Abstract: Structural feature vectors, which extracted from the harmonic waves in DEMON spectral, can be used to establish the templates for recognizing the propeller blade-number of ship propeller. However, in the recognition method based on template matching algorithm, there are some problems hard to be solved, such as relying on template library and confidence factor algorithm, containing too many constraints and unable to find missing templates. In this paper, a Deep Neural Network (DNN) based method for propeller blade-number recognition is proposed. In this method, the template library is only used when training the deep neural network, so that the problem of relying on template library and confidence factor algorithm disappears in the recognition process. In addition, by extracting the recognition error item, the missing templates can be found as the supplement of the template library. Through the tests of propeller blade-number recognition from the measured large amount of ship radiated noise data, it is confirmed that the DNN based method has higher accuracy in propeller blade-number recognition, and the recognition process is more simple and reliable.