Single-cell Three-Dimensional Reconstruction Based On Micro-vision And Deep Learning

With the development of cell biology,three-dimensional reconstruction of cells has become a hot research topic and a difficult problem nowadays.In this paper,three-dimensional reconstruction of cancer cells is studied.Due to the lack of feature points on the cell surface and the difficulty of extracting depth information,the existing three-dimensional reconstruction algorithms are difficult to reconstruct cells,and the cost of cell three-dimensional reconstruction is too high.In this paper,a three-dimensional reconstruction algorithm based on cell contour and cell three-dimensional motion information is proposed,and the motion information solving algorithm is improved several times.Firstly,we shot a large number of cell rotation videos through OEK system,which provides a data base for algorithm innovation.At the same time,according to the characteristics of cell movement in video,a new idea of three-dimensional reconstruction of cells using cell contour and cell three-dimensional motion information is proposed,which provides a direction for algorithm innovation.Secondly,using the three-dimensional motion solving algorithm based on optical flow method and presupposed model,the motion vector field of cells is preliminarily solved,but because of the fixed model,the application scope of the algorithm is very small.So we improved the algorithm to a three-dimensional motion solving algorithm based on optical flow method and diameter search.Under the condition of satisfying the hypothesis of optical flow method,the solution is very accurate,but because of the limitation of the hypothesis of optical flow method,the practical application effect of the algorithm is not satisfactory.So we further propose a single-cell three-dimensional motion solving algorithm based on deep learning,which overcomes the application limitations caused by the premise assumption of optical flow method,improves the accuracy,application range,anti-interference ability and other performance of the algorithm,and realizes the practical application of the algorithm.Finally,in order to solve the problem of large errors caused by the lack of three-dimensional geometric constraints in point cloud data.In this paper,a 3-D point cloud data fusion algorithm based on cell three-dimensional rotation axis is proposed,which greatly reduces the error.The low cost and automation of cell three-dimensional reconstruction is realized.