In order to solve the pre-processing problem of spectral data in near infrared rapid detection analysis of content of milk components, a pre-processing algorithm for score resetting (SR) of principal components (PC) in the near infrared spectrum on the basis of histogram layering mapping technology is proposed. With glucose content in the three components samples consisting of glucose, NaCl and water, and lactose content in the fresh milk samples, as the detecting objects, cumulative contribution rates of the near infrared scattering spectral PC scores are pre-processed by means of mapping by layer and by piece. Furthermore, partial least squares (PLS) regression analysis method is used for modeling, thereby test and analysis of sugar content information in corresponding near infrared spectra are completed. The results show that after SR pretreatment, the predicted deviation of the calibration curve of the milk lactose content PLS model is reduced by 23.9%, the actual prediction deviation is reduced by 27.8%, and the actual prediction deviation of the verification set is reduced by 16.7%. This SR spectral preprocessing method takes into account multi-scale information such as spectra, reference content value and component correlation to realize spectral information denoising enhancement. Therefore, false deletion of useful information can be avoided, and inadequate fitting and overfitting can be prevented.