The Research On Colonoscopy Image Classification Method Based On Deep Convolutional Neural Network

With the development of science and technology and the advancement of medical equipment,the scale of medical images is also growing rapidly.How to extract the features we need and are interested in from medical image data,so as to further process the medical image accurately and efficiently,is a test for researchers and medical workers.In recent years,neural network technology has made great breakthroughs in the field of medical imaging.Convolutional neural network,the representative of deep learning algorithm,has attracted extensive attention in the field of medical image analysis.It is of great significance to mine the deep features of the image through the deep convolutional neural network,so as to process the medical image efficiently and further assist doctors in diagnosis.Colon disease is a common and frequently occurring disease in children.Among them,colon polyps are more likely to become cancerous,which seriously affects children's physical health and the growth and development of various body systems.In this regard,we cooperate with Hunan children's hospital to annotate the colon image dataset.By exploring the performance of different neural network models for colon image classification methods,we proposed two optimization models:1.A colon image classification method based on Net-GAP model with global average pooling is proposed.We use global average pooling to improve the structure of the classical deep neural networks VGG and ResNet,and propose Net-GAP models,including VGG16/19-GAP and ResNet101/152-GAP.With the introduction of global average pooling,the parameters of the network models,especially the VGG,are greatly reduced.In addition,we also provide a new colon image dataset.Compared with the basic networks,the network models we proposed have better classification performance on the colon image dataset,and reduces the parameters of the model to a certain extent.2.A colon image classification method based on IResNet-CBAM model with dual attention mechanism is proposed.Attention mechanism is a method that can strengthen important features and ignore unnecessary features.It establishes dynamic weight parameters by making relevant and irrelevant choice of information features to strengthen key information and weaken useless information,thereby improving the effect of deep learning algorithms.We use the CBAM module composed of channel attention and spatial attention mechanisms to improve ResNet and get the IResNet-CBAM model.The experimental results show that IResNet-CBAM achieves the expected effect.Compared with many classical network models,this method has higher classification accuracy,which proves the effectiveness of the proposed model.