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.

CN122449063APending Publication Date: 2026-07-24GUANGZHOU AUTOMOBILE GROUP CO LTD
View PDF 0 Cites 0 Cited by

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

Technical Problem

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.

Method used

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.

Benefits of technology

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.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122449063A_ABST
    Figure CN122449063A_ABST
Patent Text Reader

Abstract

Embodiments of the present application provide a gas detection model training method, an electronic device and a computer readable storage medium, wherein the method comprises: obtaining first training data and second training data, wherein the first training data is data obtained by signal detection on sample gas in a preset region, and the second training data is data obtained by signal detection on target gas in a target region; pre-training a feature processing layer of an initial model based on the first training data, and fixing model parameters in the feature processing layer to obtain a pre-trained model; and training a scene adaptation layer of the pre-trained model based on the second training data to obtain a gas detection model. The present application solves the technical problem of low gas detection accuracy in related technologies.
Need to check novelty before this filing date? Find Prior Art