VLSI design and implementation of High-performance Binary-weighted convolutional artificial neural networks for embedded vision based Internet of Things (IoT)

Abstract

To overcome the disadvantages of convolutional neural networks (CNN) architectures, Binary-weighted convolutional neural networks (BCNN) architecture is proposed. CNN utilizes high precision weights, and BCNN uses binary weights. CNN requires power-hungry, massive and expensive processors while BCNN requires power-efficient processors. The proposed architecture provides high throughput and low power dissipation. Furthermore, it also reduces computational and hardware complexity, storage complexity, critical path delay, bandwidth requirements and improves accuracy. The proposed architecture is realized using Field programmable gate array (FPGA). The proposed architecture can be applied in machine learning, computer vision, and classification of motion, analysis of data, signal and image processing and subsequently the proposed architecture wholly is befitted for embedded vision-based systems that hold a low energy resource.