【摘要】 针对一次即可识别图中物体(You Only Look Once v8,YOLOv8)的模型在密集缺陷检测任务中因特征提取能力不足导致的漏检、错检等问题，文章提出SE-YOLO算法。改进后的SE-YOLO密集缺陷检测算法为了使模型关注更多维度的特征信息，使用Swin Transformer网络作为主干网络。文章引入中心化特征金字塔模块，以提取全局长距离相关性，可以尽可能地保留输入图像的局部角点区域信息。改进后的SE-YOLO密集缺陷检测算法可以更加准确地检测出缺陷的类别和位置，在密集缺陷检测任务中具有较高的精确度与鲁棒性。更多还原

【Abstract】 Aiming at the problem of missed detection and false detection caused by the lack of feature extraction ability in the dense defect detection task, the SE-YOLO algorithm is proposed for the model of You Only Look Once v8(YOLOv8). The improved SE-YOLO dense defect detection algorithm uses the Swin Transformer network as the backbone network in order to make the model focus on more dimensional feature information. The centralized feature pyramid module is introduced to extract the global long-distance correlation, and the local corner region information of the input image can be retained as much as possible. The improved SE-YOLO dense defect detection algorithm can detect the category and location of defects more accurately, and has high accuracy and robustness in dense defect detection tasks.更多还原