Webinar Student Presence System Based on Regional Convolutional Neural Network Using Face Recognition

Abstract

World health organization announce Covid-19 as a pandemic so On March 15th 2020, the social distancing has been established with working, learning, and praying from home. Webinar is one of the solutions so those activities still can be done face to face and conference-based. With webinar, users can interact each other in an online meeting from home. Student presence is part of a webinar. The purpose of this research is to design an accurate student presence with a face recognition system using R-CNN method. The object of this research is a human face with sufficient light, medium, and the face must be facing the camera. This research proposed for a webinar student presence system is using face recognition with Regional Convolutional Neural Network (R-CNN). With object detection and several scenarios used in this method, the webinar student presence system using R-CNN will be more accurate than the methods that have ever been used before. This research has done four scenarios to obtain the best parameters like 45 of total layers, test data of the whole dataset percentage as 10%, RMSProp as model op- timizer, and 0.0001 learning rate. With those parameters, it have resulted the best system performance including 99.6% accuration, 1 × 10-4 loss, 100% precision, 99% recall, and 99.5% F1 Score.

Detection of iron deficiency anemia by medical images: a comparative study of machine learning algorithms

“…Convolutional Neural Network (CNN) is a type of algorithm in machine learning that extracts and processes data in the form of pictures or images using classification [ 19 ]. CNN is composed of or contains two major components: feature extraction, which consists of distinct descriptions that aid in increasing the precision of the data to be processed and intends to extract vital information from the data, and the classification layer, which occurs after the extraction of data features via the use of fully connected neurons for transforming and the dimension of data [ 20 , 21 ].…”

“…The k-NN contains a tiny positive integer denoted by "k." In an experimental setting, a decision with a majority of neighbors is employed. An ideal case is when k = 2 [ 20 ] is allocated to the nearest neighbor in its class. The approach computes the distance between the feature vectors and their nearest neighbors and does not generate duplicates, instead producing synthetic data points that varied slightly from the actual data points.…”

Bell Pepper Leaf Disease Classification Using Fine-Tuned Transfer Learning

“…It is common practice to utilize dropouts to avoid overfitting and accelerate the learning process [36], [37]. Optimizers are algorithms that adjust the weights and biases in the neural network by reducing the distance between the network output and the target [38]. This stage determines whether it is the best model developed with the highest accuracy value at the Transfer Learning architecture stage.…”