Online Real-Time Monitoring System for Surface Defects of Hot-State Heavy Rails Based on Zero-Shot Learning and Edge-Cloud Collaborative Machine Vision

In view of the problems and difficulties in quality monitoring of heavy rail universal rolling line, a machine vision system for rail rolling defect monitoring based on image correlation and cloud edge coordination is proposed, In this system real-time images of rail underside, waist surface and tread surface by $\mathbf{4 - 6}$ high frequency cameras is acquired, and the images is transmitted to the edge server for processing through 5 G network, after correlating the frame by frame by Edge server, the defects on the rail surface can be judged by the size and change of the correlating values. The system is characterized by high speed of image recognition, high stability of image data processing and analysis, and high reliability. The image resolution can reach 0.5 $\mathrm{mm} / \mathrm{pixel}$, rail surface defects larger than $3-5 \mathrm{~mm}$ can be clearly collected and displayed. Finally, the real-time monitoring of the surface quality of hot heavy rail is realized successfully by means of the intelligent detection algorithm of surface defects of heavy rail, and the rate of missed detection and false detection in the rolling process of heavy rail is reduced.