Research On Suspension Cells Recognition And Multicellular Tracking Method Of Microscopic Images

In the field of biology and medicine,the detection of cells is a very important content.Cells are observed and recorded through a microscope,and the images are stored by a computer.How to accurately segment and identify cells from images and develop automatic cell tracking algorithms is a major problem in cell biology research.For suspension cells,the cell shape difference is small,the spatial distribution is scattered,and the imaging process contains noise and cell movement,which are all difficult problems in the process of suspension cell identification and tracking.This paper has done the following work on how to accurately identify and track suspension cells:(1)In this paper,the image is filtered and denoised by preprocessing operation.By comparing various cell segmentation algorithms,an improved watershed segmentation algorithm is proposed to segment and identify the image.This algorithm combines threshold segmentation on the basis of the watershed algorithm.The results show that this segmentation algorithm has higher segmentation accuracy.(2)This paper proposes a cell image recognition method based on deep learning.Based on the Faster RCNN framework,this paper selects the VGG16 neural network.Firstly,the improved watershed algorithm is used to initially identify the cell images to assist in the labeling of the dataset to solve the time problem caused by manual labeling.The dataset is augmented by spatially enhancing and pixel-enhancing the images.Then,the optimal model is iteratively trained by adjusting the hyperparameters,and the cell recognition results with higher accuracy are obtained.Finally,the missed detection of cells is further improved by improving the anchor box and the non-maximum suppression algorithm.(3)This paper proposes a cell tracking method based on suspension cell recognition.The cells in the cell image are tracked by constructing a graph.When tracking,firstly select the best matching cell pair between adjacent frames as the starting point of matching.Then,neighbor cells matching are performed according to the feature information between cells until all cells are matched.In this paper,the problems of mismatching caused by adherent cells and frame break caused by missed detection in cell space tracking are dealt with separately.