| In this paper,the pattern recognition method for non-destructive detection of sweet potato powder is studied by means of visible and near-infrared spectrums,near-infrared spectrums,and laser induced breakdown spectrums.By using support vector machine(SVM)model,partial least squares(PLS)model and neural network model based on spectral information,the classification model of detecting sweet potato powder’s producing area and the regression model of detecting alum content in sweet potato powder were obtained,and the rapid detection of biological information of sweet potato powder was achieved.It is of great significance to promote the rapid detection of food quality.The main contents and research results of this paper are as follows:(1)Study on classification method of sweet potato flour producing area based on spectral information.Firstly,wavelet filtering and SG smoothing algorithm are used to filter and de-noise the spectral data.Secondly,based on the full spectrum band,using SVM models,PLS models,and principal component analysis-support vector machine(PCA-SVM)models,the classification model of sweet potato flour producing area is obtained.The results show that the PCA-S VM model based on visible and near-infrared spectrum has the highest classification accuracy.(2)Study on determination method of alum content in sweet potato flour based on spectral information.Firstly,the SG smoothing algorithm,continuum removal and the neighborhood component analysis are used to do feature selection to spectral data.The LIBS standard library is used to select features of the laser induced breakdown spectrum.Secondly,based on the spectral band after feature selection,the regression model for detection of alum content in sweet potato flour was obtained by using PLS models,PCA-SVM models,and neural network models.The results show that the neural network model based on near-infrared spectrums and laser induced breakdown spectrums has better detection accuracies and higher correlation coefficients. |
PDF Full Text Request