A radar target recognition method based on adaptive label knowledge distillation

By using an adaptive label knowledge distillation method and optimizing student model training with estimation weights and temperature prediction modules, the problems of high computational load and insufficient real-time performance in radar target recognition are solved, and efficient radar target recognition results are achieved.

CN118194973BActive Publication Date: 2026-06-02XIDIAN UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
XIDIAN UNIV
Filing Date
2024-03-29
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

In existing technologies, radar target recognition methods based on deep learning suffer from problems such as large model computation, limited resources, and difficulty in meeting real-time requirements. Traditional knowledge distillation methods have failed to effectively solve the problems of low recognition rate and poor implementation caused by the imbalance of soft targets in teacher models and the sensitivity of distillation temperature.

Method used

An adaptive label knowledge distillation method is adopted. The teacher model is pre-trained by constructing a training dataset. The influence of class imbalance is reduced by estimating weights and temperature prediction modules. The temperature gradient is distilled by back learning to guide the training of student models, thereby achieving adversarial learning and course learning effects.

Benefits of technology

Effectively compressing model parameters improves radar target recognition performance, meets real-time requirements, reduces model deployment difficulty and time cost, and enhances recognition accuracy and feasibility.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118194973B_ABST
    Figure CN118194973B_ABST
Patent Text Reader

Abstract

The present application relates to a kind of radar target identification method based on adaptive label knowledge distillation, pre-training teacher model is carried out using the training data set of construction, and student model is trained by training data set and pre-training teacher model guidance, using the student model of training completion to the HRRP data of measurement carries out radar target identification, obtains target identification result.The present application realizes the compression of model using label knowledge distillation method, solves the problem that the parameter quantity of neural network radar terminal deployment is huge and difficult to meet the high real-time requirement.Problems, using the tendency score of each training sample to calculate the estimated weight of each training sample, to solve the problem that each class sample information contained in teacher model is not balanced, generate distillation temperature for each training using temperature prediction module, gradually improve the training difficulty using distillation temperature, achieve the effect of course learning, save the time cost wasted in adjusting distillation temperature parameter.
Need to check novelty before this filing date? Find Prior Art