基于改进YOLOv7模型的血细胞检测算法研究
Research on blood cell detection algorithm based on YOLOv7 improved model
周煜庭 1余华平 1肖粮钧 1何彪 1曾慧群1
作者信息
- 1. 长江大学计算机科学学院 荆州 434023
- 折叠
摘要
在医学上,血细胞计数检测是衡量人体健康与否的重要诊断方法,但是血细胞图像中存在小目标和重叠目标的检测难点.针对上述问题,提出一种改进的YOLOv7目标检测算法.通过对原始的YOLOv7网络增加全局注意力机制(GAM),提升网络的感受野,提高对小目标的检测精度;提出融合了加权双向特征金字塔网络(BiFPN)和递归门控卷积 HorNet的特征金字塔HorNet-BiFPN结构,利用其高阶空间交互作用增强网络的特征融合能力,实现对红细胞重叠区域的建模,解决对重叠红细胞的检测问题.实验结果表明,改进的YOLOv7模型的检测精确率达到了96.3%,对单张图片的检测时间达到了74 ms,对图像中的3类细胞均实现了较强的检测效果,达到了医学辅助诊断的合理性.
Abstract
In medicine,blood count detection is an important diagnostic method to measure human health,However,there are difficulties in detecting small targets and overlapping cells in blood cell images.To solve the above problems,an improved YOLOv7 object detection algorithm is proposed.By adding global attention mechanism(GAM)to the original network,improve the Receptive field of the network and the detection accuracy of small targets.A feature pyramid HorNet-BiFPN structure is proposed that combines the BiFPN network and the recursive-gated convolution HorNet.Its high-order spatial interaction is used to enhance the feature fusion capability of the network,realize the modeling of overlapping regions of red blood cells,and solve the detection problem of overlapping red blood cells.The experimental results show that the detection accuracy of the improved YOLOv7 model reaches 96.3%,the detection time of a single image is 74 ms,and the detection effect of three types of cells in the image is relatively strong,which achieves the rationality of medical assisted diagnosis.
关键词
血细胞检测/YOLOv7/重叠区域检测/全局注意力机制/HorNet-BiFPNKey words
blood cell detection/YOLOv7/overlap area detection/global attention mechanism/HorNet-BiFPN引用本文复制引用
基金项目
新疆自治区创新人才建设专项自然科学计划(自然科学基金)(2020D01A132)
出版年
2024
NETL
NSTL
万方数据