Sensory Evaluation Model Construction And Application Based On Electronic Nose/tongue

Traditional sensory evaluation is mostly conducted by artificial sensory evaluation,mainly through the human sense of smell,taste and other sensory organs to form a comprehensive sensory perception of the product.The evaluation process is time-consuming and labor-consuming,and is affected by the environment,physical condition and psychological factors of evaluation personnel.The final evaluation results are fuzzy and the error is relatively large.The electronic nose and electronic tongue developed to simulate the perception mechanism of mammals can just make up for this defect.Given that the taste of food is actually the comprehensive perception of smell and taste,the electronic nose / tongue fusion is used in sensory evaluation in this study.In this study,an electronic nose / tongue detection system was used to detect the smell and taste of apple.We analyzed sensory difference information of apple and characteristics of repetitive experimental data to ensure its stability for sensory feature information acquisition,and to provide a good data basis for the subsequent application of pattern recognition to establish sensory organ model.The original characteristic data set was obtained by fusing the sensory data of smell and taste,and the correlation between the variables was analyzed by combining the internal components and sensor responsive characteristics of apple.Principal component analysis(PCA),importance of variable projection of partial least squares(PLS)and stepwise discriminant analysis are used to reduce the dimension of the data and eliminate redundant information.Three optimized feature sets and complete feature sets were used as model to input,and sensory feature labels of apple were used as model to output.Support vector machine(SVM)and extreme learning machine were used to construct sensory identification model and sensory quantitative evaluation model.In the process of model construction,the bottom structure of four fused data feature sets was explored,and the performance of the sensory model was discussed by using the identification accuracy,fitting accuracy and error as the effectiveness evaluation indicators.Inquiry platform of the sensory evaluation of apple based on ZigBee-GPRS technology and wireless network transmission was constructed.The analysis results show that in the process of sensory identification model construction,the overall performance of the identification model constructed by extreme learning machine is better,and among this,the average correct rate of sensory difference identification of samples as the identification model based on the importance of variable projection,stepwise discriminant optimization method and original feature set is 100%.In the process of construction of sensoryquantitative evaluation model,the evaluation model constructed by support vector mechanism shows better overall performance,and the quantitative evaluation model constructed by VIP6 feature subset based on the importance of variable projection has the best performance,and the fitting accuracy reaches 0.9960.The model can be used to quantify sensory preference of apple,and the quantitative score of error only ranges from 0.05 to 0.296.The quantitative model of sensory evaluation based on electronic nose / tongue fusion was applied to inquiry platform of apple sensory evaluation,which realized the web query of retailers or secondary processors on the sensory preference of apple in consumers.The research shows that the sensory evaluation model based on electronic nose / tongue fusion and its query platform can realize the real-time information collection,identification,quantitative evaluation and preference of apple sensory characteristics.The fuzzy evaluation problem of artificial sensory evaluation is solved,and the data support is provided for the purchasing strategy of retailers and secondary processors,and reasonable purchasing budget is made according to the preferences of consumers.