This paper studies the effects of 5 spectral pretreatment methods, including Savitzky-Golay (SG) convolution smoothing, standardized normal variate (SNV), first-order derivative, multiplicative scatter correction (MSC) and wavelet denoising (WDS), on the nitrate partial least squares (PLS) measurement model. The root mean square error of estimation (RMSEE), root mean square error of prediction (RMSEP), correlation coefficient R and regression analysis F-testing for predictive values and actual values are adopted to evaluate the effects of pretreatment. The experiment results show that the prediction accuracy of the PLS model treated by the SG convolution smoothing is higher than that of other spectral methods. Five softwares of spectral pretreatment methods are compiled, achieving the functions of spectrum data acquisition and preprocessing, as well as spectral mapping and preservation. |