One-dimensional Range Profile Target Recognition Based on Few-shot Learning
Abstract
Using one-dimensional range profile(ORP) for radar target recognition has an important research value because of its easy acquisition, easy processing and fast calculation. Currently, the ORP recognition method represented by deep learning has been developed rapidly, while deep learning often relies on large, high-quality data sets. However, due to the particularity of requirements, it is difficult to obtain a large number of target data for model learning. When the amount of data is insufficient, the model is prone to overfitting. To solve this problem, this paper proposes a method for ORP target recognition based on few-shot learning. By using metric learning, a cascaded metric loss function is established to make the features of same class more clustered in the model, at the same time, increase the differentiation between different classes of features, so as to complete the task of target recognition under the condition of small samples. The simulation data of five aerial targets are used in the experiment, and the results show that this method can effectively realize the target recognition