Application of a new Restricted Boltzmann Machine to Radar Target Recognition
J. Y. Xia, X. Li, Yuyao Liu · 2016
When facing radar target recognition, the main problems focus on the data representation capability and the robustness to cope with noise. The merits of deep learning such as automatic setting for training and hierarchical extraction of features. Most of existing deep networks are related to Restricted Boltzmann machine (RBM), which has played an important role in deep learning techniques. The models election problem in RBM and its deep architecture is very intractable since both their learning and inference are highly time-consuming. As regular RBM has a restriction between visible units and hidden units, this restriction will cause reduction of the recognition probability when training data samples are degraded by noise. Fuzzy Restricted Boltzmann machine (FRBM) is a new RBM, in which the parameters of collection between visible units and hidden units are replaced by fuzzy number. In this paper, FRBM has been applied to moving airplane radar target data. Tested target contain three categories airplane HRRP data samples. The proposed FRBM can significantly reduce the number of free parameters and the degree of over fitting. Moreover, compared with RBM and traditional classification methods, it has showed better representation capacity and better robustness property when the training data are contaminated by noises.