A galaxy spectrum contains the information of the age and metallicity distribution of the stars in the galaxy. Measuring the stellar population parameters from the observed spectral data is very important for an in-depth understanding of the formation and evolution of the galaxy. LAMOST (Large Sky Area Multi-Object Fiber Spectroscopic Telescope) has observed a large number of galaxy spectra. These spectra are high-dimensional data, and there is a highly nonlinear relationship between the spectra and their physical parameters. Deep learning is suitable for processing multi-dimensional and massive nonlinear data. Therefore, a convolution neural network with 8 convolution layers +4 pooling layers +1 full connection layer is constructed based on deep learning to automatically estimate the age and metallicity of LAMOST Data Release 7 (DR7) galaxy. The experimental results show that the prediction of stellar population parameters (age and metallicity) using convolution neural network model for galaxy spectra is basically consistent with the parameter values obtained by traditional methods with an accuracy better than 0.18dex. As the signal to noise ratio ($S/N_r$) increases, the dispersion of the differences decreases. We also compare the measurement results of convolutional neural network with random forest regression model and deep neural network. The results show that the convolutional neural network is better than the other two regression models.