and dust remaining on the inner walls of containers during the packaging process can act as crystal nuclei, adsorbing unstable components of the oil and further exacerbating precipitation and flocculation. All of these precipitations are endogenous, causing the Sichuan pepper oil to become cloudy and layered, severely reducing the product's sensory quality.

[0003] Currently, the detection of endogenous precipitation in Sichuan pepper oil still relies primarily on traditional light inspection, which is highly susceptible to subjective bias by the inspectors, leading to missed detections and misjudgments. Furthermore, manual inspection is inefficient, increasing labor costs and failing to meet the real-time monitoring needs of large-scale production. To address the challenge of precipitation detection in liquid foods, some research has been reported, mainly focusing on chromatography, machine vision, Raman spectroscopy, and X-ray detection. However, these methods have the following limitations when applied to the detection of endogenous precipitation in Sichuan pepper oil: machine vision is easily affected by the color and matrix of the Sichuan pepper oil, making it difficult to effectively distinguish between precipitation and the oil phase background; Raman spectroscopy is easily interfered with by the fluorescence and background signals of pigments and other components in the Sichuan pepper oil, masking the characteristic peaks of the precipitation; X-ray equipment is expensive and requires radiation protection; chromatography requires complex pretreatment and has a long detection cycle, failing to meet the real-time monitoring needs of production lines. Currently, the lack of efficient and reliable detection technologies for endogenous precipitation in Sichuan pepper oil production and quality control severely restricts its quality control and improvement. Visible / near-infrared spectroscopy offers advantages such as speed, non-destructive testing, and suitability for liquid sample detection, showing promising application prospects in the detection of endogenous precipitation in Sichuan pepper oil. However, related research has not yet been conducted. Therefore, there is an urgent need to develop a dedicated detection method based on this technology to address the practical testing needs of the Sichuan pepper oil industry. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies for detecting endogenous precipitation in Sichuan pepper oil, such as low detection efficiency, significant subjective influence, and susceptibility to matrix interference, this invention provides a method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy, which can achieve rapid and accurate detection of endogenous precipitation in Sichuan pepper oil.

[0005] To achieve the above objectives, the present invention provides the following solution: A model for analyzing endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy was established. This model can be used to quickly identify and screen Sichuan pepper oil samples containing endogenous precipitation.

[0006] Includes the following steps: S1. Spectral data of pepper oil samples with and without endogenous precipitation were collected using a visible / near-infrared spectrometer; S2. After removing abnormal samples from the spectral data of the pepper oil samples collected in step S1, preprocessing is performed to obtain the preprocessed spectral data of the pepper oil samples. S3. Extract feature variables from the spectral data of the Sichuan pepper oil samples after preprocessing in step S2 to obtain the feature variables; S4. Using the feature variables extracted in step S3 as input variables and the presence or absence of endogenous precipitation as output variables, establish a near-infrared spectral detection model and a near-infrared spectral prediction model for the presence or absence of endogenous precipitation in Sichuan pepper oil, and obtain the final near-infrared analysis model through cross-validation. S5. Take the pepper oil sample to be tested, use a visible / near-infrared spectrometer to collect the spectrum of the pepper oil to be tested, obtain the corresponding near-infrared spectrum to be analyzed, and classify the pepper oil sample to be tested by whether there is endogenous precipitation. S6. Input the visible / near-infrared spectrum of the pepper oil sample to be tested described in step S5 into the visible / near-infrared analysis model established in S4 for processing, and output whether the pepper oil sample to be tested contains endogenous precipitation; finally, compare the output result with the classification result in S5 to verify the accuracy of the model test.

[0007] Furthermore, in step S1, the scanning wavelength of the visible / near-infrared spectrometer is 340-1100 nm, and the acquisition method is diffuse reflection mode. The specific steps are as follows: first, a reference spectrum is acquired, using air as a reference, and the moving platform and blank sample cell are controlled to translate, continuously recording more than 50 reference spectra; second, sample spectra are acquired, collecting at least 200 spectra of pepper oil samples with and without endogenous precipitation.

[0008] Furthermore, in step S2, the collected spectral data is preprocessed, including Savitzky-Golay smoothing, first derivative, second derivative, multivariate scattering correction (MSC), and standard normal variable transformation (SNV). Abnormal samples in step S2 refer to non-normal samples whose near-infrared spectral data deviate from the overall sample pattern due to experimental errors, operational mistakes, environmental interference, or defects in the sample itself, which will significantly affect the accuracy and stability of modeling.

[0009] Furthermore, the feature variable selection method in step S3 is the random forest feature importance ranking method.

[0010] Furthermore, the range of the characteristic variables selected in step S3 is 490~550 nm, 660~760 nm, 850~880 nm, and 920~1005 nm.

[0011] Further, step S4 specifically involves using the spectral feature variables obtained in step S3 as input variables X, and the categories of no endogenous precipitation and with endogenous precipitation as output variables Y{1,2}. A random forest classification model (RF) is applied to divide the dataset into training, testing, and validation sets. The training set data is used to train the random forest classification model, establishing a visible / near-infrared spectral detection model for endogenous precipitation in Sichuan pepper oil. After validating the model performance and fine-tuning the parameters using the testing set, a rapid near-infrared spectral prediction model for the detection of endogenous precipitation in Sichuan pepper oil is established. The validation set is input into the visible / near-infrared spectral prediction model for cross-validation, resulting in a visible / near-infrared analysis model.

[0012] Furthermore, in step S4, the ratio of the training set, test set, and validation set is set to (7~8):(1~2):1.

[0013] Furthermore, in step S4, the grid search method is used to optimize the key parameters of the random forest classification model and set a reasonable range of candidate parameter values: the parameters include the number of decision trees and the maximum depth of decision trees, the candidate value range of the number of decision trees is 100~500, and the candidate value range of the maximum depth of decision trees is 5~30.

[0014] Furthermore, in step S5, the presence or absence of endogenous precipitates in the Sichuan pepper oil can be classified using a light inspection method. Beneficial effects

[0015] This invention provides an application for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy. It primarily addresses the shortcomings of existing techniques for detecting the presence of endogenous precipitation in Sichuan pepper oil, such as low detection efficiency, high equipment cost, and susceptibility to matrix influence. This invention enables rapid detection of the presence or absence of endogenous precipitation in Sichuan pepper oil, with the following specific benefits: 1) This invention constructs a visible / near-infrared spectral data model for detecting endogenous precipitation in Sichuan pepper oil. This model significantly improves the spectral distinction between endogenous precipitation and Sichuan pepper oil matrix, providing reliable technical support for accurate screening of endogenous precipitation.

[0016] 2) Using the above data model, this invention is the first to apply visible / near-infrared spectroscopy to the detection of endogenous precipitation in Sichuan pepper oil. It has high detection efficiency (detection time for a single sample ≤30 seconds) and strong accuracy (model prediction accuracy reaches 97%), filling the technical gap in rapid and accurate detection of endogenous precipitation in Sichuan pepper oil.

[0017] 3) To address the spectral interference caused by complex matrices such as unsaturated fatty acids and pigments in Sichuan pepper oil, this invention utilizes specific spectral preprocessing optimization combined with variable screening of characteristic wavelengths to effectively suppress matrix interference, significantly improve the detection sensitivity of low-content endogenous precipitates, and adapt to the detection needs of Sichuan pepper oils of different purities.

[0018] 4) The random forest model is used to process visible / near-infrared spectral data, which can be directly adapted to high-dimensional spectra and has stronger resistance to matrix noise. The model has excellent generalization ability, low risk of overfitting, stable classification accuracy, and can output the importance of feature wavelengths to assist in the analysis of sedimentation features. Parameter tuning is simple and more suitable for industrial online detection.

[0019] 5) This invention is simple to operate, and the pepper oil sample does not require complicated pretreatment (only direct sampling and testing is required). It does not require operators to have a strong professional background, the testing cost is low, and the supporting equipment is highly portable. It is suitable for large-scale screening in production sites and circulation links, and provides technical support for intelligent and rapid supervision of pepper oil quality and safety. Attached Figure Description

[0020] Figure 1 This is a visible / near-infrared spectrum of Sichuan pepper oil collected in an embodiment of the present invention.

[0021] Figure 2 This is a preprocessed spectral image of Sichuan pepper oil in an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of feature wavelength extraction in an embodiment of the present invention.

[0023] Figure 4 This is a comparison chart of the predicted category and the actual category of the validation set in an embodiment of the present invention. Detailed Implementation

[0024] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as a limitation of the present invention, but rather as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0025] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Any stated value or intermediate value within a stated range, as well as each smaller range between any other stated value or intermediate value within said range, is also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0026] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0027] Various modifications and variations can be made to the specific embodiments described in this specification without departing from the scope or spirit of the invention, as will be apparent to those skilled in the art. Other embodiments derived from this specification will also be apparent to those skilled in the art. This specification and embodiments are merely exemplary. Example 1:

[0028] 1. Materials and Equipment 1.1 Sources of Spectral Data Experimental samples: 50 samples of Sichuan pepper oil were provided by Yaomazi Food Co., Ltd., including 25 pure Sichuan pepper oil samples without endogenous precipitation and 25 Sichuan pepper oil samples with endogenous precipitation. 1.2 Instruments and Software Hardware equipment: Ocean Optics USB2000+ Visible / Near Infrared Spectrometer; Software platform: MATLAB R2023a.

[0029] 2. Experimental Methods and Results A. Collect near-infrared spectral data of Sichuan pepper oil samples: Forty pepper oil samples were placed on a translation stage and analyzed using a preheated (30 min) Ocean Optics USB2000+ visible / near-infrared spectrometer. The detection mode was diffuse reflectance, with a scanning wavelength of 340-1100 nm, an ambient temperature of 20-25℃, a resolution of 5 cm⁻¹, an integration time of 70 ms, and 32 scans. Air was used as a reference, and the average spectrum was taken as one visible / near-infrared spectral data set for each sample. Samples were thoroughly mixed and the instrument baseline calibrated before each acquisition. Ten data sets were collected for each sample, resulting in 400 valid visible / near-infrared spectra (each spectrum containing absorbance values ​​at 2048 wavelengths). Figure 1 As shown.

[0030] B. Preprocessing of visible / near-infrared spectral data of Sichuan pepper oil samples: The 400 spectral data points obtained in step A were organized into a feature matrix X (400×2048) and an output variable Y (400×1). Samples without endogenous precipitation were assigned an output analogy of 1, while samples containing endogenous precipitation were assigned an output class of 2. After checking for outliers and missing values, preprocessing was performed using methods including Savitzky-Golay smoothing, MSC, and SNV. Preprocessing methods reduced the influence of sample state, stray light, light scattering, instrument response, and external environmental factors, thereby reducing errors and improving model accuracy. The preprocessed spectral data are shown below. Figure 2 As shown.

[0031] C. Extract feature variables from the preprocessed spectral data: To eliminate irrelevant variables and improve model computation speed and accuracy, a variable selection method was combined with the preprocessing and optimization of the spectral data. The variable selection method was random forest feature importance ranking, selecting 30 feature wavelengths according to importance ranking, with the wavelength range as follows: Figure 3 As shown, the specific wavelengths are 499 nm, 516 nm, 542 nm, 668 nm, 710 nm, 732 nm, 755 nm, 850 nm, 862 nm, 879 nm, 920-940 nm, and 950-1001 nm.

[0032] D. Constructing a visible / near-infrared spectral analysis model for Sichuan pepper oil: The wavelengths of the feature variables obtained in step C are imported into the Random Forest (RF) classification model. The dataset is shuffled using the randperm function, and the ratio of training, test, and validation sets is set to 7.2:1.8:1 to ensure that the class distribution in each set is consistent with the original dataset. The training set contains 288 spectra (145 for class 1, 143 for class 2), the test set contains 72 spectra (36 for class 1, 36 for class 2), and the validation set contains 40 spectra (22 for class 1, 18 for class 2). During the model building phase, the top 30 most important feature wavelengths are selected to establish a fixed feature wavelength set and saved to ensure the model's repeatability and stability. Key parameters of the random forest model, including the number of decision trees (n_estimators) and the maximum depth of decision trees (max_depth), are optimized using a grid search method. A random forest classifier (with 100 decision trees) was used for training. Parameter optimization was performed using out-of-bag error estimation to improve the model's generalization ability. The optimal parameter combination was determined to be: n_estimators=100, max_depth=20. Based on the optimal parameters, a model for detecting endogenous precipitation in Sichuan pepper oil was established on the training set, and the model's performance was validated on the test set, obtaining a visible / near-infrared spectral prediction model for rapid detection of Sichuan pepper oil. The overall prediction accuracy on the test set was 0.975, the macro F1 score was 0.971, and the area under the ROC curve (ROC-AUC) was 0.991, indicating high model prediction accuracy. The absolute differences between each index and the training set were ≤0.005, indicating stable generalization performance and no overfitting.

[0033] A completely independent validation set is input into the model as the test dataset to simulate a real-world application scenario, confirm the model's stability, and plot a comparison between the predicted and actual classes on the validation set. Figure 4 According to the results, out of the 22 samples of true category 1, 22 were correctly predicted and 0 were misclassified as category 2; out of the 18 samples of true category 2, 17 were correctly predicted and 1 was misclassified as category 1, with no missed detections, thus the final near-infrared spectral analysis model was obtained.

[0034] E. Validation of the visible / near-infrared spectral analysis model for Sichuan pepper oil: To further verify the accuracy and robustness of the final model in practical applications, near-infrared spectral data of 10 pepper oil samples (independent of the training, test, and validation sets) were collected and input into the final visible / near-infrared spectral analysis model for processing and analysis, outputting results of "containing endogenous precipitation" or "no endogenous precipitation." Simultaneously, the actual precipitation status of the 10 samples was verified using a light inspection method, comparing the consistency between the model predictions and actual detection results. The results show a good correlation between the model predictions and measured values ​​for the 10 independent samples, with an overall prediction accuracy of 0.986 and a standard deviation of 0.008, indicating that the established visible / near-infrared spectral detection model can quickly and accurately identify the presence of precipitation in liquids.

[0035] Five pepper oil samples were selected for three repeated measurements. The robustness of the model was assessed by comparing the predictive consistency of the three repeated experiments. The validation results showed that the overall accuracy of the three repeated experiments was 0.958 with a standard deviation of 0.014; the recall rate of samples containing endogenous precipitation was 0.953 with a standard deviation of 0.005, indicating that the model has good robustness and the detection stability and reliability meet the requirements of practical applications. Comparative Example 1:

[0036] Similar to Example 1, the difference is that feature variables are not extracted from the preprocessed spectrum; instead, a random forest classification model is constructed directly using the feature matrix X of all 2048 wavelengths.

[0037] The test results show that the overall prediction accuracy of the test set is 0.915, the macro F1 score is 0.908, and the area under the ROC curve (ROC-AUC) is 0.951. The absolute difference between the corresponding indicators of the training set and the test set is ≥0.02, indicating poor generalization stability and obvious overfitting. Comparative Example 2:

[0038] Similar to Example 1, the difference lies in using Support Vector Machine (SVM) to construct the classification model, selecting Radial Basis Function (RBF) as the kernel function, and optimizing the penalty coefficient C (range 0.01~100) and kernel parameter γ (range 0.001~10) through grid search combined with 5-fold cross-validation to determine the optimal parameter combination with classification accuracy as the objective. Since SVM is susceptible to dimensionality issues and computationally complex with high-dimensional spectral data, the feature matrix X needs to be reduced to 50 dimensions using PCA before being input into the model for training. The optimized SVM model achieves a classification accuracy of 95.8% on the training set, but due to the loss of some intrinsic sedimentation feature information during PCA dimensionality reduction, the accuracy on the test set and validation set drops to 82.5% and 79.0%, respectively, indicating significantly insufficient generalization ability.

[0039] In summary, this embodiment achieves rapid detection of endogenous precipitation in Sichuan pepper oil by combining visible / near-infrared spectral data with a random forest classification model. The model training set, test set, and validation set are hierarchically divided in a ratio of 7.2:1.8:1. After preprocessing, feature wavelength extraction, and parameter optimization, the overall accuracy of the test set reaches 97.5%, and the validation set exhibits stable performance and good robustness. The detection time for a single sample is ≤30 seconds, requiring no complex preprocessing, and can meet the large-scale screening needs of Sichuan pepper oil production sites and distribution channels.

[0040] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An application of visible / near-infrared spectroscopy for detecting endogenous precipitation in Sichuan pepper oil, characterized in that, Includes the following steps: S1. Spectral data of pepper oil samples with and without endogenous precipitation were collected using a visible / near-infrared spectrometer; S2. After removing abnormal samples from the spectral data of the pepper oil samples collected in step S1, preprocessing is performed to obtain the preprocessed spectral data of the pepper oil samples. S3. Extract feature variables from the spectral data of the Sichuan pepper oil samples after preprocessing in step S2, and obtain the feature variables; S4. Using the feature variables extracted in step S3 as input variables and the presence or absence of endogenous precipitation as output variables, establish a near-infrared spectral detection model and a near-infrared spectral prediction model for the presence or absence of endogenous precipitation in Sichuan pepper oil, and obtain the final near-infrared analysis model through cross-validation. S5. Take the pepper oil sample to be tested, use a visible / near-infrared spectrometer to collect the spectrum of the pepper oil to be tested, obtain the corresponding near-infrared spectrum to be analyzed, and classify the pepper oil sample to be tested by whether there is endogenous precipitation. S6. Input the visible / near-infrared spectrum of the pepper oil sample to be tested described in step S5 into the visible / near-infrared analysis model established in S4 for processing, and output whether the pepper oil sample to be tested contains endogenous precipitation; finally, compare the output result with the classification result in S5 to verify the accuracy of the model test.

2. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, characterized in that, In step S1, the scanning wavelength of the visible / near-infrared spectrometer is 340-1100 nm, and the acquisition method is diffuse reflectance mode. The specific steps are as follows: First, a reference spectrum is acquired, using air as a reference, and the moving platform and blank sample cell are controlled to translate, continuously recording more than 50 reference spectra; second, sample spectra are acquired, collecting at least 200 spectra of pepper oil samples with and without endogenous precipitation.

3. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, characterized in that, In step S2, the collected spectral data is preprocessed, including Savitzky-Golay smoothing, first derivative, second derivative, multivariate scattering correction (MSC), and standard normal variable transformation (SNV). Abnormal samples in step S2 refer to abnormal samples whose near-infrared spectral data deviate from the overall sample pattern due to experimental errors, operational mistakes, environmental interference, or defects in the sample itself, which will significantly affect the accuracy and stability of modeling.

4. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, characterized in that, The feature variable selection method in step S3 is the random forest feature importance ranking method.

5. The use of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, wherein the range of the characteristic variables selected in step S3 is 490~550 nm, 660~760 nm, 850~880 nm, and 920~1005 nm.

6. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, characterized in that, The specific steps of step S4 are as follows: using the spectral feature variables obtained in step S3 as input variables X, and the categories of no endogenous precipitation and with endogenous precipitation as output variables Y{1,2}, the dataset is divided into training set, test set and validation set using a random forest classification model (RF). The random forest classification model is trained using the training set data to establish a visible / near-infrared spectral detection model for endogenous precipitation in Sichuan pepper oil. After verifying the model performance and optimizing the parameters using the test set, a rapid near-infrared spectral prediction model for the detection of endogenous precipitation in Sichuan pepper oil is established. The validation set is input into the visible / near-infrared spectral prediction model for cross-validation to obtain the visible / near-infrared analysis model.

7. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 6, characterized in that, The ratio of the training set, test set, and validation set is set to (7~8):(1~2):

1.

8. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, characterized in that, In step S4, the grid search method is used to optimize the key parameters of the random forest classification model and set a reasonable range of candidate parameter values: the parameters include the number of decision trees and the maximum depth of decision trees, the candidate value range of the number of decision trees is 100~500, and the candidate value range of the maximum depth of decision trees is 5~30.

9. The application of the method for detecting endogenous precipitation in Sichuan pepper oil based on visible / near-infrared spectroscopy according to claim 1, characterized in that, In step S5, the presence of endogenous precipitates in the Sichuan pepper oil can be classified using a light inspection method.

Images