A vector hydrophone array signal direction of arrival estimation method based on RBM-BP is proposed to address the problems of low training efficiency and estimation accuracy in existing neural network-based direction of arrival estimation algorithm models under low signal-to-noise-ratio (SNR) conditions. Firstly, the Restricted Boltzmann Machine (RBM) algorithm is used to extract features from the signal data output by the vector hydrophone, and then the extracted high-dimensional feature data is transformed into the main low-dimensional feature data; Then use these data to train the neural network model through the Back Propagation (BP), and finally use the trained model to estimate DOA. In the simulation experiment, BP and RBM-BP models were trained under five different SNR conditions, and the performance of the two models was compared. The experimental results show that under low SNR, RBM-BP has higher DOA estimation accuracy and less training time. Overall, the performance of RBM-BP algorithm is superior to that of BP algorithm.