A two channel deep convolutional neural network based method is proposed to predict the remaining useful life of aero engines. The method is to utilize maximal information coefficient for reducing data dimensionality noise, and Kalman filter for reducing data noise on the traditional convolution neural network. The realization of data reconstruction is subjected by data slicing and setting the data slice label as the remaining useful life of the last cycle. The segmented and linear remaining useful life decay models are introduced and a method to judge the starting point of life decay is given. The two features before and during life decay as the input of the twochannel network model are used. The results of testing on NASA turbofan Engine Simulation Data Set (C MAPSS) show that the relevant indexes of this method are significantly better than that of other algorithms when the test data range is large. And this method has significant advantages in the remaining useful life prediction of aero engines.