目的 针对药包玻璃瓶缺陷检测中目标检测精度低及小目标漏检率高的问题，提出一种改进YOLOv11的药包玻璃瓶外观缺陷检测方法。方法 首先，在YOLOv11的主干网络中引入动态蛇形卷积网络，通过其自适应地关注不同缺陷特性，有效聚焦不同形状和大小的缺陷特征，增强模型对缺陷局部结构特征的提取能力;其次，在浅层网络中构建多尺度空洞注意力机制，全面捕捉并整合多尺度特征信息;最后，设计微小目标检测层，捕捉网络结构浅层特征中丰富的细节信息，进一步提高微小缺陷目标的检测能力。结果 实验结果表明，该方法在预灌封注射器数据集上的检测平均准确率达到88.38%，较基准模型提升3.8%，特别是在小目标检测上表现突出。结论 改进方法能够有效提高药包玻璃瓶缺陷的检测精度，为自动化检测领域提供一种切实可行的解决方案。

The work aims to proposea detection method for glass vial appearance defects based on an improved YOLOv11 to address the low detection accuracy and high miss detection rate for small defect targets in pharmaceutical glass vial detection. Firstly, a dynamic snake convolution network was introduced into the backbone network of YOLOv11. By adaptively focusing on different defect characteristics, it effectively concentrated on defect features of various shapes and sizes, enhancing the model's ability to extract local structural characteristics of defects. Secondly, a multi-scale dilated attention mechanism was constructed in the shallow network to comprehensively capture and integrate multi-scale feature information. Finally, a small target detection layer was added to capture rich detailed information from the shallow features of the network structure, further improving the detection capability for small defect targets. Experimental results demonstrated that the improved YOLOv11 method achieved a mean average precision of 88.38% on the prefilled syringe dataset, representing a 3.8% improvement over the baseline model, with particularly outstanding performance in small target detection. The proposed method effectively enhances the detection accuracy of pharmaceutical glass vial detects, providing a practical solution for the field of automated inspection.