This source would argue the other way

https://cissm.umd.edu/sites/default/files/2019-08/2000-UCS-C...

It seems that a warhead or decoy would fill a single pixel on the sensor until you got very close to the target, 1 km or so, at which case you have to execute a high-g maneuver in a few ms. The state of the art is a "two color" system where you could make either the warhead or the decoys any "color" you want with surface treatments and/or thermal management (worst case: put the warhead inside a shroud cooled to liquid nitrogen temperature and fire the weapon at night when it won't be illuminated by the sun.)

There was a test in the early 2000s I read about where they were able to pick out a warhead which was intermediate in properties with a set of decoys with various surface treatments. That 's great but they knew exactly what they were up against which we wouldn't know if it was a North Korean missile.

I'd have more faith in the discrimination abilities of ground-based radars in the 12 GHz range than in the "two color" focal plane imager system.