A method for improving the precision of a near-infrared data detection solid-state composite seasoning deliciousness degree model
By performing SG smoothing and first derivative preprocessing on near-infrared spectral data, and combining principal component analysis and partial least squares dimensionality reduction, an information-enhanced umami detection model was established. This solved the problem of low model accuracy in existing technologies and enabled rapid and accurate evaluation of the umami of solid compound seasonings.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI TOTOLE FOOD LTD
- Filing Date
- 2023-07-14
- Publication Date
- 2026-07-14
AI Technical Summary
Existing near-infrared spectroscopy technology has low accuracy when establishing a rapid detection model for the umami of solid compound seasonings, possibly because relying on only a single dimensionality reduction algorithm results in insufficient extraction of near-infrared data features.
SG smoothing and first derivative preprocessing are used to increase the signal-to-noise ratio of spectral data. Principal component analysis and partial least squares method are combined for dimensionality reduction. The first 6 principal component factors and partial least squares factors after dimensionality reduction are selected as independent variables. An information-enhanced detection model is established using an artificial neural network algorithm.
It improves the detection accuracy and efficiency of the model, reduces subjective errors, achieves rapid and accurate freshness evaluation, conforms to the concept of green environmental protection, and is simple to operate.
Smart Images
Figure CN117033895B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of near-infrared spectral data processing, and in particular to a method for establishing a model to improve the accuracy of near-infrared data detection of the umami of solid compound seasonings. Background Technology
[0002] Near-infrared spectroscopy is a green analytical technique that utilizes the optical characteristics of organic substances in the near-infrared spectral region to rapidly detect the properties and content of one or more chemical components within the organic compound. It offers advantages such as convenience, speed, efficiency, accuracy, and zero pollution; no sample pretreatment is required and the sample is not damaged; and it does not consume chemical reagents. In recent years, it has been widely applied in many areas, including food composition determination, food production control, food safety testing, and food sensory analysis.
[0003] With the continuous improvement of living standards, people's requirements for seasonings have gradually evolved from single-ingredient to multi-ingredient, and from low-end to mid-to-high-end. Solid compound seasoning products with functions such as flavoring, umami enhancement, and aroma addition are becoming increasingly popular in the market. For solid compound seasonings, the unique umami flavor plays a decisive role in their taste quality. Currently, the evaluation of the umami flavor of solid compound seasonings mainly relies on professional sensory evaluation methods conducted by trained and experienced evaluators. This method is time-consuming, difficult, and demanding, and the evaluation results are easily affected by many subjective factors such as environment, physical condition, and mood, often resulting in significant errors. Therefore, exploring a rapid, accurate, and comprehensive evaluation method for umami flavor is of great significance for evaluating the umami flavor of solid compound seasonings.
[0004] When applying near-infrared spectroscopy to the sensory evaluation of the umami flavor of solid compound seasonings, the near-infrared spectrum, composed of over 2000 data points, requires dimensionality reduction to enhance the reliability and accuracy of the model. Various dimensionality reduction methods exist, such as principal component analysis, partial least squares analysis, and elimination of non-information variables. While these algorithms provide excellent technical support for building rapid detection models based on near-infrared spectral data, they still exhibit relatively low accuracy when used to build rapid umami flavor detection models for solid compound seasonings. This may be due to insufficient feature extraction from near-infrared data using these dimensionality reduction algorithms alone. Therefore, this patent proposes a method for enhancing the information after near-infrared spectral dimensionality reduction, thereby increasing the reliability and accuracy of the rapid umami flavor prediction model. Summary of the Invention
[0005] The purpose of this invention is to propose a method that enhances the reliability and accuracy of a rapid prediction model for the umami flavor of solid compound seasonings by combining near-infrared spectral data with dimensionality reduction information.
[0006] To achieve the above objectives, this invention proposes a method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings, comprising the following steps:
[0007] Step 1: Collect various solid compound seasonings from the market as samples and obtain the near-infrared spectral data of the samples;
[0008] Step 2: Conduct manual evaluation of the freshness of various samples, and use the average data of the manual evaluation as the dependent variable data for establishing the freshness detection model;
[0009] Step 3: Perform SG smoothing and first derivative preprocessing on the near-infrared spectral data to increase the signal-to-noise ratio of the near-infrared spectral data; then perform principal component analysis and partial least squares dimensionality reduction on the preprocessed spectral data.
[0010] Step 4: Select the first 6 principal component factors and partial least squares factors after dimensionality reduction to form 12 new information-enhanced independent variables. Combine these with the dependent variable data and use an artificial neural network algorithm to establish a freshness detection model.
[0011] Step 5: Perform accuracy testing on the freshness detection model.
[0012] Furthermore, in step 1, after collecting solid seasonings from the market, grinding them, and then scanning them with a near-infrared instrument, the near-infrared spectral data is obtained.
[0013] Furthermore, in step 2, multiple people conduct manual evaluations of the freshness of various samples, and the average value of the evaluation data from different evaluators is used as the dependent variable data for establishing the freshness detection model.
[0014] Furthermore, in step 5, the accuracy of the freshness detection model is detected by comparing it with a control model.
[0015] Furthermore, the control model includes a first control model and a second control model;
[0016] Furthermore, the first control model uses the first 12 principal component factors after dimensionality reduction as independent variables and the average freshness data as dependent variables, and is established using an artificial neural network algorithm.
[0017] The second control model uses the first 12 partial least squares factors after dimensionality reduction as independent variables and the average freshness data as dependent variables, and is established using an artificial neural network algorithm.
[0018] Furthermore, in step 5, the accuracy detection steps for the freshness detection model are as follows:
[0019] Step 5.1: Collect 10 new solid compound seasonings from the market and obtain near-infrared spectral data and umami data:
[0020] Step 5.2: Perform principal component analysis and partial least squares dimensionality reduction on the near-infrared data from Step 5.1, respectively;
[0021] Step 5.3: Select the first 12 principal component factors and substitute them into the first control model to predict the umami data of the newly collected 10 solid compound seasonings;
[0022] The first 12 partial least squares factors were selected and substituted into the second control model to predict the umami data of 10 newly collected solid compound seasonings.
[0023] The first 6 principal component factors and partial least squares factors after dimensionality reduction were selected to form 12 information-enhanced independent variables. These variables were then substituted into the umami detection model to predict the umami data of 10 newly collected solid compound seasonings.
[0024] Step 5.4: Professional sensory evaluators conduct manual evaluations of the umami of the 10 newly collected solid compound seasonings and average the umami scores obtained by different evaluators.
[0025] Step 5.5: Calculate the relative errors between the freshness values predicted by the first control model, the second control model, and the freshness detection model and the human evaluation data, and compare the three models to see the information enhancement effect.
[0026] Compared with the prior art, the advantages of the present invention are:
[0027] 1. This invention performs SG smoothing and first-derivative preprocessing on the spectral data of solid compound seasonings to increase the signal-to-noise ratio of near-infrared spectral data. Then, principal component analysis and partial least squares dimensionality reduction are performed on the preprocessed spectral data to overcome the shortcomings of a single dimensionality reduction algorithm in extracting feature information.
[0028] 2. This invention selects the first 6 principal component factors and partial least squares factors after dimensionality reduction to form 12 new information-enhanced independent variables. The average freshness data is used as the dependent variable to establish a freshness detection model, which effectively improves the detection accuracy and efficiency of the model.
[0029] 3. The method of the present invention does not involve experiments or use chemicals in the entire detection process, does not generate chemical pollution, and conforms to the concept of green environmental protection; moreover, it is simple to operate, low in cost, easy to implement, and suitable for widespread application. Attached Figure Description
[0030] Figure 1 This is the near-infrared spectrum of the solid compound seasoning in step 1 of this embodiment of the invention.
[0031] Figure 2 This is a diagram showing the results of dimensionality reduction using artificial neural networks based on principal component analysis in an embodiment of the present invention.
[0032] Figure 3 This is a diagram showing the results of artificial neural networks based on partial least squares dimensionality reduction in an embodiment of the present invention.
[0033] Figure 4 This is the modeling result of the freshness quantitative forecasting model established based on information enhancement in the embodiments of the present invention. Detailed Implementation
[0034] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below.
[0035] This invention proposes a method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings, comprising the following steps:
[0036] Step 1: 81 solid compound seasoning samples were collected from the market, ground, and then scanned using a near-infrared spectral instrument to obtain the near-infrared spectral data of the samples. Professional sensory evaluators manually evaluated the umami of the collected solid compound seasonings, and the umami scores obtained from different evaluators were averaged to increase the objectivity of the umami data. Example data of near-infrared spectra and umami scores are shown in Table 1.
[0037] Table 1. Sample data from near-infrared spectral data
[0038] sample Deliciousness 10000 9997 9994 9991 9988 …… Sample 1 58.78 0.77095 0.7711 0.77195 0.77235 0.7731 …… Sample 2 55.36 0.85325 0.85405 0.85595 0.8575 0.85755 …… Sample 3 59.6 0.91435 0.914 0.91515 0.9151 0.9137 …… Sample 4 65.42 0.93535 0.93455 0.9342 0.9362 0.9368 …… Sample 5 65.15 0.9041 0.90295 0.9057 0.9065 0.90635 …… Sample 6 65.28 0.6809 0.68185 0.68075 0.6818 0.681 …… Sample 7 57.07 0.92595 0.92565 0.92715 0.9276 0.92775 …… Sample 8 66.44 0.7546 0.7538 0.75475 0.7553 0.7541 …… Sample 9 57.72 0.91295 0.9129 0.9129 0.91495 0.91515 …… Sample 10 53.21 1.02975 1.02835 1.029 1.0287 1.02885 …… Sample 11 63.1 1.02135 1.02235 1.023 1.0243 1.0237 …… …… …… …… …… …… …… …… ……
[0039] Step 2: SG smoothing and first-derivative preprocessing were performed on the spectral data of the solid compound seasoning to increase the signal-to-noise ratio of the near-infrared spectral data. Then, principal component analysis and partial least squares dimensionality reduction were performed on the preprocessed spectral data. Table 2-4 shows some example data of the first 12 principal component analysis factors, the first 12 partial least squares factors, and the enhanced data.
[0040] Table 2. Sample data for the first 12 principal component analysis factors
[0041]
[0042]
[0043] Table 3. Sample data for the first 12 partial least squares factors and augmented data.
[0044]
[0045] Table 4. Some example data for augmented data are shown in the table.
[0046]
[0047]
[0048] Step 3: Using the top 12 principal component factors after dimensionality reduction as independent variables and the average freshness data as the dependent variable, a fast freshness detection model 1 is established using an artificial neural network algorithm. Using the top 12 partial least squares factors after dimensionality reduction as independent variables and the average freshness data as the dependent variable, a fast freshness detection model 2 is established using an artificial neural network algorithm. From the principles of principal component analysis and partial least squares dimensionality reduction, we know that factors at the beginning contain more data information. Therefore, we select the top 6 principal component factors and partial least squares factors after dimensionality reduction to form 12 new information-enhanced independent variables, using the average freshness data as the dependent variable, and establish a fast freshness detection model 3 using an artificial neural network algorithm. The parameters of the artificial neural network algorithm are: 12 nodes in the input layer, 7 in the hidden layer, and 1 in the output layer. The learning efficiency from the input layer to the hidden layer is 0.6, and the learning efficiency from the hidden layer to the output layer is 0.5. The momentum term is 0.4, and the number of training iterations is 250,000.
[0049] Step 4: Collect 10 new solid compound seasonings from the market and obtain near-infrared spectral data and umami data. Perform principal component analysis and partial least squares dimensionality reduction on the near-infrared data. Select the first 12 principal component factors and substitute them into Model 1 to predict the umami data of the 10 newly collected solid compound seasonings. Select the first 12 partial least squares factors and substitute them into Model 2 to predict the umami data of the 10 newly collected solid compound seasonings. Select the first 6 principal component factors and partial least squares factors after dimensionality reduction to form 12 information-enhanced independent variables, substitute them into Model 3, and predict the umami data of the 10 newly collected solid compound seasonings.
[0050] Step 5: Professional sensory evaluators conducted manual evaluations of the umami flavor of the 10 newly collected solid compound seasonings, and averaged the umami flavor scores obtained by different evaluators. The relative errors between the umami flavor values predicted by Model 1, Model 2, and Model 3 and the manual evaluation data were calculated to assess the effectiveness of the information enhancement. The error data results of the three models compared using different solid compound seasoning samples are as follows:
[0051] Example 1:
[0052] This embodiment uses near-infrared spectral data from 81 solid compound seasoning samples, extracts features using principal component analysis dimensionality reduction, and models the results of a rapid umami detection model 1 based on the first 12 principal component factors combined with an artificial neural network algorithm. (The model is shown in the image.) Figure 2 As shown.
[0053] This embodiment uses near-infrared spectral data from 81 solid compound seasoning samples, extracts features using principal component analysis (PCA) dimensionality reduction, and establishes a rapid umami detection model based on the first 12 principal component factors combined with an artificial neural network algorithm. The Pearson correlation coefficient (R) between the predicted values of the rapid umami detection model and the evaluation data of professional sensory evaluators is 0.95, with an average relative error of 1.98%.
[0054] Model 1 was used to predict the umami of 10 new solid compound seasonings collected from the market. The results are shown in Table 5.
[0055] Table 5. Prediction results of principal component dimensionality reduction method
[0056]
[0057] Example 2:
[0058] This embodiment uses near-infrared spectral data from 81 solid compound seasoning samples, extracts features using partial least squares dimensionality reduction, and models the results of a rapid umami detection model 2 based on the first 12 partial least squares factors combined with an artificial neural network algorithm. (The model is shown in the image.) Figure 3 As shown.
[0059] This embodiment uses near-infrared spectral data from 81 solid compound seasoning samples, extracts features using partial least squares dimensionality reduction, and establishes a rapid umami detection model based on the first 12 partial least squares factors combined with an artificial neural network algorithm. The Pearson correlation coefficient (R) between the predicted value of the rapid umami detection model and the evaluation data of professional sensory evaluators is 0.97, and the average relative error is 1.41%.
[0060] Model 2 was used to predict the umami of 10 new solid compound seasonings collected from the market. The results are shown in Table 6.
[0061] Table 6. Prediction results of the partial least squares dimensionality reduction method
[0062]
[0063] Example 3:
[0064] This embodiment uses near-infrared spectral data from 81 solid compound seasoning samples. Features are extracted using principal component analysis and partial least squares dimensionality reduction methods. Based on the modeling results of a new information-enhanced model (Model 3) for rapid umami detection, constructed using the first six principal component factors and partial least squares factors after dimensionality reduction, and combined with an artificial neural network algorithm, the model consists of 12 independent variables. Figure 4 As shown.
[0065] This embodiment uses near-infrared spectral data from 81 solid compound seasoning samples. Principal component analysis and partial least squares (PLS) dimensionality reduction methods were used to extract features. Based on the first six principal component factors and PLS factors selected after dimensionality reduction, a new information-enhanced model of 12 independent variables was constructed and established using an artificial neural network algorithm. The Pearson correlation coefficient (R) between the predicted values of the rapid umami detection model and the evaluation data of professional sensory evaluators was 0.92, with a mean relative error of 2.39%.
[0066] Model 3 was used to predict the umami of 10 new solid compound seasonings collected from the market. The results are shown in Table 7.
[0067] Table 7 Forecast Results of Information Enhancement Methods
[0068]
[0069] A comparison of the results from the three methods shows that although principal component analysis (PCA) and partial least squares (PLS) dimensionality reduction yielded better modeling results, their average relative errors in prediction were significantly higher than those of the information enhancement method. This indicates that the information enhancement method, which uses near-infrared spectral data for dimensionality reduction, can avoid some overfitting issues in the modeling process, resulting in more accurate predictions and better model reliability and accuracy. The results are shown in Table 8.
[0070] Table 8
[0071]
[0072] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.
Claims
1. A method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings, characterized in that, Includes the following steps: Step 1: Collect various solid compound seasonings from the market as samples and obtain the near-infrared spectral data of the samples; Step 2: Conduct manual evaluation of the freshness of various samples, and use the average data of the manual evaluation as the dependent variable data for establishing the freshness detection model; Step 3: Perform SG smoothing and first derivative preprocessing on the near-infrared spectral data to increase the signal-to-noise ratio of the near-infrared spectral data; then perform principal component analysis and partial least squares dimensionality reduction on the preprocessed spectral data. Step 4: Select the first 6 principal component factors and the first 6 partial least squares factors after dimensionality reduction to form a new set of 12 information-enhanced independent variables. Combine these with the dependent variable data and use an artificial neural network algorithm to establish a freshness detection model. Step 5: Perform accuracy testing on the freshness detection model.
2. The method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings according to claim 1, characterized in that, In step 1, solid seasonings are collected from the market, ground, and then scanned using a near-infrared instrument to obtain the near-infrared spectral data.
3. The method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings according to claim 1, characterized in that, In step 2, multiple people conduct manual evaluations of the freshness of various samples, and the average value of the evaluation data from different evaluators is used as the dependent variable data for establishing the freshness detection model.
4. The method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings according to claim 1, characterized in that, In step 5, the accuracy of the freshness detection model is tested by comparing it with a control model.
5. The method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings according to claim 4, characterized in that, The control model includes a first control model and a second control model.
6. The method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings according to claim 5, characterized in that, The first control model uses the first 12 principal component factors after dimensionality reduction as independent variables and the average freshness data as dependent variables, and is established using an artificial neural network algorithm. The second control model uses the first 12 partial least squares factors after dimensionality reduction as independent variables and the average freshness data as dependent variables, and is established using an artificial neural network algorithm.
7. The method for establishing a model to improve the accuracy of near-infrared data detection of the umami flavor of solid compound seasonings according to claim 6, characterized in that, In step 5, the accuracy of the freshness detection model is tested, and the steps are as follows: Step 5.1: Collect 10 new solid compound seasonings from the market and obtain near-infrared spectral data and umami data; Step 5.2: Perform principal component analysis and partial least squares dimensionality reduction on the near-infrared data from Step 5.1, respectively; Step 5.3: Select the first 12 principal component factors and substitute them into the first control model to predict the umami data of the newly collected 10 solid compound seasonings; The first 12 partial least squares factors were selected and substituted into the second control model to predict the umami data of 10 newly collected solid compound seasonings. The first 6 principal component factors and partial least squares factors after dimensionality reduction were selected to form 12 information-enhanced independent variables. These variables were then substituted into the umami detection model to predict the umami data of 10 newly collected solid compound seasonings. Step 5.4: Professional sensory evaluators conduct manual evaluations of the umami of the 10 newly collected solid compound seasonings and average the umami scores obtained by different evaluators. Step 5.5: Calculate the relative errors between the freshness values predicted by the first control model, the second control model, and the freshness detection model and the human evaluation data, and compare the three models to see the information enhancement effect.