Abstract: Dense and cluster fog weather seriously affect the traffic safety during expressway driving. In order to reduce the probability of traffic accidents, a cross-scale feature context fusion network for highway visibility detection in foggy days is proposed, which realizes the automatic classification and early warning of visibility from a single surveillance image. This network used two parallel branches to extract the overall feature of the road and multi-scale detail feature in foggy expressway images. The cross-scale feature context fusion module was designed to calculate the correlation between scales, adaptively fused multi-scale detail feature with the overall feature of the road, and improved the discrimination ability of the network. Meanwhile, a real scenario fog image dataset was constructed, and the images were all from the surveillance videos of multiple highways in China. The experimental results show that this algorithm achieves the highest detection accuracy, and can provide technical support for intelligent management of the highway management departments.