The high-throughput satellite system, which supports Giga/Tera bps transmission, is extremely limited by the nonlinear distortions introduced by TWTA amplifier and Group delay effects of devices. In this paper, we propose two equalization algorithms based on convolutional neural network(CNN), one uses regression method while the other uses classification method. The algorithms aim to achieve more precise and robust equalization for the nonlinear distortions of the satellite channel with resnet and dropout layers reducing overfitting of the network. The proposed equalizers are compared with the current equalizers used least mean square (LMS) and traditional neural network equalizers. The simulation results show the BER performance of equalizers we design outperforms. At the same time, the better performance during different satellite channel model shows their strong robustness.