Abstract: Due to impedance mismatch, nonlinear devices and other reasons, noise floor of signals caught by broadband reconnaissance receiver is not flat, which is easy to cause weak signal undetected, so the need for noise floor is estimated. While the noise floor changes rapidly, the past noise floor estimation algorithm based on morphological filtering has estimation accuracy and large bandwidth signal undetected conflicts. This paper proposed multiscale morphological filtering, which applying different scales filter kernel in different frequency by detection the variation of noise floor estimation in different kernel scales,which increased estimation accuracy in the noise floor. Through simulation experiments, the algorithm can effectively estimate the noise floor, making the spectrum after the correction better for detection.