Abstract: The difference between the fingerprint features of radiation sources is subtle and the recognition rate is easy to decline due to the noise interference. To solve these problems, we proposed an optimization algorithm of individual recognition of radiation sources based on stacking method which can integrate the recognition results of multiple heterogeneous networks. Based on the different fingerprint features extracted by heterogeneous networks under the condition of low SNR, heterogeneous networks were able to improve the extraction ability of fingerprint features. At the same time, this paper used the heterogeneous network of EfficientNets series with small network scale and obvious structural difference as the basic network in order to avoid the excessive model scale caused by the improvement of classification accuracy. Firstly, under the condition of Gaussian channel, the network model effectively identified the spurious noise of power amplifier and then the optimization algorithm improved the overall performance of the model. The results show that this method can utilize further the differences between the fingerprint features of the signal and has a higher recognition rate for different individual radiation sources.