Research On Lung Disease Detection Technology Based On Broad Learning

With the increasing risk of lung disease in people’s daily life and the worldwide spread of COVID-19,lung disease screening has become crucial,rapid and timely diagnosis results can help effectively isolate COVID-19 and curb the rapid spread of the disease.At present,detection technology through CT images has become a focus of current research.However,there are still some challenges in the current research.On the one hand,the existing COVID-19 diagnostic method relies on a large number of redundant labeled data and time-consuming data training process to obtain satisfactory results.However,as a new epidemic,it is still challenging to obtain large clinical datasets,which will inhibit the training of deep learning models.On the other hand,the long training period of detection models based on deep learning can delay detection results and accelerate the spread of diseases,and can not timely retrain the model according to the changes of regional diseases.According to the imaging characteristics of COVID-19 lung disease,this thesis studied the above two aspects.The specific work is as follows:1.This thesis conducts research on the COVID-19 image diagnosis algorithm based on width learning system(BLS)and deep learning related networks.Firstly,a thorough study was conducted on its theoretical basis and characteristics,and the limitations of BLS were analyzed.Then,an experimental study was conducted on the classification of COVID-19 images using basic BLS.The experimental results showed that the model has significant advantages in training speed,but the testing accuracy needs to be improved.2.Due to the limitations of the basic BLS,a CT image classification algorithm of COVID-19 based on improved broad learning is proposed(K-BLS).First,in order to standardize the brightness and contrast of the image,normalize the original data and eliminate the interference area,then K-means is introduced to make BLS better adapt to high-dimensional data of lung CT images,alleviate the performance defects of randomly obtaining node weights,and establish a broad learning model related to typical feature learning.Then use the public datasets to evaluate the performance of the network.The research results show that under the condition of limited data scale,KBLS can produce more accurate diagnosis results of COVID-19 than the current technology.3.A diagnosis model integrate broad learning and deep learning into a joint framework is proposed to quickly diagnose COVID-19(FA-BLS).This model extracts the deep features of the image through the Res Net50 convolution module based on transfer learning and performs feature fusion to alleviate the problem of insufficient data.The spatial attention mechanism is introduced to retain more important information and reduce the interference of redundant content on the network.The rapid learning ability of broad learning is used to adaptively select features for rapid diagnosis.In the experiment part,this thesis discusses the rapidity and effectiveness of the model and the influence of node selection on broad learning system.A series of comparative experimental results show that this model can achieve the same level of accuracy as the deep learning model under the small-scale dataset task,and make up for the shortcomings and weaknesses of the recognition task based on deep learning.