Bathymetry is very important for the study of marine environment. The traditional laser echo algorithm can process the echo signal quickly and realize the water depth measurement. However, due to the influence of water turbidity and water depth, the laser echo signal obtained in some areas will appear weak water bottom echo signal or overlap of water surface and bottom echo, which brings challenges to the extraction of water depth information. In order to solve these problems, a Deep Learning model CNN-LSTM is proposed in this paper. Firstly, each bin value of laser echo are taken as data point, then these data points are classified into water surface points, water bottom points and noise points by deep learning method. The water depth information of the laser echo signal is calculated according to the coordinates of water surface points and water bottom points. Data points classification and bathymetric experiments are carried out with laser echo data from the South China Sea, the experimental results show that the classification accuracy of this model reaches 97.62%. At the same time, the water depth information of the laser echo signal is calculated and compared with the in-situ data. The RMSE reaches 0.46 m which is better than the single CNN、LSTM and 1D FCN models. This paper provides a good idea and scheme for the field of laser echo sounding.