Abstract: In the sensor network, the missing data is inevitable due to many of interference factors, which hinders data analysis and processing.In this research, by using the convolution neural network(CNN) which can extract and learn effective features data independently, a missing data imputation approach based on convolution neural network was proposed. Firstly, for the temporal correlation of time series and the spatial correlation between sensor nodes, the CNN imputation model was used to estimate the missing data with single dimensional correlation. Then, based on the imputation results of temporal dimension and spatial dimension, the missing data were imputed by considering the multidimensional correlation. Finally, experimental verification is performed on the real data set. Experiment results show that the imputation result considering the temporal and spatial multidimensional correlation is better than the imputation result considering the single dimensional correlation. And compared with the BP model, the validity of the convolution neural network imputation model is verified.