Abstract: In the context of complex drilling site environments, where the drilling rod targets exhibit cross-scale characteristics that are difficult to perceive accurately, thus limiting the stable operation of automated drilling rigs, a method for drill pipe recognition and localization based on SDA-YOLO and stereo vision is proposed. Building on the lightweight YOLO v8n-seg model, a multi-level feature fusion module (SDI) and an attention-based scale sequence fusion module (ASF-YOLO) are introduced, and an Inner-CIOU loss function is designed to strengthen the constraints on small-scale targets. The SDA-YOLO model is constructed. To validate the effectiveness of the method, a dataset consisting of 1600 real-field drill pipe images is created, which is divided into training and testing sets at a ratio of 8∶2. Experimental results show that the model achieves precision, recall, and mean average precision values of 99.7%, 93.5%, and 98.0%, respectively, which represent improvements of 13.2, 9.2, and 12.3 percentage points compared to the baseline YOLO v8n-seg model, significantly enhancing the recognition and segmentation accuracy of target drill pipes. Moreover, based on stereo vision theory, integrated with dual calibration and stereo rectification, the RAFT-Stereo stereo matching algorithm is introduced to perform dense disparity estimation, enabling high-precision three-dimensional localization of the drill pipe's inner diameter center, with an average absolute error of 7.82 mm. The study demonstrates that the proposed method can provide reliable visual perception support for automatic centering of drilling rig, intelligent make-up and break-out, and collaborative handling, thus promoting the automation and intelligence level of drilling rig operations.