Abstract: A practical manual neural network model was designed to make the location of target and predication. We used error backward direction propagation calculation method and established three-layer neural network to solve the problems of serious overlap of spectral peaks and big noise in the spectrum analysis. When the quantity of samples to be located is significant, We employed manual neural network method to analyze and discuss the corn’s protein content and nearinfrared spectrum. By analyzing the experimental result it is concluded that manual neural network method is superior to linear regression method and partial leastsquares method.

Key words: spectrum analysis, location and predication, manual neural network method