CN114813593A - Method for detecting total acid content of fermented grains based on hyperspectral imaging technology - Google Patents

Abstract

The invention provides a method for detecting the total acid content in grains of grains based on hyperspectral imaging technology. The method includes: collecting grains of grains samples, measuring the total acid content of grains grains; Obtain the average spectral information and color information of the region of interest; perform SNV‑SG preprocessing on the spectral information; use RC‑SPA to filter the characteristic wavelengths of the spectrum, and obtain the hyperspectral eigenvalues of each sample; fuse the eigenvalues and color information The data and the total acid content are used to construct a regression model; the hyperspectral eigenvalues of the sample to be tested are used as the input of the regression model to obtain the total acid content of the grains to be tested; the total acid content in the ROI of the grains is visually analyzed. The present invention can accurately detect the total acid content in the grain grains through the characteristics of the hyperspectral spectrum, and realize online detection.

Description

Detection method of total acid content in fermented grains based on hyperspectral imaging technology

technical field

The invention relates to a method for detecting the total acid content of fermented grains based on hyperspectral imaging technology, and belongs to the technical field of solid-state fermentation index detection in solid-state brewing.

Background technique

In the liquor brewing industry, glutinous rice grains are the raw material for liquor distillation and the direct source of flavor compounds. Among them, acid is the formation of strong-flavor liquor, and suitable acidity is conducive to the improvement of liquor saccharification and fermentation and yield. If the acidity is not enough, the flavor of liquor will not be strong enough and the taste is monotonous; but the acidity is too high, which will inhibit beneficial microorganisms ( Mainly for the growth and reproduction of yeast), which affects the liquor yield and the liquor taste is not good. At present, in the production process of traditional liquor, when judging the quality of fermented grains and the quality of fermented grains, they all rely on the sensory evaluation of the workers or the conventional physical and chemical measurement methods, which are cumbersome to operate and have low quality inspection efficiency.

For example, Chinese patent CN104007113B, the name of the invention is the detection method for the acidity of fermented grains, which records that the steps of the detection method of fermented grains acidity include: preparing a fermented grains solution, adding phenolphthalein, measuring the initial red light intensity value X with a color sensor, The quadratic function curve of the red light intensity of the cool solution Y=-9.67879*10 ^-4 *X ² +0.70671*X-17.37787 Calculate Y, according to M=XY, to calculate the red light of the cool solution when the titration end point is reached Intensity value M; add alkaline solution dropwise, mix well, detect the red light intensity value N of the solution, when N is less than M, stop the titration; calculate the acidity value of fermented grains by consuming the amount of alkaline solution. The patented method can accurately detect the acidity value of glutinous rice grains of various flavor types such as strong flavor, Maotai flavor and Qing flavor. However, the method for detecting the acidity content of fermented grains described in the patent is destructive, and the sample cannot be used again after the sample is detected. If the sample size is large, it will cause a waste of resources. In addition, the preferred soaking time is 25-40min, and the detection period is longer, in order to completely dissolve the acid substances of the grains in the water. In addition, the method for detecting the acidity content of the fermented grains described in the patent cannot realize on-line monitoring and visualize the acidity of the fermented grains samples.

In addition, the patent number CN111539920A, the title of the invention is an automatic detection of the quality of glutinous rice grains in a liquor brewing process, which also records a kind of automatic detection method for the quality of glutinous rice grains in the liquor brewing process involving the field of liquor brewing and image processing. The method includes the following steps: A. Establishing a bad grain quality evaluation model based on an adaptive fuzzy inference algorithm; B. Training the bad grain quality evaluation model based on bad grain samples with known quality classification and corresponding classification results; C. Using the post-training method The quality evaluation model of glutinous rice grains is used to detect the glutinous rice grains to be tested. The invention is suitable for the automatic detection and evaluation of the quality of the grains in the grain-mixing link in the fully automated liquor brewing process. However, the quality detection method for fermented grains described in this patent only takes image information (color, texture and other features) as input, and the image information can only reflect the surface information of fermented grains, but cannot express changes in the internal structure, so the detection accuracy will be relatively lower. And I don't think it seems correct to monitor grain quality from a texture point of view, since grains are sticky and granular, if you flip a batch of samples after they've been imaged, the batch will be Obtaining two very different texture information may lead to very different detection results of the same batch of samples.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for detecting the total acid content of glutinous rice grains based on hyperspectral imaging technology, aiming at the technical problems existing in the prior art. In this method, hyperspectral imaging technology is used, traditional imaging and spectral technology are combined, the spatial and spectral information of the sample is obtained at the same time, and the preferred characteristic spectral information is used to fuse the color information to improve the detection speed and detection accuracy of the model. The SPXY algorithm is used to divide the data set, the SNV combined with the SG algorithm is used for the preprocessing, and the RC-SPA algorithm is used to filter the characteristic wavelengths. , the hyperspectral characteristic value of the sample to be tested is used as the input of the regression model, the total acid content of the fermented grains to be tested is obtained, and the total acid content in the ROI of the fermented grains is visually analyzed, and the present invention can accurately detect the total acid content in the fermented grains The acid content overcomes the shortcomings of traditional manual identification, such as being slow and subject to subjective influence, and provides technical support for the transformation and upgrading of the industrialization of liquor brewing and the intelligent online monitoring of the fermentation state of fermented grains.

In order to realize the above purpose of the invention, the specific technical scheme of the present invention is:

The detection method for the total acid content of glutinous rice grains based on hyperspectral imaging technology includes the following steps:

1) Acquisition of sample hyperspectral images and correction of the collected images: Collect glutinous grains samples from different pits and different layers in the winery, and use the hyperspectral imaging system to collect the hyperspectral images of the samples;

2) According to GB/T12456-2021, measure the total acid content of glutinous rice grains;

3) Obtain the average spectral information and color information of the region of interest;

4) Data processing: preprocess the original spectrum, and screen out the characteristic wavelengths related to the total acid content;

5) Integrate the characteristic wavelength and image color information with the measured total acid content to establish a regression model, and evaluate the established model to determine the validity of the model;

6) Using the characteristic spectral variable to be measured as the input of the regression model to obtain the result of the total acid content of the fermented grains to be measured.

As a preferred embodiment of the present application, in step 1), a total of 128 grains of grains samples were collected from different cellars and different layers in the winery. The samples are usually collected in two different periods. For example, for the first time, we randomly collected samples from the upper, middle and lower layers of the 13 cellars in the winery at the same time (March 28). Three sampling points were randomly selected by a rotary sampler for collection, in which the depth of the pit is 2m, the sampling point of the upper layer is 0.6m away from the level pit of the pit, the sampling point of the middle layer is 1.1m away from the flat pit of the pit, and the sampling point of the lower layer is away from the level of the pit. 2m from the cellar, a total of 117 samples were obtained. During the second collection (March 29), 11 pits were randomly selected from the 13 pits collected for the first time for sampling, and each pit was randomly selected to obtain 11 samples. In the end, a total of 128 samples of glutinous rice were obtained.

The hyperspectral imaging system is mainly composed of a hyperspectral camera, an illumination system, a high-precision electronically controlled stage and a computer equipped with special processing software. The collected spectral range is 940-1730 nm, with a total of 224 bands and a spectral resolution of 3.3 nm. The exposure time of the hyperspectral camera was set to 4.02ms, the acquisition frequency was 50Hz, and the moving speed of the stage was 16.42mm/s. The 3D hyperspectral image data blocks of the grain samples were obtained by the linear push-broom method. Calibration of hyperspectral images and selection of regions of interest.

As a preferred embodiment of the present application, the acquisition of the color information in step 3) mainly adopts the first-order moment, second-order moment and third-order moment in the H, S, and V color spaces of the grayscale image of the glutinous rice sample ROI. moments as features. One sample has 9 features, and 128 samples obtain 128×9 color feature data in total.

As a preferred embodiment in this application, the preprocessing method described in step 4) is standard normal transformation combined with SG convolution smoothing (SNV-SG); the feature wavelength extraction method is regression coefficient method combined with continuous projection Algorithm (RC-SPA).

As a preferred embodiment in this application, the regression model established in step 5) is a cascade forest (CF) model; R ² and RMSE are used as evaluation indicators to determine the validity of the model.

As a preferred embodiment in the present application, step 6) includes step 7) after "using the characteristic spectral variable to be measured as the input of the regression model to obtain the result of the total acid content of the fermented grains to be measured": the obtained The total acid content was visualized, and the acidity distribution map of the grains was obtained.

Further, the method for calibrating the collected images is as follows: first, place a standard PTFE whiteboard in the imaging area, and collect a calibration image of the standard whiteboard with full white reflection; then cover the lens cover, turn off the light source, and obtain a full black calibration image; finally, the corrected hyperspectral image is calculated according to the following formula.

Where: R is the corrected image; I is the original image; W is the standard whiteboard image; B is the dark current image with the lens turned off.

Further, the selection of the region of interest specifically adopts morphological processing and binary norm calculation, with the center of the collected glutinous rice sample image as the center of the circle and 100 pixels as the radius to divide a circular area as the ROI of the sample.

The above-mentioned method is used for the detection of the total acid content in the grains.

Compared with the prior art, the present invention has the following beneficial effects:

(1) The method for detecting the total acid content of fermented grains based on the hyperspectral imaging technology disclosed in the present invention utilizes the preferred characteristic spectrum information to fuse color information for the first time to improve the detection accuracy of the model. The detection method provided by the present invention is simple to operate and does not have damage It overcomes the shortcomings of artificial sensory evaluation method to detect the quality of glutinous rice grains, which takes a long time and is affected by subjective consciousness. Moreover, the established rapid non-destructive testing method for total acid content in glutinous rice grains based on hyperspectral technology and deep learning has greatly improved the detection performance. Efficiency provides theoretical support and technical support for the intelligent development of solid-state brewing.

(2) In the technical field, the present invention adopts hyperspectral imaging technology to detect the total acid content of glutinous rice grains, and utilizes the unique advantages of hyperspectral imaging technology, and uses each pixel in the sample ROI as the input of the optimal detection model , combined with pseudocolor data processing to visualize the distribution of total acid content across all pixels. The change of the corresponding content of each pixel is represented by the change of color, and the distribution of the total acid content of glutinous rice grains in the sample ROI is intuitively reflected.

(3) In terms of data input, the present invention utilizes the preferred characteristic spectrum information to fuse color information to improve the detection accuracy of the model. In the collected hyperspectral images, the image information can reflect the external characteristics of the fermented grains. Because the color of the fermented grains in different fermentation states is different, the color moments of the images in the HSV space are very different; The information can reflect the internal components and structural characteristics of grain grains, which will lead to great differences in the spectral reflectance of grain grains in different fermentation periods at different wavelengths. Compared with a single data set, the fusion of image and spectral information contains more comprehensive feature information that can clearly distinguish the fermentation of fermented grains, and can achieve the purpose of accurate detection.

(4) In data processing, the present invention adopts SNV in conjunction with SG to preprocess the spectrum, removes noise more, and enhances the information related to the composition; Effectively retains the wavelength with the highest correlation with the total acid content and eliminates the problem of collinearity between wavelengths, greatly reducing the number of wavelengths and achieving the purpose of rapid detection.

(5) In the aspect of model establishment, the present invention develops a deep learning model-CF to detect the total acid content of fermented grains. The CF model has fewer hyperparameter settings, is more efficient, and is capable of performing representation learning, showing excellent performance even with only small-scale training data. The model can obtain more accurate and stable detection results than conventional machine learning algorithms.

(6) The present invention proposes a non-destructive, rapid and accurate method for detecting the total acid content of fermented grains. The method is helpful for monitoring the fermentation of glutinous rice grains, and has guiding significance for the timely adjustment of process parameters in the brewing process of liquor, and provides an alternative method for traditional detection methods.

Description of drawings

Fig. 1 is the schematic flow sheet of the detection method of the total acid content of glutinous rice grains based on hyperspectral imaging technology in the present invention;

Fig. 2 is the selection flow chart of hyperspectral ROI in the present invention;

Fig. 3 is the HSV map of a sample in the embodiment 1 of the present invention;

Fig. 4 is the spectral reflectance image of fermented grains sample of the present invention;

Fig. 5 is a visualized cloud image of grains of grains with different total acid contents of the present invention.

Detailed ways

The embodiments of the present invention are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments may be combined with each other under the condition of no conflict.

It should be noted that, in order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention. , not all examples.

Thus, the following detailed description of the embodiments of the invention is not intended to limit the scope of the invention as claimed, but rather to represent only selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The raw materials, equipment and methods used in the present invention, unless otherwise specified, are the common raw materials, equipment or methods in the field.

Example 1:

The detection method for the total acid content of glutinous rice grains based on hyperspectral imaging technology includes the following steps:

1. Collection of glutinous rice samples

The glutinous grains samples collected in this example were taken from a winery in Yibin, Sichuan, China. This collection is divided into two times. For the first collection, we randomly collected the upper, middle and lower glutinous grain samples of 13 cellars in the winery at the same time (March 28). Type sampler randomly selects three sampling points for collection, of which the depth of the pit is 2m, the upper sampling point is 0.6m away from the pit level cellar, the middle level sampling point is 1.1m away from the pit pit level cellar, and the lower layer sampling point is 2m away from the pit pit level cellar. , a total of 117 samples were obtained. During the second collection (March 29), 11 pits were randomly selected from the 13 pits collected for the first time for sampling, and each pit was randomly selected to obtain 11 samples. In the end, a total of 128 samples of glutinous rice were obtained.

2. Acquisition and correction of hyperspectral images

A push-broom hyperspectral imaging system in the laboratory was used to obtain hyperspectral images of fermented grains. The system is mainly composed of a hyperspectral camera (Finnish FX17 series), an illumination system (OSRAM, German), a high-precision electronic control platform and a computer equipped with special processing software (LUMO-scanner). The collected spectral range is 940-1720 nm with a spectral resolution of 3.4 nm, with a total of 224 bands. The exposure time of the hyperspectral camera was set to 4.02 ms, the acquisition frequency was 50 Hz, and the moving speed of the stage was 16.42 mm/s.

The collected spectral image is corrected, and the formula for black and white correction is shown in formula (1).

where R is the corrected reflectance image; I is the original hyperspectral image; W is the standard whiteboard image; D is the dark current image with the lens off.

In order to improve the modeling accuracy, the present invention adopts morphological processing and binary norm calculation to remove the background of the hyperspectral image, and divides a circular region of interest ( ROI), the specific process is shown in Figure 2.

3. Determination of total acid content

Determination of total acid content of fermented grains (GB/T12456-2021) is mainly based on the principle of acid-base neutralization, and is determined by potentiometric titration with PH indicating the end point. First, weigh the sample with an electronic balance, the accuracy is accurate to 0.001g, set the volume to a polyplug measuring cylinder, let it stand for 30 minutes and shake it 2-3 times, filter it and mix it with distilled water, and titrate to pH with 0.1mol/L NaOH standard solution A reading of 8.2 is indicated, the titration is stopped and the reading is recorded. The formula for calculating the total acid content of the sample is shown in formula (2):

X=c×V×100×(100/20)×(1/10) (2)

In the formula: X is the total acid content of the fermented grains sample (mmol/10g); c is the concentration of NaOH standard solution (mol/L); V is the volume of NaOH consumed by the sample solution (mL); 20 is the volume of absorbed filtrate ( mL); 100 is the sample dilution volume (mL); 10 is the sample volume (mL).

3. Extract spectral data and color data

According to formula (3), the average spectral value of all pixels in the ROI area of each sample is calculated as the spectral data of the sample, so that 128 samples can obtain the original spectral data of 128*224 (number of samples * number of wavelengths); The original spectral curve is shown in Figure 4. The spectral information in the image is mainly the presentation of the combined frequency and frequency doubling absorption of the hydrogen-containing groups (such as C-H, O-H and N-H, etc.) In addition, the content of chemical components in different quality grains of grains is also different, and these differences are used to cause changes in the absorption peaks of specific wavelength bands of the spectrum.

where: is the average spectral value of the selected ROI; m is the number of pixels inside the selected ROI; n is the number of bands of the hyperspectral data of glutinous rice; .

In this embodiment, color moments are used to represent color features. The colors of glutinous rice grains are mainly light yellow, light brown and reddish brown. first and third order moments. The first moment is the mean, representing the average intensity of the color components; the second moment is the color variance, representing inhomogeneity; the third moment is the skewness of the color components, representing asymmetry. Figure 3 is the HSV diagram of one of the samples, and their calculation formulas are as follows. One sample has 9 features, and 128 samples obtain a total of 128×9 color feature data.

In the formula, represents the first-order moment, N represents the number of pixels, represents the i-th color component of the j-th pixel (the color components of the three color channels H, S, and V), represents the second-order moment, and represents the third-order moment. moment.

4. Build the detection model

Step 1: Division of experimental samples

Before data modeling, the data set is usually divided into training set and test set. The training set data is mainly used for model establishment, and the test set data is mainly used to test the detection effect of the model, and the optimal model is selected according to the detection effect. The sample data is divided into training set (105) and test set (23) by SPXY algorithm. The statistical results of the total acid content of the fermented grains samples are shown in Table 1.

Table 1 Statistical results of total acid content of glutinous rice grains samples

Step 2: Spectral Preprocessing

During spectral data acquisition, spectra are often subject to various disturbances, such as electrical noise, background noise, baseline drift, and radio scattering, which can cause spectral changes and affect the reliability of multivariate calibration models. Therefore, in order to reduce the influence of these information factors on the modeling and improve the accuracy of the model, the present invention adopts the SNV combined SG algorithm to preprocess the data before establishing the model. SNV mainly eliminates the influence of surface scattering and optical path variation on the diffuse reflectance spectrum. The principle is to assume that the spectral data obeys a normal distribution, subtract the average value of the spectral curve from the spectral value at each wavelength, and then divide by The standard deviation formula of this curve is calculated, and the specific formula is:

in:

where X _isnv is the SNV corrected spectrum of sample i, X _i is the original spectral matrix of sample i, X _i,j is the spectral matrix of all samples, m is the number of bands, X _i is the spectral mean of sample i, σ _i is the standard deviation of the spectrum.

SG can smooth the spectrum, which can reduce noise after smoothing. S-G convolution smoothing is similar to moving window smoothing, the difference is that it uses polynomials to perform polynomial least squares fitting on the data in the moving window, which is essentially a A weighted average method, emphasizing the central role of the center point. However, when using this method, you need to pay attention to the selection of the width of the moving window. If it is too small, the noise will not be filtered out, and if it is too large, the effective information will be smoothed out. Calculated as follows:

where X _isg is the SG-corrected spectrum of sample _i , w is the window size, and hi is the smoothing coefficient.

Step 3: Data dimensionality reduction processing

The collected hyperspectral data has a lot of redundant and collinear information, and the large amount of data can easily lead to the problems of complex model and large amount of calculation. Therefore, the present invention adopts the RC-SPA combination algorithm to extract the characteristic wavelength of the spectrum. The RC-SPA algorithm includes the following steps:

Firstly, the regression model of total acid content and average spectral reflectance in fermented grains is established by partial least squares regression. Since the regression coefficient of each wavelength in the regression expression represents the contribution ratio of each wavelength, the greater the absolute value of the coefficient, the greater the The greater the impact of the regression model. The invention selects the wavelength corresponding to the regression coefficient whose absolute value is greater than 1000 in the regression equation established by the spectral data and the total acid content as the selected wavelength, and obtains a total of 22 wavelength combinations; Optimizing, removing collinear wavelengths, SPA is a forward variable selection method that performs a simple projection operation in a vector space to remove redundancy, obtain subsets of valuable variables and solve collinearity problems. The obtained new variable is the variable with the largest projection value on the orthogonal subspace with the previously selected variable among all the remaining variables. The specific operation steps of the SPA algorithm are as follows:

S3-1: Denote the initial iteration vector as x _k(0) , the number of extracted characteristic wavelengths as N, and the spectral matrix as column J. Randomly select a column of the spectral matrix (column j), and assign the value of column j to x _j , denoted as x _k(0) ;

S3-2: Denote the set of remaining column vector positions as s,

S3-3: Calculate the projection of x _j to the remaining column vectors, respectively,

S3-4: Select the spectral wavelength of the maximum projected value,

S3-5: Let x _j =p _x , j∈s, n=n+1, if n<N, go back to step S3-2 for cyclic calculation;

S3-6: The wavelength finally extracted is: {x _k(n) =0,...,N-1}, according to the detection results of the training set data, k(0) and N corresponding to the smallest RMSE value are the optimal Initial variables and number of variables. Finally, four spectral wavelengths of 1538.3nm, 1566.8nm, 1595.2nm and 1659.4nm are obtained as the characteristic spectral wavelengths that affect the total acid content of fermented grains, and the hyperspectral characteristic values of each sample are obtained from these four characteristic spectral wavelengths.

Step 4: Build the detection model

Cascaded Forest (CF) is an ensemble learning method based on decision trees, which processes feature information layer by layer, similar to neural networks, but the CF model has fewer hyperparameter settings and is more efficient, even with small-scale training data. excellent performance. Each cascade in CF contains 2 random forests and 2 full random forests, each of which is composed of regression trees. In addition, during the expansion process of the intermediate cascade, the newly expanded cascade is evaluated on a validation set and terminated if there is no significant performance gain. In this way, terminating the model early can not only effectively avoid the overfitting of the model, but also avoid the model from being too complicated.

The applicant used Partial Least Squares (PLSR), Support Vector Regression (SVR) and CF algorithms to establish detection models of fermented grains hyperspectral data and total acid content, respectively. Hyperspectral data is the fusion of eigenvalues and image color data. The root mean square error (RMSEC) of the training set, the root mean square error of the test set (RMSEP), the coefficient of determination of the training set (R _C ² ), the coefficient of determination of the test set (R _P ² ), and the residual prediction deviation (RPD) were used ) to evaluate the performance of the model. In addition, the robustness of the model is assessed by the absolute difference between RMSEC and RMSEP (AB_RMSE), the smaller the value, the more stable the model seen. RPD represents the relative detection performance of the model: a value of RPD between 1.5 and 2 indicates that the model has poor detection ability, between 2 and 2.5 indicates that effective detection can be achieved, and between 2.5 and 3 or greater indicates that the model has very good detection ability. For good detection accuracy, the best model should have high R _P2 value, RPD value and low RMSEP, ^AB_RMSE .

The modeling results under the same preprocessing and the same eigenvalues are shown in Table 2. As shown in Table 2, the RPD value of PLSR is 1.4768, and the values of RMSEC and RMSEP are 0.9824 and 0.8656, respectively. The detection effect is poor, and it is not suitable for the detection of total acid content of grains. The values of RPD and AB_RMSE of the SVR model are respectively 5.9139 and 0.1520, the detection effect is better than PLSR, second to CF, the RPD value of CF model is 6.4744, the value of AB_RMSE is 0.0407, the detection accuracy is the highest, and the stability is the best. Therefore, the present invention selects the CF model to construct the detection model of the total acid content of fermented grains.

Table 2 Modeling results of CF, SVR and PLSR

Step 5: Visualization

Substitute the data of each pixel in the ROI into the model for detecting the total acid content of glutinous rice grains, and obtain the total acid content value of each pixel point. Then, the content value of each pixel is stretched to the grayscale range of 0-255, and the visual distribution cloud map of the total acid content of each fermented grain is obtained according to the principle of Jet chromaticity band. Figure 5 is a visualization of the distribution of three different total acid contents, which can intuitively see the distribution of total acid and the difference of each pixel point.

The real and detected values of the final test set are as follows:

Example 2:

In order to prove the stability and generalizability of the model, the present invention adopts the samples of glutinous rice grains from different fermentation periods in the Yibin winery for detection and analysis. On the 24th day and on the 32nd day, the fermented grains collected in four different fermentation stages, a total of 120 samples, according to the detection method used in Example 1, the evaluation indicators of the final model are R _C ² =0.9982, RMSEC = 0.0255, R _P2 = ^0.9897 , RMSEP=0.0446, RPD=6.9853.

In the process of determining this invention scheme, there have been cases with large errors, such as:

1. At the beginning, we did not preprocess the spectrum, but used the original spectrum for subsequent modeling analysis. Due to the presence of a large number of noise signals in the spectrum, the modeling results were poor. The evaluation indicators at this time are: R _C ² =0.8764, RMSEC=0.0931, R _P ² =0.7621, RMSEP=0.1278, RPD=1.5445.

2. Spectral optimization is not performed using feature extraction algorithm, but full wavelength is used for modeling. Due to the large amount of redundant information in the spectrum, the modeling speed is slow, and the modeling effect is not ideal. The specific evaluation indicators are R _C2 =0.8595, RMSEC= _0.5809 , ^Rp2 =0.8743, RMSEP= ^0.6022 , RPD=2.0602.

The above are only typical embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the present invention. within.

Claims (10)

1. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology is characterized by comprising the following steps of:

1) acquiring a hyperspectral image of a sample and correcting an acquired image: collecting fermented grain samples of different layers of different cellars in a winery, and collecting hyperspectral images of the samples by using a hyperspectral imaging system;

2) measuring the total acid content of the fermented grains according to GB/T12456-2021;

3) acquiring average spectral information and color information of an interested area;

4) data processing: preprocessing an original spectrum, and screening out characteristic wavelengths related to the total acid content;

5) fusing characteristic wavelength and image color information with the measured total acid content to establish a regression model, evaluating the established model, and judging the effectiveness of the model;

6) and taking the characteristic spectrum variable to be detected as the input of the regression model to obtain the result of the total acid content of the fermented grains to be detected.

2. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology as claimed in claim 1, wherein in step 1), the number of the fermented grain samples of different layers of different cellars in a winery is 128, the fermented grain samples are collected in two different periods, the fermented grain samples of the upper, middle and lower layers of 13 cellars in the winery are collected for the first time, a rotary sampler is adopted for each layer to randomly select three sampling points for collection, wherein the depth of each cellar is 2m, the upper sampling point is 0.6m away from the cellars of the cellars, the middle sampling point is 1.1m away from the cellars of the cellars, the lower sampling point is 2m away from the cellars of the cellars, and 117 samples are obtained in total; and in the second collection, 11 cellars are randomly selected from the 13 cellars collected for the first collection for sampling, and each cellar randomly selects one position to obtain 11 samples.

3. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology as claimed in claim 1, wherein the hyperspectral imaging system in the step 1) mainly comprises a hyperspectral camera, an illumination system, a high-precision electronic control objective table and a computer with processing software; the collected spectral range is 940-1730nm, 224 wave bands are totally formed, and the spectral resolution is 3.3 nm; the exposure time of the hyperspectral camera is 4.02ms, the acquisition frequency is 50Hz, the moving speed of the objective table is 16.42mm/s, a three-dimensional hyperspectral image data block of the fermented grain sample is obtained in a linear array push-broom manner, and the acquired hyperspectral image is calibrated and an interested area is selected.

4. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology according to claim 1, wherein the obtaining of the color information in the step 3) mainly adopts a first moment, a second moment and a third moment of a gray image of a fermented grain sample ROI in H, S, V color space as features; there are 9 features in one sample, and 128 samples result in 128 × 9 color feature data.

5. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology according to claim 1, wherein the pretreatment method in the step 4) is standard normal transformation combined with SG convolution smoothing; the characteristic wavelength extraction method is a regression coefficient method combined with a continuous projection algorithm.

6. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology as claimed in claim 1, wherein the regression model established in the step 5) is a cascade forest model; with R ² RMSE is used as an evaluation index to determine the effectiveness of the model.

7. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology as claimed in claim 1, wherein the method further comprises the following steps of 7): visualizing the obtained total acid content to obtain the acidity distribution map of the fermented grains.

8. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology as claimed in claim 1, wherein the method for correcting the acquired image comprises the following steps: firstly, a standard polytetrafluoroethylene white board is placed in an imaging area, and a full white reflection calibration image of the standard white board is collected; then, covering the lens cover, and closing the light source to obtain a completely black calibration image; finally, calculating the corrected hyperspectral image according to the following formula;

in the formula: r is the corrected image; i is the original image; w is a standard whiteboard image; b is a dark current image with the lens turned off.

9. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology as claimed in claim 1, wherein the region of interest is selected by adopting morphological processing and binary norm calculation, and a circular region is divided by taking the center of the collected fermented grain sample image as a circle center and taking 100 pixels as a radius as a ROI of the sample.

10. The method for detecting the total acid content of the fermented grains based on the hyperspectral imaging technology according to any of the claims 1 to 9, which is used for detecting the total acid content of the fermented grains.

Images