IMPLEMENTATION OF DEEP LEARNING NEURAL NETWORK LENET5 ON STM32 MICROCONTROLLER FOR IMAGE RECOGNITION

Abstract

The advent of smart mobile devices, along with the explosion of Internet-based applications and services, has led to the birth of a new computing paradigm – edge computing. Along with the current expanding trend of artificial intelligence applications, deploying artificial intelligence and deep learning applications on edge computing platforms is a prominent trend. This paper will investigate the ability to execute deep learning models using the convolutional neural network LeNet5 for deep learning problems implemented on low-power microcontrollers based on ARM architecture. We present the process of designing and implementing the handwritten digit recognition problem on the STM32 development board. We use Google Colab and Python language to train the convolutional neural network model, then map the trained model to execute on the STM32F411 microcontroller development board with the use of X-Cube-AI tool. The experimental results show that the implementation on the microcontroller achieves nearly the same performance as that on the general purpose computers.