In this paper we discussed the nonlinear increase feature of seismic activity frequency before large earthquakes. By applying the cumulative growth model of seismic activity we found that the occurrence of small earthquakes approximately obeys Poisson's distribution law in a steady state, in which the cumulative frequency of earthquakes shows non-linear behavior before large earthquakes. Then by using the earthquake catalogues of strong and weak earthquakes from China Earthquake Networks Center, we obtained the exponential growth examples of cumulative number of earthquakes before 15 M ≥ 7.0 earthquakes occurred in Chinese Mainland from 1970. We found that there exists the nonlinear growth characteristic of annual-scale in cumulative number of small earthquakes before M ≥ 7.0 earthquakes. The short-term nonlinear characteristic of monthly-scale existed in cumulative number of small earthquakes before some M ≥ 7.0 earthquakes, which can be used as a short-medium-term indicator for earthquake prediction.