Comparison of the Efficiency of a Neural Network for Image Recognition on Microcontrollers

Вантажиться...
Ескіз

Дата

2024

Науковий керівник

Назва журналу

Номер ISSN

Назва тому

Видавець

КПІ ім. Ігоря Сікорського

Анотація

The paper is devoted to comparing two popular models of 32-bit microcontrollers for working with neural networks for object recognition. The target devices were the ESP32 and STM32 microcontrollers, on which an artificial neural network was deployed, written using the Python programming language and the TensorFlow library. Micropython was chosen as the operating system for the microcontrollers. The paper compares the performance of the ESP32 and STM32 microcontrollers for object detection using a neural network and their classification. The image recognition time and the percentage of correctly classified objects were compared depending on the number of neuron layers and the number of training epochs within these networks. The article shows that the number of layers and training epochs directly affects the accuracy of object classification in the image. The obtained results show that increasing the number of layers of the neural network increases the overall accuracy of object recognition using the studied neural network, increasing the number of training epochs logarithmically increases the accuracy of recognition and classification within the neural network, but at the same time, increasing the number of neuron layers leads to an increase in the total recognition time. The difference in the obtained results for the accuracy of image recognition of microcontrollers differs within 5%.

Опис

Ключові слова

microcontroller, neural network, epoch, training, classification, мікроконтролер, нейромережа, епоха, навчання, класифікація

Бібліографічний опис

Sharuiev, R. D. Comparison of the Efficiency of a Neural Network for Image Recognition on Microcontrollers / R. D. Sharuiev, P. V. Popovych // Мікросистеми, Електроніка та Акустика : науково-технічний журнал. – 2024. – Т. 29, № 2(127). – С. 300851.1-300851.7. – Бібліогр.: 26 назв.