Industrial pallet identification based on improved YOLOv5

Abstract Pallet recognition is a critical technology for industrial unmanned forklifts, yet accurately locating pallet holes using depth cameras remains challenging due to complex industrial environments. This paper proposes an improved YOLOv5 (named YOLOv5-GE) to recognize and locate the pallet hole position. In the YOLOv5-GE, the ECA (Efficient Channel Attention) module is introduced after the CSP module of the backbone network, and the CBS module of the neck network is replaced by the GSC (Ghost-Shuffle Convolution) module. YOLOv5-GE outperforms the baseline YOLOv5 by 0.71% in mAP@0.5, 8.55% in mAP@0.5:0.95, and 11.27% in FPS. These advancements make YOLOv5-GE particularly suitable for real-time pallet hole recognition in complex industrial settings.