In the process of actual radar observations, the influence of noise could inevitably generates breakpoint and induces defocusing in the time-frequency image of micro-motion target obtained by narrowband data, which leads to the inaccuracy of micro-motion feature extraction. In this paper, the frequency diversity is utilized in feature extraction of micro-motion time-frequency image. For stepped frequency radar, the wideband radar data is transformed into narrowband data near each frequency point, and then a set of feature can be obtained by the data near each frequency point, then fusion process is done. Measurement and numerical result show the stability and accuracy of feature extraction are improved.