Classification of vocal cough for COVID-19 detection by Spectrogram images using deep learning

Abstract

This work, which we present in our master thesis, is to develop a diagnostic support system that enables the detection of COVID-19 disease using human voice images. Since an accurate diagnosis of COVID-19 disease is a difficult task that requires a series of clinical examinations and tests to verify the signs and symptoms of the disease. The objective of this project is to design an automated diagnostic system for the early detectionofCOVID-19diseaseusingschematicimagesofacousticsignals. Themaingoal is to differentiate between COVID-19 patients and healthy individuals. The proposed system is based on two main steps: (1) feature extraction from the spectrograph and (2) classification by deep learning. The discriminative features we selected include two parameters: (1) the feature extraction process and (2) the neural network model parameters. The classification process is based on six deep learning classifiers and models: (1) (Alex Net), (Res Net5), (VGG 16), (VGG 19), and (ANN). Feature extraction is performed by simulation software: MATLAB Mathworks, in addition to the classification process performed in simulation by SPYDER software. The COVID-19 disease database we were able to obtain after manual structuring against apreviousdatabasewasusedinourexperiments. Theperformancemeasuresusedinthis study are: Accuracy, the loss curve and the Accuracy curve. The results obtained are different.