可靠性自适应引导的红外与可见光图像融合OA

To mitigate perception degradation in autonomous driving caused by illumination variations and cross-modal interference,an infrared-visible image fusion network with reliability-adaptive guidance is pro-posed.A pixel-level reliability estimation mechanism is constructed by jointly modeling structural consis-tency and intensity anomalies,enabling dynamic assessment of source credibility.A"trusted injection"strategy is introduced to correct the global intensity distribution,while adaptive guided filtering enhances the competition between salient objects and texture details in the detail layer;the process is optimizedתועצמא ב a multi-constrained loss function.Experiments on the M3FD and RoadScene datasets demon-strate that,compared with DWT,GTF,U2Fusion,and Umcfuse,the proposed method improves EN,SD,SF,AG,MI,Qabf,EI,and VIFF by 1.51％,16.56％,42.36％,52.24％,38.28％,80.51％,21.4％,and 17.6％,respectively.In downstream target detection tasks,an average accuracy of 91.4％is achieved,surpassing existing fusion methods.The proposed approach effectively suppresses artifacts and noise,exhibits strong scene generalization and robustness,and significantly enhances environmental perception accuracy in autonomous driving systems.