https://github.com/tszydlo/FogML-Zephyr

https://github.com/tszydlo/FogML-Zephyr

tinyml zephyr-rtos

Last synced: 16 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/tszydlo/FogML-Zephyr
• Owner: tszydlo
• License: apache-2.0
• Created: 2022-06-23T11:41:55.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-12-15T15:03:37.000Z (11 months ago)
• Last Synced: 2024-07-30T21:08:19.083Z (3 months ago)
• Topics: tinyml, zephyr-rtos
• Language: C
• Homepage:
• Size: 85 KB
• Stars: 2
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# FogML-Zephyr

The application recognizes gestures an IoT device performs based on the built-in accelerometer. It uses the classification algorithms and programming tools provided by FogML:

* FogML-SDK [https://github.com/tszydlo/fogml_sdk]

* FogML tools [https://github.com/tszydlo/FogML]

The application is implemented for the Zephyr-OS operating system. The tests were carried out on a Thingy:91 device containing nrf9160 and manufactured by Nordic Semiconductor.

## Classification

The accelerometer data processing pipeline is presented in the figure.

![Classification pipeline](./doc/classification.png)

In the example, gestures are classified using the random forest algorithm. The generated classifier source code is in the `src/fogml_generated/random_forest_model.c` file.

For the device to recognize other gestures, it is necessary to collect training data, assign labels, and train the model. The entire process is described in Jupyter Notebook, along with training data in the `tools` directory. Notepad can be run in eg Google Colab or Anaconda.

## Anomaly detection

The example uses the modification of the K-Means algorithm and z-score to detect anomalies - deviations from training data. Model generation for this purpose is described in Jupyter Notebook located in the `tools` directory.

## Building

The application is implemented for the Zephyr OS operating system. Since the application is for Thingy:91 device, we use NRF Connect SDK 2.0.0 and VSCode plugin, which simplifies the process of building, flashing and debugging applications for Zephyr. The application can also be built directly using the `west` tool and the base Zephyr.

The application uses the basic drivers available in the operating system, and porting to another device should be relatively easy.

## Bibliography

Please mention us and cite our papers if you use this work and find it useful.

```

@inproceedings{FogMLSzydlo2018,

  author    = {Tomasz Szydlo and

               Joanna Sendorek and

               Robert Brzoza{-}Woch},

  editor    = {Yong Shi and

               Haohuan Fu and

               Yingjie Tian and

               Valeria V. Krzhizhanovskaya and

               Michael Harold Lees and

               Jack J. Dongarra and

               Peter M. A. Sloot},

  title     = {Enabling Machine Learning on Resource Constrained Devices by Source

               Code Generation of the Learned Models},

  booktitle = {Computational Science - {ICCS} 2018 - 18th International Conference,

               Wuxi, China, June 11-13, 2018, Proceedings, Part {II}},

  series    = {Lecture Notes in Computer Science},

  volume    = {10861},

  pages     = {682--694},

  publisher = {Springer},

  year      = {2018},

}

```