Gas detection model training method, electronic device, and computer-readable storage medium
By pre-training the feature processing layer in the gas detection model and fine-tuning the scene adaptation layer in the target region, combined with staged training and two-sample validation, the problem of low accuracy of the gas detection model in dynamic environments is solved, enabling rapid deployment and stable operation across scenarios, and improving detection accuracy and response efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU AUTOMOBILE GROUP CO LTD
- Filing Date
- 2026-04-20
- Publication Date
- 2026-07-24
AI Technical Summary
Existing gas detection models cannot adapt to dynamic changes in the field environment, resulting in low detection accuracy. In particular, they lack feature adaptation capabilities when faced with temperature and humidity fluctuations, cross-interference of multiple gases, or sensor aging.
By acquiring training data from preset and target regions, the feature processing layer of the pre-trained model is fixed with parameters, and the scene adaptation layer is fine-tuned only in the target region. A phased training strategy and a two-sample verification mechanism are adopted to achieve rapid deployment and on-demand upgrades across scenarios, thereby improving the model's transfer efficiency and generalization ability.
Without requiring retraining the entire model, it improves the accuracy and environmental adaptability of gas detection, ensures the stability and reliability of the system, supports edge deployment, reduces data dependency and communication overhead, and improves response efficiency and security.
Smart Images

Figure CN122449063A_ABST