| As a raw material for pasta,flour has an irreplaceable role in daily diet.In order to attract consumers,many manufacturers add additives to the flour to improve the flouriness and appearance.However,it is reported that azodicarbonamide(ADA),the most commonly used reinforcing agent,will be harmful to human health if taken in excess.At present,chemical methods are generally used to detect the ADA content in flour.However,this detection method has some disadvantages such as low efficiency and not suitable for field detection.Near-infrared detection technology can achieve fast,efficient,non-destructive testing.Although the original spectrum has overlapping peaks and the detection process is susceptible to environmental influences,but using stoichiometry can solve the above problems.Therefore,in this paper near-infrared detection technology combined with RBF-NN were put forword to the quantitative detection of ADA content in flour.In this paper,the spectral data of ADA samples were collected by near-infrared spectrometer,and the abnormal samples generated in the experiment were removed by monte carlo method.The remaining samples are divided into sample sets by KS and nKS methods.The comparison uses the trend-down method,the standard normal variable transformation method,the first derivative,the second derivative,and a combination thereof to preprocess the spectral data.Then,the correlation coefficient method is used to extract six feature bands for establishing SVM and RBF-NN models.It is finally determined that the standard normal variable transformation method combined with the first derivative method to preprocess the data and establish the RBF-NN model has the highest prediction accuracy.The coefficient of determination are 0.9997,prediction root mean square error and relative analysis error of the model are 0.5938 and 110.3772,respectively,And through the human-computer interaction software to achieve the visualization of the detection system. |
PDF Full Text Request