Research And System Implementation Of Red Blood Cell Recognition Classification Based On BP Neural Network

The classification and identification based on erythrocyte morphological features is great significance to medical research.In order to research the classification and recognition of the blood red blood cell,the most hospitals and research institutes observe the morphological of erythrocytes by using the microscope,and statistic its shape.This method only give a rough estimation of the red blood cell morphology and quantity and other related information to do the statistics and records.However,if you wanted to keep accurate records of the red blood cells in area,shape,roundness and other information,it’s difficult for person to do this word.Furthermore it is also a great effort and physical exertion for medical and scientific research workers.If we have a sound red blood cell identification and classification system,which can automatically do accurate morphological analysis,chart information statistics and so on.This will greatly improve the accuracy of erythrocyte detection and the operational efficiency of medical technicians.This paper designs and develops the system for the identification and classification of red blood cell from the actual testing needs of hospitals and research institutes.Specific research includes the following six aspects:1.The module for Red cell image preprocessing and enhancement is essential in most image processing systems.In the system developed in this paper,we mainly use the binary filter operator to smooth the image,then use the BM3 D algorithm to remove the Gaussian white noise of the image and improve the contrast between the image feature target and the background by sharpening the image,and sharpen the edge of the image.The preprocessing and enhancement modules are designed to improve image quality and pave the way for the next segmentation.2.Human peripheral blood smear red blood cells may be overlapped and not obvious in regional characteristics and so on,which will cause adverse effect in image segmentation.Therefore,the segmentation algorithm of red blood cell image is very important,after comparing the threshold segmentation,Canny segmentation and Watershed segmentation method.According to the large number of non-discriminant way for segmentation algorithms,and compare the results of segmentation.The segmentation result based on the marker-based watershed segmentation algorithm is generally optimal.This algorithm can effectively divide most of the red blood cells,and also ensure the edge feature information not lost.In the aspect of erythrocytecharacterization,the characteristics,including circumference,area,circle degree,rectangle and Fourier shape descriptor of erythrocytes,were obtained.These features are shape feature operators with certain degree of discrimination.3.The red blood cell was identified and classified by BP neural network,and the red blood cell shape data were normalized and crossed,and other operations.Finally,the BP neural network recognition rate is above rate of 93% is obtained.4.According to the task of the algorithm,the algorithm is divided into functional modules which are independent and cooperative to each other,and maintain the scalability of the algorithm.The system uses three-stage interface,this interface is flexible and can be customized,very much in line with software development requirements.The functional function modules of the system are coded and tested,and the results of erythrocyte detection are displayed graphically.5.Synthesize the BP neural network and decision tree in the red blood cell recognition accuracy,computational complexity and algorithm scalability,get the conclusion that the former is better than the latter.