In terms of performance, which one is faster, 2D raycast, collission or distance sorting?

I'm met with a 2D situation where I need to be able to detect when objects come into a vision cone (think enemy vision cone in stealth games).

Would this best be done with a cone-like object with collision, a 2D raycast or FindWithTag then converting to polar coordinates of the object's local space and checking whether the angle and distance fit for each object with the correct tag?