Research On Lung Image Classification Method Based On Inception-ResNet-v2

With the development of people’s lives,the pathogenicity rate of lung tissue also increases.Chest X-ray examination is an important way to judge lung disease.However,the manual examination of chest radiographs has problems such as heavy workload,strong professionalism,and dependence on the clinical diagnosis experience of physicians.Faced with these problems,the current main solution is to use computer-aided diagnosis based on big data and artificial intelligence.Most of the existing artificial intelligence algorithms to assist in the diagnosis of lung diseases not only have problems such as poor image preprocessing,poor classification performance of image analysis,and low accuracy of results,but also the judgment rate of existing algorithms for newly discovered lung diseases Low,such as new coronary pneumonia.Therefore,this topic first studies the method of removing the ribs of chest radiographs,and then uses the Inception-ResNet-v2 combined network to analyze the chest images to better solve the above problems.Finally,the model of this topic is used in the diagnosis of new coronary pneumonia.The main work of this subject is as follows:(1)This topic uses a rib subtraction scheme based on the U-Net network of lung images.In the preprocessing stage of the chest radiograph,the local contrast enhancement technology is first used to increase the contrast of the lung lobes of the chest radiograph,and then the ribless chest radiograph is generated through the U-Net network.This method of rib subtraction increases image quality and network classification accuracy.(2)In this topic,the Inception-ResNet-v2 network is selected as the backbone network.The Inception block structure in the network can increase the sparsity and width of the network.At the same time,the combination of multi-size convolution kernels can not only expand the receptive field,but also increase the network calculation speed.This multi-channel parallel structure can strengthen the lung disease analysis capabilities of the model.(3)This topic optimizes the structure of the backbone network.First,the model integrates the Inception block and the residual block,and the jump connection can reduce the probability of performance regression in the complex deep model.Then this topic makes relevant adjustments to the network structure to increase the adaptability of the network to data.The adjusted combined network can improve the recognition performance of lung diseases.(4)This subject studies the image recognition of new coronary pneumonia.First,we studied the typical cases of new coronary pneumonia,using high-precision CT images as the research object,and then segmented the lung parenchyma based on the CT images,and finally applied the improved Inception-ResNet-v2 network to the auxiliary diagnosis of new coronary pneumonia.Finally,this topic tests the performance of the model through experiments.The experimental results show that the improved Inception-ResNet-v2 network has higher performance and recognition accuracy.The rib subtraction method can help the model improve the classification performance,and the network has a higher experimental resolution in the recognition of new coronary pneumonia.