Abstract: The non-stationary and nonlinearity of vibration signal of complex turbine rotor will seriously affect the state identification of turbine rotor. In order to ensure the safe operation of turbine rotor, a convolutional neural network (CNN) state recognition method based on symmetrized dot pattern (SDP) feature fusion was proposed. Based on the SDP analysis method, the method conducted feature fusion of signals from all directions and positions of the turbine rotor, obtained the SDP diagram of fused characteristics, and finally performed image recognition of fused characteristics SDP based on CNN to realize rotor fault state recognition. Compared with other state recognition methods, this method improved the representation difference of different state characteristics, and then improved the learning effect and recognition accuracy. At the same time, the experimental results show that compared with other state recognition methods, this method has the highest accuracy of rotor vibration state recognition, reaching 96%.