Convolutional neural networks have been widely used in SAR target recognition. However, due to the small number of target samples in SAR images and coherent speckle noise in images, the networks cannot fully learn the deep features of samples, which exerts a certain impact on the recognition performance of the networks. To address the above problems, this study proposes a data fusion-based target recognition method. The algorithm firstly suppresses noise and extracts edge information of the original image and then fuses the processed two types of feature information. It expands the single-channel grey-scale image fusion to a two-channel image as the training sample and constructs a convolutional neural network model with high- and low-layer features fused, which uses the attention mechanism to enhance the learning of useful features. The experimental results reveal that the method demonstrates excellent performance in the recognition of different target models on the MSTAR dataset.