Abstract: Bolts, as common fasteners, are widely used in various applications. To address the issues of insufficient feature extraction and decreased detection accuracy caused by small bolt targets and complex background interference, an improved YOLOv11n bolt defect detection method is proposed. Firstly, to enhance the model's feature extraction capability, the CBAM attention mechanism is introduced. Secondly, in order to improve the balance between computational efficiency and accuracy in the feature extraction phase, the MobileNetV3 network is embedded into the backbone network structure. Additionally, CIoU is replaced with Alpha-IoU to improve bounding box regression accuracy. Finally, to address task interference caused by shared features in the original detection head for classification and regression tasks, the YOLO detection head is decoupled. Training and testing are conducted on a self-made dataset. Experimental results show that this method achieves a detection accuracy and mean average precision of 88.7% and 82.5%. Respectively, representing improvements of 4.3% and 3.4% compared to YOLOv11n, enhancing detection accuracy and robustness in complex scenarios.