Chinese chestnut browning detection by near infrared spectroscopy and scaled conjugate gradient back propagation neural network

In order to realize the non - destructive detection of Chinese chestnut browning, near infrared spectroscopy ( NIR) with the brand of 12000～4000cm-1 was used to acquire the spectra of shelled and unshelled chestnuts of“Mao Ban Hong ”with the different browning grade. The original near infrared spectra data were processed by Savitzky - Golay smoothing and standard normal variate ( SNV ) transforming. Then, principal component analysis ( PCA) was applied to extract the characteristic information of the spectrum, and the back propagation neural network based on the scaled conjugate gradient algorithm ( SBP) was set up by the principal components as the input.The results showed that the recognition of browning level for unshelled chestnuts was the best when the principal component's number was 8 for SBP, and the accuracy for the training and testing samples were 100% and 98.7% , respectively.For the shelled chestnuts, 10 was the best number of principal components for SBP neural network, and the recognition for the training and testing samples were 65.3% and 64.4% respectively. Finally, the comparison between the traditional back propagation neural network based on gradient descent algorithm ( GBP) and Radial basis function neural network ( RBF) was proceeded. The results from the validation samples showed that the recognition of SBP for discriminating unshelled and shelled chestnuts was 100% and 66.7% , respectively.SBP was better than GBP and RBF for the discrimination of browning for Chinese chestnut.