An attribute-driven dynamic model optimization method and system
By controlling the dynamic activation of attribute recognition and self-training, and combining attribute association rules and user corrections, the attribute recognition model is optimized, solving the problem of illogical recognition results in existing technologies, improving recognition accuracy and adaptability, and realizing the model's self-learning and optimization.
Patent Information
- Application Number
- CN202511501840.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-21
- Publication Date
- 2026-01-20
- Estimated Expiration
- 2045-10-21
AI Technical Summary
Existing deep learning-based attribute recognition methods suffer from illogical and common-sense results in practical deployments, leading to invalid alarms and erroneous events, and low recognition accuracy.
By acquiring algorithm configuration parameters, the activation and deactivation of attribute recognition and self-training are controlled. Initial attribute information is obtained using object detection and attribute recognition models, and post-processing is performed based on preset attribute association rules to generate final attribute information. User corrections are received and model training is performed. Attribute alarm rules, coexistence rules, and mutual exclusion rules are introduced for refined processing.
It improves the accuracy and adaptability of attribute recognition, reduces false alarms, enhances the real-time performance and interactivity of the system, ensures high quality of labeled data, and enables the model to learn and optimize itself.
Smart Images

Figure CN120976684B_ABST
Abstract
Citation Information
Patent Citations
Construction site personnel attribute detection method, system and device and medium
CN114140745A
Interstitial tumor mitotic cell recognition and counting method based on deep learning
CN118918096A