A dataset and a benchmark model for simultaneous detection of oil well drill pipes and their coupling gaps.

Real-time automatic visual detection of oil well drill pipes and their coupling gaps can improve drilling efficiency. Training such a deep learning system requires a large number of labelled images. Our work presents two main contributions: (1) a publicly available dataset comprising 1000 accurately labelled images of drill pipes and their coupling gaps; and (2) a cascaded segmentation model consists of two YOLOv11 modules for simultaneous detection of drill pipes and their coupling gaps. The images were collected from Hubei Jianghan Petroleum Instrument Co., Ltd. We manually labelled 500 raw images with the drill pipes. In addition, we manually annotated the coupling gaps of drill pipes in the 500 cropped images of drill pipe regions. To our knowledge, this is the first publicly available relatively largest dataset in this field, and we hope it can promote the development of related research and applications.