The Identification Study Of Food Crispness Based On Acoustic Signal

Crispness is an important index in food texture evaluation,and there are few studies on the identification of the crispness of different kinds of foods.The purpose of this study is to identify the crispness of different kinds of foods based on acoustic signal.Carrot,White radish,potato,sweet potato,Fuji apple and crystal pear which were wet crispy foods were chosen as the samples of this study.The acoustic signal characteristics and mechanical crispness were used to construct discrimination model and cloud model.After that,the best model was chosen to identify food crispness.Finally,the microscopic morphological parameters of the samples were acquired to explain crispness differences from the microscopic perspective.The main contents and conclusions of this study are summarized as follows:1.The acquisition and analysis of the mechanical and acoustic signals of different samples.CT3 texture analyzer,sound transducer and computer were used to record the acoustic signals.The recorded acoustic signals were denoised,and then acoustic signal characteristics were extracted,including sound intensity,maximum short-time frame energy,waveform index and amplitude difference,which were time-domain characteristics.And the frequency domain characteristics include the peak value of power spectral density and power spectral density characteristic.Mechanical crispness was also measured using CT3 texture analyzer.2.Establishing crispness classification model based on acoustic signals.(1)The mechanical crispness of the samples was classified.There were two classification methods in this study,which were divided into six categories and three categories respectively.The dimension of acoustic signal characteristics was reduced by principal component analysis.The principal component was used as the input of crispness discrimination model,and the output of the discrimination model was the two classification methods.The construction methods of the discrimination model included Fisher discriminant method,MLP neural network and RBF neural work.And six different kinds of discrimination models were obtained.(2)The six discrimination models were compared with each other.The results showed that when the samples were classified into three categories,the accuracy was much larger than the condition which the samples were classified into six categories.The reason might be that there were little differences among samples when they were classified into six categories.However,when the RBF neural network was used to establish the discrimination model,the accuracy of modeling set and validation set was 82.7% and 85.4%,which was largest among the six models.Therefore,it was better to use RBF neural network to construct the discrimination model,and the samples should be classified into three categories.3.Comparing crispness based on cloud model and acoustic signal(1)To identify the samples more clearly,the maximum short-time energy,waveform index,sound intensity and power spectral density characteristic which were highly correlated with crispness were remained,and the remained acoustic signal characteristics were combined with the cloud model method to compare crispness.According to the results of the one-dimensional cloud model,the two-dimensional cloud model and the comprehensive cloud model,the crispness order of the six samples was carrot>potato>sweet potato>white radish>Fuji apple>crystal pear.The dispersion degree of the carrot,potato,white radish and sweet potato was large,however,the dispersion degree of Fuji apple and crystal pear was small.(2)The verification experiment including the mechanical experiment and sensory evaluation was carried out.The results of the verification experiment proved the cloud model was suitable for food crispness evaluation,and the cloud model method was superior to the traditional mechanical crispness evaluation and sensory evaluation.(3)The cloud model method was compared with point two,and the results proved the cloud model method was superior.Finally,the cloud model method was chosen to discriminant different food crispness.4.The relationship between microstructures and food crispness.(1)The microscopic images of different samples were recorded with a microscope.The microstructure parameters which were area,perimeter and Feret's diameter were remained after ANOVA and correlation analysis.The larger the crispness,the smaller the microstructure parameters were.Among the six samples,the microstructure parameters of crystal pear were largest,and the microstructure parameters of carrot were smallest.Therefore,the crispness of crystal pear was smallest,and the crispness of carrot was largest.(2)The microstructure parameters of the samples were used to predict crispness.Clustering analysis and principal component analysis were used to reduce the dimension of the microstructure parameters respectively.One-dimensional linear model and BP neural network were combined with the dimensionality reduction parameters respectively to predict crispness.A total of four prediction models were obtained.After comparison,the optimal model was constructed by BP neural network,and the dimensionality reduction method was principal component analysis.The average relative error of the verification experiment was 4.79%.In this study,some typical vegetables and fruits were used as the research object,and acoustic signals were used to identify food crispness,which provide a theoretical basis for food crispness rating.In addition,this study analyzed the reasons for different crispness of different samples from the microstructure aspect,and provided the research basis for food crispness from microstructure aspect.