AN IMPROVED BOOSTING-BIPLS MODELS BASED ON WEIGHT ADJUSTMENT FOR SOIL HEAVY METAL CONTENT PREDICTION

Dong Ren, Jun Shen, Shun Ren, Kai Ma, and Xinting Yang

Keywords

XRF, heavy metal, Boosting-BiPLS, spectral similarity

Abstract

Heavy metal pollution in soil has got more and more attention, and X-ray fluorescence spectroscopy analysis is a widely used method for heavy metal content in soil. The establishment of accurate model is helpful for the rapid detection of heavy metal content. Firstly, eight spectrum pretreatment methods are used before modelling, and the pre-processing method of least squares which improved multi-scatter correction is chosen. Secondly, Boosting-backward interval partial least squares (Boosting-BiPLS) model is established which combines several basic models with different characteristics into a strong one to solving the “building nesting effect of BiPLS. Then from bias oriented, an improved Boosting-BiPLS model is proposed, which the weight of samples is adjusted on the basis of the relative deviation of the samples and the weight of base models is dynamically calculated by the spectral similarity. Finally, to prove the effectiveness of the improved model, the improved Boosting-BiPLS model is compared with the traditional Boosting-BiPLS model. The results show that the correlation coefficients of the five heavy metal elements of the improved Booting-BiPLS model are all about 0.99, and the average relative deviations are all <10%, with the prediction accuracy of Boosting-BiPLS improved by more than 50%. Moreover, the model is more stable.

Important Links:



Go Back