Abstract: In view of the influence of time variability of the spectrum feature of radiated noise caused by the target motion on the stability of target classification, a spectral shift invariant feature extraction method based on the time frequency image accumulation transform is proposed. It can not only extract the line spectrum signal submerged in the strong noise, but also can give the parameter information of the spectrum line in real time, and meantime combine the parameters of the spectrum line. Based on the principle of auditory feature recognition, the target features are extracted by the anti-frequency-shift modeled octave trianglar filtering method. The simulation and actual data processing show that the proposed features help the detection device overcome the time-varying influence of the spectrum caused by the unknown complex motion of the target and improve the stability of the classification features.