Radar Intra-Pulse Signal Modulation Classification Based on Omni-Dimensional Dynamic Convolution

In this work, the omni-dimensional dynamic convolution (ODConv) layer based network (OD-CNN) with focal loss function is applied to the radar intra-pulse signal modulation classification, which greatly improves the classification accuracy. Compared to the convolution layer, the ODConv layer employs a novel multi-dimensional attention mechanism to learn four types of attentions along four dimensions of the kernel space in a parallel manner, which further improves the feature mining ability of the model. In order to illustrate the superior modulation classification ability of the proposed model, it is compared with the other three CNN based models. Simulation results demonstrate that the proposed model not only has the highest classification accuracy, but also uses the least number of training parameters and floating point operations (FLOPs).