System design of neurocomputers on microcontrollers in conditions of limited resources
The development of software tools for electronic equipment has led to the development and widespread use of neural network technology. They are used for processing and making decisions based on the received information, which is not discrete, but has a polymorphic essence. Processing entity data and computing decisions requires significant amounts of computing power and device operating memory. This problem does not allow the widespread use of neural network technologies in portable devices and devices based on microcontrollers. The aim of the article – adapt neural network technology for use on portable environments and microcontroller-based electronic devices. The chosen method of implementing a neural network based on the resource-saving Hamming algorithm, and the optimized program code in the C language made it possible to significantly reduce the requirements for the hardware of the device on which this technology can be implemented. The analysis of modern microcontrollers allowed us to choose and apply the optimal power-to-energysaving microcontroller-STM32, which allowed us to implement a simplified neural network on its basis. The developed algorithm was implemented on a debug board with an STM32 microcontroller in a device that allows you to recognize handwritten numbers entered from the touch screen. The created portable device for recognizing handwritten numbers is applicable as a module in other electronic equipment products. The prospects of using the latest variants of implementing neurocomputers on microcontrollers are shown.