Research On Automatic Acquisition And Recognition Technology Of Micro Leucocyte Images

The acquisition and recognition of micro leucocyte images has always been an important research field in medical image data processing.The number and percentage of different leukocyte types in the blood of a normal person are different from those with a disease.Clinically,the medical staff usually judges whether a person is normal or suffering from a certain disease based on the quality and count of human leukocytes.However,the complicated and repetitive acquisition and identification of microscopic leukocyte images consume a lot of manpower and material resources.Computer aided diagnosis can not only guarantee the objectivity and accuracy of medical diagnosis,but also save the time and energy of medical professionals,it has a good application value and practical significance.In this paper,the automatic acquisition and recognition technology of micro leucocyte images has been deeply studied,The main tasks of this paper are as follows:Firstly,an eight-neighborhood auto-focusing algorithm based on plane fitting under a fully automated microscope is proposed.An eight-neighborhood fast focus algorithm based on plane fitting is used to automatically acquire leukocyte images under a microscope.In this paper,the edge detection operator is improved based on the edge information of the microcell image,and a four-module operator is proposed.Compared with the traditional focusing operator,the analysis results show that the eight neighborhood operator evaluation function has the single peak and sharpness in the focus of the microcell image.At the same time,according to the characteristics of blood smears,we propose a focusing strategy based on plane fitting.Compared with human vision,the experimental results show that the fitting degree is very good.Secondly,a method for automatic identification of white blood cell microscopic images based on an improved convolutional neural network is proposed.In this paper,the image denoising technique and K-means color clustering algorithm are used to preprocess and segment the microscopic leukocyte images,Then we simplify the convolutional neural network(LeNet5)structure based on the handwritten font set of the mnist_uint8.The improved convolutional neural network id used as a classifier to identify microscopic leukocyte images.At the same time,simulation experiments are done in MatlabR2014a environment in this paper.The simulation results show that the improved convolutional neural network is superior to the original convolutional neural network in speed,and the accuracy of microscopic leukocyte recognition is 96?.