Using the Radar cross section (RCS) to independently recognize the ship targets can strengthen the autonomous decision-making ability of the combat platform and promote the unmanned combat OODA cycle. In order to mine the deep features of RCS sequence data and realize high-precision and efficient ship RCS recognition, Bi-directional gated recurrent unit net (BiGRU-Net) was established, which reduces the impact of long-term dependency problems on recurrent neural network, while ensuring recognition efficiency. The experimental results show that compared with other recognition models, the recognition accuracy of BiGRU-Net is increased by 1.43%, 14.37%, which can effectively improve the recognition accuracy of RCS under different noise environments while ensuring the recognition efficiency.