har-feature-augmentation

Model architectures and data for feature-augmented CNNs deployed on TinyML for use in HAR applications

Dependencies

• TensorFlow 2 Framework (Inc. with Google Colab)
• Arduino TFLite Library

File Structure

X is used to as a placeholder for specific architectures or model version files.

```

data.csv // Model hyperparams and experimental data harserialcom.py // Modified 'TestOverSerial' Script LICENSE README.md

arduino harmodelX // Model type (e.g. R, AB ...) . arduinomain.cpp
. hardetectionmodel.cpp // File containing model weights . hardetectionmodel.h harmodel.ino // Main file mainfunctions.h
modelsettings.h // Global parameter definitions testdata.cpp // One frame of test data testdata.h

   data
           data_combined.csv      // Data sent to MCU over Serial
           serial_test_config.json // Config for har_serial_com.py

models X // ZIP files containing models used for study X1228.zip // Specific model version ZIP files (1-8) . X24216.zip . X38332.zip . X416464.zip X5326128.zip X6648256.zip X712811512.zip X8256161024.zip

python // Used to create NN models cnnuci2f.ipynb cnnuci2rf.ipynb cnnucir.ipynb

ucidata // Data for train, val and test datasets alldata.csv alldataquant.csv alldatatest.csv alldataval.csv answers.csv answerstest.csv answersval.csv features.csv testfeatures.csv valfeatures.csv

```

Workflow

• Modify all necessary file directories in the python files, and the serial port in har_serial_com.py.
• Run the .ipynb files in Google Colab. To run the file, a Google Drive must be connected containing the files in uci_data.
• This will automatically generate har_detection_model.cpp and test_data.cpp files.
• Download the ZIP file and copy the .cpp and .h files to the Arduino project directory matching the model type (R,F or A).
• Connect an Arduin
• Compile using the Arduino IDE.
• Copy har_serial_com.py into the Scripts subfolder of the Arduino TFLite Library
• Open the command line at the directory of the installed Arduino TFLite library.
• Run the command python .\scripts\har_serial_com.py --example har_model --verbose all (more info)