Abstract: To mine hidden information from massive AIS trajectory data and provide a scientific basis for decision-making of marine fishery management departments, this paper proposes a marine fishing vessel density prediction method based on deep learning and fusion of spatial-temporal features. Firstly, the driving area of fishing vessels is grided according to fishing vessel trajectory dataset. Secondly, high-density fishing vessel areas are selected for study to avoid data sparsity. Thirdly, the fishing vessel distribution data is constructed into a three-dimensional matrix of spatial and temporal fusion. Finally, the convolutional recurrent neural network model is used to capture spatial and temporal features, while the convolutional neural network is stacked to enhance the learning of spatial features. The experiment was specifically tested with real fishing vessel trajectory data of the East China Sea. Results showed that the predicted values of fishing vessel density were very close to the true values, with an average absolute error of 4×10^-4. It indicates that the model can better fit the distribution characteristics of fishing vessel density, which improve effectively the accuracy and robustness of fishing hotspot prediction.