Intelligent Detection of Suspicions of COVID-19 Using X-Ray Imaging

Organization, Extraction, Classification and Prediction of Age in Facial Images using Convolutional Neuronal Network

“…The images of faces have been obtained from 7 public databases such as: age-gender-data, age-prediction, all-agefaces, appa-real, face-age, fg-net and utkface in total 58554 images were obtained divided into 6 categories, which are: from 8 to 15, from 15 to 23, from 23 to 35, from 35 to 44, from 44 to 57, from 57 to more years, this age division is made because the applied ranges cover most of the age of a person, in addition this division covers as much data as possible from the databases obtained; in figure 2 we can see some of the images with which we will work [3].…”

“…It was debated among various mathematical models because the most precise is required for the implementation in the research work, it was concluded in the approach of a Convolutional Neuronal Networks (CNN) implemented in Python using keras and TensorFlow [3], this neural network has 6 stages of operation, these are:…”

“…In addition, between layers 4 and 5, an Elatten layer was added, which oversees changing the magnitudes of the tensor, so that the modified or adjusted magnitude is the same as the number of resources contained in the tensor. [3]  Phase 5 and 6: They are layers composed of the Dense, Batch Normalization and Dropout layers. In the end, after phase 6 you have an output layer, which in this situation is the Dense layer, which is a layer fully connected to each of the neurons of the previous layer.…”

“…In the end, after phase 6 you have an output layer, which in this situation is the Dense layer, which is a layer fully connected to each of the neurons of the previous layer. [3] It should be noted that each of the mentioned layers have as activation functionality to Rectified Linear Unit (ReLU), except for the last layer whose activation functionality is the SoftMax functionality, in Figure 4 we have the possibility of seeing the scheme with the layers of the proposed model. Once the model has been built, it should be compiled, for that we use the categorical_crossentropy functionality as a loss functionality, in addition to this an Adam optimizer was used; When the compiled model is obtained, the training is carried out using the images of the Training folder indicated above, finally the model is saved in the Hierarchical Data Format (HDF), through the h5 expansion.…”

“…There are various difficulties for physical recognition such as the equality of the twins, obstruction in the face, amorphous faces, etc. [3], technological difficulties such as the quality of the image, the electronic equipment that captures the image, the movement that is being exerted at that moment of the capture, the light of the reflection, etc., and the most important factor that is the software that must be robust, fast and capable of preprocessing, processing, extracting characteristics, identifying and verifying the identity of the captured face [4].…”