Based on baseline is a smooth curve and under
the collected spectrum, a robust penalized quantile regression
with B-spline basis has been proposed to baseline estimation.
Then an iterative reweighted method has been adopted for
quantile regression optimization. Instead of man tuning the
hyperparameter in penalized quantile regression, augmented
Lagrangian method is applied to hyperparameter optimization.
Experiments on simulated and real data sets show that our
method is more effective in baseline correction than other
baseline estimation methods in simulated data set. For real
data set, the calibration results after the baseline correction
step are better than other preprocessing and baseline correction
methods.