Radar one-dimensional range image target identification method based on deep belief network

The invention belongs to the radar technology field, and concretely relates to a radar one-dimensional range image target identification method based on a deep belief network. Frequency domain features and subspace features of a target one-dimensional range image are combined, formed new feature vectors serve as input data to train and identify a network, and the combined feature vectors carries out expansion on sample information to facilitate the improvement of the precision of a model. In a constructed deep belief network containing three hidden layers, each layer input is formed by parallel connection of output of the former two layers. The structure enables learned network parameters to depend on feature information of different layers instead of independent relying on the current layer, so the expression capability of output features on original information is further improved and the identification rate is increased. Moreover, a model of the invention is utilized to carry out an identification test on