Research On The Application Of Female Cancer Medical Image Recognition Based On Transfer Learning

With the rapid development of the Internet and the continuous maturity and progress of computer technology,image processing technology in computer technology has been widely used.With the further research on machine learning,its application in medical images in biomedical engineering is becoming more and more important.Nowadays,the incidence of cancer continues to rise,and society needs to pay more attention to the prevention and treatment of cancer.Breast cancer and cervical cancer are the first and second cancers in women.The incidence of breast cancer and cervical cancer is increasing year by year.The age of patients is getting younger and younger,which endangers the health of women all over the world.Early diagnosis,discovery and treatment are the most effective ways to improve survival rate.However,the manual diagnosis and screening has abnormal complexity and complexity,leading to cancer screening can not be popularized.The two kinds of cancer images studied in this paper,mammogram and pathological image of cervical cancer cells,although they are at different stages of diagnosis,have common points from the point of view of simple image classification.Compared with normal images,they have more prominent points.Using the features of shape,texture and color,image classification learning algorithm can modify the classifier by iteration.Finally,the second classification of pictures is realized.Based on the pre-trained convolutional neural network,this paper introduces the transfer learning to realize the classification and recognition of medical images.Pre-training depth neural network extracts the generalized image features.Based on the generalized features,migration learning combines and abstracts the existing generalized features with new data sets,and then generates new data set features,i.e.the features of specific target data,and uses these features to realize image classification.There are many more abstract features extracted by transfer learning,that is,the dimension of feature space is very high.Therefore,in the case of small medical image data sets,migration learning is easy to cause over-fitting,that is,the generalization ability of well-trained deep neural networks is poor.This paper classifies medical images in low-dimensional feature space by feature dimensionality reduction and support vector machine(SVM)model.In this paper,migration learning is applied to classify mammograms and pathological images of cervical cancer cells,and feature dimensionality reduction and SVM are applied to classify images of cervical cancer cells with small data sets.The experimental results show that the transfer learning has better classification results when the training set is relatively large.At the same time,feature dimensionality reduction based on transfer learning combined with SVM classification method can achieve better data classification results than simple transfer learning when the training set is small.