Research on Improving the Safety of Construction Workers FPN-PAN-YOLOv5s Method for Crack Detection in Glass Melting Furnace Crowns

To address the shortcomings of existing methods and better detect various cracks appearing on the crown of glass melting furnaces, a network model in the “You Only Look Once” (YOLO) series, YOLOv5s, was used to extract features from furnace arch images. To address the issue of insufficient utilization of shallow feature information in this method, a path aggregation algorithm was adopted to optimize YOLOv5s based on a feature pyramid network (FPN). Then, a furnace arch crack detection method, based on an FPN-path aggregation network (PAN)-YOLOv5s, was constructed. The results indicate that the crack detection method for a glass furnace roof based on the FPN-PAN-YOLOv5s can achieve crack recognition in complex environments, providing effective support for the safety detection of glass furnace roofs and ensuring the safe operation of workshop workers.