一种基于YOLO目标检测模型的线上动态学习模型
By using an online dynamic learning model based on the YOLO object detection model, combined with a text-image interaction feature enhancement network of a large language model and a teacher-student model, automatic label generation and high-quality object data augmentation, the problems of object detection model adaptability to new data and labeling cost are solved, achieving efficient self-supervised training and improved detection accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2025-07-17
- Publication Date
- 2026-07-17
AI Technical Summary
Existing object detection models perform poorly when faced with new data, are prone to catastrophic forgetting problems, and labeling data is tedious, time-consuming, and difficult to guarantee accuracy. Traditional dynamic learning methods increase storage costs.
We employ an online dynamic learning model based on the YOLO object detection model, combined with a text-image interaction feature enhancement network of a large language model and a teacher-student model, automatic label generation and high-quality object data augmentation, and self-supervised training by constraining the model output through a loss function.
It effectively alleviates the problem of catastrophic forgetting, reduces data storage costs, enhances the model's adaptability and learning ability to dynamic environments, and improves detection accuracy and generalization ability.
Smart Images
Figure CN120894646B_ABST