A real-time vehicle-mounted image defogging method, system and storage medium

By employing a two-stage coupling optimization method, the dehazing knowledge of the teacher model is transferred to the lightweight student model, and optimization is performed in the closed-loop domain adaptive stage. This solves the problems of high model complexity and insufficient generalization ability in existing technologies, and achieves efficient and reliable image dehazing processing under adverse weather conditions.

CN122265101APending Publication Date: 2026-06-23DONGFENG MOTOR GRP
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
DONGFENG MOTOR GRP
Filing Date
2026-03-12
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

Existing image dehazing methods have limited generalization ability under real atmospheric conditions, and their complex model structures make them difficult to meet the requirements of real-time processing in vehicles. Furthermore, in the absence of real paired data, it is difficult to construct effective unsupervised cross-domain constraints, resulting in insufficient visual realism and physical rationality of the dehazing results.

Method used

A two-stage coupled optimization method is adopted. The dehazing knowledge of the teacher model is transferred to the lightweight student model through the structured knowledge transfer stage, and the student model is optimized in the closed-loop domain adaptation stage. The visual-language model is introduced as a semantic supervision source by combining supervised loss and feature alignment loss to ensure that the dehazing result conforms to the physical laws of the natural scene and human cognitive priors.

Benefits of technology

It significantly reduces the computational complexity and number of parameters of the model, improves the generalization ability and perception robustness in harsh weather environments, ensures the semantic consistency and physical credibility of the defogging results, and meets the real-time processing requirements of vehicles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122265101A_ABST
    Figure CN122265101A_ABST
Patent Text Reader

Abstract

The application discloses a kind of real-time vehicle image defogging method, system and storage medium, belong to image processing technical field.The method includes offline training stage and online inference stage.Offline training stage first executes structured knowledge transfer, utilizes the clear-foggy image pair of source domain with annotation, the defogging knowledge of pre-trained teacher model is transferred to lightweight student model, and preliminary student model is obtained;Then closed-loop domain self-adaptation is executed, and the real fog image pair in target domain is used to optimize the preliminary student model in double layers, the inner layer maintains the basic performance of source domain, and the outer layer updates the parameters of target domain by dynamic smoothing mechanism, to obtain the final defogging model.Online inference stage inputs the fog image collected by vehicle camera in real time into the final defogging model, and outputs corresponding defogging image and transports to downstream perception module.The application effectively overcomes the domain offset problem while significantly reducing the model calculation overhead.
Need to check novelty before this filing date? Find Prior Art