Specific emitter identification (SEI) technique means that the relationship between the signal and the individual of the radiation source is judged by the difference in the signal of the transmitter response. Both the traditional methods and new emerging methods to make the neural networks for SEI rely on signal samples with category information. However, in conventional practice, the signal samples with category information are difficult to acquire. In order to solve this problem, this paper introduces a density peak clustering (DPC) algorithm in unsupervised learning to achieve SEI without classless information signal samples. Since the performance of the DPC algorithm is greatly influenced by the artificial input parameterdc, this paper utilizes the diffusion equation and the kernel density estimation improved algorithm for classifing the data without the need of manual input parameters. The algorithm proposed in this paper is good in effects, and reliable and effectiveness in algorithm.