基于Mask R-CNN的荧光编码微球图像检测方法  

Fluorescence Encoded Microsphere Image Detection Based on Mask R-CNN

作　　者：刘铭赫 史再峰 罗韬[2] 王溥萌 姚素英 Liu Minghe;Shi Zaifeng;Luo Tao;Wang Pumeng;Yao Suying(School of Microelectronics,Tianjin University,Tianjin 300072,China;College of Intelligence and Computing,Tianjin University,Tianjin 300072,China;Tianjin Key Laboratory of Imaging and Sensing Microelectronic Technology,Tianjin 300072,China)

机构地区：[1]天津大学微电子学院,天津300072 [2]天津大学智能与计算学部,天津300072 [3]天津市成像与感知微电子技术重点实验室,天津300072

出　　处：《南开大学学报（自然科学版）》2022年第1期40-46,共7页Acta Scientiarum Naturalium Universitatis Nankaiensis

基　　金：国家自然科学基金(62071326);天津市自然科学基金(17JCYBJC15900)。

摘　　要：为降低荧光编码微球技术的应用成本,提出了一种基于Mask R-CNN目标检测算法的荧光编码微球图像检测方法.首先基于TensorFlow和Keras深度学习框架搭建Mask R-CNN网络模型,整体网络由特征提取网络,候选区域生成网络和分支处理网络3部分构成;通过有标注定性图像样本集训练网络模型,并使用合成图像实现训练集数据增强;将待检测定性图像样本输入训练完成的网络模型获得定性图像的语义掩膜.实验结果表明,对于单色和双色微球定性实验图像,平均检测准确度分别达94.17%和95.96%,可实现荧光编码微球定性图像的边界框检测、分类以及语义掩膜生成.To reduce the application cost of fluorescence encoded microsphere technology,a fluorescence encoded microsphere image detection method based on Mask R-CNN object detection method is proposed.Firstly,the Mask R-CNN network model is built based on TensorFlow and Keras deep learning framework.The Mask R-CNN consists of feature extraction network,proposal region generation network and branch processing networks.The network model is trained by annotated qualitative image sets,and the training set data enhancement is implemented using the synthetic images.The qualitative image samples are detected using a trained network model to obtain semantic masks.The experimental results show that for the qualitative images of single color and two-color microsphere images,the average detection accuracy is 94.17%and 95.96%,respectively.The boundary box detection,classification and semantic mask generation of fluorescence encoded microspheres are realized by this method.

关 键 词：目标检测 荧光编码微球 深度学习 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]