Nondestructive testing method for comprehensive quality indexes of wine grapes
Through spectral analysis technology and machine learning models, the problem of detecting comprehensive quality indicators of wine grapes is solved, efficient non-destructive testing of wine grapes is achieved, and detection accuracy and discrimination ability are improved.
Patent Information
- Application Number
- CN202510133443.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-06
- Publication Date
- 2025-06-24
AI Technical Summary
The prior art is difficult to effectively monitor and evaluate the comprehensive quality indicators of wine grapes, resulting in the inability to accurately determine the optimal harvest time.
The grape optical acquisition system is built using spectral analysis technology, and efficient non-destructive detection of the quality of wine grapes is achieved through pre-processing spectral data, detecting physical and chemical indicators, normalization processing, feature extraction and machine learning model training.
Through deep learning algorithms, the analysis ability of abnormal spectra is enhanced, combined with the advantages of traditional and deep learning algorithms, efficient detection of various indicators of wine grapes is achieved, and the ability to judge grape quality and detection accuracy are improved.
Smart Images
Figure CN120195115A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical fields of spectral analysis and physical and chemical index detection and analysis, and specifically to a non-destructive detection method for comprehensive quality indexes of wine grapes. Background Art
[0002] Grapes are a globally popular crop and are valued for their economic significance and nutritional contributions. The quality of grapes depends on primary metabolites such as sugars and acids, as well as secondary metabolites such as phenolic compounds. The ripening process of grapes involves changes including the accumulation of phenolic substances in the peel, an increase in sugar content, a decrease in acid content, and a gradual decrease in tannin content in the seeds after color change. This decrease is relatively rapid at the beginning of color change and relatively stable during subsequent ripening. Therefore, during the grape ripening process, the changes in various indexes are diverse rather than uniform. Monitoring these changes throughout the ripening process is a prerequisite for determining the optimal harvest time.
[0003] In recent years, the introduction of chemometrics has greatly expanded the application prospects of spectral technology in evaluating the internal quality of fruits. By analyzing the spectral characteristics of fruits, information related to quality indexes can be extracted. Spectral analysis usually requires minimal sample processing and has the advantages of simplicity, rapidity, and non-destructiveness compared to traditional physical and chemical measurement methods. In addition, traditional methods require not only expensive chemical reagents but also a large amount of time for detection. In contrast, spectral analysis does not require submission to laboratory procedures, thus shortening the analysis time and facilitating process optimization.
[0004] In recent years, spectral technology has been effectively applied to the development of various fruit quality detection methods and has been used for mangoes, avocados, and citrus fruits. In the field of single-parameter prediction, hyperspectral imaging technology has been used to determine the soluble solid content in Red Globe grapes; near-infrared spectroscopy has been used to predict the texture parameters and soluble solid content of whole berries; there has also been research on the influence of handheld near-infrared spectrometers on grape spectra, considering temperature and light, and predicting the soluble solid content.
[0005] Regarding multi-index quality, visible light-near infrared (Vis-NIR) hyperspectral imaging technology has been used to detect tomato quality, and a comprehensive quality index (CQI) has been introduced; based on near-infrared spectroscopy for detecting the quality of mangoes, oranges, and avocados, a new fruit quality index (FQI) has been designed. Summary of the Invention
[0006] The purpose of the present invention is to provide a non-destructive detection method for comprehensive quality indexes of wine grapes to solve the problems raised in the prior art.
[0007] To achieve the above purpose, the present invention provides the following technical solution: A non-destructive detection method for comprehensive quality indexes of wine grapes, the steps of which are as follows:
[0008] Step 1: Build a grape optical acquisition system to collect the optical signals of grapes and generate a spectral waveform L.
[0009] Step 2: Preprocess the spectrum.
[0010] Step 3: Detect the physical and chemical indexes of the collected grape samples.
[0011] Step 4: Normalize the obtained physical and chemical indexes of grapes.
[0012] Step 5: Extract features for specific spectral information to obtain specific spectral bands.
[0013] Step 6: Use partial least squares algorithm, support vector machine, and convolutional neural network to perform modeling training on the spectral bands obtained in Step 3.
[0014] Step 7: After obtaining the machine learning model, use the trained machine learning model and deep learning model to predict the grape quality respectively, and then fuse the prediction results of the two models. Adopt a simple voting method. When the prediction results of the two models are the same, accept the prediction result; when the prediction results of the two models are different, determine the final prediction result according to the historical accuracy of the model or other weight assignment methods. Fuse the features extracted by the machine learning algorithm and the features learned by the deep learning model to form a new feature vector, and use the new feature vector to train a new classification model, another machine learning model or a simple linear classifier to improve the discrimination ability of the grape quality. In the feature fusion process, use the principal component analysis method to reduce the dimension of the fused features, remove redundant information, improve the training efficiency and accuracy of the model, and evaluate the model.
[0015] As an optimal technical solution, in Step 1, build a grape optical acquisition system by the way of diffuse transmission, and transmit the collected grape optical signals to the spectrometer through optical fibers to generate the spectral waveform L of the fruit.
[0016] As an optimal technical solution, in Step 2, preprocess the spectrum, including standard normal variate transformation, multiplicative scatter correction, first derivative, second derivative, Savitzky-Golay convolutional smoothing, Savitzky-Golay convolutional smoothing + first derivative (SG + 1D).
[0017] As an optimal technical solution, in Step 3, detect the physical and chemical indexes of grapes, including total phenols, total flavonoids, flavan-3-ols, tannins, total reducing sugars, soluble solids content, total acids, pH.
[0018] As an optimal technical solution, in Step 4, normalize the obtained physical and chemical indexes of grapes and calculate through formula (1).
[0019] Xni = (Xi - Xmin) / (Xmax - Xmin) (1)
[0020] Xni is the value after the index is normalized by the maximum and minimum values; Xi is the measured value of the index; Xmin is the minimum value of a specific index among all the tested materials; Xmax is the maximum value of a specific index among all the tested materials.
[0021] As a preferred technical solution, in step five, feature extraction is performed on specific spectral information to obtain a specific spectral band.
[0022] As a preferred technical solution, in step six, the spectral band obtained in step five is modeled and trained using the partial least squares algorithm, support vector machine, and convolutional neural network.
[0023] As a preferred technical solution, in step seven, after obtaining the machine learning model, the model is evaluated.
[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: By using the deep learning algorithm, the analysis ability of abnormal spectra is enhanced, and the advantages of traditional and deep learning algorithms are combined. The quality of grapes is discriminated and detected from two dimensions of machine learning algorithms and deep learning algorithms. By normalizing various indicators of wine grapes (white varieties), a comprehensive index (AQI) is established to perform high-efficiency and non-destructive quality detection on wine grapes (white varieties). BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a schematic flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0027] As Figure 1 shown, the present invention provides the following technical solutions, and the steps are as follows:
[0028] Step 1: Build a grape optical acquisition system to collect the optical signal of grapes and generate a spectral waveform L;
[0029] Step 2: Preprocess the spectrum;
[0030] Step 3: Detect the physical and chemical indexes of the collected grape samples;
[0031] Step 4: Normalize the obtained physicochemical indexes of grapes;
[0032] Step 5: Extract features for specific spectral information to obtain specific spectral bands;
[0033] Step 6: Use partial least squares algorithm, support vector machine, and convolutional neural network to perform modeling training on the spectral bands obtained in Step 3;
[0034] Step 7: After obtaining the machine learning model, use the trained machine learning model and deep learning model to predict the grape quality respectively, and then fuse the prediction results of the two models. Adopt a simple voting method. When the prediction results of the two models are the same, accept the prediction result; when the prediction results of the two models are different, determine the final prediction result according to the historical accuracy of the model or other weight assignment methods. Fuse the features extracted by the machine learning algorithm and the features learned by the deep learning model to form a new feature vector, and use the new feature vector to train a new classification model, another machine learning model or a simple linear classifier to improve the discrimination ability of grape quality. In the feature fusion process, use the principal component analysis method to reduce the dimension of the fused features, remove redundant information, improve the training efficiency and accuracy of the model, and evaluate the model, evaluate the model.
[0035] In Step 1, build a grape optical acquisition system through the way of diffuse transmission, and transmit the collected grape optical signal to the spectrometer through optical fiber to generate the spectral waveform L of the fruit.
[0036] In Step 2, preprocess the spectrum, including standard normal variate transformation, multiplicative scatter correction, first derivative, second derivative, Savitzky-Golay convolutional smoothing, Savitzky-Golay convolutional smoothing + first derivative (SG + 1D).
[0037] In Step 3, detect the physicochemical indexes of grapes, including total phenols, total flavonoids, flavan-3-ols, tannins, total reducing sugars, soluble solids content, total acids, pH.
[0038] In Step 4, normalize the obtained physicochemical indexes of grapes, and calculate through Equation (1)
[0039] Xni = (Xi - Xmin) / (Xmax - Xmin) (1)
[0040] Xni is the value after normalization by the maximum and minimum values; Xi is the measured value of this index; Xmin is the minimum value of a specific index among all the tested materials; Xmax is the maximum value of a specific index among all the tested materials.
[0041] In step five, feature extraction is performed on specific spectral information to obtain specific spectral bands.
[0042] In step six, the spectral bands obtained in step five are modeled and trained using partial least squares algorithm, support vector machine, and convolutional neural network.
[0043] In step seven, after obtaining the machine learning model, the model is evaluated.
[0044] The spectral detection described by this method is equipped with a high-resolution micro-spectrometer and has a built-in circuit capable of synchronously triggering and controlling a xenon lamp. A standard whiteboard is used (as a white reference). A calibration operation (standard white reference definition) is performed before each data acquisition.
[0045] The grapes are placed on the standard whiteboard for spectral acquisition. Each sample is measured 10 times from both sides (sunny side and shady side), and the spectral data obtained for each sample is analyzed using the Euclidean distance (Askanazi and Grinberg).
[0046] The data point with the minimum distance from other data points is selected as the representative data for each sample.
[0047] Physicochemical indexes of grape fruits to be detected: total sugar, soluble solids content (SSC), reducing sugar, total acid, total phenolic content in skin (TPN), total phenolic content in seeds (TPD), total flavan-3-ols, total flavonoids, pH, skin tannins, seed tannins.
[0048] Establishment of the comprehensive index AQI: Considering that the measurement units of various indexes are different, it is necessary to normalize the magnitudes of the selected parameters to a consistent scale before the summarization process, which is accomplished by mapping the values to a predetermined range using min-max normalization (Eq.1).
[0049] Perform min-max normalization on the physicochemical indexes: Xni = (Xi - Xmin) / (Xmax - Xmin). Xni is the value of the index after normalization by the maximum and minimum values; Xi is the measured value of this index; Xmin is the minimum value of a specific index among all the measured materials; Xmax is the maximum value of a specific index among all the measured materials.
[0050] The factor analysis method is used to identify the latent factors and dominant factors of wine grapes (white varieties) and evaluate the comprehensive quality of wine grapes (white varieties).
[0051] The analysis involves the Kaiser-Meyer-Olkin (KMO) and Bartlett's sphericity test. The higher the KMO value, the higher the reliability of the factor analysis data. If Bartlett's sphericity test produces a significance level of 0.00, a strong correlation between the data sets can be inferred. Factors are determined based on eigenvalues greater than 1 or cumulative contribution exceeding 80%.
[0052] Spectral data processing: During the spectral acquisition process, various environmental factors can introduce various types of noise, which may affect the accuracy of modeling and prediction precision (Adesokan et al., 2023; Li et al., 2023b).
[0053] The present invention adopts five preprocessing methods: multiplicative scatter correction (MSC), Savitzky-Golay smoothing (SG), first derivative (FD), SG+FD, and detrending (DT).
[0054] Calibration set and prediction set: To ensure that the distributions of the calibration set and the prediction set are consistent with the true distribution, this method uses the sample set division based on the successive projections algorithm for joint x-y distances (SPXY). This algorithm considers both spectral features and physicochemical indexes in sample selection to ensure that the calibration set contains the range of the prediction set.
[0055] Using the SPXY algorithm, the CS and MK grape samples are divided into a calibration set and a prediction set in a ratio of 3:1.
[0056] Modeling method: It is planned to use three different prediction models, namely partial least squares regression (PLSR), support vector machine regression (SVR), and convolutional neural network (CNN); the spectral data is used as the input, and the AQI is used as the target variable.
[0057] Model evaluation: The calibration root mean square error (RMSEC), prediction root mean square error (RMSEP), calibration determination coefficient (R2 c), prediction determination coefficient (R2 p), and residual prediction deviation (RPD) are used to evaluate the accuracy and stability of the model.
[0058] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.
Claims
1. A non-destructive testing method for comprehensive quality indexes of wine grapes, characterized in that: The steps are as follows: Step 1: Build a grape optical collection system to collect grape optical signals and generate a spectrum waveform L; Step 2: pre-process the spectrum; Step 3: Conduct physical and chemical index testing on the collected grape samples; Step 4: normalizing the acquired grape physical and chemical indicators; Step 5: Extract features based on specific spectral information to obtain specific spectral bands; Step 6: Use partial least squares algorithm, support vector machine and convolutional neural network to model and train the spectral bands obtained in step 3; Step 7. After obtaining the machine learning model, use the trained machine learning model and deep learning model to predict the quality of grapes respectively, and then fuse the prediction results of the two models. Use a simple voting method. When the prediction results of the two models are consistent, accept the prediction result; when the prediction results of the two models are inconsistent, determine the final prediction result based on the historical accuracy of the model or other weight distribution methods, fuse the features extracted by the machine learning algorithm and the features learned by the deep learning model to form a new feature vector, and use the new feature vector to train a new classification model, another machine learning model or a simple linear classifier to improve the ability to distinguish the quality of grapes. In the feature fusion process, use the principal component analysis method to reduce the dimensionality of the fused features, remove redundant information, improve the training efficiency and accuracy of the model, and evaluate the model.
2. The nondestructive testing method for comprehensive quality indexes of wine grapes according to claim 1, characterized in that: In step 1, a grape optical collection system is built by diffuse transmission, and the collected grape light signal is transmitted to the spectrometer through optical fiber to generate the fruit's spectral waveform L.
3. The nondestructive testing method for comprehensive quality indexes of wine grapes according to claim 1, characterized in that: In step 2, the spectrum is preprocessed by standard normal transformation, multivariate scattering correction, first-order derivative, second-order derivative, Savitzky-Golay convolution smoothing, and Savitzky-Golay convolution smoothing + first-order derivative (SG+1D).
4. The method for nondestructive testing of comprehensive quality indicators of wine grapes according to claim 1, characterized in that: In step three, the physical and chemical indicators of grapes are tested, including total phenols, total flavonoids, flavan-3-ols, tannins, total reducing sugars, soluble solids content, total acid, and pH.
5. The method for nondestructive testing of comprehensive quality indicators of wine grapes according to claim 4, characterized in that: In step 4, the obtained grape physical and chemical indicators are normalized and calculated by formula (1): Xni=(Xi-Xmin) / (Xmax-Xmin) (1) Xni is the value of the index after normalization by the maximum and minimum values; Xi is the measured value of the index; Xmin is the minimum value of a specific index among all the tested materials; and Xmax is the maximum value of a specific index among all the tested materials.
6. The nondestructive testing method for comprehensive quality indexes of wine grapes according to claim 1, characterized in that: In step five, feature extraction is performed on specific spectral information to obtain specific spectral bands.
7. The method for nondestructive testing of comprehensive quality indicators of wine grapes according to claim 1, characterized in that: In step six, the spectral bands obtained in step five are modeled and trained using partial least squares algorithm, support vector machine, and convolutional neural network.
8. The nondestructive testing method for comprehensive quality indexes of wine grapes according to claim 1, characterized in that: In step seven, after obtaining the machine learning model, the model is evaluated.