Face recognition is a commonly used biometric identification method, but in a complex environment, due to the low quality of imaging devices, the obtained face image has low resolution and contains little information. In this paper, an adaptive learning algorithm based on convolutional neural network is used. This algorithm uses each layer of convolutional neural network to effectively integrate local and global features, and at the same time reduces the problem of gradient correlation decline caused by the increase of network layers. In this paper, a SuperResolution Network (SRN) method is used to up-sample the trained image, and Bicubic sampling is used to improve the quality of the trained image. After the reconstruction of SRN, the recognition rate of ResNet (Residual Network) is obviously improved under the condition of low resolution, which proves that the method used in this paper is effective.