ABSTRACT
Classification of the use of masks today is very necessary regarding the pandemic period that is not over yet. This is mainly to break the chain of transmission of the COVID-19 virus from one person to another. From the literature that has been studied, the Convolutional neural network method can be used to distinguish the types of masks used in society. The advantage of the Convolutional Neural Network is that it can recognize objects with a fairly high level of accuracy, but it has a weakness, namely that the training process time is still quite high. This is the author's concern by doing a custom layer on the convolutional neural network. In addition, the addition of data augmentation is done to increase the number of data variations. The result used 18-34 custom layers in an average of around 97.93%, with an average computation time for the training process of about 1 minute 83 seconds. The resulting classification errors using Mean Absolute Error is 0,0163 © 2021 IEEE.
ABSTRACT
Coronavirus is named because of the structure of the crown on its body surface. The effect of the Coronavirus for sufferers is a disturbance in the respiratory system. Several days later, the disruption from the lung infection got worse. To identify the cause of this disease, the doctor performs a computed tomography scan and manually observes the changes that occur in the lungs through an X-ray. Image identification using machine learning is the latest trend these days to assist medical analysts. If the number of patients treated is large enough, this is very helpful in the analysis. The choice of Convolutional Neural Network is due to the many architectural algorithms being developed at this time. This method works with multiple layers. But the drawback is that the computation time for the training process takes a long time. The purposed way in this research is a custom layer using 18-34 layers. There is four class in the test, namely Normal lung conditions, COVID19, bacterial pneumonia, and viral pneumonia. Data augmentation is used to add variation to data. The results showed that the method offered could be used to identify pneumonia with an average identification accuracy of 98.7% - 100%. The average value of error the MSE 18-34 layer is 0.0539, RMSE 0.1981, and MAE 0.0319. The average consumption time for the training process is 2.25 seconds. The best accuracy calculation is obtained at 34 layers with the Adaptive Moment Estimation optimizer with a computation time of around 1 minute 48 seconds. © 2020 IEEE.
ABSTRACT
COVID19 is a pandemic of infectious diseases caused by a coronavirus. This virus is a new variant found in Wuhan, China, in December 2019. Symptoms felt by COVID patients19, in general, are cold, the body feels tired, and dry cough. However, some patients may experience nasal congestion, runny nose, sore throat, or diarrhea. Medically, to identify this disease, visual radiological observation is carried out. The development of computer technology helps to process data through image processing. At this stage, the convolutional neural network is the latest and in-depth machine learning machine used to classify images. Observations did on X-Ray Images with four classes. Namely lung Normal condition 234 files, exposed to COVID 43 files, exposed to bacterial 242 files, and exposed to virus 148 files. Preprocessing did use auto contrast to improve image sharpness. Data augmentation was exposed to increase the amount of data variation. In addition to the X-Ray dataset, this research also uses two classes of COVID and NON-COVID on the CT-Scan dataset. The results were using 34-layers, resulting in an average accuracy of 99.25% and on 26-layer an average accuracy of 97.86%. The training time needed is 1 minute and 15 seconds. Average Error results for 34-layer is MSE 0.0237, RMSE 0.1441 and MAE 0.0120. It is 50% better than the 26 layer shows an average MAE of 0.00351. © 2021 American Institute of Physics Inc.. All rights reserved.