The quantitative analysis model for crude protein in corn straw was constructed in this paper

and the selection method for spectrum characteristic band was discussed and verified.Firstly

107 samples were preprocessed

then a DB2 wavelet transform method with default threshold 4-level decomposition was used to reconstruct the spectra after removing two abnormal samples.The determination coefficient of cross validation(R2CV) for crude protein model was increased from 0.788 9 to 0.920 8 after pretreatment.Interval partial least squares(IPLS) and its improved methods

i.e.backward interval partial least squares(BIPLS) and synergy interval partial least squares(SIPLS) were adopted to select the characteristic bands

that IPLS and it's improved approach BIPLS

SIPLS could locate the characteristic bands more effectively and accurately compared with principal component analysis(PCA)

competitive adaptive reweighted sampling(CARS)

correlation coefficient(CC)

genetic algorithm(GA)

moving windows partial least squares methods(MWPLS).When SIPLS was using 30 band interval

the optimal model validation results were obtained in the band ranges of 10 128-10 398 cm-1 and 11 196-11 462 cm-1

with the correlation coefficient of validation set is 0.978 4

the determination coefficient R-square of prediction set is 0.957 2 and the root mean square error of prediction set of 0.221 1.IPLS method exhibited a better real time accuracy

and it and has a certain practicability in data support for the determination of ammoniation and alkalization of corn straw.