指纹识别是身份认定和同一认定中重要的方法之一，但由于其特征尺寸较小、分布密集并且存在一定的漏检以及误检等问题。针对以上问题，提出了一种基于双层路由注意力机制（bi-level routing attention, BRA）的指纹二级特征检测方法。在Yolov8中嵌入BRA注意力机制、从而减少在指纹特征检测过程中出现的漏检以及误检等问题并实现更灵活的内容感知；调整YOLOv8的网络结构，针对指纹特征添加小尺寸目标检测层。实验结果表明YOLOv8-B网络模型的平均精准度mAP@0.5提升了4.3%，mAP@0.5:0.95提升了8.5%，分别达到98.2%和74.9%。并且检测速度基本保持不变，能够有效地检测指纹的二级特征，降低误检、漏检等问题的发生。

Fingerprint recognition is one of the important methods for identification or authentication, but there are some problems such as missing and false detection due to its small feature size and dense distribution. To solve the above problems, a fingerprint secondary feature detection method based on BRA(bi-level routing attention mechanism) was proposed. The BRA attention mechanism was embedded in Yolov8 to alleviate the problems of missing and false detection during fingerprint feature detection to enable a more flexible content awareness. The network structure of YOLOv8 was adjusted and a small-size target detection layer for fingerprint features was added. The experimental results show that the average accuracy of the YOLOv8-B network model Is increased, with mAP@0.5 increased by 4.3% and mAP@0.5:0.95 increase by 8.5%, reaching 98.2% and 74.9% respectively. And the detection speed remained basically unchanged, which can effectively detect the second-level characteristics of fingerprints and reduce the problems such as false and missing detection.