A passive domain adaptive target recognition method

By employing a knowledge distillation method based on a teacher-student model, and combining source domain category relationships with target domain data, the method optimizes category relationships and independence processing, thereby resolving the issue of inconsistent category relationships in source-free domain adaptation and improving the classification accuracy of the target domain.

CN120495769BActive Publication Date: 2026-04-10UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-05-09
Publication Date
2026-04-10

AI Technical Summary

Technical Problem

Existing passive domain adaptation methods suffer from inconsistent class relationships and blurred boundaries due to domain offset in the target domain, making it difficult to effectively improve classification accuracy.

Method used

By employing the teacher-student model in knowledge distillation, and through source domain category relationship optimization and target domain data fusion, the teacher-student model is trained to improve category discrimination. By utilizing the optimized category relationship of the teacher model and the category independence loss function of the student model, adaptive target recognition in the passive domain is achieved.

Benefits of technology

It improves the classification accuracy of the target domain, overcomes the adverse effects of domain offset, and enhances the clarity and discriminability of class boundaries.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120495769B_ABST
    Figure CN120495769B_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of target recognition, and specifically discloses a passive field target adaptive recognition method, which is characterized in that source domain data and target domain data have the same category set, a source domain model is trained by using the source domain data to obtain a model classifier, and the source domain classifier is used to calculate the source domain category relationship; on this basis, a teacher-student model in knowledge distillation is trained and used to realize passive field adaptive target recognition, specifically: the teacher model is used to learn the category information of the target domain, and the source domain category relationship is introduced to make it compatible with the category information of the target domain, so that the category relationship more in line with the actual data style of the target domain is obtained. Meanwhile, the feature representation of each category is refined in the training, the similarity between the categories is reduced, and the discrimination of the categories is improved, so that the boundary line between different categories is clearer. The application improves the classification accuracy of the model.
Need to check novelty before this filing date? Find Prior Art