Red Blood Cell Detection And Counting Based On Convolutional Neural Network

Erythrocyte detection and counting is a medical clinical examination project which is used for the purpose of detecting blood cell disease and its related diseases and supporting the segmentation,classification and tracking of erythrocytes.It is also an important branch of medical image processing research based on cell image processing and analysis technology.In recent years,with the development of image processing technology and machine learning methods,red blood cell detection and counting methods emerge new theories and methods constantly.Especially,deep learning technology made breakthroughs in image detection,segmentation and identification.In this context,against some problems and challenges of current cell detection and counting research,this paper conducted the following study.1.Against the problem of cell adhesion overlap in cell images,this work proposed a cell counting method based on deconvolution regression network with the cell images characterized by Gaussian kernel for single cells.Unlike the traditional methods,the deconvolution regression network here uses a unique codec-style depth neural network model.In the design of the tag image,each cell in the original image is used with one slightly smaller than the cell size of the Gaussian kernel to represent.In the convolution part extracting features of cell images,deconvolution part recovering the image size and feature.So,building an end to end cell counting solution.2.Learn from the Faster R-CNN network model,we proposed a new cell detection method based on the Faster R-CNN network characterized by the advanced features of cell images.Unlike most simple reliance on regional candidate box and scoring methods for evaluation of detection models,for cell detection,this paper constructs a method based on cell images advanced features and regional featured box,which generated by using convolutional neural networks method,and applied supervised features for image classification expressed into cells detection,successfully.