Research On Fungus Recognition Algorithm Of Leucorrhea Optical Image Based On Machine Learning

Vaginal disease is one of the high incidence of women,of which leucorrhea routine examination is one of the most common examination items,and vaginal cleanliness is one of the most important indicators.Vaginal cleanliness is mainly through judging the types and quantity of fungus in leucorrhea.At present,most hospitals at home and abroad rely on the method of artificial microscopy to observe the types of formed points in the field of view of the microscope and the approximate number of each,and then obtain the cleanliness index based on experience.The traditional method lacks a quantitative standard,and has the disadvantages of low detection efficiency and low accuracy.In view of this,this thesis proposes an automatic fungus recognition algorithm for leucorrhea microscopic image,which can accurately and quickly determine the number and type of bacteria such as leaf mold in leucorrhea microscopic images.The principal component analysis,Haar feature,direction gradient histogram and machine learning are uesed in this algorithm to accurately identify bacteria such as leaf mold and bacilli in leucorrhea microscopic images.The main contents of this study are as follows:Firstly,it introduces the shortcomings and shortcomings of the traditional artificial leucorrhea routine examination and the academic significance and social significance of the intelligent identification research of leucorrhea microscopic images,and further describes the research status of medical microscopic image automatic recognition at home and abroad.Secondly,the instrument for obtaining the microscopy image of leucorrhea and its parameters are introduced.Then,it introduces how to preprocess the acquired leucorrhea microscopy image to accurately obtain the area to be detected,this step includes grayscale,image Smooth,morphological processing,image segmentation.Then,the obtained region to be detected is identified by a geometric feature such as circularity and area to identify cocci and bacilli.Finally,combining the Gabor feature,LBP feature,Haar feature and Hog feature of the image,three kinds of machine learning methods,decision tree,Adaboost and SVM,are used to train the classification model of mold.Due to the high dimension of some features,the PCA algorithm is used to reduce dimension.The different classifiers are used to train different image features.The cross-validation method is used to obtain the average performance of the classification model obtained by training each classifier.Analyse the training results by using accuracy,sensitivity and specificity as indicators.On this basis,This thesis proposes an algorithm that combines multiple features and votes on the results of classifier recognition.After relevant theoretical analysis,700 leucorrhea microscopy images were tested using an improved algorithm.The identification rate of molds was as high as 96.7%,and the false detection rate and missed detection rate were all controlled below 4%.The recognition rate of bacillus and cocciis is close to 100%,meeting the requirements of clinical indicators.