Adaptive transfer learning framework for dense prediction of human activity recognition

doi:10.19682/j.cnki.1005-8885.2019.0028

中国邮电高校学报(英文) ›› 2019, Vol. 26 ›› Issue (5): 1-10.doi: 10.19682/j.cnki.1005-8885.2019.0028

• Artificial Intelligence • 下一篇

Adaptive transfer learning framework for dense prediction of human activity recognition

Zhang Zhao^1,Zhang Yong ^{1 (*),} Teng Ying lei¹, Guo Da¹, Deng Hai qin²

1. School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
2. AIdong Super AI (Beijing) Company Limited, Beijing 100007, China

收稿日期:2019-01-21 修回日期:2019-04-09 出版日期:2019-10-31 发布日期:2019-11-06
通讯作者: ZhangYong, E-mail：yongzhang@bupt.edu.cn E-mail:yongzhang@bupt.edu.cn
作者简介:Corresponding author: ZhangYong, E-mail：yongzhang@bupt.edu.cn
基金资助:
国家重大科技专项

Adaptive transfer learning framework for dense prediction of human activity recognition

Zhang Zhao^1,Zhang Yong ^{1 (*),} Teng Ying lei¹, Guo Da¹, Deng Hai qin²

1. School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
2. AIdong Super AI (Beijing) Company Limited, Beijing 100007, China

Received:2019-01-21 Revised:2019-04-09 Online:2019-10-31 Published:2019-11-06
Contact: ZhangYong, E-mail：yongzhang@bupt.edu.cn E-mail:yongzhang@bupt.edu.cn
About author:Corresponding author: ZhangYong, E-mail：yongzhang@bupt.edu.cn
Supported by:
State Major Science and Technology Special Projects

摘要/Abstract

摘要： Human activity recognition (HAR) for dense prediction is proven to be of good performance, but it relies on labeling every point in time series with the high cost. In addition, the performance of HAR model will show significant degradation when tested on the sensor data with different distribution from the training data, where the training data and the test data are usually collected from different sensor locations or sensor users. Therefore, the adaptive transfer learning framework for dense prediction of HAR is introduced to implement cross-domain transfer, where the proposed multi-level unsupervised domain adaptation (MLUDA) approach combines the global domain adaptation and the specific task adaptation to adapt the source and target domain in multiple levels. The multi-connected global domain adaptation architecture is proposed for the first time, which can adapt the output layer of the encoder and the decoder in dense prediction model. After this, the specific task adaptation is proposed to ensure alignment of each class centroid in source domain and target domain by introducing the cosine distance loss and the moving average method. Experiments on three public human activity recognition datasets demonstrate that the proposed MLUDA improves the prediction accuracy of target data by 20% compared to the source domain pre-trained model and it is more effective than the other three deep transfer learning methods with an improvement of 10% to 18% in accuracy.

关键词: transfer learning, human activity recognition, dense prediction, global domain adaptation, specific task adaptation

Abstract: Human activity recognition (HAR) for dense prediction is proven to be of good performance, but it relies on labeling every point in time series with the high cost. In addition, the performance of HAR model will show significant degradation when tested on the sensor data with different distribution from the training data, where the training data and the test data are usually collected from different sensor locations or sensor users. Therefore, the adaptive transfer learning framework for dense prediction of HAR is introduced to implement cross-domain transfer, where the proposed multi-level unsupervised domain adaptation (MLUDA) approach combines the global domain adaptation and the specific task adaptation to adapt the source and target domain in multiple levels. The multi-connected global domain adaptation architecture is proposed for the first time, which can adapt the output layer of the encoder and the decoder in dense prediction model. After this, the specific task adaptation is proposed to ensure alignment of each class centroid in source domain and target domain by introducing the cosine distance loss and the moving average method. Experiments on three public human activity recognition datasets demonstrate that the proposed MLUDA improves the prediction accuracy of target data by 20% compared to the source domain pre-trained model and it is more effective than the other three deep transfer learning methods with an improvement of 10% to 18% in accuracy.

Key words: transfer learning, human activity recognition, dense prediction, global domain adaptation, specific task adaptation

Zhang Zhao, Zhang Yong, Teng Ying lei, Guo Da, Deng Hai qin. Adaptive transfer learning framework for dense prediction of human activity recognition[J]. The Journal of China Universities of Posts and Telecommunications, 2019, 26(5): 1-10.

参考文献

【1】Rezaie H, Ghassemian M. An adaptive algorithm to improve energy efficiency in wearable activity recognition systems. IEEE Sensors Journal, 2017, 17(16): 5315-5323.

Verma V K, Lin W Y, Lee M Y, et al. Levels of activity identification & sleep duration detection with a wrist-worn accelerometer-based device. Proceedings of the 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC’17), Jul 11-15, 2017, Seogwipo, Republic of Korea. Piscataway, NJ, USA: IEEE, 2017: 2369-2372.

González S, Sedano J, Villar J R, et al. Features and models for human activity recognition. Neurocomputing, 2015, 167: 52-60.

Sarkar J, Vinh L T, Lee Y K, et al. GPARS: A general-purpose activity recognition system. Applied Intelligence, 2011, 35(2): 242-259.

Zhao M M, Yue S C, Katabi D, et al. Learning sleep stages from radio signals: A conditional adversarial architecture Proceedings of the 34th International Conference on Machine Learning (ICML’17), Aug 6-11, 2017, Sydney, Australia. 2017: 4100-4109.

Lee S M, Sang M Y, Cho H. Human activity recognition from accelerometer data using convolutional neural network. Proceedings of the 2017 IEEE International Conference on Big Data and Smart Computing (BigComp'17), Feb 13-16, 2017, Jeju, Republic of Korea. Piscataway, NJ, USA: IEEE, 2017: 131-134.

Panwar M, Dyuthi S R, Prakash K C, et al. CNN based approach for activity recognition using a wrist-worn accelerometer. Proceedings of the 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC’17), Jul 11-15, 2017, Republic of Korea. Piscataway, NJ, USA: IEEE, 2017: 2438-2441.

Liu G C, Liang J H, Lan G J, et al. Convolution neutral network enhanced binary sensor network for human activity recognition. Proceedings of the 2016 IEEE SENSORS, Oct 30-Nov 3, 2016, Orlando, FL, USA. Piscataway, NJ, USA: IEEE, 2016: 3p.

Yao R, Lin G S, Shi Q F, et al. Efficient dense labelling of human activity sequences from wearables using fully convolutional networks. Pattern Recognition, 2018, 78: 252-266.

Zhang Y, Zhang Y, Zhang Z, et al. Human activity recognition based on time series analysis using U-Net. arXiv preprint arXiv:1809.08113, 2018.

Varamin A A, Abbasnejad E, Shi Q F, et al. Deep auto-set: A deep auto-encoder-set network for activity recognition using wearables. Proceedings of the 15th International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services (MobiQuitous’18), Nov 5-7, 2018, New York, NY, USA. New York, NY, USA: ACM, 2018: 246-253.

Pan S J, Yang Q. A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10): 1345-1359.

Tzeng E, Hoffman J, Zhang N, et al. Deep domain confusion: Maximizing for domain invariance. arXiv:1412.3474, 2014.

Long M S, Cao Y, Wang J. Learning transferable features with deep adaptation networks. Proceedings of the 32nd International Conference on Machine Learning (ICML’15), Jul 6-11, 2015, Lille, France. 2015: 97-105.

Long M S, Zhu H, Wang J M, et al. Deep transfer learning with joint adaptation networks. Proceedings of the 34th International Conference on Machine Learning (ICML’17), Aug 6-11, 2017, Sydney, Australia. 2017: 10p.

Ganin Y，, Lempitsky V . Unsupervised domain adaptation by backpropagation. arXiv:1409.7495v2, 2014.

Chen W H, Cho P C, Jiang Y L. Activity recognition using transfer learning. Sensors and Materials, 2017, 29(7): 897-904.

Khan M A A H, Roy N, Misra A. Scaling human activity recognition via deep learning-based domain adaptation. Proceedings of the 2018 IEEE International Conference on Pervasive Computing and Communications (PerCom’18), Mar 19-23, 2018, Athens, Greece. Piscataway, NJ, USA: IEEE, 2018: 9p.

Rokni S A, Nourollahi M, Ghasemzadeh H. Personalized human activity recognition using convolutional neural networks. arXiv:1801.08252v1,2018.

Bux Sargano A, Wang X F, Angelov P, et al. Human action recognition using transfer learning with deep representations. Proceedings of the 2017 International Joint Conference on Neural Networks (IJCNN'17), May 14-19, 2017, Anchorage, AK, USA. Piscataway, NJ, USA: IEEE, 2017: 463 - 469.

Xie S, Zheng Z, Chen L, et al. Learning semantic representations for unsupervised domain adaptation. Proceedings of the 35th International Conference on Machine Learning (ICML’18), Jul 11-13, 2018, Stockholm, Sweden. 2018: 5419-5428.

Chavarriaga R, Sagha H, Calatroni A, et al. The opportunity challenge: A benchmark database for on-body sensor-based activity recognition. Pattern Recognition Letters, 2013, 34(15): 2033-2042.

Barshan B, Yüksek M C. Recognizing daily and sports activities in two open source machine learning environments using body-worn sensor units. Computer Journal, 2013, 57(11): 1649-1667.

Reiss A, Stricker D. Introducing a new benchmarked dataset for activity monitoring. Proceedings of the IEEE 16th International Symposium on Wearable Computers (ISWC’12), Jun 18-22, 2012, Newcastle, UK. Piscataway, NJ, USA: IEEE, 2012: 108-109.

Adaptive transfer learning framework for dense prediction of human activity recognition

Adaptive transfer learning framework for dense prediction of human activity recognition

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 0

编辑推荐

Metrics

本文评价