Evaluation of the impact of physical adversarial attacks on deep learning models for classifying covid cases