I don't think you can. You would probably need a spectrograph. The problem is that each color has overlap from each other color along with different quantum efficiency, which is how many photons the sensor records compared to how many photons there originally were. Also, each color is recording a large range of wavelengths within it's range, not all of which are equally captured by the sensor.
If you knew the QE of your sensor, and were able to take very good calibration and test frames you MIGHT be able to get an approximation, but I don't know how accurate it would be. Perhaps a bit better than using your eyeball and saying "It looks deep red, so it's X temperature".