Model optimization of near- infrared spectroscopy and back propagation artificial neural network for identifying the geographical origin of Tremella fuciformis

Near- infrared spectroscopy in combination with artificial neural network was used to identify the geographical origin of tremella fuciformis. A total of 120 samples from Sichuan province and Fujian province were studied.After being pre- treated with average deviation and first derivative,the dimension of near- infrared absorption spectroscopy data were reduced and applied to develop classification models by principal components analysis and back propagation artificial neural network.The results showed that the cumulative contribution of first three principal components was 100%,but identification accuracy was 88.3% by principal components analysis.Thus the artificial neural network was further used to optimize the structure of classification model. Under 2 output layers and 11 hidden layers,the identification accuracy reached 100%.The study demonstrated that near- infrared absorption spectroscopy based on artificial neural network can be used as an accurate and rapid technique for identification of geographical origin of tremella fuciformis. Models builded by this study can help building geographical indications and monitoring quality for raw materials of food.