Research And Application Of Machine Learning In Infant’s Stool Image Classification

Machine learning is a branch of artificial intelligence,it is also the core of artificial intelligence,and it is the fundamental way to make computers intelligent.Image classification is a more common task in machine learning.It can be divided into supervised learning and unsupervised learning according to the presence or absence of supervision.With the rise of big data and the Internet,machine learning has achieved certain results in many areas such as image recognition,semantic recognition,and autonomous driving.This paper mainly studies the classification of infant’s stool image data.The research content can be divided into three modules.The infant stool image is first pre-treated,including low-bright stool image enhancement,data enhancement of stool images,and region of interest acquisition of stool images.The baby’s stool images are then classified using traditional machine learning classifiers(K-nearest neighbors and support vector machines),61.16% and 76.33% accuracy respectively.Finally,using the migration learning model to classify infant feces images,a higher accuracy rate is obtained.Combined with the classification algorithm in this paper,this paper designs and implements a mobile recognition system for infantile early stool images for early diagnosis of infantile biliary atresia.The main research contents of the article are as follows:1)Due to the characteristics of the baby stool image itself,Firstly,image enhancement of the baby’s stool image of brightness is performed to obtain detailed information of the image.Data is then enhanced on the stool image data,and the method used in the text is random cropping.Finally,in order to ignore the influence of unnecessary background information on the classification results,we obtain the image regions we are interested in for all the stool image data.2)The traditional machine learning classifier(KNN,SVM)was used to classify the stool images and compare their evaluation indexes.The experimental results show that the KNN classifier has better classification effect on stool images than the SVM classifier.Add a set of control experiments,compared with the presence or absence of data enhancement,the classifier has no effect on the classification results of the stool image.The experimental results show that the classification of the classifier after data enhancement is better.3)Use the migration learning model(VGG16,Res Net50,and Inception V3)to classify stool images,Comparing the accuracy,ROC curve and AUC value of the classification model of different models,analyze the results of each model test,the best classification model is obtained.The experimental results show that Inception V3 model had the best classification effect,with accuracy and AUC reaching 91.67% and 0.918 respectively.4)Use Linux operating system with Android development platform,a mobile recognition system for early infant fecal images was designed and implemented to identify infant feces stool images(sick,normal).And the higher recognition accuracy is obtained,which verifies the feasibility of the algorithm.