Intelligent Commodity Settlement System based on Embedded Equipment and Convolutional Neural Network

Abstract

The rapid development of deep neural networks makes unmanned supermarket solutions based on computer vision possible. However, the computational complexity of convolutional neural networks is much higher than traditional algorithms, and the limitations of limited resources on embedded devices cannot meet real-time requirements. This article proposes an intelligent commodity settlement system based on embedded devices and deep learning. It uses CenterNet network and heterogeneous convolution filters to fuse. The initial layer convolution kernel is designed as a heterogeneous kernel to solve the detection of large differences in commodity scales. Experimental results show that the improved network structure has an average detection accuracy improvement of 3.2% compared to the original network structure, and the IoU index is increased by 3.1%, which can meet the real-time commodity recognition requirements of embedded devices.