Implementation of Deep Learning Model on Low-Cost Embedded System (ARM Cortex-M)

This repository is part of my MSc thesis, focusing on deploying a YOLO-based deep learning model for object detection on low-cost microcontrollers, specifically the STM32H743 Nucleo board with ARM Cortex-M7 architecture.

Target Hardware

• Microcontroller: STM32H743 Nucleo Board
• Architecture: ARM Cortex-M7

Development Tools & Software

• IDE & Deployment Tools: STM32CubeIDE, X-Cube-AI
• Deep Learning Frameworks: Keras, TensorFlow, TensorFlow Lite (TFLite), Darknet

Repository Structure

• imgs/ — Contains images and diagrams used for documentation and visualization.
• Srcs/ — Contains source code for STM32 which was configured and generated by STM32 Cube IDE.
• Docs/ — Includes documentation, tutorials, and references for deploying ML/DL on ARM Cortex-M microcontrollers.
• Dataset/ — Stores datasets used for training and evaluation.
• cfgs/ — Configuration files for the deep learning model, including network architecture and inference settings.
• Demo/ — Contains parsing scripts and example applications to test the deployed model on STM32.

References

• Model - YOLOFastestv1
• Dataset from Robolow - Hard Hat Universe Dataset
• Model Quantization - Tensorflow Lite Conversion