Application of Neural Network in the Analysis of Near-Infrared Spectra

The main problem in the spectrum analysis technology is the difficulty in locating the target which will affect the predication and the analysis. How to choose the right mathematic model becomes the key point in the spectrum analysis. This paper designs the practical manual neural network model to locate the target and predicate. This paper uses error backward direction propagation calculation method and establishes three-layer neural network to solve the problems such as the spectrum peaks overlap seriously and noise is big in the spectrum analysis. When the quantity of samples to be located the target is significant, employ manual neural network method to analyze and discuss the corn´s protein content and near-infrared spectrum. By analyzing the experimental result this paper concludes that manual neural network method performs better than linear regression method and partial least-squares method and obtains ideal result.