There are several ways, depending on whether the baseline drift is linear or nonlinear. I would start with the detrend function, and if that didn’t give an acceptable result, then consider a low-order polynomial using polyfit and polyval.
There are several other ways to deal with it, including using a bandpass (or highpass) digital filter with a very low low-frequency cutoff. (If possible, I prefer the filter because it gives me more control over the frequency content of my filtered signal. If you want to design a filter, do an fft of your signal first, so that you can determine where best to define your cutoff frequencies.)
