Radar Intra–Pulse Signal Modulation Classification with Contrastive Learning

The existing research on deep learning for radar signal intra–pulse modulation classification is mainly based supervised leaning techniques, which performance relies a large number of labeled samples. To overcome this limitation, self–supervised framework, contrastive (CL), combined with the convolutional neural network (CNN) and focal loss function proposed, called CL––CNN. A two–stage training strategy adopted by CL–CNN. In first stage, model pretrained using abundant unlabeled time–frequency images, data augmentation used to introduce positive–pair negative–pair samples learning. second fine–tuned classification, only uses small images. simulation results demonstrate that CL–CNN outperforms other models traditional methods in scenarios Gaussian noise impulsive noise–affected signals, respectively. addition, proposed also shows good generalization ability, i.e., performs well