YOLO11n-Screw: A Recognition Method for Aircraft Screw Tail Marks

Abstract Aircraft screws are manufactured from various materials depending on their specific applications and are identified by tail markings. To enable the automated sorting of screws removed during aircraft maintenance and repair, a lightweight and high-precision object detection model, YOLO11n-Screw, is proposed for the recognition and classification of these screw tail markings. The model incorporates the RepNCSPELAN4 backbone network, which effectively reduces the number of parameters while maintaining high performance. An improved LSCD detection head structure enhances the model's accuracy in localizing key regions and strengthens its recognition capabilities. Additionally, the integration of the Cascaded Group Attention (CGA) mechanism allows for more effective focus on the subtle features of screw tails within complex backgrounds. Experimental results demonstrate that YOLO11n-Screw performs exceptionally well in detecting aircraft screw tail markings, achieving a precision of 97.5%, a mean average precision (mAP) of 96.2%, and an F1 score of 0.94. Furthermore, the model’s parameter count is reduced to 2.25×10⁶M, and its size is compressed to 4.9 MB. Compared to the original YOLO11n model, the improved version shows significant advancements in detection precision, model lightweighting, and adaptability to industrial scenarios, offering strong potential for engineering applications.