The invention discloses an initial point detection algorithm of a transient part of a radio frequency signal in a radio frequency fingerprint identification technology. The detection algorithm includes the steps of signal acquisition, interception of transient process waveform of each waveform, signal normalization, envelope extraction, envelope filtering, envelope partitioning, Fourier transform of each block of normalized envelope signal, calculation of frequency domain flatness of frequency domain amplitude, time domain flatness and time frequency flatness of each normalized envelop signal, and detection of minimal values of the time and frequency domain flatness as an initial pint of a transient signal. Based on time frequency and frequency domain characteristics, the transient initial point of a radio frequency signal of a transmitter is detected, thereby effectively extracting fingerprint features, and then a classifier is utilized to classify wireless devices, thereby enabling the radio frequency fingerprint identification technology to be better applied to the field of wireless network security. The initial point detection algorithm of the transient part of the radio frequency signal in the radio frequency fingerprint identification technology has a relatively high identification accuracy rate and relatively short classification time.