To obtain a low-noise

high-accuracy and good self-adaptvity model

it is essential to preprocess near infrared（ NIR） spectrum and optimize wave length.Experiment ensures ash content in wheat flour as research object and improve robustness of partial least squares method（ PLS） model by combining genetic algorithm（ GA）-wavelength-optimization with spectral pretreatment. Contrast the three indexes

R~2

RMSEC and RMSEP

by different data preprocessing methods

130 samples are selected randomly to establish spectral data pretreatment +GA+PLS model.According to the result

R~2 is increased from 70.31 % to 97.46 %

RMSEC is reduced from 0.077 5 to 0.022 6

RMSEP is reduced from 0. 0996 to 0. 021 3. It is feasible to quantitatively analyze on wheat flour ash content by GA and spectral preprocessing

prediction ability and precision of the obtained model is much more higher.