Research on the Flavor Quality Model of Chicken Essence Seasoning Based on PCA and PSO-SVM Algorithm

Abstract: Objective: To propose a flavor quality model of chicken essence seasoning based on principal component analysis and particle swarm optimization-support vector machine algorithm
.
Method: Taking 6 kinds of chicken essence seasoning flavor sensory data as the research object, the principal component analysis of the flavor components of chicken essence seasoning was carried out to reduce the data dimension; the particle swarm optimization algorithm was used to obtain the best parameters of the support vector machine, and the support vector machine was used Complete the training and classification of the flavor data of chicken essence seasoning
.
Results: Compared with other traditional models, the model proposed in this paper has higher accuracy and high practical value
.

Keywords: chicken seasoning; PCA; PSO-SVM algorithm; support vector machine; quality control model.

With the continuous improvement of people's living standards, people have higher and higher requirements for food
.
Chicken essence seasoning is an important food seasoning, and its flavor quality largely determines the taste of food [1]
.
Traditional methods generally use instrumental analysis and manual evaluation for flavor quality control
.
However, the data obtained by instrumental analysis can only be used to build a flavor database of chicken essence seasoning ingredients.
The way of manual evaluation is limited by the experience of the tasters, and is subjective and expensive, which is not conducive to the sensory evaluation of large quantities of data
.

Principal Component Analysis (PCA) is a data mining technique in multivariate statistics
.
It is mainly used to reduce the dimensionality of the data.
Under the premise of not losing the main component information, fewer variables are selected to replace the original variables, in order to achieve the purpose of data redundancy and speed up the calculation speed
.
Support Vector Machine (SVM) is a pattern recognition algorithm based on small samples.
It can not only overcome the large-capacity requirements of traditional neural networks, but also effectively solve the problem of dimensionality disasters and local minimums.
Excellent performance in solving nonlinear problems [2]
.

The composition of chicken essence seasoning is complex and diverse, and there is a non-linear relationship and a large amount of data redundancy [3]
.
In response to the above problems, this paper proposes a flavor quality model of chicken essence seasoning based on principal component analysis (PCA) and particle swarm optimization-support vector machine (PSO-SVM)
.
Among them, the PCA algorithm effectively solves the data redundancy between the components and reduces the data dimension
.
The PSO-SVM algorithm solves the problems of traditional neural network complexity and high data quality requirements, and through the parameter optimization of the PSO algorithm [4], the SVM algorithm has a better classification and recognition effect
.
The model can effectively classify different chicken essence seasonings under limited data sample training, so as to establish a better chicken essence seasoning flavor quality model
.

1 The
algorithm design steps of the flavor quality model based on the PCA and PSO-SVM algorithm are as follows:
①The normalized data preprocessing is performed on the sample data of chicken essence seasoning
.
This facilitates the identification of component information and avoids the phenomenon of "big numbers eat decimals"; ②Perform principal component analysis on the normalized data, reduce the dimensionality of the data, and obtain a set of linearly independent principal components to represent the pattern of the sample Characteristics, the sample contribution rate is greater than 85%; ③Using the particle swarm optimization algorithm to obtain the optimal parameter sum of SVM, so that the classification effect of SVM is the best; ④The data after principal component analysis is imported into SVM for training, and samples are selected Perform predictive analysis
.

2 Experimental results and analysis
2.
1 Data source and test platform The
data in this paper comes from a set of chicken essence seasoning data
.
There are 6 types of samples, and the sample label names are K524, G101, G822, K365, K362, and K862
.
There are 16 samples of each chicken essence seasoning, a total of 96 samples
.
The first 10 samples of each chicken essence seasoning are used as the training set of SVM, and the remaining samples are used as the test set
.
That is, there are 60 samples in the training set and 36 samples in the test set
.

Use matlab2012a combined with python2.
7 mixed programming, and use the libsvm toolkit to simulate
.

2.
2 PCA analysis
After normalizing the chicken essence seasoning data to [0,1], use Matlab to perform PCA analysis on the data to obtain the feature contribution rate and cumulative contribution rate between the main components
.

The cumulative contribution rate of the first two principal components is about 92.
66%, which can contain most of the flavor information of the original data, so the first and second principal components are selected as new variables for observation and analysis
.

2.
3 The prediction classification of PSO-SVM model is
known from principal component analysis.
The first two main factors are selected as the input of SVM, 60 samples are used as training set, and 36 samples are used as test set
.
The accuracy of the cross-validation method is used as the fitness function value of PSO to find the best parameter values ​​bestc and bestg of SVM
.

In the selection of the kernel function of SVM, the radial basis kernel function is selected as the kernel function of the SVM in this paper
.
Use 4 kinds of kernel functions to make the optimal hyperplane (decision boundary) of SVM classification (SVC) [5], as shown in Figure 1
.

In order to further test the performance of the PCA and PSO-SVM models, this paper uses ordinary support vector machine classifiers, PCA-based improved support vector machine classifiers and BP neural network classifiers (BPNN) for comparative analysis, and the recognition accuracy is used as an evaluation.
The standard of the performance of the detection mode, the higher the recognition accuracy, the better the performance of the corresponding model
.
The recognition effects of the four different models are shown in Table 1
.
It can be seen from Table 1 that the recognition accuracy of the PCA and PSO-SVM classifier models reached 98.
87%, which is much higher than the PCA-SVM classifier, support vector machine classifier (SVM) and BP neural network classifier (BPNN)
.
It shows that the classification and prediction effect of the chicken essence seasoning flavor quality model based on PCA and PSO-SVM algorithm proposed in this paper is good
.
3 Conclusion The
advantages of the chicken essence flavor quality model based on the PCA and PSO-SVM algorithm proposed in this paper are reflected in the category selection, which breaks through the limitation of only 2 to 3 categories in many literatures, and 6 kinds of chicken essence seasonings are selected as samples
.
First, the sample data is normalized and preprocessed, and then PCA is used to reduce the dimensionality of the data
.
Then use PSO to optimize the parameters of SVM, select the RBF kernel function as the kernel function of SVM, and obtain an SVM model with the highest classification rate
.
The classification and recognition rate of the model is compared with other methods (such as BP neural network, traditional SVM, etc.
), and the results show that the method in this paper has a better classification prediction effect
.
This model well overcomes the shortcomings of poor real-time and poor reproducibility in traditional chicken essence flavor analysis methods, and provides a fast and effective method for the flavor quality control of chicken essence seasoning
.

References
[1] Qin Lan, Li Fenghua, Tian Huaixiang, et al.
Application of electronic tongue in the analysis of taste difference of chicken essence seasoning[J].
Chinese Condiments,2014,39(10):132-135.

[2]Yu Daoyang Qi Gongmei, Qu Dingjun, et al.
Trace multi-component gas detection method based on SVM and PCA[J].
Pattern Recognition and Artificial Intelligence,2015,28(8):720-727.

[3]Qin Lan, Li Fenghua ,Tian Huaixiang, et al.
Research on the correlation between artificial sensory evaluation and electronic nose sensory analysis of chicken essence seasoning[J].
Food and Machinery,2014,30(4):11-13.

[4]Hu Wang,Li Zhishu .
A kind of more Simplified and efficient particle swarm optimization algorithm[J].
Journal of Software,2007(4):861-868.

[5]Wu Guifang,He Yong.
Cashmere raw material variety identification analysis based on principal component analysis and support vector machine[J].
Spectroscopy and Spectral Analysis, 2009,29(6):1541-1544.