Traditional radar detection methods generally detect the target of a range cell as a single target, but do not estimate the number of targets in the range cell. Aiming at this research vacancy, a target number estimation method based on deep residual learning is proposed. The method converts the radar signal into time-frequency graph and inputs it to the trained deep residual network. The residual network can accurately estimate the number of radar targets according to the difference of time-frequency graph between single target and multiple targets. The residual network can then output an estimate of the number of radar targets. Simulation results show that this method can estimate the number of radar targets effectively.