Near-infrared Detection of Straw Cellulose by Orthogonal Signal Correction and Partial Least Squares

Abstract

Orthogonal signal correction (OSC)and partial least squares(PLS)were used during the pretreatment of straw to reduce environmental noise and prediction models were established for near-infrared detection of straw cellulose. Tests were run with soybean stalk as the object of study. Test results indicated that compared to a model established using a traditional denoising method, the determination coefficients for calibration set models established by second derivative+smoothing and OSC were 0.9318595 and 0.9328905 respectively while the root mean square error for calibration (RMSEC) were 0.6762902 and 0.6696454. For an OSC-PLS regression model with a factor of 8, the relative standard deviation of a prediction model was less than 5%. In the OSC denoising process, the root mean square error fluctuated with the increasing number of PLS factors. Compared to the second derivative-smoothing denoising, OSC-PLS denoising removed the non-correlated variation from spectra and improved interpretation ability of variation while the analysis and convergence were expedited. It was therefore concluded that OSC-PLS denoising could be used to realize the rapid and accurate near-infrared detection of straw cellulose.