Research On Image Classification Method Combining Visual Attention Mechanism And Deep Learning

With the development of the times and the improvement of medical level,medical institutions will regularly replace advanced medical imaging equipment,which is accompanied by a blowout growth of medical imaging data.It takes a long time to train a professional medical imaging doctor,so there is currently a shortage of supply in the industry.Chest X-ray examination and computed tomography(CT)examination play a very important role in the auxiliary diagnosis of lung diseases.However,the manual reading work has problems such as strong subjectivity,time-consuming,and error-prone doctors when they are tired.Therefore,it is necessary to propose an intelligent recognition method for lung images based on deep learning.Different from general natural images,both X-ray images of COVID-19 and CT images of lung cancer have the characteristics of high similarity between categories,which increases the difficulty of image classification.In view of the characteristics of this data,this paper proposes two image classification methods based on attention mechanism:1.A COVID-19 chest X-ray image classification method based on multi-scale channel feature fusion network MCFF-Net is proposed.This part combines the attention mechanism with the deep convolutional neural network,proposes a parallel channel attention feature fusion module PCAF,and builds the MCFF-Net model based on this module.With different depths and classifiers,a total of 9 models are proposed.This paper builds a three-category dataset Dataset-A and a four-category dataset Dataset-B based on the public image dataset to verify the effectiveness of the model.Experiments have verified that MCFF-Net has good classification performance for chest X-ray images of COVID-19,surpassing a series of classical deep learning models and other research methods in the same field.2.A classification method of lung cancer CT images based on MCFF-SRM-Net is proposed.In order to verify the generalization performance of MCFF-Net on different types of images,this paper uses the public dataset Chest CT-Scan Images Datasets for validity verification.At the same time,this paper introduces a style-based recalibration module SRM,which is interspersed in the MCFF-Net model based on the characteristics of the soft attention mechanism,in order to improve the learning ability for the style features of images.After experimental verification,MCFF-SRM-Net also has excellent performance in lung cancer CT image classification,and achieves high accuracy.It also proves that MCFF-Net has strong generalization ability in image classification.