This paper proposes a control strategy based on deep learning for a quadrotor to suppress the unknown ground effects during the landing procedure. Due to the complexity of the grounds effects, it is very difficult to obtain the accurate dynamic model. To solve this issue, we set up a compensation model for the ground effects in the landing procedure by using the learning ability of the deep neural network (DNN). Then the super--twisting method is combined with the DNN to formulate a nonlinear robust adaptive landing control strategy which is able to suppress the ground effects and drive the quadrotor to its desired landing point accurately. The Lyapunov based stability analysis and the spectral normalization are employed to prove the stability of the closed loop system, and finite--time convergence of landing error is also achieved. Real-time flight experimental results are included to show the good landing control performance of the proposed control strategy.