In abstract case, you'd probably want to use xy color coordinates from xyY color space to specify LED color. Or even wavelength + spread, as you've tried. Those were the models that were used in one microLED-related research project I was part of.

Using those in images however would be another can of worms, you'll need some kind of physics-based rendering with good HDR tone mapping, as the human perception of light-emitting object against reflective background is highly non-linear.

For your use case it would probably be better to develop a standardized testing setup and just take RGB or XYZ coordinates from an image taken with a calibrated camera. Something like this:

* "standard gray" material surface

* a hole for LED with some kind of light pipe of standard shape

* uniform diffuse lighting of intensity computed to be proportional to your LED power measured in a standard way.

In this way it should be possible to create a catalog of LEDs useful for designing products.