Fluid Cells Based On Convolution Neural Network Image Visible Part Of Feature Recognition Method Research

Medical examination,the urine test is a very important inspection items.It is a long-term research topic to judge how to use the modern computer technology to effectively identify the visible components of urine specimens for the diagnosis of urine test.In order to accurately identify the target components using a computer,there are usually several key techniques:electron microscopy auto focus specimens,removal of image noise,computer segmentation,feature extraction and recognition.Real life has been the use of digital image technology to achieve the above technology,in which the extraction and identification of features in general from the shape and texture point of view,the integrated use of such as invariant moment method,Fourier transform method,As well as BP neural network method,have achieved some success,but also have their own shortcomings.Depth learning is a relatively hot research direction,there are some advantages of shallow learning,which convoluted neural network is a more mature depth of the learning network,in all areas have a certain research and application.Based on the principle of convolution neural network,the paper studies the tangible components of humoral cells on the AVE mirror test platform.The paper first introduces the origin of the topic,and then summarizes the current research situation from three aspects: feature extraction,depth learning and convolution neural network.The general practice of feature extraction is to obtain a descriptive semantic model from the appearance of image form,and then establish a mathematical model by semantic model transformation.The mathematical model has a certain degree of discrimination on the sample,and can not interfere with each other through training And whether the test can achieve a satisfactory degree of recognition.The main work of this thesis is:(1)This paper introduces the principle and structure of artificial neural network,compares the basic structure of artificial neural network and the advantages of other feature recognition methods,and takes BP neural network as the typical,and explains how BP neural network is Image recognition,improved the identification method of body fluid based on BP neural network,and analyzed its identification of advantages and disadvantages.(2)Convolution neural network as a key part of this paper,its structure and shallow artificial neural network is different,training methods is also difficult.In this paper,we improve the LeNet-5 network model,make the training library and test library suitable for convolution training,find the sensitivity of each layer of the network,and then use the error to adjust the network weights and thresholds step by step,and finally get the red and white blood cells The recognition error curve.By comparison,it is found that convolution neural network has someadvantages in feature extraction.