Abstract:
In December 2019, the first infected new coronavirus case (COVID-19) was discovered in Hubei,
China. The COVID-19 pandemic has spread to every country on the planet, and has significantly
affected every aspect of our daily life. There are a limited number of COVID-19 test kits available
in hospitals, lack of radiologist, and a shortage of equipment due to the increasing cases daily.
Chest X-rays are the common method used for the diagnosis of this COVID-19. Examining chest
X-rays is a difficult process even for a professional radiologist. Because of incorrect diagnosis and
treatment, the troublesome way of detecting COVID-19 has resulted in the loss of life. There is a
need to increase the accuracy of diagnosis. Early, fast and accurate diagnosis of COVID-19 could
help to reduce the morbidity and mortality associated with this illness.
This study aims at the detection and classification as a quick alternative diagnosis option of patients
affected by COVID-19 based on their chest X-rays to prevent it from spreading among people.
The convolutional neural network and histograms of oriented gradients are designed to accomplish
the difficult task of detecting disease such as COVID-19, assisting medical experts in their
diagnosis and treatment. We aimed to examine the diagnostic accuracy of hybrid CNN and HOGbased feature extraction for the diagnosis of COVID-19 by evaluating chest X-ray images and
Support Vector Machine for classification. The dataset is collected from the University of Gondar
referral hospital and the Kaggle disease image repository. After collecting datasets, image
preprocessing, image segmentation and image augmentation techniques are applied to improve the
performance of COVID-19 disease identification. We also apply anisotropic diffusion filtering to
remove noise, histogram equalization to balance the intensity of an image, and YOLOv3 for input
chest X-ray image detection. We were successful in doing with the help of Keras (using
TensorFlow as a backend) and Python. After the DCCNet model is evaluated, we have achieved
99.9% training accuracy and 98.3% test accuracy and we got 100% training accuracy and 98.5%
test accuracy in HOG. After evaluating the hybrid model achieved 99.97% for training and 99.67%
for testing to detect and classify COVID-19 which is better than s
