Specific emitter identification is playing an increasingly important role in the military and civilian sectors. With the rapid development of deep learning technology, the recognition performance of SEI method has been significantly improved. However, these methods often rely on a large number of radiation source sample data, and perform poorly in practical application scenarios with limited sample size. To solve this problem, this study proposes a novel deep learning network model CRCPA-GCN to achieve SEI in small-sample scenarios. The model integrates CPCA and GCNet attention modules in a multi-layer complex convolutional neural network, and the recognition performance in small-shot scenarios is significantly improved by using the methods of class reconstruction and adversarial training. In this study, a series of experiments were carried out on public datasets and compared with the current mainstream SEI networks. Experimental results show that the proposed CRCPA-GCN network model achieves an accuracy of 95.81% under the learning condition of 20 shot, which is better than other mainstream SEI networks and performs well in robustness.