Electromagnetic Signal Classification Based on Deep Sparse Capsule Networks
In complex electromagnetic environments, electromagnetic signal classification rates are low as long time have to be the cost to extract features. To cope with the issue, in this paper, an electromagnetic signal classification method is proposed based on deep sparse capsule networks. In the proposed method, received signals are frequency reduced and sampled processing first. Subsequently, a cross ambiguity function based on linear canonical transformation, a cross ambiguity function based on linear canonical domain, and higher-order spectrum are estimated, respectively. The maximum value of each section of the cross ambiguity function is combined with the maximum value of equally spaced cross sections of higher order amplitude spectrum to obtain the two-dimensional feature information. Finally, electromagnetic signals are classified by the deep sparse capsule networks. The simulation results show that the proposed method not only has good classification performance but also can automatically get a hierarchical feature representation by learning. Moreover, the corresponding time cost can be effectively reduced.