Research On Radar Target Recognition Technology Under The Condition Of Lack Of Knowledge

Due to complicated radar scenes and difficulty in obtaining target data,radar target recognition has always had the problem of lack of knowledge reflected in the modeling process of feature extraction and lack of information,which has always hindered the development of radar target recognition.Based on the diversified data augmentation methods and model optimization strategies of meta-learning,we are committed to exploring radar efficient target recognition algorithm under the condition of lack of knowledge.The main work of this paper is as follows:1.Radar target recognition technology based on data augmentation strategy aims to increase supervision information to achieve more accurate and effective learning,thereby improving radar target recognition rate.We not only use traditional data augmentation strategies to extend data,but also generate new radar target samples based on deep generation model,such as Generative Adversarial Nets.Next,a data enhancement algorithm based on image translation model is proposed.MUNIT is used as the basic model,which increases the amount of information and richness of radar target sample.Meanwhile,based on the real dataset,we construct different degrees of lack of knowledge by changing the amount of sam ple to conduct a complete and effective experimental verification of the algorithm.2.Radar target recognition technology based on model optimization strategy of meta-learning aims to improve model learning method and network structure to achieve more efficient learning.We first introduce the idea of meta-learning with metric learning,which improves CNN learning method and loss function.Secondly,it combines meta-learning with transfer learning and explores the feasibility of transfer methods for few shot target recognition.Then,in view of practical problems faced by meta-learning applied to radar few shot target recognition,a Prototype Metric Network is proposed.Finally,combining meta-learning and transfer learning,a comprehensive solution for few shot target recognition is proposed.This solution can provide the most appropriate few shot learning algorithm according to the actual application scenarios of radar.Moreover,some guidance for engineering applications is put forward.