Analysis Of Vibration Signal During Chewing Of Dry-Crisp Food

In the various texture characteristics of food,crispness is one of the most common indicators for evaluating food quality.For dry-crisp food,its crispness directly characterizes the merits of its quality.At present,the method of sensory evaluation of food crispness is subjective,the equipment of mechanics measurement method is expensive,and the method of acoustic measurement is carried out from air conduction sound,and the characteristic information is not comprehensive.Therefore,in this paper we collected the acoustic signal and bone vibration signal simultaneously when the subjects chewed the dry-crisp food,and analyzed the vibration signal,extracted time domain,frequency domain eigenvalues of signal,built the dry-crisp food crispness identification and prediction model based on acoustic signal and bone vibration signal eigenvalues.Provide a new way to research the crispness of dry-crisp food.The main contents and conclusions of this paper are as follows:1.Collection of vibration signal and sensory evaluation of sample crispnessChose three kinds of representative dry-crisp foods(Want Want Sen bei,Le ba baked potato chips,Hsu Fu Chi rock burning pellets pancake)for moisture treatment.According to the different treatment time,three kinds of samples were obtained with 8,8,and 6 crispness gradients respectively.From the 20 candidates,the eight candidates(male to female ratio 1: 1)with good oral environment,good bone vibration signal and qualified sensory crispness evaluation were selected as the subjects.Set the suitable sampling frequency,sound sampling resolution,bone vibration signal acquisition time,bone vibration signal acquisition mode and bone vibration signal acquisition position and so on in the vibration signal acquisition.Subjects evaluated sensory crispness of all crispness gradients samples of three kinds of dry-crisp foods,did the molar chewing test,the incisor biting test and the bone conduction headphone test.Effectively collected the acoustic signal and bone vibration signal generated when the subjects chewed the samples.2.Extraction and selection of eigenvalues of acoustic signal and bone vibration signalAfter doing the zero adjustment and elimination of the trend item for bone vibration signal,coif5 wavelet base,the three-layer decomposition number were chosen for wavelet default threshold denoising to acoustic signal and db2 wavelet base,the three-layer decomposition number were chosen for wavelet default threshold denoising to bone vibration signal.The eigenvalues of the pre-processed vibration signal were extracted.13 time domain eigenvalues,4 power spectrum eigenvalues and 43 one third octave valid values were extracted from acoustic signal.11 time domain eigenvalues,4 power spectrum eigenvalues and 31 one third octave valid values were extracted from bone vibration signal.Using the variance analysis method to analyze the crispness evaluation results of the samples,it was found that the sensory crispness values of the three kinds of samples were significantly changed during the moisture absorption process(P<0.05).The Pearson correlation analysis was carried out on the extracted time domain,the power spectrum eigenvalues respectively with the corresponding sample sensory crispness values.The results showed that there were 15 characteristic parameters with significant correlation with sensory crispness values among three kinds of samples,and the characteristic parameters of bone vibration signal were more than the characteristic parameters of acoustic signal(12> 3),indicating that bone vibration signal provides more feature information about the sensory crispness of dry-crisp food than acoustic signal.3.Analysis of acoustic signal and bone vibration signalCross correlation analysis was used between acoustic signal and bone vibration signal,among bone vibration signal of three bone vibration signal acquisition positions.The analysis showed that there is a certain similarity among waveforms of vibration signal,but the similarity was low(0 <most of Rxy <0.50),indicating that acoustic signal and bone vibration signal during chewing of dry-crisp food contained different and texture-related feature information.The influence of sample types and moisture treatment time on the 15 characteristic parameters(X1,X2,X3...X15)was analyzed by multiple factors and multiple variables variance analysis.The results showed that at the 0.05 level,the effect of sample types was significant for the other characteristic parameters except X3 characteristic parameters(P <0.05).In addition to X2,X3,X5,X9,X14 characteristic parameters,moisture treatment time had significant effect on the other characteristic parameters(P <0.05).In addition to X1,X2,X3,X9,X14 characteristic parameters,sample types×moisture treatment time had significant effect on the other characteristic parameters(P <0.05).The principal component analysis of three kinds of samples' selected 15 characteristic parameters was carried out.The contribution rates of two principal components PC1 and PC2 were 72.49% and 22.39% respectively,and the cumulative contribution rate was 94.88%,which effectively achieved the purpose of reducing the characteristic parameter data dimensions.In the bone conduction headphone test,discrimination rate analysis of bone vibration signal acquisition positions,crispness gradients and sample types showed that the differentiate effect between forehead position and zygomatic arch position was bad,but subjects can make an effective identification on mastoid position.Subjects had poor distinction among sample crispness gradients and better distinction among sample types.The discrimination rate of molar chewing was generally higher than that of incisor biting among three kinds of samples.4.Establishment of dry-crisp food crispness modelThe linear kernel function was selected to construct the support vector machine identification model of each sample.The one third octave valid values selected by the line chart effect were import,the quality of sample crispness(good quality of crispness and reduced quality of crispness)were export.The results of model test showed that models' most recognition rate of corresponding kinds of samples' crispness quality was? 90%.The recognition effect was ideal,indicating that the support vector machine models had good application performance in the crispness quality judgment of dry-crisp food.Two principal components of 15 characteristic parameters,the eigenvalues of acoustic signal in 15 characteristic parameters were independent variables,and the sensory crispness value of samples were taken as the dependent variables to establish the sample integrated multiple linear regression model and the sample individual multiple linear regression model.The fit of regression of the integrated model was not high(R2 <0.6),and the average relative error of three kinds of samples crispness prediction was more than 7%.The fit of regression of individual models was generally higher(R2> 0.9),and the average relative error of crispness prediction was lower than that of the integrated model,which was about 5%.For the same kind of samples,both the sample integrated model and the sample individual model of each kind of samples,the model which combineed acoustic signal and bone vibration signal characteristic information was better than the model which only used the acoustic signal characteristic information(the average relative error reduced by about 1%).In this study,the support vector machine identification models were obtained to distinguish the crispness quality and the multiple linear regression models were obtained to predict the sensory crispness value of three kinds of dry-crisp food.