Reach on Discrimination Method of Fresh Jujube Variety Using Wavelet Transform and Artificial Neural Network Model

Abstract

In order to discriminate quickly varieties of fresh jujube samples and find the effective method of discrimination spectroscopy. Firstly, eight different pretreatment methods are used for pretreatment data. Secondly, wavelet transform is applied to compress the data. Finally, three discriminate models of artificial neural network are used to respectively discriminate fresh jujube variety and compare these three discriminate methods. Experimental results show that using wavelet transform to compress data can improve the artificial neural network learning rate as many as at least 18 times. Besides, the best pretreatment method of discriminate fresh jujube is Savitzky-Golay filter and Multiple Scattering Correction, and the identification rate can reach to 95%. After selecting the appropriate pretreatment method, and discriminate three and four kinds of fresh jujube samples, the model of three kind artificial neural networks can work and the identification rate can reach to 95%. If the kinds of fresh jujube are enlarged, only SVM neutral network can work and the identification rate can also reach to 95%.