[superkuh]

It depends on the type of observations being made. If it is done with a single dish with a single antenna feed then to make images you just have to know when and where you are pointing while you record the power.

But most images you see of radio astronomy are done with interferometers. This means you use multiple pairs of antenna. Each pair will respond to a point source going across the sky with the interference pattern modulating the voltage at the receiver that follows an acos pattern relative to the distance between the antennas in the pair.

The multiple pairs of antenna all each contribute a 2D fringe pattern. The interference makes the receivers record what is essentially the fourier transform of the sky. So some pair might have sine wave of a particular wavelength and angle, and another pair another angle. By combining all these 2D patterns of waves the sum ends up being an "image" of the sky (convolved with the primary and sidelobe beams).

That gets you a "DIRTY" image, and it really is dirty. The beam of an interferometer is a complex thing. So after that multiple types of cleaning algorithms may be used to use prior knowledge to pull just the actual signal out of the noisy, gunked up "image".

So in the end each pixel is really the result of a somewhat arbitrary algorithm picking out the bright bits after taking the FFT of the sky of each baseline pair and combining them.

If you want to know more I recommend running the python (cross platform) interferometry simulator https://launchpad.net/apsynsim while watching the video lecture series at https://www.youtube.com/watch?v=0TwnZhiEc3A and then https://www.youtube.com/watch?v=mRUZ9eckHZg