Research On Modeling,Optimization,Identification And Synesthesia Of Machine Taste Based On Nerve Conduction Mechanisms

Machine taste is a bionic detection technology that simulates the working principles of biological taste systems and realizes the decision-making analysis of taste substances with the aid of taste sensor arrays and data processing models.The physiological channel model based on neural conduction mechanisms has obvious biological structural characteristics and functional properties,providing a tool to explain the working mechanisms of sensory systems.In addition,the physiological channel model can be combined with machine taste to obtain real perceptual information,so that the output of the machine taste system can be closer to perceptual patterns of humans and thus realize the effective analysis of taste substances.Therefore,this study proposed physiological channel models of machine taste based on human nerve conduction mechanisms to enhance the bionics of the machine taste system and achieve effective taste analysis.The main works were as follows:（1）The taste and olfactory detection data of different food materials were acquired using the machine taste and machine olfactory technologies,respectively.The data obtained were used for the validation of universality and effectiveness of the proposed physiological channel models.In addition,the human detection data of the materials were acquired by the physiological sensory and used for sensory evaluation and olfactory-taste synesthesia effect simulation.（2）A computational model of taste pathways（CMTP）was proposed for human taste nerve conduction mechanisms to enhance the bionics of the machine taste system and achieve effective taste identification.First,CMTP described the dynamic characteristics of conduction pathways for human taste information from the tongue to the brain cortex.Second,the output of crucial CMTP modules showed chaotic characteristics,which validated the rationality of crucial modules.Third,when stimulated,the output of CMTP nodes presented fast-response ability and1/f characteristics,which reflected the bionic performance of CMTP.Then,the output of CMTP nodes exhibited fast-response ability and chaotic characteristics under the input of machine taste data of beer,tea and apple samples.This indicates that CMTP enhanced the bionics of the machine taste system and made processed results close to human perception patterns.Finally,compared with the classification results of multiple identification models and ablation studies,CMTP achieved effective classification of beer,tea,and apple.（3）To solve the problem of the limited physiological significance for CMTP parameters,Hebbian-habituation learning rules（HHLRs）were proposed for the perception mechanisms of biological nervous systems to optimize CMTP parameters for effective classification and sweetness evaluation of taste substances.First,HHLRs optimized node connection weights of the pattern recognition module for the CMTP to make these weights biologically meaningful.Second,the effectiveness of HHLRs was verified based on the simulation results of dynamic characteristics for the CMTP node output before and after optimization.The results indicated that,compared with the unoptimized CMTP,the 1/f characteristics and synchronization of the node output for the optimized CMTP were improved,and the bionic performance of the optimized CMTP was enhanced.This verified the effectiveness of HHLRs.Finally,the optimized CMTP was used for substance classification and sweetness evaluation of machine taste.Compared to the unoptimized CMTP,signal preprocessing,and pattern recognition models,the optimized CMTP achieved the best classification and evaluation performance.（4）A taste classification method based on CMTP and a convolutional neural network（CNN）was proposed.This method achieved effective taste classification in data of different sizes.First,in the taste classification of small-sample data,CMTP combined with CNN achieved better classification performance than several recognition models,including the accuracy of 96.00%and 96.67%,Kappa coefficients of 94.95%and 95.77%,and area values of receiver operating characteristic curves of 0.9750 and 0.9792.Second,in the taste classification of large-sample data,the CMTP combined with a time-channel expansion（TCE）was used for data augmentation of machine taste.CMTP-TCE achieved better performance than traditional recognition and other data augmentation models,thus verifying the effectiveness of CMTP-TCE augmentation.Finally,a dot-product attention mechanism（DPAM）was combined with a residual network（Res Net）to identify the CMTP-TCE output for effective taste classification of large-sample data.The best classification results of CMTP combined with DPAM-Res Net were obtained for large-sample data and were reflected by the taste classification and ablation studies of augmented data in different multiples.（5）An olfactory-taste synesthesia model（OTSM）was proposed to achieve effective flavor classification by modeling olfactory-taste nerve conduction mechanisms.First,the dynamic characteristics of conduction pathways for human olfactory-taste information were described by the OTSM.Second,when stimulated,the output of OTSM nodes had 1/f characteristics and synchronization.This showed that the OTSM had bionic performance.Third,OTSM was used for flavor substance classification and achieved better identification results than those of single machine olfactory,single machine taste,and machine olfactory-taste fusion systems under multiple classification models.Fourth,OTSM combined machine olfactory and machine taste to investigate the olfactory-taste synesthesia effect,which included nasal,olfactory-taste coexistence,posterior nasal（NOTCPNEs）,halo and horn effects.In the study of NOTCPNEs,different perceptual intensities of three stages for the NOTCPNEs were reflected by recognition results of the OTSM output.The NOTCPNEs were revealed by recognition results of the OTSM output.In the analysis of halo and horn effects,there was a high correlation between the OTSM output and halo/horn effect sensory evaluation scores,and correlation analysis results revealed halo and horn effects.Finally,the best performance over multiple signal preprocessing and pattern recognition models in flavor identification application,including an accuracy of 95.56%,a Kappa coefficient of 94.17%,and an F₁-score of 95.58%,was acquired by the OTSM.In conclusion,the CMTP and OTSM proposed on human nerve conduction mechanisms acquired effective human perception information,enhanced the bionics of the machine taste system,and made the processing results of the machine taste system close to human perception patterns.In addition,CMTP and OTSM achieved effective taste and flavor classification,respectively.It improved the decision performance of the machine taste system.