The Wavelet And Neural Network In The Application Of Near Infrared Spectrum Inversion Model

Near Infrared Spectroscopy (NIR) technology is a growing test method in recent years, with the characters of efficient, rapid and non-invasive. The key of the technique's application is to establish an accurate prediction, high anti-interference ability analysis model. With the rapid development of information processing technology, wavelet transform (wavelet) and artificial neural network (ANN) have made a wide range of applications on N1RS for their unique advantages. However, the combination of them which is applied on NIRS analysis of soil and crops have been no reports yet.This paper, which is based on NIRS and information processing technology, take the NIRS on moisture of yellow-brown soil as subject of study. First, do preprocessing to the data by wavelet transform, and then build the inverse model for the moisture testing through artificial neural network. Relevant content and conclusions are as follows:(1) In this paper, wavelet transform is selected to make preprocessing to the data. The spectrum of 400?2400nm is decomposed by the level of 9and the wavelet function of db2, and then a series of wavelet coefficients are got. The part of low-frequency is composed by most of the background information. At the same time, the noise information is the major part of high-frequency. Finally, the useful information is mainly located in the middle-frequency of details. Cd5 taken out as the input of the model is comparatively appropriate. It has also compressed the data dimension while getting the useful information.(2) Under the premise of the data which is pretreated, artificial neural network is used to build the inverse models. BP neural networks and RBF neural networks are used respectively. Using the networks to predict the unknown samples, the correlation coefficient (R) between inverse and measured values are 0.846 and 0.802. It can be seen in the building of inverse model about the moisture content of yellow-brown soil, BP neural network is superior to RBF network.(3) The research methods have applied to NIRS of corps'canopy, and some progress has been made.