Classification Of Pneumonia Type Image Based On Convolution Neural Network

In recent years,medical image processing has become a hot research topic in the field of computer vision.The diagnosis of pneumonia type mainly depends on the experience of doctors.Hospital needs to set up a special department and personnel to determine the image of pneumonia,so time-consuming and laborious.In addition,a number of pneumonia CT images are very similar,easy to cause the doctor to determine the error.To this end,the existing methods of image feature extraction and target recognition are studied.It is found that the traditional image processing method is low in classification and recognition rate of pneumonia,mainly because of the artificial selection of image features can not be perfect representative of the target.Deep learning is a hot direction in the field of machine learning in recent years.Convolutional neural network(CNN),as a representative network of deep learning,has the ability of autonomous learning,and has the invariance of displacement,scaling and distortion.It can be used to predict the type of pneumonia by training the convolution neural network with a large number of labeled pneumonia data.Although convolution neural network algorithm has many advantages,but there are also some problems,such as lead to over fitting.To this end,without affecting the correct rate of the recognition of pneumonia,the convolutional neural network classification algorithm was improved.Therefore,a classification algorithm based on convolution neural network is proposed,which consisted of three convolution layers,three subsampling layers and one fully connected layer.In addition the dropout method and elastic gradient descent method are carried out on the convolution layer.The clinical experimental results show that the algorithm can accurately classify CT images of different pneumonia than general research recognition algorithm,such as adaboost and svm algorithms,and the revised convolution neural network can prevent over-fitting phenomena in training data.However,The improved algorithm needs to be trained for a long time.But it also reduces the convergence time of the neural network after the optimization of the gradient descent.