Electronic Tongue Detection System Based On Deep Learning And Its Application In Drink Analysis

Currently,consumers have an increasing demand for drinks quality and safety.The traditional drink detection and analysis methods have many problems,such as single detection components,difficult operation and poor accuracy,so it is urgent to find a convenient and fast detection method.Electronic tongue is a novel type of bionic taste detection instrument that has emerged in recent years.It uses sensor technology to imitate human taste sensing organs,and combines pattern recognition technology to realize qualitative or quantitative detection of different drinks.Pattern recognition technology plays a decisive role in the accuracy of system detection.The current pattern recognition technology has some problems such as inadequate adaptability and insufficient generalization capacity.Therefore,this paper establishes various deep learning models,and uses different drinks for detection and verification.The detailed study contents are as follows:(1)A combined model based on electronic tongue and Generative Adversarial Networks(GAN)-Convolutional Denoising Autoencoder(CDAE)-Extreme Learning Machine(ELM)is preposed to realize rapid traceability detection of coffee from different origin.Aiming at the problems of low accuracy and poor generalization ability of deep learning model caused by the insufficient number of original data samples of electronic tongue detection.GAN is used to expand the data scale of training samples to improve the stability of the system.According to the characteristics of complex electronic tongue output signal,large dimension and many noises,CDAE is used to extract the features of electronic tongue signal in low dimensional feature space to improve the expression ability of key features;finally,ELM is used to classify and identify the extracted feature information,and a coffee origin traceability detection and analysis model is constructed.The results show that the designed model has better performance compared with traditional models.(2)A pattern recognition method combined electronic tongue and Self-adaptive Attentional Residual Convolution Autoencoder(SARCAE)is proposed to detect and analyze different botanical honey.Based on the autoencoder,it combines convolution operation and autoencoder which can accomplish the unsupervised feature extraction of original signals without the participation of labels.The embedding of the residual learning block and Self-adaptive Attention Mechanism(SAM)alleviates the gradient disappearance of the model and obtains more detailed feature information from the original signal,respectively.Experiment results show that,compared with traditional models,SARCAE has better feature extraction and classification capabilities,and has better anti-noise scarcity capabilities and anti-data scarcity capabilities.(3)A pattern recognition method which combined electronic tongue and the Inception Convolutional Autoencoder based on Meta-learning strategy(MICAE)is proposed to quickly distinguish different grades of Longjing tea.In order to acquire more feature information,the modified inception structure is added to the convolutional autoencoder to improve the nonlinear representation ability of the network.In order to improve the adaptation ability of the model in the new data environment,the meta-learning pre-training strategy is introduced to obtain better initial parameters of the model.The experiment result shows that the model guided by meta-learning has better learning ability of new task features.The result shows that the designed model can accurately distinguish different grades of Longjing tea than traditional models.The above research provides a novel research method for the application of deep learning in the intelligent sensory system field,and also provides an effective method for drink detection.