Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0
According to the Industry 4.0 paradigm, all objects in a factory, including people, are equipped with communication capabilities and integrated into cyber-physical systems (CPS). Human activity recognition (HAR) based on wearable sensors provides a method to connect people to CPS. Deep learning has...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2018-07-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | http://www.mdpi.com/1424-8220/18/7/2146 |
| _version_ | 1828233867490754560 |
|---|---|
| author | Xiaochen Zheng Meiqing Wang Joaquín Ordieres-Meré |
| author_facet | Xiaochen Zheng Meiqing Wang Joaquín Ordieres-Meré |
| author_sort | Xiaochen Zheng |
| collection | DOAJ |
| description | According to the Industry 4.0 paradigm, all objects in a factory, including people, are equipped with communication capabilities and integrated into cyber-physical systems (CPS). Human activity recognition (HAR) based on wearable sensors provides a method to connect people to CPS. Deep learning has shown surpassing performance in HAR. Data preprocessing is an important part of deep learning projects and takes up a large part of the whole analytical pipeline. Data segmentation and data transformation are two critical steps of data preprocessing. This study analyzes the impact of segmentation methods on deep learning model performance, and compares four data transformation approaches. An experiment with HAR based on acceleration data from multiple wearable devices was conducted. The multichannel method, which treats the data for the three axes as three overlapped color channels, produced the best performance. The highest overall recognition accuracy achieved was 97.20% for eight daily activities, based on the data from seven wearable sensors, which outperformed most of the other machine learning techniques. Moreover, the multichannel approach was applied to three public datasets and produced satisfying results for multi-source acceleration data. The proposed method can help better analyze workers’ activities and help to integrate people into CPS. |
| first_indexed | 2024-04-12T19:46:07Z |
| format | Article |
| id | doaj.art-1850c4ccc4ab46299958937070af78d1 |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-04-12T19:46:07Z |
| publishDate | 2018-07-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-1850c4ccc4ab46299958937070af78d12022-12-22T03:18:57ZengMDPI AGSensors1424-82202018-07-01187214610.3390/s18072146s18072146Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0Xiaochen Zheng0Meiqing Wang1Joaquín Ordieres-Meré2Department of Industrial Engineering, ETSII, Universidad Politécnica de Madrid, 28006 Madrid, SpainSchool of Mechanical Engineering and Automation, Beihang University (BUAA), Beijing 100083, ChinaDepartment of Industrial Engineering, ETSII, Universidad Politécnica de Madrid, 28006 Madrid, SpainAccording to the Industry 4.0 paradigm, all objects in a factory, including people, are equipped with communication capabilities and integrated into cyber-physical systems (CPS). Human activity recognition (HAR) based on wearable sensors provides a method to connect people to CPS. Deep learning has shown surpassing performance in HAR. Data preprocessing is an important part of deep learning projects and takes up a large part of the whole analytical pipeline. Data segmentation and data transformation are two critical steps of data preprocessing. This study analyzes the impact of segmentation methods on deep learning model performance, and compares four data transformation approaches. An experiment with HAR based on acceleration data from multiple wearable devices was conducted. The multichannel method, which treats the data for the three axes as three overlapped color channels, produced the best performance. The highest overall recognition accuracy achieved was 97.20% for eight daily activities, based on the data from seven wearable sensors, which outperformed most of the other machine learning techniques. Moreover, the multichannel approach was applied to three public datasets and produced satisfying results for multi-source acceleration data. The proposed method can help better analyze workers’ activities and help to integrate people into CPS.http://www.mdpi.com/1424-8220/18/7/2146deep learningdata preprocessingHuman Activity Recognition (HAR)Internet of things (IoT)Industry 4.0 |
| spellingShingle | Xiaochen Zheng Meiqing Wang Joaquín Ordieres-Meré Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0 Sensors deep learning data preprocessing Human Activity Recognition (HAR) Internet of things (IoT) Industry 4.0 |
| title | Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0 |
| title_full | Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0 |
| title_fullStr | Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0 |
| title_full_unstemmed | Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0 |
| title_short | Comparison of Data Preprocessing Approaches for Applying Deep Learning to Human Activity Recognition in the Context of Industry 4.0 |
| title_sort | comparison of data preprocessing approaches for applying deep learning to human activity recognition in the context of industry 4 0 |
| topic | deep learning data preprocessing Human Activity Recognition (HAR) Internet of things (IoT) Industry 4.0 |
| url | http://www.mdpi.com/1424-8220/18/7/2146 |
| work_keys_str_mv | AT xiaochenzheng comparisonofdatapreprocessingapproachesforapplyingdeeplearningtohumanactivityrecognitioninthecontextofindustry40 AT meiqingwang comparisonofdatapreprocessingapproachesforapplyingdeeplearningtohumanactivityrecognitioninthecontextofindustry40 AT joaquinordieresmere comparisonofdatapreprocessingapproachesforapplyingdeeplearningtohumanactivityrecognitioninthecontextofindustry40 |