Abstract：In radar high-resolution range profile target recognition, the representation of data can influence the performance of the classifier significantly. In order to make the data representation more separable, this paper proposes a variational auto-encoder with disentangled representation. By modeling the common representation between different class and the distinctive representation respectively, after optimizing the parameter of the network, it can extract a more separable representation of the data. The measured HRRP data are used to show the effectiveness and efficiency of the algorithm.
