A long-tail radiation source individual identification method and device based on domain generalization and a storage medium

By constructing a multi-source domain dataset and introducing a gradient reversal mechanism and class balance loss, the radiation source individual identification model is optimized, solving the identification difficulties caused by long-tail distribution and neighborhood offset, and achieving high-precision and robust radiation source individual identification.

CN122046035BActive Publication Date: 2026-07-21HANGZHOU DIANZI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HANGZHOU DIANZI UNIV
Filing Date
2026-04-20
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing radiation source individual identification technologies struggle to achieve high-precision and robust radiation source individual identification under conditions of long-tail distribution and domain offset, especially for rare categories where the identification accuracy is insufficient.

Method used

By constructing a multi-source domain dataset, introducing a gradient reversal mechanism and class balance loss, and combining a feature extractor, classifier, and domain discriminator, the overall objective function is optimized to improve the model's generalization ability and recognition accuracy.

Benefits of technology

Without requiring prior information about the target domain, the model effectively improves the accuracy and robustness of individual radiation source identification in complex environments, especially significantly improving the identification performance of rare categories.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122046035B_ABST
    Figure CN122046035B_ABST
Patent Text Reader

Abstract

The application discloses a long-tail radiation source individual identification method and device based on domain generalization and a storage medium. The method is characterized in that: a source domain data set with multiple categories in a long-tail distribution is obtained; batch training data is sampled therefrom; a feature extractor is used to extract feature representation, which is simultaneously input into a classifier and a domain discriminator; then, based on a classification prediction result, a domain prediction result combined with a gradient inversion mechanism, and a class balance loss, an integrated overall objective function is constructed; all model parameters are jointly optimized by minimizing the objective function, an optimized identification model is obtained, and the optimized identification model is finally used for high-precision individual identification of target domain radiation source signals. Through collaborative optimization of classification accuracy, class balance and domain generalization capability, the application effectively solves the problem of insufficient identification performance of the model on an unknown target domain in a complex scene where training data is in a long-tail distribution and there is domain shift.
Need to check novelty before this filing date? Find Prior Art