Please use this identifier to cite or link to this item:
|Title:||Radio-frequency (RF) sensing for deep awareness of human physical status||Authors:||Lim, Jun Wei||Keywords:||Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Engineering::Computer science and engineering::Computer applications::Life and medical sciences
|Issue Date:||2022||Publisher:||Nanyang Technological University||Source:||Lim, J. W. (2022). Radio-frequency (RF) sensing for deep awareness of human physical status. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/157024||Abstract:||Radio-frequency has been gaining its popularity over the years due to its ability to transmit data remotely. Radar is the product of radio-frequency and is able to detect an object’s distance and velocity. The radar has many potentials such as detecting vital signs which may be implemented in hospitals for reading patients that are not able to take a reading through traditional means such as burnt patients. Reading vital signs of these patients with radar would help as it does not require the equipment to have any contact with the patient. The purpose of this project is to apply deep learning on raw data that is produced by the current hardware which has the purpose of reading human vital signs through the means of radar. Currently data from this device has reading of interference as well apart from the actual reading. In this project, models with two different purposes are proposed to help identify actual reading from interferences and extract respiration rate from the raw data files that have been produced by the radar. Two models, Convolution Neural Network and Recurrent Neural Network are used in this project to determine which model would produce a better result to achieve the purpose of this project. Different batch sizes along with different dataset and intake size were tested to determine which configuration is best suited for training. Although Recurrent Neural Network is better suited for time series data such as radar data, this is not the case in this project. In the model with the purpose of predicting actual reading from interference, Convolution Neural Network has performed better than Recurrent Neural network which is shown with confusion matrix later in the report. With actual reading being 0.0278% of the whole dataset, Convolution Neural Network was still able to predict 52% of actual reading correctly. In the model with the purpose of predicting the respiration rate from a given bin, Convolution Neural Network has outperformed Recurrent Neural Network with the lowest val loss of 0.738 while Recurrent Neural Network has its lowest val loss at 1.48.||URI:||https://hdl.handle.net/10356/157024||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Student Reports (FYP/IA/PA/PI)|
Updated on Nov 27, 2022
Updated on Nov 27, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.