Research And Application Of Parasite Egg Recognition Model Based On Transfer Learning

With the improvement of people’s living standards and health awareness,human parasitic diseases have been better suppressed.However,China has a vast territory and uneven development.In some areas,the problem of food borne parasites is prominent,and the prevention control of parasitic diseases is not optimistic.In the process of prevention and treatment of parasitic diseases,the focus is how to quickly detect and diagnose.At present,the common detection and diagnosis methods of parasitic diseases are mainly based on the collection of excreta and the use of microscope to judge by manual inspection.The efficiency of this method is not high and the accuracy is greatly affected by human factors.How to effectively combine the existing image processing technology and machine learning method to realize the detection of parasitic diseases is great significance to the construction of intelligent medical system.With the development of image processing technology and artificial intelligence technology,an image-based rapid detection method of human parasitic diseases has attracted the attention of relevant researchers.However,from the current research progress,there are still several shortcomings: 1)The image of parasitic eggs obtained by microscopic medical imaging is often a weak boundary image with impurities.The classical recognition algorithm is difficult to obtain the ideal recognition effect.2)Because most human parasitic diseases have been well suppressed in many areas of China,it is relatively difficult to collect a large number of sample data.In the case of relatively lack of sample data,improving the accuracy and efficiency of the recognition algorithm is still a problem to be solved.Therefore,this study introduces the transfer learning mechanism,and gives a parasite egg recognition method.In the case of small samples,it gets better recognition accuracy.The main work of this thesis is as follows:(1)In order to solve the problem of target location in data processing and model recognition,a parasite egg location method based on ORB and CN feature descriptors is proposed.Then,according to the characteristics of the parasite egg image,the color difference segmentation standard is defined.The candidate regions of the image are divided and the parasite egg is described by the fusion feature descriptor.The results of feature matching are used for accurate positioning,and then the areas suspected of parasite eggs in the tested images are quickly extracted,which provides support for the improvement of subsequent recognition efficiency.(2)To solve the problem that it is difficult to train a parasite egg recognition model with high recognition accuracy and strong robustness with small samples,a parasite egg recognition model based on transfer learning Res Net-18-FF is proposed through the experimental analysis of typical deep learning models VGG-16、 Res Net-18 and Goog Le Net,which balances the relationship between the number of samples and recognition accuracy.On this basis,in order to further overcome the problem of insufficient number of samples,a mechanism based on the original sample through correlation transformation is introduced.The experiment shows that these methods and techniques can improve the recognition accuracy.(3)Based on the above research and combined with the actual needs of the hospital for parasitic disease detection,a set of parasite egg recognition platform based on transfer learning is designed and implemented.At present,the platform has been used in related hospitals and achieved good results.