The Design And FPGA Implementation Of Target Detection Hardware System Based On Asynchronous Design Method

With the rapid development of machine learning,target detection technology based on deep convolutional neural network has been widely used.However,due to the complexity and diversity of application scenarios,as well as the increasingly high requirements for target detection hardware devices,low power consumption and miniaturization have become an important research direction in this field.In the past decades,synchronous circuit was the main direction in the field of digital circuit design.However,with the natural advantages of low power consumption,strong adaptability and strong anti-interference ability,event-driven asynchronous circuit design method has drawn a lot of attention and interest,and ushered in a major development opportunity in recent years.In this paper,we take the advantages of low power consumption and strong adaptability of asynchronous circuit to design a target detection hardware system based on asynchronous design method and implement it on FPGA.Firstly,we briefly introduce the characteristics of asynchronous circuit,and propose the overall research scheme of target detection system according to the analysis of design requirements and hardware platform resources.Secondly,the newly developed Mobile Net-YOLOV3 target detection algorithm in literature is modified and optimized in this work.Thirdly,we present the architecture of the whole system which is divided into modules and implemented one by one.Finally,we test these modules on Vivado platform of Xilinx,and verify the function of the target detection hardware system integrated on Viretex-7VX485 T FPGA.In addition,we also analyze and compare its power consumption and performance with other works.The result shows that the proposed hardware system has good performance in speed and energy efficiency ratio.In particular,the asynchronous computing module has a significant reduction in power consumption when compared with the synchronous circuit in the same scenario,which proves the feasibility and effectiveness of the asynchronous design method to achieve low power target detection hardware system.