Application Of Artificial Taste Analysis For Food Quality Inspection

People pay more attention to food quality inspection and quality control with increasing life quality. The detection methods for food quality are required to be not only accurate and objective, but also rapid and convenient. Thus, it may be very important to introduce new detection means and investigate their feasibility in food quality inspection and quality control. Artificial taste technique (electronic tongue, E-tongue) as a new technique is one of the ways matching these requirements, which could be used to analyze and recognize the taste of liquid. The researches in the artificial taste field have been focused on three main aspects:the developments of sensor and sensor array, the optimizations and comparisons of pattern recognition methods, and the applications to various analytical tasks. Besides the development of sensors with excellent performance, appropriate pattern recognition methods are crucial for successful applications of E-tongues. The key advantage of the artificial taste is that it may be possible to assist, substitute, even surpass the human sensory, and may be suitable to detect the quality attributions about the taste of food. The study scopes of this paper will focus on the optimizations and comparisons of pattern recognition methods and the applications to the food industry. The study offers theory and data bases to open up the application ways of artificial taste technique.It is necessary for the E-tongue field to introduce novel and more effective pattern recognition methods developed in the machine learning community, in order to overcome the problems of pattern recognition methods usually used. In this paper, for the first time, RF is introduced and investigated for the E-tongue data processing, and the performance of RF in classification (three data sets) and regression (three data sets) will be focused on. Furthermore, the comparisons of the performance of RF with currently more popular methods, BPNN and SVM, have been done. RF with5-fold cross-validation and without data preprocessing or feature selection/extraction, exhibits better classification and prediction performance than other two classical methods—BPNN and SVM, especially for unbalanced, multiclass and small sample data sets. The average correct rates (CR) on CV sets of the three data sets performed by BPNN, SVM and RF were83.35%,58.06%and98.83%, respectively. The average Q2on CV sets of the three data sets performed by BPNN, SVM and RF were0.886,0.353and0.913, respectively. Moreover, for the independent testing sets, RF also gives satisfactory prediction results (average CR and Q2were99.75%and0.932, respectively), which means the prediction models constructed by the E-tongue signal features and RF can actually be employed for the sample inspection.Whether the physicochemical indexes of food are correlated with the sensor response signals of the E-tongue is the basis of applying the E-tongue to analyze food. In this paper, the feasibilities of the E-tongue for food quality inspection are studied by using10%orange juices of different brands and Chinese vinegars of different types and different grades as examples through combining the measurement results of the E-tongue and the main physicochemical indexes related with sensory quality (taste quality) of food, with the purpose of offering a feasible assistant means for the sensory evaluation (taste evaluation). The results indicate that:(1) the physicochemical indexes of food are significant correlated with the sensor response signals of the E-tongue, which means the E-tongue can be used to detect the taste quality of foods and the related physicochemical indexes can be used to explain the results obtained by the E-tongue;(2) the E-tongue can differentiate10%orange juices of different brands and Chinese vinegars of different types and different grades;(3) the first principle component can be explained as sweetness in PCA (principle component analysis) and the first and second canonical variables can be explained as sweetness and the ratio of sugar and acid in CDA (canonical discriminant analysis), respectively, besides, these indexes are ranged in an increasing sequence along the positive directions of the coordinates;(4) the E-tongue can predict the class memberships and grades of the unknown samples by constructing models, such as RF classification models;(5) PCA and CDA have some advantages when differentiate samples, but cluster analysis and SIMCA (soft independent modeling of class analogy) have advantages when compare (imitate) the same types of products with competitors and can initially judge whether the diversities exist in the taste quality between products. All these results confirm the E-tongue can be applied for food quality inspection according to the sensory profiles (taste properties) of food, which also prove that the E-tongue has potential for the application to food sensory inspection.An analytical method based on artificial taste and random forest has been established and investigated when being applied for food quality control. The investigated cases include the recognition of food geographical origin, the identification of food under different storage conditions, the quality stability study of different batches of food, the quality stability study of food under storage period, the determination of shelf life and storage times of food. The results show’artificial taste technique coupled with different data analysis methods can fulfill all those tasks. These are basic researches for the application of the established analytical method in the food quality control field and open up a new research field for the artificial taste technique. At same time, this established analytical method provides a new means for food research. The methods of the quality control chart plotted based on principle component (PC) and the plot of PC vs time were also proposed in this paper.