Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (12): 191-198.DOI: 10.3778/j.issn.1002-8331.2110-0224

• Graphics and Image Processing • Previous Articles     Next Articles

Improved YOLO Framework Blood Cell Detection Algorithm

WANG Yufeng, LI Dahai   

  1. School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China
  • Online:2022-06-15 Published:2022-06-15



  1. 江西理工大学 信息工程学院,江西 赣州 341000

Abstract: In order to solve the problems of low accuracy, wrong detection and missing detection of traditional target detection algorithms in blood cell detection task, a target detection algorithm YOLO-Att based on improved YOLO framework is proposed. Based on the YOLO framework, a multi-scale residual enhancement module is added to the backbone network to improve the utilization rate of feature information by combining with the feature level of low-level information-rich network. An attentional gating structure embedding model is designed to obtain more high quality information of main features. At the same time, Focal loss is used to replace the cross entropy in the original loss function to improve the weight of positive and negative samples and accelerate the convergence rate of the model. Finally, [K]-means++ clustering algorithm is used to optimize the anchor frame of the target to further improve the detection accuracy. Compared with the existing target detection algorithms such as Faster-RCNN, SSD and YOLOv4, YOLO-Att increased mAP to 66.32% in the BCCD detection task of the universal blood cell data set, and the detection rate reaches 85.4 ms, which is more in line with the real-time detection task of blood cell.

Key words: blood cell test, YOLO, multiscale residuals block, attentional mechanism

摘要: 为解决传统目标检测算法在血细胞检测任务中出现的检测精度偏低、错检及漏检等问题,提出了一种基于YOLO框架的改进目标检测算法YOLO-Att,该算法在YOLO框架结构的基础上,在骨干网络中增加了一种多尺度残差增强模块,结合低层信息丰富网络的特征层次,进而提高特征信息利用率;并设计了一种注意力门控结构嵌入模型,以获取更多高质量的主要特征信息;同时使用Focal loss代替原损失函数中的交叉熵,提高正负样本权重,加快模型收敛速度;采用[K]-means++聚类算法对目标进行锚框优化,进一步提升检测准确率。相较于现有的Faster-RCNN、SSD以及YOLOv4等目标检测算法,YOLO-Att在通用血细胞数据集BCCD检测任务中,将mAP提高至66.32%,检测速率达到了85.4?ms,更符合血细胞检测任务的实时性。

关键词: 血细胞检测, YOLO, 多尺度残差块, 注意力机制