Negative Selection by Clustering for Contrastive Learning in Human Activity Recognition

This repository is designed to implement the idea of "Negative Selection by Clustering for Contrastive Learning in Human Activity Recognition" in this paper. For more details, please see https://arxiv.org/abs/2203.12230 or https://ieeexplore.ieee.org/document/10026419

Requirements

This project code is done in Python 3.8 and third party libraries.

TensorFlow 2.x is used as a deep learning framework.

The main third-party libraries used and the corresponding versions are as follows:

• tensorflow 2.3.1

• tensorflow_addons 0.15.0

• numpy 1.18.5

• scipy 1.5.0

• scikit-learn 0.23.1

• sklearnex

Running

This demo can be run with the following command:

python main.py

Code Organisation

The main content of each file is marked as follows:

• Augment.py: This file contains a variety of sensor data augmentation methods.
• TPN.py: This file contains the network structure of TPN
• main.py: This file contains the details of ClusterCLHAR and the methods how it works.

Experiment

Results of different divisions in self-supervised experiments.

MotionSense:

Test subjects	Train/Val subjects	F1-score
1 - 5	rest	95.16
6 - 10	rest	90.29
11 - 15	rest	86.76
16 - 19	rest	89.13
20 - 24	rest	90.95
1, 9,12, 17, 19	rest	88.99
2, 6, 10, 14, 22	rest	90.23
3, 4, 5, 13, 16	rest	94.26
8, 18, 21, 23, 24	rest	87.28
11, 15, 20, 7	rest	93.98

UCI-HAR:

Test subjects	Train/Val subjects	F1-score
1 - 6	rest	94.00
7 - 12	rest	89.59
13 - 18	rest	89.25
19 - 24	rest	97.95
25 - 30	rest	93.00
9, 10, 16, 18, 24, 28	rest	83.73
1, 5, 13, 17, 25, 29	rest	95.33
2, 3, 6, 12, 14, 23	rest	91.05
4, 19, 22, 26, 27, 30	rest	96.61
7, 8, 11, 15, 20, 21	rest	96.45

Citation

If you find our paper useful or use the code available in this repository in your research, please consider citing our work:

@ARTICLE{10026419,
  author={Wang, Jinqiang and Zhu, Tao and Chen, Liming Luke and Ning, Huansheng and Wan, Yaping},
  journal={IEEE Internet of Things Journal}, 
  title={Negative Selection by Clustering for Contrastive Learning in Human Activity Recognition}, 
  year={2023},
  volume={10},
  number={12},
  pages={10833-10844},
  doi={10.1109/JIOT.2023.3239945}
}

Reference

• https://github.com/iantangc/ContrastiveLearningHAR

• https://github.com/google-research/simclr.

• https://github.com/diheal/resampling