Deep Learning Based Residual Neural Network for Digestive Endoscopy

摘要

The deep learning is becoming more and more popular nowadays which has participated in plenty of areas. Recently, a growing number of work can be applied to the deep learning including the medical area. The residual neural network has been proposed for many years, but no one apply it to the digestive endoscopy. The constantly improved research on deep learning can make the residual neural network work better than before. The residual neural network has higher accuracy than the normal neural network. In this research, the pictures are represented by RGB model and the pictures would be preprocessed by neural network in order to exact their features. Because the original pictures are extremely big and the residual neural network is a large model so that it can not handle such big data, the original pictures must be preprocessed to smaller size so that the neural network was applied to it. Reducing the resolution has been tried to preprocess the pictures but this way is not effective, the accuracy is not satisfactory. Furthermore, the residual neural network is linear processing to the data so that it combine with the normal neural network would have better results. Also, the residual neural network overcomes the degradation in normal neural network so that the results can be better.