Noise-equivalent power (NEP) is a measure of the noise inherent in an electromagnetic radiation detector's (e.g., CCD pixel element's) noise (e.g., thermal noise), specifically the power (e.g., watts) of a signal that equals its inherent noise (i.e., signal-to-noise ratio of 1), which quantifies and aspect of the device's sensitivity. Often the NEP is cited per square-root of the bandwidth to which it applies, in units of watts/Hz1/2. The noise sensitivity of devices is likely to be at least somewhat frequency-dependent, and I imagine the above number is cited for the frequency-regime (band) for which the device is intended and serves as an overall average. For more detail, a device's NEP can be cited or graphed by frequency. The practice of citing the power divided by the square root of the bandwidth seems to suggest the resulting value is expected to be (somewhat) constant over the device's intended frequency-regime.
Noise temperature seems like a measure of the same aspect of a device, but, by name, suggests noise shaped like the Planck function.