Deep learning has grown rapidly in recent years, with excellent results for many computer vision
applications, such as image classification and object detection. One aspect of the increased
popularity of deep learning is its ability to mitigate the need for handcrafted features. It reduces
the manual work of identity recognition and facilitates automatic processing. Inspired by the
advantages of the hierarchical feature extraction of deep learning, this work investigates the
development of a Convolutional Neural Network (CNN) algorithm to solve the problem of Covid 19 detection based on deep learning techniques. The proposed system takes raw grey scale x-ray
images as input and classifies the images to Covid-19 or non-Covid-19. The data consists of grey
scale chest x-ray images of normal people, Covid-19 patients, and other images for pneumonia
and normal patients. We started with an extensive literature review of the latest Artificial
Intelligence and computer vision techniques for Covid-19 detection. The data is available from
open source. Deep transfer learning using pre-trained CNN – such as VGG-16, VGG-19,
RESNET-50, Inception V3 and Exception – were used. The results obtained were promising for
most of networks. Some layers were added to the pretrained networks to improve performance.
Multiple design iterations were conducted to improve the model's performance. The best design
model achieved better performance compared to state-of-the-art pre-trained models using same
data. Indeed, after a training phase of less than 100 epochs, our model utilizing VGG19 network
was capable of providing 87.08% accuracy, 87.50% Precision, 87.00% Recall and 87.25 F1. The
most challenge of this work was data sample number and this challenge is due to non-availability
of reliable data.