Justification for choosing a single-board hardware computing platform for a neural network performing image processing

This article describes testing the operation of the YOLO2 neural network on single-board platforms and processors. The task was to find a computing platform on which the neural network could perform the processing of the images presented to it as quickly as possible. In this case, interference in the operation of the remaining modules of the system should have been minimized. In addition, the task was to minimize the solution price. Based on the results of the comparison, the Neural Compute Stick 2 hardware platform was chosen as a computing platform for the tasks with neural networks. In the future, the selected module can be used instead of the NVIDIA 1070 ti video card for the task of clustering and segmenting objects using the YOLO2 real-time network. This choice will simplify the design of the hardware-software complex and minimize its cost.