I just got some rapideye images and I would like to know if there is need of any preprocessing before applying classification to it? I know that there isn't any automated way of, for example, applying DOS, but my question is: is it actually needed? Based on the comment below: I need to apply either min distance or max likelihood, in order to identify built up areas in multiple images (28 to be exact)
You need good reflectance values if 1) you want to measure some biophysical characteristics of the surface or 2) you want to generalize your approach over large areas.
The radiometric corrections with level 3A are only basic corrections (providing 1/100 radiances), but not reflectance values. If you want to apply an algorithm trained on one image to other images, then you should convert your DN into reflectance values which will reduce the variability due to Sun-Earth distance and solar angle. This is the minimum required preprocessing step (from radiance to top of atmosphere reflectance) in order to have similar statistics on each image, and it is quite simple (information is available in the metadata). The second preprocessing (from top of atmosphere to top of canopy), would require more inputs (information about the atmosphere at the time of acquisition or reference object on each image). This could improve your classification, but I think that it is not necessary in your case.