Information Interaction Of Multi-sensor Array In Electronic Tongue And Quantitative Analysis For Indica Rice Taste

As one of the main intelligent sensory instruments,electronic tongue has been widely used in the food field because of its fast,simple and real-time analysis advantages.At present,electronic tongue is mainly used to distinguish and identify the overall differences of samples,but it still cannot achieve the quantitative analysis.The voltammetric signal of multi-sensor array in electronic tongue is lack of proper decomposition and transformation,and it is stuck on reading the vertex and inflection point of the signal as the feature value,which leads to the insufficient extraction of the useful weak information contained in it,and thus limits the ability of the sensor array to identify small differences.The main reason that electronic tongue cannot achieve the quantitative analysis is that information interaction of multi-sensor array in electronic tongue is not clear.In this paper,a voltammetric electronic tongue developed by intelligent sensory technology laboratory was the basis,information entropy was used in the combination with multidimensional data,indica rice was taken as the application object,the study on information interaction of multi-sensor array in electronic tongue was carried out so as to realize the quantification.The main conclusions were as follows:(1)In the aspect of extracting the feature vector,wavelet packet decomposition based on similarity entropy and Fourier transform were used to extract the feature vector without multicollinearity.The results showed that the best wavelet basis functions of all electrodes in the sensor array were db N series,that of Ag electrode was db2,Pt,Au and W electrode were db4,Pd and Ti electrode were db6,and the best decomposition layer was 4.The original voltammetric signal was decomposed into detail signal,and then the feature vector was extracted by Fourier transform.In order to judge the validity of the feature vector by the multicollinearity,the overlapping and redundant parts in the original data were deleted.According to the correlation analysis with the physicochemical indexes of rice,the correlation between the feature vector and alkali spreading value,gel consistency,amylose,protein and starch was relatively high.In this section,the interference of collinearity on interactive information was excluded.(2)In the aspect of window feature extraction,on the basis of the feature vector,the continuous window feature extraction and characterization method based on relative entropy was used to extract the feature matrix from the multi-sensor array signal in electronic tongue,and the effective detail signal with interaction information was found.On the basis of the feature vector,the signal-to-noise relative entropy was calculated by using the concept of relative entropy.Through the analysis on different window operation methods,it could be concluded that continuous window operation was more suitable for the feature matrix extraction of the sensor array than partition window operation.The continuous window feature extractions between the electrodes and frequencies were characterized by dial color block characterization image.The results showed that the top 6 to 8 detail signals contained the interaction information,which was called as the effective detail signals.According to the correlation analysis with the physicochemical indexes of rice,the effective detail signal was highly correlated with gel consistency,amylose,protein and starch.In this section,the interaction information was located and the effective detail signal containing the interaction information was confirmed.(3)In the aspect of the enhancement for discrimination and recognition,on the basis of the effective detail signal,the multi-sensor array in electronic tongue based on information entropy could accurately distinguish and recognize rice origin and type.The sensing entropy of single electrode and between the electrodes in multi-sensor array was defined using the principle of information entropy,and the unit sensing vector and the interactive sensing vector were constructed.The results showed that six sensing entropies could be used for the identification of rice origin effectively,and all the interactive sensing vectors could be used for the discrimination of rice types.Support Vector Machine(SVM)and K-Nearest Neighbor(KNN)classifier were used.The results showed that the training and prediction accuracy of SVM for origin identification were 89.0% and 82.9% respectively,and the training and prediction accuracy for type discrimination were 96.0% and 88.6% respectively.In this section,the interaction information contained in the effective detail signal could enhance the discrimination and recognition ability of the sensor array.(4)In the aspect of quantitative analysis,on the basis of the effective detail signal,the multi-sensor array in electronic tongue based on joint entropy and mutual information could effectively quantitatively predict the physicochemical indexes of rice using multidimensional data.Based on the effective detail signal,the multidimensional interaction matrix was constructed.Convolution Neural Network(CNN)model,Back Propagation Neural Network(BPNN)model and CNN+BPNN joint model were used to establish a quantitative model with multidimensional interaction matrix as the input.The results showed that the training and prediction accuracy of CNN+BPNN joint model were 87.0%?92.0% and 82.9%?89.5%,respectively.The interaction between the electrodes of multi-sensor array transmitted more quantifiable information than that between the frequencies.Multidimensional interaction matrix was strongly correlated with gel consistency,amylose,protein and starch,proving that the combination of information entropy and multidimensional data played an important role in the quantitative ability of the model.In this section,it was made clear that interaction information could realize the effective quantification ability of the sensor array.(5)In the aspect of quantitative analysis for sensory evaluation,based on the multidimensional interaction matrix through information entropy and multidimensional data,the multi-sensor array could be used to evaluate rice taste.Taking artificial taste score as the target,the prediction accuracy of tandem evaluation model was as high as91.4%.The correlation between rice taste and physicochemical index was analyzed.The results showed that there was a strong nonlinear correlation between rice taste and physicochemical indexes such as gel consistency,amylose,protein,starch and chalkiness.Combined with the results of quantitative analysis for physicochemical index,the interaction information was related to gel consistency,amylose,protein and starch,which verified the effective quantitative ability of multi-sensor array in electronic tongue.In this section,it was confirmed that interaction information could realize the sensory evaluation ability of the sensor array,and verified its quantitative ability.In this paper,information entropy was innovatively used as an effective tool for the study on information interaction and multi-dimensional quantification of multisensor array in electronic tongue.The interaction information of the sensor array was found and its quantification function was realized.This paper provided a theoretical reference for intelligent sensory evaluation and technical support for quantitative analysis of electronic tongue.