基于YOLOv4⁃tiny模型的细胞图像识别技术研究  

Cell image recognition system based on YOLOv4⁃tiny model

作　　者：柴媛媛 CHAI Yuanyuan(Basic Courses Department,Logistics University of People′s Armed Police Force,Tianjin 300309,China)

机构地区：[1]中国人民武装警察部队后勤学院基础部,天津300309

出　　处：《现代电子技术》2022年第9期46-49,共4页Modern Electronics Technique

基　　金：2019年中国人民武装警察部队后勤学院校级项目:面向野战环境检验医学的显微视觉关键技术研究(WHJ201903);2021年中国人民武装警察部队后勤学院校级项目:基于课程思政的计算机基础课程在线教学模式探索与研究(WHJY202110)。

摘　　要：根据细胞的形态特征进行病理分析是现代医疗健康领域常用的技术手段,传统的细胞识别及分类存在易疲劳、效率低、医师水平及主观因素带来的不确定性等问题。为此,提出基于YOLOv4⁃tiny模型的细胞图像识别技术。在Jetson Nano人工智能平台上设计开发了面向细胞的智能检测系统,通过加入Dropout改进了YOLOv4⁃tiny轻量化网络模型,有效防止了训练数据过度拟合的问题,实现了基于细胞形状特征的精准识别。实验结果表明,该系统的细胞检测准确率可高达99%,能够大幅提高细胞在显微镜下的检测精度及检测效率,促进了人工智能技术在医学检测领域的应用。Making pathological analysis according to the morphological characteristics of cells is a common technological means in the field of modern medical and health.In the traditional cell recognition and classification,there exists some deficiencies,such as susceptible to fatigue,low efficiency and uncertainty caused by different doctor levels and subjective factors.Therefore,a cell image recognition technology based on YOLOv4⁃tiny model is proposed.On the artificial intelligence platform Jetson Nano,a cell⁃oriented intelligent detection system is designed and developed.By adding Dropout,the YOLOv4⁃tiny lightweight network model is improved,which effectively prevents the over fitting of training data,so as to realize accurate recognition based on cell morphological characteristics.The experimental results show that the cell detection accuracy of the system can reach as high as 99%,and the system can greatly improve the detection accuracy and detection efficiency of cells under the microscope,and promote the application of artificial intelligence technology in the field of medical detection.

关 键 词：细胞图像识别 YOLOv4⁃tiny模型 智能检测 目标识别 网络模型改进 病理分析 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]