MeatNet: An FPGA-based Machine Learning tool for in-situ food quality tracking using sensor fusion

This repository contains the code used in the paper entitled An FPGA-based Machine Learning tool for in-situ food quality tracking using sensor fusion from MDPI Biosensors.

If you find it useful, please consider cite us:

• Enériz, D.; Medrano, N.; Calvo, B. An FPGA-Based Machine Learning Tool for In-Situ Food Quality Tracking Using Sensor Fusion. Biosensors 2021, 11, 366. https://doi.org/10.3390/bios11100366

• @article{En_riz_2021, title={An FPGA-Based Machine Learning Tool for In-Situ Food Quality Tracking Using Sensor Fusion}, volume={11}, ISSN={2079-6374}, url={http://dx.doi.org/10.3390/bios11100366}, DOI={10.3390/bios11100366}, number={10}, journal={Biosensors}, publisher={MDPI AG}, author={Enériz, Daniel and Medrano, Nicolas and Calvo, Belen}, year={2021}, month={Sep}, pages={366}}

Requirements

In requirements.txt you can find the list of required packages and their versions used in the development of this project.

It is also recommended to run the notebooks in a Jupyter environment where the vivado_hls command is enabled, since the HLS simulations are automatically launched from Python using the .tcl scripts, that load the HLS-related files (MeatNet-XXX.X).

Contents

This work is contained in the four notebooks included in the repo. In consumption.ipynb you can find the steps to generate the World's consumption meat consumption, which motivated this work.

All the steps to download, visualize, and preprocess the dataset are contained in the virtualization.ipynb, where the training process is also contained. After this, the main results of regression and classification of the trained models are also included.

Finally, in implementation.ipynb you will find the fixed-point datatype selection process, the retraining and the final HLS simulation. Then, there is a guide to generate the FPGA-implemented results, whose results are compared with the rest of the model stages.

The fourth notebook is fpga.ipynb, which contains the custom driver and the FPGA-performance test. This notebook must be uploaded and run in the PYNQ-Z2 file system (or whichever your board is).