Convolutional Neural Networks (CNN) and DBSCAN Clustering for SARs-CoV Challenges: Complete Deep Learning Solution

Abstract

AbstractEarly diagnosis of Covid-19 is a challenging task requiring congruous clinical medical imaging, which is a time- consuming process and suffers from accuracy problems due to variations between different laboratory results. The clinical symptoms of Covid-19 show resemblance with acute respiratory distress syndrome. The major clinical symptoms linked with this disease are fever, cough, headache, migraine, and breathlessness. Despite tremendous research going on, knowing the way of transmission and its early detection remains a mystery. There is no treatment as of now for this virus, so a lot of unprecedented containment and mitigation policies such as closure of business places, schools, and colleges, marriage gathering restrictions, transport restrictions, and social distancing are being employed. These policies are able to limit the transmission of covid-19, but are not always feasible. Steps must be taken to slow down the spread of this virus and make an early diagnosis of infection to save lives. This paper gives a clear idea about the introduction of Covid-19, its symptoms, post covid-19 symptoms, challenges posed by the virus, and proposed solutions for its early detection to slow down its rate of transmission. Methods: The proposed solution includes a clustering algorithm for massive contact tracing that helps to slow down the transmission rate, and automatic virus detection and classification network known as Convolutional Neural Networks (CNN) based upon deep domain transfer learning. Results: The pre-trained model VGG-19 is used and the hyperparameters of the model are tuned as per classification requirement by exploiting the concept of deep domain transfer learning. The model is implemented on publicly available chest radiography images and the system classifies the dataset as covid and non-covid images. The CNN achieves \(97.35\%\) accuracy outperforming all the existing methods. A new concept of employing nanocoating and nano sprays is also introduced in the paper. Conclusion: To discuss various critical challenges posed by Covid-19. Addressing those issues and proposing various solutions. Proposing clustering machine learning algorithm for massive contact tracing. Developing automatic covid-19 detection and classification system based upon automatic feature detection. Providing solutions based upon nanotechnology to slow down transmission. KeywordsGlobal average poolingNNLUCNNAMsgradSGDADAMHybrid parallelismMax-pooling