首页|基于改进YOLOv7模型的血细胞检测分类

基于改进YOLOv7模型的血细胞检测分类

扫码查看
目的 探讨改进YOLOv7算法在血细胞图像不同类型细胞自动检测分类中的应用,以提高血细胞识别分类的准确度.方法 将滑动窗口变换器模块引入YOLOv7,同时采用加权双向特征金字塔网络结构,使网络能够获取并传递更加丰富的特征信息,使用斯库拉交并比损失代替完全交并比损失,实现更加精准的目标框定位.结果 通过不同算法在BCCD血细胞数据集上展开实验可得,改进的YOLOv7模型对红细胞、白细胞和血小板的识别准确度分别达到89.3%、98.5%和91.5%,平均准确度达93.1%,相比于原YOLOv7模型提升了2.6%.通过与已发表的血细胞人工智能检测算法进行对比可知,本文算法具有更高的准确度.结论 改进的YOLOv7模型可以有效应用于血细胞识别分类任务,为血细胞的检测提供重要的参考价值.
Blood Cell Detection and Classification Based on Improved YOLOv7 Model
Objective To explore the application of improved YOLOv7 algorithm for the automatic detection and classification of different types of blood cells in blood cell images,to improve the accuracy of blood cell recognition and classification.Methods The swin transformer module was integrated into the YOLOv7,coupled with the adoption of the weighted bidirectional feature pyramid network structure,which enabled the network to acquire and propagate richer feature information.The SCYLLA-IoU loss function was employed to replace the conventional complete IoU loss,resulting in more precise target bounding box localization.Results Experimental evaluations conducted on the BCCD blood cell dataset showcased that the improved YOLOv7 model achieved recognition accuracies of 89.3%,98.5%,and 91.5%for red blood cells,white blood cells,and platelets,respectively.The mean average precision reached 93.1%,which demonstrated a 2.6%improvement over the original YOLOv7 model.Comparative analysis with other published artificial intelligence-based blood cell detection algorithms revealed the superior accuracy of the proposed algorithm.Conclusion The improved YOLOv7 model proves effective for blood cell recognition and classification tasks,which provides significant value in the domain of blood cell detection.

blood cell detectionYOLOv7neural networksweighted Bi-directional feature pyramid network(Bi-FPN)SCYLLA-IoU(SIoU)

刘涛、李明、马金刚

展开 >

山东中医药大学 医学信息工程学院,山东 济南 250355

血细胞检测 YOLOv7 神经网络 加权双向特征金字塔网络 斯库拉交并比损失函数

2022年山东省研究生优质教育教学资源项目

SDYAL2022041

2024

中国医疗设备
中国整形美容协会

中国医疗设备

CSTPCD
影响因子:0.825
ISSN:1674-1633
年,卷(期):2024.39(9)