The invention provides a wavelet decomposition binary denoising method for the ultrahigh-frequency signal waveform of the partial discharge. The method comprises the steps of 1) arranging a partial-discharge ultrahigh-frequency sensor on to-be-detected power equipment, and acquiring the ultrahigh-frequency signal of the partial discharge by detection equipment; 2) according to a selected mother wavelet and a decomposition level, conducting the multi-scale wavelet decomposition treatment on the signals of the partial discharge to obtain a wavelet coefficient and a signal waveform at each scale; 3) in order to reduce the signal waveform distortion of the partial discharge, processing the coefficient of the wavelet decomposition, calculating the mathematical morphological gradient of the cumulative energy function of signals at each scale, calculating the steepness of the mathematical morphological gradient, comparing the steepness with a preset threshold, deeming the signals at the above scale contain a partial discharge signal and keeping the wavelet coefficient at the above scale to be constant on the condition that the steepness is larger than the preset threshold; or deeming the signals at the above scale only contain noise signals and resetting the wavelet coefficient at the above scale to be zero on the condition that the steepness is not larger than the preset threshold; 4) reconstructing the wavelet according to the processed wavelet coefficient to obtain the denoised ultrahigh-frequency signal of the local discharge.