This repository contains the code I developed for my Master's Degree in Computer Engineering thesis. I graduated with honours at the University of Padova, where I was supervised by professor Fabio Vandin.
You can find my thesis at the following link: http://tesi.cab.unipd.it/62146/
Please refer at the following repository: https://github.com/DavideBuffelli/TrASenD for a more updated version, and a link to a paper version of this work.
If you use the code in this repository, please cite my thesis:
A Deep Learning Model for Personalized Human Activity Recognition, Buffelli D., University of Padova, 2019
or the following paper:
@article{buffelli2021attentionbased,
title={Attention-Based Deep Learning Framework for Human Activity Recognition with User Adaptation},
author={Davide Buffelli and Fabio Vandin},
year={2021},
journal={IEEE Sensors Journal}
}
-
deepSense.py My implementation of the DeepSense model.
-
pre-processing This folder contains the files that pre-process the data from the HHAR dataset.
-
transferLearning This folder contains the custom DeepSense model I developed. This model adapts to a specific user, improving the accuracy of the predictions.
-
tests The folder contains the files implementing the tests that have been done.
- Python 3.x
- NumPy package
- TensorFlow 1.5.x (haven't tested it with other versions)
- HHAR Dataset - Heterogeneity Activity Recognition Data Set.
- DeepSense: A Unified Deep Learning Framework for Time-Series Mobile Sensing Data Processing - The DeepSense original paper by Shuochao Yao, Shaohan Hu, Yiran Zhao, Aston Zhang, Tarek Abdelzaher.
- DeepSense - The code created by the authors of the paper.
- HHAR-Data-Process - The code for the pre-processing created by the authors of DeepSense.
- TensorFlow - TensorFlow official website.
- NumPy - NumPy official website.
Refer to the the file LICENCE.