Deep learning model-based berry tea quality spectrum sensing detection method

Abstract

The invention discloses a deep learning model-based berry tea quality spectrum sensing detection method, which comprises the following steps of: acquiring near infrared spectrum data of a berry tea sample by using a portable spectrum sensing device, and preprocessing the acquired spectrum data; preparing a berry tea powder sample from the berry tea subjected to spectrum collection by using a grinding mill; preparing tea soup from the berry tea powder sample by using a solvent extraction method; determining the physical and chemical values of tea polyphenol and dihydromyricetin as quality indexes of the berry tea based on a conventional method; and preprocessing near infrared spectrum data of the berry tea, and inputting the preprocessed near infrared spectrum data into the detection quantitative model to quickly obtain a near infrared spectrum quality index detection result. According to the method, the spectral information of the raspberry tea sample is directly acquired by utilizing the portable spectrum sensing device, a complicated pretreatment process is not needed, and meanwhile, accurate and rapid prediction of the raspberry tea quality index is further realized by utilizing the deep learning model, so that an economical and feasible path is provided for sustainable development of the raspberry tea industry.

Description

Technical Field

[0001] The present invention relates to the field of rapid detection of tea quality, and in particular to a berry tea quality spectral sensing detection method based on a deep learning model. Background Art

[0002] As a natural food preservative, dietary supplement, and raw material source for new functional beverages, berry tea has been widely welcomed by consumers in recent years. Studies have shown that appropriate intake of berry tea has antibacterial, anti-inflammatory, blood vessel softening, liver protection, and immunity-enhancing effects on the body. These functions of berry tea are mainly related to internal chemical components such as tea polyphenols, dihydromyricetin, amino acids, and various trace elements. These internal organic compounds are also closely related to the quality of berry tea, and their content will directly affect the quality and health care functions of berry tea. The content of tea polyphenols in berry tea can reach 43.58%, which is more than three times that of ordinary green tea. Dihydromyricetin is the most important flavonoid compound in berry tea and is considered to be the main material basis for the pharmacological effects of berry tea. It has been confirmed that the higher the quality grade of berry tea, the higher the content of these beneficial ingredients.

[0003] Existing methods for testing the quality of berry tea include human sensory panel testing and instrumental analysis technology. Human sensory panel testing requires a rigorous and cumbersome procedure, and in particular requires tea experts to evaluate berry tea based on a series of predefined physical and chemical indicators to ultimately determine its quality. Instrumental analysis techniques include gas chromatography, high performance liquid chromatography, gas chromatography-mass spectrometry, and high performance liquid chromatography-mass spectrometry. As an auxiliary technology for quality assessment, this method can ensure the objectivity and accuracy of the results, but it requires expensive equipment and cumbersome sample pretreatment. Therefore, there is an urgent need to establish a sensitive, economical, efficient, and environmentally friendly detection method to achieve rapid detection of berry tea quality. Summary of the invention

[0004] The purpose of the present invention is to provide a berry tea quality spectral sensing detection method based on a deep learning model to overcome the problems of complex operation and long detection time in existing detection technologies.

[0005] In order to achieve the above object, the present invention provides a berry tea quality spectral sensing detection method based on a deep learning model, the method comprising:

[0006] S1. Berry tea spectrum collection: Use a spectral sensor device to obtain spectral data of different berry tea samples, and use the Derivative Spectroscopy derivative spectrum and Savitzky-Golay smoothing algorithm of the MATLAB third-party intelligent library to process the collected near-infrared spectrum to reduce noise and baseline drift;

[0007] S2. Conventional determination of berry tea quality physical and chemical indicators: first, a berry tea powder sample is prepared from the berry tea after the spectrum is collected using a grinder, then a solvent extraction method is used to prepare tea soup from the berry tea powder sample, and finally the physical and chemical values ​​of the berry tea quality indicators are determined using conventional methods;

[0008] S3. Rapid detection of tea polyphenols and dihydromyricetin in berry tea: Based on the Deep Learning Toolbox framework, a one-dimensional convolutional neural network (CNN), a long short-term memory network (LSTM) and a CNN-LS TM berry tea quality prediction quantitative model were built, and the spectral data after derivative and smoothing preprocessing in S1 were input into the quantitative model to quickly output the detection results of tea polyphenols (TP) and dihydromyricetin (DMY) in berry tea.

[0009] Preferably, the parameters for collecting the near-infrared spectrum of berry tea using a portable spectral sensing device are: spectral range 900-1700nm, spectral resolution 10nm, integration time 5-10ms, averaging times 5-8 times, and ambient temperature 23-26°C.

[0010] Preferably, the noise and baseline drift in the spectrum are processed using the Derivative Spectroscopy derivative spectrum and Savitzky-Golay smoothing algorithm of the MATLAB third-party intelligent library, wherein the derivative order of the derivative spectrum is 1-2, the window width is 5-11 (set to an odd number), the polynomial order of the Savitzky-Golay smoothing algorithm is 2-3, and the window width is 5-15.

[0011] Preferably, the process of using a grinder to prepare a berry tea powder sample from the berry tea after collecting the spectrum specifically includes:

[0012] Put an appropriate amount of berry tea into a small grinder and grind it using the grinder. The grinder grinds each type of berry tea according to the principle of "press for 3 seconds and stop for 1 second". Repeat this operation 8-10 times to initially obtain powdered berry tea. Take the crushed powdered berry tea and place it on a 40-60 mesh sieve. At the same time, shake the sieve evenly to finally obtain a fine powdered berry tea sample.

[0013] Preferably, the process of preparing tea soup from the berry tea powder sample by solvent extraction specifically includes:

[0014] Preparation of tea soup for tea polyphenol determination: weigh 0.1990-0.2009g of berry tea powder sample into a centrifuge tube, add 4-5mL of 70% methanol aqueous solution preheated at 70±1℃, and stir evenly with a glass rod. Immediately transfer to 70±1℃ water bath for extraction, cool to room temperature after extraction, transfer to centrifuge for centrifugation (3500r / min, 10min) after balancing; extract the residue once with 70% methanol aqueous solution, repeat the above operation, combine the extracts and make up to volume, mix evenly, filter with 0.45μm membrane, and obtain tea soup sample.

[0015] Preparation of tea soup for dihydromyricetin determination: weigh 0.4990-0.5009g of berry tea powder sample into a centrifuge tube, add 8-10mL of methanol aqueous solution, weigh and record, mix well and place in a 50℃ ultrasonic instrument for ultrasonic extraction for 30 minutes, take out and cool to room temperature and weigh again, use pure methanol to keep constant weight before ultrasonic extraction, mix well and filter using a 0.45μm membrane to obtain a tea soup sample.

[0016] Preferably, the content of determining the physicochemical value of tea polyphenols in berry tea by ultraviolet spectrophotometry based on conventional methods specifically includes:

[0017] Prepare gallic acid working solution with concentrations of 10, 20, 30, 40, 50, 60, 80, 100, and 120 μg / mL; dilute the tea soup with water and shake well, take 0.5-1.0 mL of the gallic acid working solution, water, and the diluted tea soup, respectively, and then add the folin phenol solution and sodium carbonate solution to each test tube in a volume ratio of 5:4, dilute to volume with water and shake well, and place at room temperature for 50-65 minutes. Before using the ultraviolet spectrophotometer to measure tea polyphenols, the photometric measurement parameters are set as follows: the detection wavelength is 765 nm, the number of repeated measurements is 3, and the photometric mode is absorbance.

[0018] Preferably, the content of determining the physicochemical value of dihydromyricetin in berry tea using high performance liquid chromatography based on conventional methods specifically includes:

[0019] Prepare dihydromyricetin working solution with concentrations of 100, 200, 300, 400, 500, and 700 μg / mL; dilute the tea soup with the initial mobile phase (acetonitrile: 0.1% phosphoric acid aqueous solution = 15:85) at a ratio of 1:10, and filter again using a 0.45 μm membrane after mixing. Before using a high-performance liquid chromatograph to measure dihydromyricetin, replace the chromatographic column with a C18 column and create a new detection method, with an elution gradient of 0-12 min, 85% B; 12-20 min, 85-70% B; 20-25 min, 70-85% B; 25-35 min, 85%, and the instrument detection parameters are set as follows: column temperature of 25°C, detection wavelength of 291 nm, flow rate of 1 mL / min, injection volume of 10-20 μL, and use an ultraviolet detector for determination while saving records.

[0020] Preferably, the function expression of the CNN quantitative prediction model is as follows:

[0021] CNN(S tea1 , S tea2 ···S teaN )= TP S (1)

[0022] CNN(S tea1 , S tea2 ···S teaN )= DMY S (2)

[0023] Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S; the berry tea spectrum data is used as input and the corresponding TP data is used as output to train and predict the CNN model;

[0024] Preferably, the CNN model structure includes an input layer, 2 convolutional layers, 2 batch normalization layers, 2 Dropout layers, an activation function layer, 2 pooling layers, 1 fully connected layer and an output layer; the parameter dimensions of the one-dimensional convolutional layer are 32 and 64 respectively, the convolution kernel is 3, the convolution kernel size is 1*1, and the convolution step is 1; the dropout layer discard ratio is set to 0.2 and 0.3 respectively, that is, 20% and 30% of the neurons are randomly discarded; the activation function is the ReLU function; the pooling kernel sizes of the pooling layer are 3 and 5 respectively; the number of fully connected layers is 1, which is used to output 1 feature.

[0025] Preferably, the functional expression of the LSTM quantitative prediction model is as follows:

[0026] LSTM(Stea1 , S tea2 ···S teaN )= TP S (3)

[0027] LSTM(S tea1 , S tea2 ···S teaN )= DMY S (4)

[0028] Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S; using the berry tea spectrum data as input and the corresponding TP data as output, the LSTM model is trained and predicted;

[0029] Preferably, the LSTM model structure includes an input layer, a hidden layer, an activation function layer, a fully connected layer and an output layer; the number of hidden layers is a first LSTM layer, a second LSTM layer and a corresponding first Dropout layer, a second Dropout layer; the number of neurons in the first LSTM layer is 4, and the Dropout layer randomly discards neurons with a discard ratio of 0.2; the number of neurons in the second LSTM layer is 16, and the Dropout layer randomly discards neurons with a discard ratio of 0.3; the activation function is a ReLU function; the number of fully connected layers is set to 1 layer to output 1 feature.

[0030] Preferably, the function expression of the CNN-LSTM quantitative prediction model is as follows:

[0031] CNN-LSTM(S tea1 , S tea2 ···S teaN )= TP S (5)

[0032] CNN-LSTM(S tea1 , S tea2 ···S teaN )= DMY S (6)

[0033] Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S; the berry tea spectrum data is used as input and the corresponding TP data is used as output to train and predict the CNN-LSTM model;

[0034] Preferably, the CNN-LSTM structure is an input layer, a sequence folding layer, two convolutional layers, an activation function layer, a sequence anti-folding layer, a flattening layer, two LSTM layers, two Dropout layers, a fully connected layer, and an output layer; the parameter dimensions of the one-dimensional convolutional layer are 16 and 32 respectively, the convolution kernel is 2, the convolution kernel size is 1*1, and the convolution step is 1; the activation function is a ReLU function; the number of neurons in the LSTM layer is 4 and 8 respectively; the discard ratio of the Dropout layer is set to 0.2 and 0.3 respectively; the number of fully connected layers is set to 1 layer to output 1 feature.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. This application uses a portable spectral sensor device to directly obtain the spectral information of the berry tea sample. Compared with the conventional berry tea quality detection method, this method does not require a complicated pre-treatment process and effectively improves the efficiency of berry tea quality detection.

[0037] 2. Compared with conventional machine learning models, deep learning models have the advantages of end-to-end learning and automatic extraction of data features, which improves detection efficiency and accuracy.

[0038] 3. The deep one-dimensional convolutional neural network model has a powerful convolutional layer, which can deeply mine the characteristics of berry tea spectral data, accurately analyze the complex nonlinear relationship between berry tea quality and near-infrared spectrum, and achieve more accurate prediction of berry tea quality.

[0039] 4. The deep long short-term memory network is a deep learning framework with a large number of hidden layers, which can alleviate the problem of gradient disappearance, better capture the complex patterns and relationships in the data, and deeply explore the relationship between the quality of berry tea and different spectral bands, thereby enhancing the robustness of the model.

[0040] 5. The deep one-dimensional convolutional neural network - long short-term memory network has both convolutional layers and LSTM layers. It can identify key spectral bands and then process global dependencies through the LSTM layer. It can more comprehensively analyze the complex relationships in spectral data, and can effectively avoid overfitting, thereby improving the overall performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1A schematic diagram of a method flow of an embodiment of the present application;

[0043] Figure 2 This is a schematic diagram of the standard curve of different concentrations of gallic acid and dihydromyricetin in the examples of this application;

[0044] (a) is a schematic diagram of the standard curve of gallic acid, and (b) is a schematic diagram of the standard curve of dihydromyricetin;

[0045] Figure 3 This is a schematic diagram of the one-dimensional convolutional neural network model structure of an embodiment of the present application;

[0046] Figure 4 This is a schematic diagram of the structure of a long short-term memory neural network model according to an embodiment of the present invention;

[0047] Figure 5 This is a schematic diagram of the structure of a one-dimensional convolutional neural network-long short-term memory neural network model according to an embodiment of the present invention;

[0048] Figure 6 This is a schematic diagram of the detection results of berry tea quality using the one-dimensional convolutional neural network model in Example 1 of the present application;

[0049] (a) is the relationship between the predicted value of tea polyphenols and its true value in the one-dimensional convolutional neural network model, (b) is the relationship between the predicted value of dihydromyricetin and its true value in the one-dimensional convolutional neural network model;

[0050] Figure 7 This is a schematic diagram of the detection results of berry tea quality by the long short-term memory network model in Example 2 of the present application;

[0051] (a) is the relationship between the predicted value of tea polyphenols in the long short-term memory network model and its true value, (b) is the relationship between the predicted value of dihydromyricetin in the long short-term memory network model and its true value;

[0052] Figure 8 This is a schematic diagram of the detection results of the one-dimensional convolutional neural network-long short-term memory network model on the quality of berry tea in Example 3 of the present application;

[0053] (a) is the relationship between the predicted value of tea polyphenols and its true value in the one-dimensional convolutional neural network-long short-term memory network model, (b) is the relationship between the predicted value of dihydromyricetin and its true value in the one-dimensional convolutional neural network-long short-term memory network model; DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0056] Embodiment 1

[0057] like Figure 1 FIG. 1 is a schematic diagram of the method flow of this embodiment, and the steps include:

[0058] S1 uses a spectral sensing device to obtain spectral data of different berry tea samples and preprocesses the obtained near-infrared spectral data.

[0059] The specific contents include: using a portable spectral sensor device to collect spectral information of all berry tea samples at 900-1700nm, and the parameters for spectral data collection are: spectral resolution 10nm, integration time 5ms, 8 average times, and ambient temperature 25℃; after the collection is completed, the Derivative Spectroscopy derivative spectrum and Savitzky-Golay smoothing algorithm of the MATLAB third-party intelligent library are used to process the collected near-infrared spectrum to reduce noise and baseline drift. Among them, the derivative order of the Derivative Spectroscopy derivative spectrum is 1, the window width is 7, and the polynomial order of the Savitzky-Golay smoothing algorithm is 3, and the window width is 9.

[0060] S2 determines the physicochemical values ​​of tea polyphenols (TP) and dihydromyricetin (DMY), the quality indicators of berry tea, based on conventional methods.

[0061] The content of preparing the berry tea powder sample by using a grinder for the berry tea after collecting the spectrum specifically includes:

[0062] 1) 80 g of berry tea was put into a small grinder. The grinder was used to grind each type of berry tea according to the principle of "press for 3 seconds and stop for 1 second". This operation was repeated 10 times to initially prepare powdered berry tea.

[0063] 2) Place the crushed powdered berry tea on a 60-mesh sieve and shake the sieve evenly to obtain a fine powdered berry tea sample.

[0064] The content of preparing tea soup by solvent extraction method specifically includes:

[0065] 1) Preparation of tea soup for tea polyphenol determination: First, weigh 0.2 g of berry tea powder sample, then add 5 mL of 70% methanol aqueous solution, extract in a 70°C water bath for 10 min, cool to room temperature, and centrifuge at 3500 r / min for 10 min to obtain a supernatant. Repeat the above operation for residue extraction, combine the extracts and make up to volume with water, mix well, and filter using a 0.45 μm membrane to obtain a tea soup sample.

[0066] 2) Preparation of tea soup for dihydromyricetin determination: First, weigh 0.5 g of berry tea powder sample, add 10 mL of 70% methanol aqueous solution, weigh and record, then place in a 50°C ultrasonic instrument for ultrasonic extraction for 30 min, take out and cool to room temperature and weigh again, use pure methanol to keep constant weight before ultrasonic extraction, mix well and filter using a 0.45 μm membrane to obtain a tea soup sample.

[0067] The content of the determination of the physicochemical value of tea polyphenols in berry tea by ultraviolet spectrophotometry specifically includes:

[0068] 1) Prepare gallic acid working solutions with concentrations of 10, 20, 30, 40, 50, 60, 80, 100, and 120 μg / mL to obtain the absorbance standard curve equation and correlation coefficient for tea polyphenol content detection, such as Figure 2 shown.

[0069] 2) First, the tea soup was diluted 100 times with water, and then 1.0 mL of the gallic acid working solution, water and the diluted tea soup were taken, and then the Folin solution and sodium carbonate solution were added to each test tube in a volume ratio of 5:4, and the volume was fixed with water and shaken, and placed at room temperature for 60 minutes. Before using the ultraviolet spectrophotometer to measure tea polyphenols, the photometric measurement parameters were set as follows: the detection wavelength was 765 nm, the number of repeated measurements was 3, and the photometric mode was absorbance.

[0070] The content of the determination of the physicochemical value of dihydromyricetin in berry tea by high performance liquid chromatography specifically includes:

[0071] 1) Prepare dihydromyricetin working solutions with concentrations of 100, 200, 300, 400, 500, and 700 μg / mL, and obtain the peak area standard curve equation and correlation coefficient for dihydromyricetin content detection, such as Figure 2 shown.

[0072] 2) First, the tea soup is diluted with the initial mobile phase (acetonitrile: 0.1% phosphoric acid aqueous solution = 15:85) at a ratio of 1:10, and after mixing, it is filtered using a 0.45 μm membrane. Before using a high performance liquid chromatograph to measure dihydromyricetin, the chromatographic column is replaced with a C18 column and a new detection method is created, the elution gradient is 0-12min, 85% B; 12-20min, 85-70% B; 20-25min, 70-85% B; 25-35min, 85%, and the instrument detection parameters are set as follows: column temperature is 25°C, detection wavelength is 291nm, flow rate is 1mL / min, the injection volume is 10-20μL, and a UV detector is used for determination while keeping records.

[0073] S3 constructs a quantitative model for berry tea quality detection based on the pre-processed near-infrared spectral data and physicochemical values.

[0074] A one-dimensional convolutional neural network (CNN) quantitative prediction model is built based on the Deep Learning Toolbox framework. Its function expression is as follows:

[0075] CNN(S tea1 , S tea2 ···S teaN )= TP S (1)

[0076] CNN(S tea1 , S tea2 ···S teaN )= DMY S (2)

[0077] Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S;

[0078] In this embodiment, the established CNN network structure is as follows: Figure 3As shown, it includes an input layer, 2 convolutional layers, 2 batch normalization layers, 2 Dropout layers, an activation function layer, 2 pooling layers, a fully connected layer and an output layer. The input layer of the CNN network model receives near-infrared spectral data, extracts features through each layer of convolution and pooling operations, and outputs feature vectors to the fully connected layer, and uses the fully connected layer to map these extracted features to the output layer. The parameter dimensions of the one-dimensional convolutional layer are 32 and 64, the convolution kernel is 3, the convolution kernel size is 1*1, and the convolution step is 1; the dropout layer discard ratio is set to 0.2 and 0.3, that is, 20% of the neurons are randomly discarded; the activation function is the ReLU function; the pooling kernel size of the pooling layer is 3 and 5, respectively; the number of fully connected layers is 1, which is used to output 1 feature.

[0079] The preprocessed near-infrared spectral data and physical and chemical values ​​are divided into a training set and a prediction set in a ratio of 3:2, and input into the CNN network model quantitative model to obtain the prediction results. Figure 6 The relationship between the predicted values ​​and true values ​​of tea polyphenols and dihydromyricetin in the constructed CNN network model is shown. For tea polyphenols, Rp is 0.9080, RMSEP is 1.4078, and RPD is 2.4064; for dihydromyricetin, Rp is 0.9299, RMSEP is 1.2494, and RPD is 2.5807. This shows that the established CNN network model has good generalization performance.

[0080] Embodiment 2

[0081] Firstly, a portable spectral sensor device was used to collect the spectral information of berry tea in the range of 900-1700nm. The parameters for near-infrared spectral data acquisition were: spectral resolution 10nm, integration time 5ms, averaging times 8 times, and ambient temperature 25℃. After the acquisition was completed, the Derivative Spectroscopy derivative spectrum and Savitzky-Golay smoothing algorithm of the MATLAB third-party intelligent library were used to preprocess the spectrum. The derivative order of the Derivative Spectroscopy derivative spectrum was 1, the window width was 7, and the polynomial order of the Savitzky-Golay smoothing algorithm was 3, and the window width was 9.

[0082] After collection, the physicochemical values ​​of tea polyphenols (TP) and dihydromyricetin (DMY), the quality indicators of berry tea, were determined using conventional methods.

[0083] Step 1: The specific steps for preparing the berry tea powder sample using a grinder include:

[0084] 1) 80 g of berry tea was put into a small grinder. The grinder was used to grind each type of berry tea according to the principle of "press for 3 seconds and stop for 1 second". This operation was repeated 10 times to initially prepare powdered berry tea.

[0085] 2) Place the crushed powdered berry tea on a 60-mesh sieve and shake the sieve evenly to obtain a fine powdered berry tea sample.

[0086] Step 2: The specific steps of preparing tea soup by solvent extraction method include:

[0087] 1) Preparation of tea soup for tea polyphenol determination: First, weigh 0.2 g of berry tea powder sample, then add 5 mL of 70% methanol aqueous solution, extract in a 70°C water bath for 10 min, cool to room temperature, and centrifuge at 3500 r / min for 10 min to obtain a supernatant. Repeat the above operation for residue extraction, combine the extracts and make up to volume with water, mix well, and filter using a 0.45 μm membrane to obtain a tea soup sample.

[0088] 2) Preparation of tea soup for dihydromyricetin determination: First, weigh 0.5 g of berry tea powder sample, add 10 mL of 70% methanol aqueous solution, weigh and record, then place in a 50°C ultrasonic instrument for ultrasonic extraction for 30 min, take out and cool to room temperature and weigh again, use pure methanol to keep constant weight before ultrasonic extraction, mix well and filter using a 0.45 μm membrane to obtain a tea soup sample.

[0089] Step 3: The specific steps of using UV spectrophotometry to determine the physicochemical value of tea polyphenols in berry tea include:

[0090] First, prepare gallic acid working solutions with concentrations of 10, 20, 30, 40, 50, 60, 80, 100, and 120 μg / mL, and dilute the tea soup 100 times with water, then take 1.0 mL of the gallic acid working solution, water, and diluted tea soup, respectively, and then add the folin solution and sodium carbonate solution to each test tube at a volume ratio of 5:4, dilute with water and shake well, and place at room temperature for 60 minutes. Before using the ultraviolet spectrophotometer to measure tea polyphenols, the photometric measurement parameters are set as follows: the detection wavelength is 765 nm, the number of repeated measurements is 3, and the photometric mode is absorbance.

[0091] Step 4: The specific steps of using high performance liquid chromatography to determine the physicochemical value of dihydromyricetin in berry tea include:

[0092] First, dihydromyricetin working solution with a concentration of 100, 200, 300, 400, 500, and 700 μg / mL was prepared, and the tea soup was diluted with the initial mobile phase (acetonitrile: 0.1% phosphoric acid aqueous solution = 15:85) at a ratio of 1:10, and filtered using a 0.45 μm membrane after mixing. Before using a high performance liquid chromatograph to measure dihydromyricetin, the chromatographic column was replaced with a C18 column and a new detection method was established, with an elution gradient of 0-12min, 85% B; 12-20min, 85-70% B; 20-25min, 70-85% B; 25-35min, 85%, and the instrument detection parameters were set as follows: column temperature of 25°C, detection wavelength of 291nm, flow rate of 1mL / min, injection volume of 10-20μL, and ultraviolet detector was used for determination while keeping records.

[0093] Afterwards, a quantitative model for detecting the quality of berry tea is constructed based on the pre-processed near-infrared spectral data and physicochemical values.

[0094] A long short-term memory network (LSTM) quantitative prediction model is built based on the Deep Learning Toolbox framework. Its function expression is as follows:

[0095] LSTM(S tea1 , S tea2 ···S teaN )= TP S (3)

[0096] LSTM(S tea1 , S tea2 ···S teaN )= DMY S (4)

[0097] Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S;

[0098] The LSTM model structure is as follows Figure 4As shown, it includes an input layer, a hidden layer, an activation function layer, a fully connected layer and an output layer; the input layer of the LSTM model receives near-infrared spectral data, performs nonlinear transformation on the input data through the hidden layer, deeply mines the spectral information and extracts key features, outputs feature vectors to the fully connected layer, and uses the fully connected layer to map these extracted features to the output layer. The number of hidden layers is the first LSTM layer, the second LSTM layer and the corresponding first Dropout layer, the second Dropout layer; the number of neurons in the first LSTM layer is 4, and the Dropout layer randomly discards neurons with a discard ratio of 0.2; the number of neurons in the second LSTM layer is 16, and the Dropout layer randomly discards neurons with a discard ratio of 0.3; the activation function is the ReLU function; the number of fully connected layers is set to 1 layer to output 1 feature.

[0099] The preprocessed near-infrared spectral data and physical and chemical values ​​are divided into a training set and a prediction set in a ratio of 3:2, and input into the LSTM network model quantitative model to obtain the prediction results. Figure 7 The relationship between the predicted values ​​and true values ​​of tea polyphenols and dihydromyricetin in the constructed LSTM network model is shown. For tea polyphenols, Rp is 0.9205, RMSEP is 1.2494, and RPD is 2.5807; for dihydromyricetin, Rp is 0.9446, RMSEP is 0.8550, and RPD is 3.0731. This shows that the established LSTM network model has excellent generalization performance.

[0100] Embodiment 3

[0101] First, the specific contents of collecting berry tea spectral information and preprocessing using a portable spectral sensor device include:

[0102] 1) Near infrared spectroscopy data acquisition parameter settings: spectral acquisition range 900-1700nm, spectral resolution 10nm, integration time 5ms, average times 8 times, ambient temperature 25℃;

[0103] 2) After the acquisition is completed, the spectrum is preprocessed using the Derivative Spectroscopy derivative spectrum and Savitzky-Golay smoothing algorithm of the MATLAB third-party intelligent library. The derivative order of the Derivative Spectroscopy derivative spectrum is 1, the window width is 7, and the polynomial order of the Savitzky-Golay smoothing algorithm is 3, and the window width is 9.

[0104] After collection, the physicochemical values ​​of tea polyphenols (TP) and dihydromyricetin (DMY), the quality indicators of berry tea, were determined using conventional methods.

[0105] Process 1: The specific steps of preparing berry tea powder sample using a grinder include:

[0106] 1) 80 g of berry tea was put into a small grinder. The grinder was used to grind each type of berry tea according to the principle of "press for 3 seconds and stop for 1 second". This operation was repeated 10 times to initially prepare powdered berry tea.

[0107] 2) Place the crushed powdered berry tea on a 60-mesh sieve and shake the sieve evenly to obtain a fine powdered berry tea sample.

[0108] Process 2: The specific steps of preparing tea soup by solvent extraction method include:

[0109] 1) Preparation of tea soup for tea polyphenol determination: First, weigh 0.2 g of berry tea powder sample, then add 5 mL of 70% methanol aqueous solution, extract in a 70°C water bath for 10 min, cool to room temperature, and centrifuge at 3500 r / min for 10 min to obtain a supernatant. Repeat the above operation for residue extraction, combine the extracts and make up to volume with water, mix well, and filter using a 0.45 μm membrane to obtain a tea soup sample.

[0110] 2) Preparation of tea soup for dihydromyricetin determination: First, weigh 0.5 g of berry tea powder sample, add 10 mL of 70% methanol aqueous solution, weigh and record, then place in a 50°C ultrasonic instrument for ultrasonic extraction for 30 min, take out and cool to room temperature and weigh again, use pure methanol to keep constant weight before ultrasonic extraction, mix well and filter using a 0.45 μm membrane to obtain a tea soup sample.

[0111] Process 3: The specific steps of determining the physicochemical value of tea polyphenols in berry tea using UV spectrophotometry include:

[0112] First, prepare gallic acid working solutions with concentrations of 10, 20, 30, 40, 50, 60, 80, 100, and 120 μg / mL, and dilute the tea soup 100 times with water, then take 1.0 mL of the gallic acid working solution, water, and diluted tea soup, respectively, and then add the folin solution and sodium carbonate solution to each test tube at a volume ratio of 5:4, dilute with water and shake well, and place at room temperature for 60 minutes. Before using the ultraviolet spectrophotometer to measure tea polyphenols, the photometric measurement parameters are set as follows: the detection wavelength is 765 nm, the number of repeated measurements is 3, and the photometric mode is absorbance.

[0113] Process 4: The specific steps of determining the physicochemical value of dihydromyricetin in berry tea using high performance liquid chromatography include:

[0114] First, dihydromyricetin working solution with a concentration of 100, 200, 300, 400, 500, and 700 μg / mL was prepared, and the tea soup was diluted with the initial mobile phase (acetonitrile: 0.1% phosphoric acid aqueous solution = 15:85) at a ratio of 1:10, and filtered using a 0.45 μm membrane after mixing. Before using a high performance liquid chromatograph to measure dihydromyricetin, the chromatographic column was replaced with a C18 column and a new detection method was established, with an elution gradient of 0-12min, 85% B; 12-20min, 85-70% B; 20-25min, 70-85% B; 25-35min, 85%, and the instrument detection parameters were set as follows: column temperature of 25°C, detection wavelength of 291nm, flow rate of 1mL / min, injection volume of 10-20μL, and ultraviolet detector was used for determination while keeping records.

[0115] Afterwards, a quantitative model for detecting the quality of berry tea is constructed based on the pre-processed near-infrared spectral data and physicochemical values.

[0116] A one-dimensional convolutional neural network-long short-term memory network (CNN-LSTM) quantitative prediction model is built based on the Deep Learning Toolbox framework. Its function expression is as follows:

[0117] CNN-LSTM(S tea1 , S tea2 ···S teaN )= TP S (5)

[0118] CNN-LSTM(S tea1 , S tea2 ···S teaN )= DMY S (6)

[0119] Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S;

[0120] The CNN-LSTM model structure is as follows Figure 5As shown, the CNN-LSTM structure is an input layer, a sequence folding layer, two convolutional layers, an activation function layer, a sequence anti-folding layer, a flattening layer, two LSTM layers, two Dropout layers, a fully connected layer, and an output layer; the input layer of the CNN-LSTM model receives near-infrared spectral data, folds the input data into a form suitable for convolution operation through the sequence folding layer, extracts features of the spectral information through layer-by-layer convolution and outputs feature vectors, the sequence anti-folding layer converts the output feature vectors back into a sequence form suitable for LSTM processing, flattens the output data into a one-dimensional array using the flattening layer and outputs it to the LSTM layer, the LSTM layer processes the long-term dependencies in the sequence data and outputs feature vectors to the fully connected layer, and the fully connected layer is used to map these extracted features to the output layer. The parameter dimensions of the one-dimensional convolution layer are 16 and 32 respectively, the convolution kernel is 2, the convolution kernel size is 1*1, and the convolution step is 1; the activation function is the ReLU function; the number of neurons in the LSTM layer is 4 and 8 respectively; the dropout ratio of the Dropout layer is set to 0.2 and 0.3 respectively; the number of layers of the fully connected layer is set to 1 layer to output 1 feature.

[0121] The preprocessed near-infrared spectral data and physical and chemical values ​​are divided into a training set and a prediction set in a ratio of 3:2, and input into the CNN-LSTM network model quantitative model to obtain the prediction results. Figure 8 The relationship between the predicted values ​​and true values ​​of tea polyphenols and dihydromyricetin in the constructed CNN-LSTM network model is shown. For tea polyphenols, Rp is 0.9216, RMSEP is 1.1605, and RPD is 2.5978; for dihydromyricetin, Rp is 0.9723, RMSEP is 0.7103, and RPD is 4.3137. This shows that the established CNN-LSTM network model has excellent generalization performance.

[0122] The embodiments described above are only descriptions of the preferred methods of the present application, and are not intended to limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements made to the technical solutions of the present application by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present application.

Claims

1. A berry tea quality spectral sensing detection method based on a deep learning model, characterized in that: The following steps are involved: S1. Berry tea spectrum collection: Use a spectral sensor device to obtain spectral data of different berry tea samples, and use the Derivative Spectroscopy derivative spectrum and Savitzky-Golay smoothing algorithm of the MATLAB third-party intelligent library to process the collected near-infrared spectrum to reduce noise and baseline drift; S2. Conventional determination of berry tea quality physical and chemical indicators: first, a berry tea powder sample is prepared from the berry tea after the spectrum is collected using a grinder, then a solvent extraction method is used to prepare tea soup from the berry tea powder sample, and finally the physical and chemical values ​​of the berry tea quality indicators are determined using conventional methods; S3. Rapid detection of tea polyphenols and dihydromyricetin in berry tea: Based on the Deep Learning Toolbox framework, a one-dimensional convolutional neural network CNN, a long short-term memory network LSTM and a one-dimensional convolutional neural network-long short-term memory network CNN-LSTM berry tea quality prediction quantitative model were built, and the spectral data after derivative and smoothing preprocessing in S1 was input into the quantitative model to quickly output the detection results of tea polyphenols TP and dihydromyricetin DMY in berry tea.

2. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: The parameters for spectral data acquisition in step S1 are: spectral range 900-1700 nm, spectral resolution 10 nm, integration time 5-10 ms, averaging times 5-8 times, and ambient temperature 23-26°C.

3. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: In step S1, the derivative order of the Derivative Spectroscopy is 1-2, the window width is 5-11, and is set to an odd number. The polynomial order of the Savitzky-Golay smoothing algorithm is 2-3, and the window width is 5-15.

4. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: The content of using a grinder to prepare a berry tea powder sample from the berry tea after the spectrum is collected in step S2 specifically includes: Put an appropriate amount of berry tea into a small grinder and grind it using the grinder. The grinder grinds each berry tea according to the principle of "press for 3 seconds and stop for 1 second". Repeat this operation 8-10 times to initially prepare powdered berry tea. Place the crushed powdered berry tea in a 40-60 mesh sieve and shake the sieve evenly to obtain a fine powdered berry tea sample.

5. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: The steps of preparing tea soup by solvent extraction in step S2 specifically include: 1) Preparation of tea soup for tea polyphenol determination: Weigh a preset weight of berry tea powder into a centrifuge tube, add methanol solution for heating and extraction, cool to room temperature, centrifuge at low speed, and extract the supernatant into a volumetric flask. Repeat the above operation for the residue, combine the extracts and make up the volume, shake well, and filter using a filter membrane to obtain a tea soup sample. 2) Preparation of tea soup for dihydromyricetin determination: Weigh a preset weight of berry tea powder into a centrifuge tube, add methanol aqueous solution, record the weight and place it in an ultrasonic instrument for ultrasonic extraction, take it out and cool it to room temperature and weigh it again, use pure methanol to keep the weight before ultrasonic extraction constant, mix well and filter it with a filter membrane to obtain a tea soup sample.

6. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 5, characterized in that: In step 1), the berry tea powder sample with a preset weight of 0.1990-0.2009 g is added with a methanol aqueous solution with a volume of 4-5 mL, and is extracted in a 70±1°C water bath for 10 min. After cooling to room temperature, the supernatant is extracted by low-speed centrifugation. The residue extraction is repeated once according to the above operation, and the extracts are combined and fixed to volume with water. After mixing evenly, the mixture is filtered using a 0.45 μm membrane to obtain a tea soup sample. The concentration of the methanol aqueous solution is 70%; The low speed centrifugal rotation speed is 3500r / min; The centrifugation time is 10 min.

7. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 5, characterized in that: In step 2), the preset weight of the berry tea powder sample is 0.4990-0.5009g, the volume of the methanol aqueous solution added is 8-10mL, the weight is weighed and recorded, after mixing, it is placed in a 50±1℃ ultrasonic instrument for ultrasonic extraction, taken out and cooled to room temperature, and weighed again, and the weight before ultrasonic extraction is constant with pure methanol, and after mixing, it is filtered using a 0.45μm membrane to obtain a tea soup sample; The concentration of the methanol aqueous solution is 70%; The concentration of pure methanol is 99.9%; The ultrasonic extraction time is 30 min.

8. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: The conventional method used in step S2 is: using ultraviolet spectrophotometry to determine the physical and chemical values ​​of tea polyphenols in berry tea, specifically including: Prepare gallic acid working solution with concentrations of 10, 20, 30, 40, 50, 60, 80, 100, and 120 μg / mL; dilute the tea soup with water and shake well, take 0.5-1.0 mL of the gallic acid working solution, water, and diluted tea soup respectively, and then add the folin phenol solution and sodium carbonate solution to each test tube at a volume ratio of 5:4, dilute to volume with water and shake well, and place at room temperature for 50 min-65 min. Before using the ultraviolet spectrophotometer to measure tea polyphenols, the photometric measurement parameters are set as follows: the detection wavelength is 765 nm, the number of repeated measurements is 3, and the photometric mode is absorbance; The concentration of the Folin phenol solution is 0.1 mol / L; The concentration of the sodium carbonate solution is 7.5%; The dilution multiple of the tea soup is 100.

9. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: The conventional method used in step S2 is: the content of determining the physicochemical value of dihydromyricetin in berry tea by high performance liquid chromatography specifically includes: Prepare dihydromyricetin working solution with concentrations of 100, 200, 300, 400, 500, and 700 μg / mL; dilute the tea soup with the initial mobile phase according to the preset ratio, mix well, and filter again using a 0.45 μm membrane. Before using a high performance liquid chromatograph to measure dihydromyricetin, replace the chromatographic column with a C18 column and create a new detection method with an elution gradient of 0-12 min, 85% B; 12-20 min, 85-70% B; 20-25 min, 70-85% B; 25-35 min, 85%. The instrument detection parameters are set as follows: column temperature of 25°C, detection wavelength of 291 nm, flow rate of 1 mL / min, injection volume of 10-20 μL, and use a UV detector for measurement while keeping records; The initial mobile phase is acetonitrile: phosphoric acid aqueous solution with a volume ratio of 15:85; The preset ratio is that the volume ratio of tea soup: initial mobile phase is 1:9; The tea soup is diluted by 10 times; The B is a phosphoric acid aqueous solution with a concentration of 0.1%; The purity of the acetonitrile is 99.9%.

10. The method for detecting berry tea quality by spectral sensing based on a deep learning model according to claim 1, characterized in that: The function expression of the CNN quantitative prediction model constructed in step S3 is as follows: CNN(S tea1 , S tea2 ···S teaN )= TP S (1) CNN(S tea1 , S tea2 ···S teaN )= DMY S (2) Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S; the berry tea spectrum data is used as input and the corresponding TP data is used as output to train and predict the CNN model; The CNN model structure includes an input layer, two convolutional layers, two batch normalization layers, two Dropout layers, an activation function layer, two pooling layers, a fully connected layer and an output layer; the parameter dimensions of the one-dimensional convolutional layer are 32 and 64 respectively, the convolution kernel is 3, the convolution kernel size is 1*1, and the convolution step is 1; the dropout layer discard ratio is set to 0.2 and 0.3 respectively, that is, 20% and 30% of the neurons are randomly discarded respectively; the activation function is the ReLU function; the pooling kernel size of the pooling layer is 3 and 5 respectively; the number of fully connected layers is 1, which is used to output 1 feature; The function expression of the LSTM quantitative prediction model constructed in step S3 is as follows: LSTM(S tea1 , S tea2 ···S teaN )= TP S (3) LSTM(S tea1 , S tea2 ···S teaN )= DMY S (4) Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S; using the berry tea spectrum data as input and the corresponding TP data as output, the LSTM model is trained and predicted; The LSTM model structure includes an input layer, a hidden layer, an activation function layer, a fully connected layer and an output layer; the number of hidden layers is a first LSTM layer, a second LSTM layer and a corresponding first Dropout layer, a second Dropout layer; the number of neurons in the first LSTM layer is 4, and the Dropout layer randomly discards neurons with a discard ratio of 0.2; the number of neurons in the second LSTM layer is 16, and the Dropout layer randomly discards neurons with a discard ratio of 0.3; the activation function is a ReLU function; the number of fully connected layers is set to 1 layer to output 1 feature; The function expression of the CNN-LSTM quantitative prediction model constructed in step S3 is as follows: CNN-LSTM(S tea1 ,S tea2 ···S teaN )= TP S (5) CNN-LSTM(S tea1 ,S tea2 ···S teaN )= DMY S (6) Among them, S teaN Refers to the spectrum of the Nth berry tea, TP S Refers to the TP value corresponding to the spectrum S, DMY S Refers to the DMY value corresponding to the spectrum S; the berry tea spectrum data is used as input and the corresponding TP data is used as output to train and predict the CNN-LSTM model; The CNN-LSTM structure is an input layer, a sequence folding layer, two convolutional layers, an activation function layer, a sequence anti-folding layer, a flattening layer, two LSTM layers, two Dropout layers, a fully connected layer, and an output layer; the parameter dimensions of the one-dimensional convolutional layer are 16 and 32 respectively, the convolution kernel is 2, the convolution kernel size is 1*1, and the convolution step is 1; the activation function is a ReLU function; the number of neurons in the LSTM layer is 4 and 8 respectively; the discard ratio of the Dropout layer is set to 0.2 and 0.3 respectively; the number of layers of the fully connected layer is set to 1 layer to output 1 feature.

Images