Covid-19 has a high rate of transmission and contagion, and the early identification of new cases helps to prevent the transmission of the virus. This becomes a challenge, as the tests applied are usually done manually and take time. Initial studies have found cases of patients who have abnormalities on chest X-rays indicating the occurrence of this disease. With the study of these images, it is possible to create models with machine learning resources to automatically identify cases of COVID-19. In this study, experiments are carried out with convolutional neural network architectures to identify and evaluate a model of COVID-19 case detection with high precision in a fast and automatic way. Through the experiments carried out it is possible to affirm that the diagnosis by image of cases of severe acute respiratory syndrome based on X-ray exams is possible, having been observed results in which the model reaches 96 % accuracy when analyzing chest X-ray with three possible diagnoses in the experiments performed. The differential of this work in relation to the literature is the identification of multiple categories in the diagnosis, which is not observed in the studies studied
