szu-chi / implement-mnist-handwritten-digit-classification-on-mcu

This repository demonstrate how to implement CNN on MCU by using STM32 Cube-AI.

Implement MNIST Handwritten Digit Classification on MCU

This repository demonstrate how to implement CNN on MCU by using STM32 X-Cube-AI.

Environment
OS	Win11
MCU	NUCLEO-F411RE
Python	3.9.12
Keras	2.6.0
Tensorflow	2.8.0
X-Cube-AI	7.2.0
IDE	VScode, STM32CubeIDE

Workflow:

1. Using Keras build classification model.
2. Create STM32 project.
3. Testing classification performance on MCU

Using Keras build classification model.

python3 train_model.py

Model Summary

Layer (type)	Output Shape	Param #
input_1 (InputLayer)	[(None, 28, 28, 1)]	0
conv2d (Conv2D)	(None, 8, 8, 8)	208
conv2d_1 (Conv2D)	(None, 3, 3, 16)	1168
flatten (Flatten)	(None, 144)	0
dense (Dense)	(None, 10)	1450

Total params: 2,826
Trainable params: 2,826
Non-trainable params: 0

Create STM32 project.

Step 1 Create New STM32 project:

1. Select NUCLEO-F411RE
   
2. Create STM32 Project
   

Step 2 Setup STM32 Peripherals:

1. X-Cube-AI

• Click Software Packs -> Select Components
  
• Enable X-CUBE-AI Core
  
• Add Network
  • Software Packs -> STMicroelectronics.X-CUBE-AI.7.2.0 -> Add network
  • Change model name to digit_cnn
  • Click Browse button select pre-trained model digit_classifier_CNN.h5
  • Software Packs -> STMicroelectronics.X-CUBE-AI.7.2.0 -> Analyze
• Software Packs -> STMicroelectronics.X-CUBE-AI.7.2.0 -> Show graph\
  

2. Activate TIM11

3. Enable Virtual Com Port Global Interrupt

4. Enable generate peripheral initalization as a pair of '.c/.h' file per peripheral

5. Generate code and Replace Scirpt

• Core/Inc/main.h
• Core/Src/main.c
• Core/Src/stm32f4xx_it.c

6. Upload program to STM32

Testing classification performance on MCU

This repository testing architecture shown as below figure.

python3 testing_on_mcu.py

Demo Video Link:

• [Using Keras build classification model]
• [Create STM32 project]
• [Testing classification performance on MCU]

Reference:

• TensorFlow 2 quickstart for beginners
• TensorFlow 2 quickstart for experts
• TinyML: Gettting Started with STM32 X-CUBE-AI
• X-CUBE-AI Expansion Package Embedded Documentation