Abstract: In order to improve the performance of classification system and further improve the classification accuracy rate of ship radiated noise, a multi-feature fusion classification method based on deep neural network is proposed in this paper. In this method, several different features of the ship radiated noise are extracted firstly, and the extracted features together are used to train a multiple-input deep neural network, so that the network can jointly learn the feature parameters directly and realize the feature fusion through the input branches and connection layers of the neural network, and then the ship radiated noise is classified. The spectral feature, Mel cepstrum coefficients and power spectral feature of the ship radiated noise are extracted for feature deep learning, and the accuracy rate of multi-feature fusion classification method is compared with that of the classification method on a single feature. The experimental results show that the deep learning based multi-feature fusion classification method can effectively improve the accuracy rate of ship radiated noise classification and it is a feasible classification method.