Neural network implementation in microcontroller with ARM Cortex-M4 core

Problem klasifikacije rukom pisanih brojeva riješen je pomoću višeslojne neuronske mreže. Mreža je trenirana u Python programskom jeziku na MNIST skupu podataka u kojem se nalaze slike rukom pisanih znamenaka. Nakon faze učenja, neuronska mreža je testirana na testnom skupu podataka. Uz pomoć spremljenih težina i bias vrijednosti, ista takva mreža kreirana je u C programskom jeziku. Na samome kraju navedena mreža je implementirana u mikroupravljač s ARM Cortex-M4 jezgrom te su testirane brzina i točnost mreže.Problem of classification of handwritten digits is solved with multilayer neural network. Network was trained in Python programming language on MNIST dataset which contains pictures of handwritten digits. After learning phase, neural network was tested on test dataset. With saved weights and biases, the same network is created in C programming language. At the very end, aforementioned network was implemented into the microcontroller with ARM Cortex-M4 core and speed and precision of network were tested