CN115062656B - Tea polyphenol content prediction method and device based on electronic nose signal space domain - Google Patents
Tea polyphenol content prediction method and device based on electronic nose signal space domain Download PDFInfo
- Publication number
- CN115062656B CN115062656B CN202210658122.4A CN202210658122A CN115062656B CN 115062656 B CN115062656 B CN 115062656B CN 202210658122 A CN202210658122 A CN 202210658122A CN 115062656 B CN115062656 B CN 115062656B
- Authority
- CN
- China
- Prior art keywords
- electronic nose
- tea
- nose signal
- tea sample
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a tea polyphenol content prediction method and device based on an electronic nose signal space domain, wherein the method comprises the following steps: 1. acquiring an electronic nose signal of a tea sample and measuring the content of tea polyphenol; 2. extracting response values of the tea sample electronic nose signals in a stable state as tea sample electronic nose signal characteristics; 3. based on the tea polyphenol content of the tea sample, optimizing the signal characteristics of the electronic nose of the tea sample by utilizing the maximum information coefficient to obtain a corresponding optimized electronic nose signal of the tea sample; 4. converting the tea sample preferential electronic nose signal into a spatial domain image by a code conversion method; 5. and training and testing the spatial domain image based on the convolutional neural grid CNN to obtain a prediction result of the tea polyphenol content. The invention can convert the one-dimensional time sequence of the electronic nose signal into a two-dimensional image, and can excavate the spatial domain characteristics of the electronic nose signal, and simultaneously, the CNN network can be utilized to realize timely and accurate prediction of the tea polyphenol content.
Description
Technical Field
The invention relates to the field of nondestructive testing and image processing, in particular to a tea polyphenol content prediction method and device based on an electronic nose signal space domain.
Background
At present, a plurality of detection methods for tea polyphenol content exist, the traditional chemical detection method has the advantages of high destructiveness, complex operation, time and labor waste, the electronic nose is used as an odor sensor, the smell of a human body can be simulated, the objectivity is high, the response time is short, the detection speed is high, the accuracy is high, and the like, and the complex pretreatment process and the destructiveness of the traditional chemical method are not needed. The electronic nose signal analysis gradually becomes an important technical means for estimating the content of tea polyphenol, and effective feature extraction is carried out on the electronic nose signal, so that the prediction precision and efficiency can be improved, and the feature extraction of the electronic nose signal is mainly shown in two aspects of time domain and frequency domain features at present. The time domain features mainly extract characteristic values such as instantaneous values, average values, area values and the like of signal changes, can represent instantaneous characteristics, overall response trend and the like of signals, can extract frequency domain characteristics after original time domain signals are transformed into frequency domains through Fourier transformation or wavelet transformation, and can represent main stream characteristics and overall levels of sensor signals. However, the method ignores the details of the signal change of the electronic nose, loses the associated information between the time sampling points of the original signals, and loses certain prediction precision.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides a tea polyphenol content prediction method and device based on an electronic nose signal space domain, so that the advantages of deep learning in the image field can be fully utilized, and the potential space domain characteristics of the electronic nose signal can be excavated, so that the prediction precision of the tea polyphenol content can be effectively improved.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
the invention discloses a tea polyphenol content prediction method based on an electronic nose signal space domain, which is characterized by comprising the following steps of:
step 1, acquiring an electronic nose signal of a tea sample and measuring the content of tea polyphenol, wherein the electronic nose signal comprises m sensor arrays, each sensor array comprises a plurality of sampling points, and each sampling point corresponds to a response value of the electronic nose signal;
step 2, extracting response values of the tea sample electronic nose signals in a stable state according to response trend of the tea sample electronic nose signals, so as to obtain m electronic nose signal characteristics;
step 3, optimizing the m electronic nose signal characteristics by using the maximum information coefficient based on the tea polyphenol content of the tea sample to obtain optimized electronic nose signals corresponding to the H optimized characteristics;
step 4, converting the tea sample preferential electronic nose signal into a spatial domain image through a code conversion method;
step 5, dividing the spatial domain image of the tea sample electronic nose signal constructed after the electronic nose signal is encoded into a training set spatial domain image and a test set spatial domain image;
step 6, training the training set spatial domain image of the tea sample electronic nose signal based on a CNN model to obtain a trained tea polyphenol content prediction model;
and 7, processing the spatial domain image of the test set of the tea sample to be predicted by using the trained tea polyphenol content prediction model, and outputting a prediction result of the tea polyphenol content.
The method for predicting the tea polyphenol content based on the electronic nose signal space domain is also characterized in that the step 3 comprises the following steps:
step 3.1, respectively forming a corresponding two-dimensional data set by the signal characteristics of each electronic nose of a tea sample and the tea polyphenol content of the tea sample;
step 3.2, dividing a scatter diagram formed by any two-dimensional data set into grids with different rows and columns, thereby obtaining grids with different division forms;
step 3.3, calculating mutual information values corresponding to grids in each division form respectively, and carrying out normalization processing on each mutual information value;
step 3.4, taking the normalized maximum mutual information value as the maximum information coefficient value of the corresponding two-dimensional data set;
and 3.5, sorting the maximum information coefficients of the m two-dimensional data sets in a descending order, and selecting the electronic nose signal characteristics corresponding to the first H maximum information coefficients as H preferred characteristics.
The step 4 includes:
step 4.1, carrying out normalization processing on the optimized tea sample electronic nose signal to obtain a normalized tea sample optimized electronic nose signal;
step 4.2, converting the normalized tea sample preferential electronic nose signal under the Cartesian coordinate system into a polar coordinate system by using an inverse trigonometric function formula, so as to obtain the radius and the angle corresponding to each tea sample preferential electronic nose signal under the polar coordinate; wherein, the radius represents a time stamp, and the angle represents a response value of the tea sample, preferably an electronic nose signal;
step 4.3, after the angle value of each preferable electronic nose signal of the tea sample under the polar coordinate system is respectively added with the angle values of other sampling points, the cosine value of the addition result is taken to form a gram and angle field matrix corresponding to the preferable electronic nose signal;
step 4.4, after the angle value of each preferable electronic nose signal of the tea sample under the polar coordinate system is respectively added with the angle values of other sampling points, the sine value of the addition result is taken to form a Graham difference angle field matrix corresponding to the preferable electronic nose signal;
step 4.5, respectively carrying out weighted average operation on the gram and the angular field matrix and the gram difference angular field matrix of each preferential electronic nose signal of the tea sample to obtain a weighted average fusion matrix of the preferential electronic nose signals of the corresponding tea sample;
and 4.6, taking a weighted average fusion matrix obtained by each preferable electronic nose signal of the tea sample as one channel of the image, so as to obtain a spatial domain image.
The step 6 comprises the following steps:
step 6.1, performing convolution operation on the training set spatial domain image by utilizing convolution check to obtain a training set feature map of the training set spatial domain image;
step 6.2, performing downsampling operation on the training set feature map by using a pooling layer to obtain training set spatial domain features of the electronic nose signals of the tea samples subjected to dimension reduction;
step 6.3, integrating the training set spatial domain features of the electronic nose signals of the tea samples subjected to dimension reduction by utilizing a full-connection layer to obtain a predicted value of the tea polyphenol content of the tea samples;
step 6.4, training the CNN model by using a gradient descent method, calculating a MSE loss function for updating model parameters, and stopping training when the loss function converges, so as to obtain a trained tea polyphenol content prediction model;
the invention relates to a tea polyphenol content prediction device based on an electronic nose signal space domain, which is characterized by comprising the following components: an acquisition unit, a feature extraction unit, a preference unit, a spatial domain image generation unit, a training unit and a prediction unit, wherein,
the acquisition unit is used for acquiring an electronic nose signal of a tea sample and measuring the content of tea polyphenol, wherein the electronic nose signal comprises a plurality of sensor arrays, each sensor array comprises a plurality of sampling points, and each sampling point has a response value;
the characteristic extraction unit is used for extracting a response value of each electronic nose signal of the tea sample in a steady state according to a response trend of the electronic nose signal of the tea sample and taking the response value as a characteristic of the electronic nose signal of the tea sample;
the optimizing unit is used for optimizing the characteristics of the tea sample electronic nose signal by utilizing the maximum information coefficient based on the tea polyphenol content of the tea sample to obtain a tea sample optimizing electronic nose signal corresponding to the optimizing characteristics;
the spatial domain image generation unit is used for encoding the tea sample, preferably the electronic nose signal, into a spatial domain image;
the training unit is used for training the spatial domain image of the tea sample electronic nose signal based on a CNN model to obtain a trained tea polyphenol content prediction model;
the prediction unit is used for processing the spatial domain image of the tea sample to be predicted by using the tea polyphenol content prediction model and outputting a prediction result of the tea polyphenol content.
The tea polyphenol content prediction device provided by the invention is characterized in that the spatial domain image generation unit comprises the following processes:
normalizing the optimized tea sample electronic nose signal to obtain a normalized tea sample optimized electronic nose signal;
converting the normalized tea sample preferential electronic nose signal under the Cartesian coordinate system into a polar coordinate system by utilizing an inverse trigonometric function formula, thereby obtaining the radius and the angle corresponding to each tea sample preferential electronic nose signal under the polar coordinate; wherein, the radius represents a time stamp, and the angle represents a response value of the tea sample, preferably an electronic nose signal;
adding the angle value of each preferable electronic nose signal of the tea sample in the polar coordinate system with the angle values of other sampling points respectively, and then taking the cosine value of the addition result to form a gram and angle field matrix corresponding to the preferable electronic nose signal;
adding the angle value of each preferable electronic nose signal of the tea sample in the polar coordinate system with the angle values of other sampling points respectively, and then taking the sine value of the addition result to form a gram difference angle field matrix corresponding to the preferable electronic nose signal;
respectively carrying out weighted average operation on the gram and the angular field matrix and the gram difference angular field matrix of each preferential electronic nose signal of the tea sample to obtain a weighted average fusion matrix of the preferential electronic nose signals of the corresponding tea sample;
and taking a weighted average fusion matrix obtained by each preferable electronic nose signal of the tea sample as one channel of the image, thereby obtaining a spatial domain image.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the method, the one-dimensional sequence of the electronic nose signal is encoded into the two-dimensional image, the tea polyphenol content deep learning prediction model based on the spatial domain features is constructed, the advantages of deep learning in the image field are fully utilized, the potential spatial domain features of the electronic nose signal are excavated, and therefore the prediction precision of the tea polyphenol content is improved;
2. according to the invention, the electronic nose technology is adopted to collect the fragrance of tea, and the tea polyphenol content is directly predicted by constructing a model, so that a tea sample is not damaged, the prediction efficiency is greatly improved, and the real-time performance is realized;
3. according to the invention, the weighted average operation is carried out on the gram and angular field matrixes and the gram difference angular field matrixes, and the complementary information of the two matrixes is combined, so that the repeated information of the spatial domain characteristics can be reduced or inhibited, and the spatial domain characteristics of the electronic nose signals can be utilized to the greatest extent.
Drawings
FIG. 1 is a flow chart of a method for predicting tea polyphenol content based on electronic nose signal space domain;
FIG. 2 is a transcoded image based on an electronic nose signal in accordance with the present invention;
fig. 3 is a schematic diagram of a tea polyphenol content predicting device based on an electronic nose signal space domain.
Detailed Description
In this embodiment, referring to fig. 1, a method for predicting the tea polyphenol content based on the electronic nose signal space domain is performed according to the following steps:
step 1, acquiring an electronic nose signal of a tea sample and measuring the content of tea polyphenol, wherein the electronic nose signal comprises a plurality of sensor arrays, each sensor array comprises a plurality of sampling points, and each sampling point has a response value;
in this embodiment, a PEN 3-type portable electronic nose is adopted, the electronic nose includes ten sensors, sensitive compounds corresponding to each sensor are different, three kinds of green tea samples of Houkui, huangshan Mao Feng and Liu an Guapian are obtained, 90 green tea samples are obtained, 5g of each green tea sample is placed in a 100mL glass beaker for sealing and placing for 30min, sampling interval time is 1s, sensor cleaning time is 100s, sensor return time is 10s, and sampling time is 75s.
Step 2, extracting response values of the tea sample electronic nose signals in a stable state according to response trend of the tea sample electronic nose signals, so as to obtain m electronic nose signal characteristics;
step 3, optimizing the m electronic nose signal characteristics by using the maximum information coefficient based on the tea polyphenol content of the tea sample to obtain optimized electronic nose signals corresponding to the H optimized characteristics;
step 3.1, respectively forming a corresponding two-dimensional data set by the signal characteristics of each electronic nose of the tea sample and the tea polyphenol content of the tea sample;
step 3.2, dividing a scatter diagram formed by the two-dimensional data set into grids with different rows and columns, so as to obtain grids with different division forms;
step 3.3, calculating mutual information values corresponding to grids in each division form respectively, and carrying out normalization processing on each mutual information value;
step 3.4, taking the normalized maximum mutual information value as the maximum information coefficient value of the corresponding two-dimensional data set;
and 3.5, sorting the maximum information coefficients of the two-dimensional data set in a descending order, and selecting the characteristics of the electronic nose signals corresponding to the first H maximum information coefficients, so as to select the first H electronic nose signals.
In this embodiment, 3 electronic nose signals are preferable from among 10 electronic nose signals. Let the vector of all sample points of each electronic nose signal time domain feature be a, the vector of all sample points of tea polyphenol content be b, the calculation formula of mutual information I (a, b) between the two is shown as formula (1):
in the formula (1), p (a) represents the probability density of the time domain feature a, p (b) represents the probability density of the tea polyphenol content b, and p (a, b) represents the joint probability density function of the time domain feature a and the tea polyphenol content b, so that the calculation of MIC is realized by a discrete method. Assuming that D (a, b) is a two-dimensional data set formed by a time domain feature a and a tea polyphenol content b, dividing a two-dimensional space into X sections and Y sections in the X, Y directions to form x×y grids G, and defining that a set of grids G based on different division forms is Ω:
in the formula (2), d|g represents the distribution of the data set D on the divided grid G. I * (D, x, y) represents the mutual information of the x-row and y-column grid.
The maximum standardization of the data set D under different division modes is formed into a feature matrix M (D), as shown in a formula (3):
in the formula (3), min { x, y } represents the minimum x row and y column grid, M (D) x,y Representing the normalized mutual information value of the x-row y-column grid.
Solving the maximum value of the feature matrix in the upper limit value of the grid division number to obtain the maximum information coefficient, wherein the maximum information coefficient is shown in the formula (4):
in the formulas (4) and (5), B (n) is a grid division number upper limit value, and MIC (D) represents a maximum normalized mutual information value of a grid of a non-line division form.
Step 4, converting the tea sample preferential electronic nose signal into a spatial domain image through a code conversion method;
step 4.1, carrying out normalization processing on the optimized tea sample electronic nose signal to obtain a normalized tea sample optimized electronic nose signal;
step 4.2, converting the normalized tea sample preferential electronic nose signal under the Cartesian coordinate system into a polar coordinate system by using an inverse trigonometric function formula, so as to obtain the radius and the angle corresponding to each tea sample preferential electronic nose signal under the polar coordinate; wherein, the radius represents a time stamp, and the angle represents a response value of the tea sample, preferably an electronic nose signal;
step 4.3, after the angle value of each preferable electronic nose signal of the tea sample under the polar coordinate system is respectively added with the angle values of other sampling points, the cosine value of the addition result is taken to form a gram and angle field matrix corresponding to the preferable electronic nose signal;
step 4.4, after the angle value of each preferable electronic nose signal of the tea sample under the polar coordinate system is respectively added with the angle values of other sampling points, the sine value of the addition result is taken to form a Graham difference angle field matrix corresponding to the preferable electronic nose signal;
step 4.5, respectively carrying out weighted average operation on the gram and the angular field matrix and the gram difference angular field matrix of each preferential electronic nose signal of the tea sample to obtain a weighted average fusion matrix of the preferential electronic nose signals of the corresponding tea sample;
and 4.6, taking a weighted average fusion matrix obtained by each preferable electronic nose signal of the tea sample as one channel of the image, so as to obtain a spatial domain image.
In the present example, for a one-dimensional time series of electronic nose signals, x= { X 1 ,x 2 ,···,x i ,···,x n Equal proportion mapping each sampling point signal of the electronic nose signal to [ -1,1 }]The interval is represented by the following formula (5) and formula (6):
X′＝[x′ 1 ,x′ 2 ,···,x′ i ,···,x′ n ],-1≤x i ′≤1 (6)
in the formula (5) and the formula (6), x i For the response value of the ith sample point, x i 'is the response value of the ith after normalization, n is the total sampling point number, and X' is the time sequence after normalization.
Mapping the obtained new electronic nose signal time sequence to a polar coordinate system, and obtaining an angle corresponding to each sequence point by using an inverse trigonometric function, wherein the sequence value range under the polar coordinate is [0, pi ], as shown in the formula (7) and the formula (8):
x″ i ＝arccos(x′ i ),-1≤x′ i ≤1 (7)
X″＝[x″ 1 ,x″ 2 ,···,x″ n ],0≤x″ 1 ≤π (8)
in the formulas (7) and (8), x i The ith response value of the electronic nose signal is converted into an angle value under the polar coordinate, and X' is a new sequence of converting the electronic nose time sequence into the polar coordinate.
Adding the time sequences under the polar coordinate system, taking cosine values, and summarizing to obtain a gram and an angular field matrix GASF, wherein the gram and the angular field matrix GASF are shown as a formula (9):
adding the time sequences under the polar coordinate system, and collecting sine values to obtain a Gramm difference angle field matrix GADF, wherein the Gramm difference angle field matrix GADF is shown as a formula (10);
carrying out weighted average summation on the two matrixes to obtain a target matrix Fusion, wherein the target matrix Fusion is shown in a formula (11);
and taking a weighted average fusion matrix obtained by the time sequence of each electronic nose signal of the tea sample as an image channel to obtain a spatial domain image. The tea sample electronic nose signal is encoded into a two-dimensional image, the time dependence of the electronic nose signal can be maintained through the transformation, the texture information and the color distribution information of the two-dimensional image can reflect the potential spatial domain information of the one-dimensional signal, meanwhile, the technical effect of improving the accuracy and the efficiency of the tea polyphenol predicted content is achieved by utilizing a CNN model, and the encoded image of the electronic nose signal is shown in fig. 2.
Step 5, dividing a tea sample electronic nose signal space domain image constructed after electronic nose signal coding into a training set space domain image and a test set space domain image;
step 6, training a spatial domain image of a tea sample electronic nose signal training set based on a CNN model to obtain a trained tea polyphenol content prediction model;
step 6.1, performing convolution operation on the training set spatial domain image by utilizing convolution check to obtain a training set feature map of the training set spatial domain image;
step 6.2, performing downsampling operation on the training set feature map by using a pooling layer to obtain the space domain features of the tea sample electronic nose signal training set after dimension reduction;
step 6.3, integrating spatial domain features of the electronic nose signal training set of the tea sample subjected to dimension reduction by utilizing the full-connection layer to obtain a predicted value of the tea polyphenol content of the tea sample;
step 6.4, training the CNN model by using a gradient descent method, calculating a MSE loss function for updating model parameters, and stopping training when the loss function converges, so as to obtain a trained tea polyphenol content prediction model;
in the example of the present invention, a CNN network is used, the first convolution layer has 128 filters with a size of 3 x 3, a stride of 1, then an average pooling layer of 3 x 3, the second convolution layer has 64 filters with a size of 3 x 3, a stride of 1, then an average pooling layer of 3 x 3, the second convolution layer has 32 filters with a size of 3 x 3, a stride of 1, then an average pooling layer of 3 x 3, the activation function used by each convolution layer is relu, dropout is used after each pooling layer to prevent overfitting, and finally the predicted value of tea polyphenol content is output through the full connection layer. The model was set to a Batch size (batch_size) of 24, training times (epochs) of 200, an optimizer of Adam was used, and a loss function of mse was used.
And 7, processing the spatial domain image of the test set of the tea sample to be predicted by using the trained tea polyphenol content prediction model, and outputting a prediction result of the tea polyphenol content.
Referring to fig. 3, in this embodiment, a tea polyphenol content prediction apparatus based on an electronic nose signal space domain includes: an acquisition unit, a feature extraction unit, a preference unit, a spatial domain image generation unit, a training unit and a prediction unit, wherein,
the electronic nose signal comprises a plurality of sensor arrays, each sensor array comprises a plurality of sampling points, and each sampling point has a response value;
the characteristic extraction unit is used for extracting a response value of each electronic nose signal of the tea sample in a steady state according to the response trend of the electronic nose signal of the tea sample and taking the response value as the characteristic of the electronic nose signal of the tea sample;
the optimizing unit is used for optimizing the characteristics of the tea sample electronic nose signal by utilizing the maximum information coefficient based on the tea polyphenol content of the tea sample, and obtaining a tea sample optimizing electronic nose signal corresponding to the optimized characteristics;
the spatial domain image generating unit is used for encoding the tea sample optimal electronic nose signal into a spatial domain image through a code conversion method; specifically, the spatial domain image generating unit performs normalization processing on the optimized tea sample electronic nose signal to obtain an optimized tea sample electronic nose signal; secondly, converting the normalized tea sample preferential electronic nose signal under the Cartesian coordinate system into a polar coordinate system by using an inverse trigonometric function formula, thereby obtaining the radius and the angle corresponding to each tea sample preferential electronic nose signal under the polar coordinate; wherein, the radius represents a time stamp, and the angle represents a response value of the tea sample, preferably an electronic nose signal; then, adding the angle value of each preferable electronic nose signal of the tea sample in the polar coordinate system with the angle values of other sampling points respectively, and then taking the cosine value of the addition result to form a gram and angle field matrix corresponding to the preferable electronic nose signal; then, adding the angle value of each preferable electronic nose signal of the tea sample in the polar coordinate system with the angle values of other sampling points respectively, and then taking the sine value of the addition result to form a gram difference angle field matrix corresponding to the preferable electronic nose signal; respectively carrying out weighted average operation on the gram and the angular field matrix and the gram difference angular field matrix of each preferential electronic nose signal of the tea sample to obtain a weighted average fusion matrix of the preferential electronic nose signals of the corresponding tea sample; and finally taking a weighted average fusion matrix obtained by each preferable electronic nose signal of the tea sample as one channel of the image, thereby obtaining a spatial domain image.
The training unit is used for training the CNN model based on the spatial domain image to obtain a trained tea polyphenol content prediction model;
the prediction unit is used for processing the spatial domain image of the tea sample to be predicted by using the tea polyphenol content prediction model and outputting a prediction result of the tea polyphenol content.
The system fully utilizes the advantage of deep learning in the image field, not only maintains the time dependence of the electronic nose signal, but also digs the potential spatial domain characteristics of the electronic nose signal, thereby improving the prediction accuracy of the tea polyphenol content.
Claims (4)
1. A tea polyphenol content prediction method based on an electronic nose signal space domain is characterized by comprising the following steps:
step 1, acquiring an electronic nose signal of a tea sample and measuring the content of tea polyphenol, wherein the electronic nose signal comprises m sensor arrays, each sensor array comprises a plurality of sampling points, and each sampling point corresponds to a response value of the electronic nose signal;
step 2, extracting response values of the tea sample electronic nose signals in a stable state according to response trend of the tea sample electronic nose signals, so as to obtain m electronic nose signal characteristics;
step 3, optimizing the m electronic nose signal characteristics by using the maximum information coefficient based on the tea polyphenol content of the tea sample to obtain optimized electronic nose signals corresponding to the H optimized characteristics;
step 4, converting the tea sample preferential electronic nose signal into a spatial domain image through a code conversion method;
step 4.1, carrying out normalization processing on the optimized tea sample electronic nose signal to obtain a normalized tea sample optimized electronic nose signal;
step 4.2, converting the normalized tea sample preferential electronic nose signal under the Cartesian coordinate system into a polar coordinate system by using an inverse trigonometric function formula, so as to obtain the radius and the angle corresponding to each tea sample preferential electronic nose signal under the polar coordinate; wherein, the radius represents a time stamp, and the angle represents a response value of the tea sample, preferably an electronic nose signal;
step 4.3, after the angle value of each preferable electronic nose signal of the tea sample under the polar coordinate system is respectively added with the angle values of other sampling points, the cosine value of the addition result is taken to form a gram and angle field matrix corresponding to the preferable electronic nose signal;
step 4.4, after the angle value of each preferable electronic nose signal of the tea sample under the polar coordinate system is respectively added with the angle values of other sampling points, the sine value of the addition result is taken to form a Graham difference angle field matrix corresponding to the preferable electronic nose signal;
step 4.5, respectively carrying out weighted average operation on the gram and the angular field matrix and the gram difference angular field matrix of each preferential electronic nose signal of the tea sample to obtain a weighted average fusion matrix of the preferential electronic nose signals of the corresponding tea sample;
step 4.6, taking a weighted average fusion matrix obtained by each preferable electronic nose signal of the tea sample as a channel of the image, so as to obtain a spatial domain image;
step 5, dividing the spatial domain image of the tea sample electronic nose signal constructed after the electronic nose signal is encoded into a training set spatial domain image and a test set spatial domain image;
step 6, training the training set spatial domain image of the tea sample electronic nose signal based on a CNN model to obtain a trained tea polyphenol content prediction model;
and 7, processing the spatial domain image of the test set of the tea sample to be predicted by using the trained tea polyphenol content prediction model, and outputting a prediction result of the tea polyphenol content.
2. The method for predicting tea polyphenol content based on electronic nose signal space domain according to claim 1, wherein the step 3 comprises:
step 3.1, respectively forming a corresponding two-dimensional data set by the signal characteristics of each electronic nose of a tea sample and the tea polyphenol content of the tea sample;
step 3.2, dividing a scatter diagram formed by any two-dimensional data set into grids with different rows and columns, thereby obtaining grids with different division forms;
step 3.3, calculating mutual information values corresponding to grids in each division form respectively, and carrying out normalization processing on each mutual information value;
step 3.4, taking the normalized maximum mutual information value as the maximum information coefficient value of the corresponding two-dimensional data set;
and 3.5, sorting the maximum information coefficients of the m two-dimensional data sets in a descending order, and selecting the electronic nose signal characteristics corresponding to the first H maximum information coefficients as H preferred characteristics.
3. The method for predicting tea polyphenol content based on electronic nose signal space domain according to claim 1, wherein the step 6 comprises:
step 6.1, performing convolution operation on the training set spatial domain image by utilizing convolution check to obtain a training set feature map of the training set spatial domain image;
step 6.2, performing downsampling operation on the training set feature map by using a pooling layer to obtain training set spatial domain features of the electronic nose signals of the tea samples subjected to dimension reduction;
step 6.3, integrating the training set spatial domain features of the electronic nose signals of the tea samples subjected to dimension reduction by utilizing a full-connection layer to obtain a predicted value of the tea polyphenol content of the tea samples;
and 6.4, training the CNN model by using a gradient descent method, calculating a MSE loss function for updating model parameters, and stopping training when the loss function converges, so as to obtain a trained tea polyphenol content prediction model.
4. Tea polyphenol content prediction device based on electronic nose signal space domain, characterized by comprising: an acquisition unit, a feature extraction unit, a preference unit, a spatial domain image generation unit, a training unit and a prediction unit, wherein,
the acquisition unit is used for acquiring an electronic nose signal of a tea sample and measuring the content of tea polyphenol, wherein the electronic nose signal comprises a plurality of sensor arrays, each sensor array comprises a plurality of sampling points, and each sampling point has a response value;
the characteristic extraction unit is used for extracting a response value of each electronic nose signal of the tea sample in a steady state according to a response trend of the electronic nose signal of the tea sample and taking the response value as a characteristic of the electronic nose signal of the tea sample;
the optimizing unit is used for optimizing the characteristics of the tea sample electronic nose signal by utilizing the maximum information coefficient based on the tea polyphenol content of the tea sample to obtain a tea sample optimizing electronic nose signal corresponding to the optimizing characteristics;
the spatial domain image generation unit is used for encoding the tea sample, preferably the electronic nose signal, into a spatial domain image;
the spatial domain image generation unit includes the following processes:
normalizing the optimized tea sample electronic nose signal to obtain a normalized tea sample optimized electronic nose signal;
converting the normalized tea sample preferential electronic nose signal under the Cartesian coordinate system into a polar coordinate system by utilizing an inverse trigonometric function formula, thereby obtaining the radius and the angle corresponding to each tea sample preferential electronic nose signal under the polar coordinate; wherein, the radius represents a time stamp, and the angle represents a response value of the tea sample, preferably an electronic nose signal;
adding the angle value of each preferable electronic nose signal of the tea sample in the polar coordinate system with the angle values of other sampling points respectively, and then taking the cosine value of the addition result to form a gram and angle field matrix corresponding to the preferable electronic nose signal;
adding the angle value of each preferable electronic nose signal of the tea sample in the polar coordinate system with the angle values of other sampling points respectively, and then taking the sine value of the addition result to form a gram difference angle field matrix corresponding to the preferable electronic nose signal;
respectively carrying out weighted average operation on the gram and the angular field matrix and the gram difference angular field matrix of each preferential electronic nose signal of the tea sample to obtain a weighted average fusion matrix of the preferential electronic nose signals of the corresponding tea sample;
taking a weighted average fusion matrix obtained by each preferable electronic nose signal of the tea sample as one channel of the image, thereby obtaining a spatial domain image;
the training unit is used for training the spatial domain image of the tea sample electronic nose signal based on a CNN model to obtain a trained tea polyphenol content prediction model;
the prediction unit is used for processing the spatial domain image of the tea sample to be predicted by using the tea polyphenol content prediction model and outputting a prediction result of the tea polyphenol content.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210658122.4A CN115062656B (en) | 2022-06-10 | 2022-06-10 | Tea polyphenol content prediction method and device based on electronic nose signal space domain |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210658122.4A CN115062656B (en) | 2022-06-10 | 2022-06-10 | Tea polyphenol content prediction method and device based on electronic nose signal space domain |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115062656A CN115062656A (en) | 2022-09-16 |
| CN115062656B true CN115062656B (en) | 2023-08-11 |
Family
ID=83199596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210658122.4A Active CN115062656B (en) | 2022-06-10 | 2022-06-10 | Tea polyphenol content prediction method and device based on electronic nose signal space domain |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115062656B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108694375A (en) * | 2018-03-30 | 2018-10-23 | 天津大学 | A kind of image conversion white wine recognition methods can be used for polyelectron nose platform |
| CN110133049A (en) * | 2019-04-10 | 2019-08-16 | 浙江大学 | Tea grades fast non-destructive detection method based on electronic nose and machine vision |
| CN110378229A (en) * | 2019-06-19 | 2019-10-25 | 浙江大学 | A kind of electronic nose data characteristics selection method based on filter-wrapper frame |
| CN112370015A (en) * | 2020-10-30 | 2021-02-19 | 复旦大学 | Physiological signal quality evaluation method based on gram angular field |
| CN113158980A (en) * | 2021-05-17 | 2021-07-23 | 四川农业大学 | Tea leaf classification method based on hyperspectral image and deep learning |
| CN114048769A (en) * | 2021-11-08 | 2022-02-15 | 太原科技大学 | Multi-source multi-domain information entropy fusion and model self-optimization method for bearing fault diagnosis |
| EP3997457A1 (en) * | 2019-07-12 | 2022-05-18 | Commissariat à l'énergie atomique et aux énergies alternatives | Detection system for an electronic nose allowing a physicochemical classification of odors and electronic nose comprising such a system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102823236B (en) * | 2010-06-09 | 2016-01-20 | 富士胶片株式会社 | Imaging device and image processing method |
-
2022
- 2022-06-10 CN CN202210658122.4A patent/CN115062656B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108694375A (en) * | 2018-03-30 | 2018-10-23 | 天津大学 | A kind of image conversion white wine recognition methods can be used for polyelectron nose platform |
| CN110133049A (en) * | 2019-04-10 | 2019-08-16 | 浙江大学 | Tea grades fast non-destructive detection method based on electronic nose and machine vision |
| CN110378229A (en) * | 2019-06-19 | 2019-10-25 | 浙江大学 | A kind of electronic nose data characteristics selection method based on filter-wrapper frame |
| EP3997457A1 (en) * | 2019-07-12 | 2022-05-18 | Commissariat à l'énergie atomique et aux énergies alternatives | Detection system for an electronic nose allowing a physicochemical classification of odors and electronic nose comprising such a system |
| CN112370015A (en) * | 2020-10-30 | 2021-02-19 | 复旦大学 | Physiological signal quality evaluation method based on gram angular field |
| CN113158980A (en) * | 2021-05-17 | 2021-07-23 | 四川农业大学 | Tea leaf classification method based on hyperspectral image and deep learning |
| CN114048769A (en) * | 2021-11-08 | 2022-02-15 | 太原科技大学 | Multi-source multi-domain information entropy fusion and model self-optimization method for bearing fault diagnosis |
Non-Patent Citations (1)
| Title |
|---|
| 基于光谱-空间特征的黄茶多酚含量估算模型;杨宝华等;《光谱学与光谱分析》;第41卷(第3期);第936-942页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115062656A (en) | 2022-09-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107529222B (en) | WiFi indoor positioning system based on deep learning | |
| CN109784552B (en) | Re-ESF algorithm-based construction method of space variable coefficient PM2.5 concentration estimation model | |
| CN109636171A (en) | A kind of comprehensive diagnos and risk evaluating method that regional vegetation restores | |
| CN111369057A (en) | Air quality prediction optimization method and system based on deep learning | |
| CN111242377A (en) | Short-term wind speed prediction method integrating deep learning and data denoising | |
| CN113935249B (en) | Upper-layer ocean thermal structure inversion method based on compression and excitation network | |
| CN115423163A (en) | Method and device for predicting short-term flood events of drainage basin and terminal equipment | |
| CN116186587A (en) | Edge end fault diagnosis method and system based on depth migration | |
| CN116363521A (en) | Semantic prediction method for remote sensing image | |
| CN111951289A (en) | BA-Unet-based underwater sonar image data segmentation method | |
| CN117233870B (en) | Short-term precipitation set forecasting and downscaling method based on multiple meteorological elements | |
| CN116796799A (en) | Method for creating small-river basin flood rainfall threshold model in area without hydrologic data | |
| CN115062656B (en) | Tea polyphenol content prediction method and device based on electronic nose signal space domain | |
| CN116363536B (en) | Unmanned aerial vehicle inspection data-based power grid infrastructure equipment defect archiving method | |
| CN116050460B (en) | Air temperature data spatial interpolation method based on attention neural network | |
| CN111488974A (en) | Deep learning neural network-based ocean wind energy downscaling method | |
| CN116704350A (en) | Water area change monitoring method and system based on high-resolution remote sensing image and electronic equipment | |
| CN102880753A (en) | Method for converting land utilization spatial characteristic scale based on fractal dimension | |
| CN115952743A (en) | Multi-source precipitation data collaborative downscaling method and system coupled with random forest and HASM | |
| CN115545296A (en) | LIP-TCN-LSTM-based urban waterlogging water level short-term prediction method and system | |
| CN115049922A (en) | Method and system for detecting change of remote sensing image | |
| CN109741239B (en) | Multi-spatial-scale information extraction method based on soil quality parameters | |
| CN113919533A (en) | Air quality tracing and forecasting method based on artificial intelligence | |
| CN113610302B (en) | Surface temperature downscaling method based on gradual clustering and cyclic neural network | |
| CN117878928B (en) | Wind power prediction method and device based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |