http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
LSTM-Based Human Fall Detection using Thermal Array Sensor
Adinda Riztia Putri,Goodness Oluchi Anyanwu,Mareska Pratiwi Maharani,Jae Min Lee,Dong-Seong Kim 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.6
Accidental fall may lead to numerous serious and deadly injures. Existing fall detection systems mostly use cameras and are considered a privacy-intrusive approach. Thermal array sensors are considered a privacy-friendly device that does not raises discomforts for users. In this study, we simulate a fall detection system using a thermal array sensor with three different algorithms: CNN, LSTM, and CNN-LSTM. Our result shows that LSTM has the best accuracy among other algorithms by 99.96%.
Fast Sensor Data Recovery using Multi-Directional LSTM for Industrial Wireless Sensor
Adinda Riztia Putri,Mareska Pratiwi Maharani,Ade Pitra Hermawan,Jae Min Lee,Dong-Seong Kim 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.11
Missing sensor values are common in Wireless Sensor Network due to its unpredictable and fragile nature. While missing sensor values may lead to several serious problems in the industrial environment where the sensors are deployed, it is inevitable at some points. Fast sensor data recovery using multi-directional LSTM is introduced to address this problem where it can handle missing sensor values and replace them with the predicted values from the previous sensor data. Our proposed method shows a promising result of low prediction loss while also achieve fast computation time of RMSE by 0.11, MAE by 0.07, MAPE by 0.01, and computing time by 0.341 ms.
IoT Sensor Data Prediction over Distributed Wireless Sensor Network
Adinda Riztia Putri,Made Adi Paramartha Putra,Jae Min Lee,Dong-Seong Kim 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.2
Traditional wireless sensor network (WSN) architecture generally exploited a single sink node as an intermediary device between sensors and the cloud. Meanwhile, the current trend of implementing a deep learning model on WSN to handle missing data values will be such a heavy workload for the sink node. This study presented the analysis of distributing the WSN architecture into fog nodes while also comparing the deep learning models that suit the distributed environment. Our implementation result shows that MLP achieve the best result of performing the data prediction by loss prediction of 0.13 RMSE, 0.09 MAE, 0.07 MAPE, and also low computational cost of 37% CPU usage on distributed computing WSN.
Ahmad Zainudin,Adinda Riztia Putri,Goodness Oluchi Anyanwu,Cosmas Ifeanyi Nwakanma,Dong-Seong Kim,Jae-Min Lee 한국통신학회 2021 한국통신학회 학술대회논문집 Vol.2021 No.11
In this paper, we proposed a hybrid-Long Short-Term Memory (LSTM) deep learning algorithm for thermal sensor-based Human Activity Recognition (HAR). Edge computing is characterized with Deep Learning (DL) computational capability as well as real-time response which is a requirement for non-intrusive HAR application. Applying DL on edge devices is more challenging due to the limited computational capability. Hence, a low-cost computational CNN-LSTM model is proposed in this work. Based on the simulation results, the proposed approach achieved a computational time of 0.5543 ms. which outperforms other algorithms.
Towards Delay Aware Data Recovery using Deep Learning in Embedded IoT Application
Made Adi Paramartha Putra,Adinda Riztia Putri,Mark Verana,Dong-Seong Kim,Jae-Min Lee 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.2
Data recovery technique in internet of things (IoT) network able to improve the devices longevity by reducing data retransmission. Most of existing research mainly focused on improving the prediction accuracy based on simulation work. Therefore, this study evaluate the data recovery in real world environment using deep learning (DL) and embedded devices by considering the processing time delay. The main goals is to ensure the low latency by applying data recovery in IoT network with low computational resource. A low complexity of deep neural network (DNN), convolutional neural network (CNN), and long short-term memory (LSTM) network were compared. Based on experimental results, the DNN network model is able to maintain low processing delay with average of 5.18 ms and 98.85% accuracy to recover single missing data.
Resolution-Aware Deep Learning Model for Emergency Communication in Smart Homes using Thermal Sensor
Goodness Oluchi Anyanwu,Cosmas Ifeanyi Nwakanma,Adinda Riztia Putri,JeongHan Kim,Gihwan Hwang,Jae-Min Lee,Dong-Seong Kim 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.2
Due to the development in sensor technologies and smart homes, Emergency and Activity Detection (EAD) has become a growing research issue as there is a need to support safety and security in homes. In this work, the impact of three different thermal sensor resolutions was investigated for EAD. The design of the system includes three parts: data acquisition, EAD and the emergency alert system. An alert system is considered reliable if the sensing model can mitigate the introduction of noise by the sensors or noisy environments. Research in this domain has seen the adoption of sensors with different resolutions. However, not much work has been done in developing resolution-aware models considering the impact of sensor resolutions on both the quality of data and the performance of the classification models. In this work, a CNN model was developed for EAD from datasets of various sensors with diverse resolutions. The results showed that the proposed model exhibited resilience in handling the error that may occur from the impact of sensor resolution for classification of normal daily living activity and emergency in a smart home.