Traditional methods for catenary bolt recognition are inefficient, and there is limited research on existing bolt localization approaches. To address these issues, a novel bolt recognition and localization method is proposed, combining an improved YOLOv8n model with an RGB-D camera. The YOLOv8n object detection model is enhanced by adding a small target detection layer to improve feature fusion; an SE attention mechanism is introduced to highlight key features and improve detection accuracy; a BiFPN module is incorporated to enhance multi-scale feature fusion efficiency; lightweight SCDown convolutions are used to reduce the model's parameter count while maintaining detection performance; and the WIoU-v3 loss function is adopted to accelerate model convergence and improve regression precision. Based on the symmetrical properties of bolts and image processing techniques, a new method is developed for bolt localization, integrating the object recognition model with depth information from the RGB-D camera to obtain the bolt's 3D center coordinates and spatial pose. Experimental results show that the improved recognition model achieves an mAP@0.5 of 90.7% on the test set, which is 2.1% higher than the original model and reduces the parameter count by 8.0%. The bolt localization method effectively locates catenary bolts, validating the reliability of the proposed approach.