Research And Implementation Of Video Detection Algorithm For Schistosome Miracidia In Complex Background

Parasitic diseases distribute widely and bring great threat to the majority of working people and lowered immunity children.The pathological changes and clinical manifestations caused by different kinds of parasites are different,but the most reliable diagnostic methods are consistent,which can be diagnosed by naked eye.Image processing and pattern recognition have been developed rapidly and widely used in various fields in recent years.Taking schistosomiasis as an example,this thesis proposes and implements a method for automatic detection of parasitic video,including the following four aspects:(1)Usually,the video background obtained by microscope is complex since the detected sample solution of schistosomiasis japonica contains much dung slag,water insects,water floating debris and other impurities.Besides,we would like to achieve the real-time miracidium video detection,so algorithm complexity should not be too high.By studying the moving target detection algorithm and characteristics of complex background miracidium video,we put forward improvement measures,combining background difference and interframe difference and morphological processing,which can accurately obtain the moving objects in video.(2)Because of the limited depth of the microscope used by medical workers in current,the schistosome video often has a defocus phenomenon.Especially under a large miracidia sample quantity condition,the recognition accuracy will be seriously affected.Aiming at this phenomenon,the third chapter studied the defocus blurred image restoration algorithm.This chapter introduced the spread function of optical system and summarized the traditional defocused image degradation model.And then we restored the defocused images in Schistosoma japonicum miracidium video by combining the methods of the Wiener filtering restoration,spatial domain restoration and nonlinear restoration.After the anti fuzzy processing,recognition accuracy is greatly improved.(3)We studied a two-category problem of Schistosoma japonicum miracidium.Its feature extraction and classifier design are also important contents in this thesis.We used the boundary chain code to represent and describe the connected area of moving targets.And the geometric feature and motion feature of Schistosoma japonicum miracidium are extracted as the basis for classification.We studied the principle of SVM classifier and the related parameters.We also analyzed the influence of Gaussian kernel function parameter and penalty factor in SVM classifier for the classification performance.Through the statistical analysis of the experimental results,we obtained a classifier,which can achieve the anticipated target.By learning and testing samples,we obtain a good result.(4)Finally,by modular design,using VS2015 integrated development environment and C/C++ programming language,we realized the video detection system of Schistosoma japonicum miracidium.Using this system,we tested a large number of different background videos and analyzed the experimental results.The final results show that the system can realize real-time detection of Schistosoma japonicum miracidium video accurately and stably.At present,this system has been put into use.