本平台为互联网非涉密平台,严禁处理、传输国家秘密或工作秘密

基于Savitzky-Golay平滑插值的烟草近红外光谱模型转移

Tobacco near infrared spectral model transfer based on Savitzky-Golay smooth interpolation

  • 摘要: 为了解决不同近红外光谱仪器子机与主机光谱成像波数不一致的问题,通过使用Zero、Slinear、Quadratic、Cubic和Nearest 5种插值方法对光谱进行处理,研究提出一种基于Savitzky-Golay平滑插值(SG-Inter)的方法,使子机光谱与主机光谱插值后波数一致,有效解决了光谱波数不一致导致主机模型无法转移的问题。结果表明:利用SG-Inter插值处理后的光谱数据与原始光谱重合度高,且模型转移后的指标预测均方根误差(RMSE)最小。该方法将光谱数据平滑预处理与插值变换同步计算,相对简化了模型转移的工作流程,有效提升了主机模型的通用适配性。

     

    Abstract: In order to solve the problem of inconsistent spectral wavenumbers between slave and host of near infrared spectrometers, spectra were treated by five interpolation models: Zero, Slinear, Quadratic, Cubic and Nearest. A new smooth interpolation method based on Savitzky-Golay (SG-Inter) was proposed to make the wavenumbers of the slave spectra consistent with those of the host spectra after interpolation, and the transfer failure of the host model caused by inconsistent spectral wavenumbers was effectively avoided. The results showed that the spectra processed with the SG-Inter interpolation had a higher consistency degree with the original spectra, the RMSEs (Root mean square errors) of indexes predicted by the model after transfer were the lowest. This method synchronized the calculation of spectrum smooth pre-processing and interpolation transformation, which simplified the workflow of model transfer and effectively improved the general adaptability of the host model.

     

/

返回文章
返回