الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 106 |  |  |
| | | from | 106 | | |
Abstract
Deep Learning Neural Networks build the core of new research in many fields. Their application covers a wide variety of fields including computer vision, natural language processing, health care and automatic predictions. The aim of this thesis is to investigate and analyze a variety of deep learning network techniques used in the biomedical field, especially concerning COVID_19 and pneumonia. This study aims to reach the best recognition rate in pneumonia and COVID_19 diagnosing systems through the optimization of the Convolutional Neural Networks. Specifically, the improvement of the recognition rate which leads to increasing the diagnosis rate of these crucial diseases. In order to find a new model that gives better diagnosis rate for COVID_19 and pneumonia, firstly, we preprocessed the input image, and then the preprocessed image was classified using different convolutional neural networks. Finally, Random Search and Hyperband optimization were chosen and performed on the Residual Networks and Xception Nets in order to get a new model by finding the best hyperparameters of this model and hence higher diagnosis rate. The effectiveness of these two optimization techniques in enhancing the diagnosis rate were compared. The results showed that using either Random Search or Hyperband optimization for ResNet and Xception Net hyperparameters tuning to find a new optimized model, give better accuracies for diagnosing pneumonia and COVID_19. In conclusion, these results suggest that tuning hyperparameters automatically for convolutional networks without many tedious manual trials give new models that are able to classify images with higher accuracies than traditional methods.