Browse By Repository:

 
 
 
   

Accelerating the convolutional neural networks(CNN) using FPGA

Tham, Wei Jian (2020) Accelerating the convolutional neural networks(CNN) using FPGA. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

[img]
Preview
Text (Full Text)
Accelerating the convolutional neural networks(CNN) using FPGA.pdf - Submitted Version

Download (3MB) | Preview

Abstract

The convolutional neural network (CNN) is inspired by the behavior of optic nerves in the living creatures and also has a huge application in video surveillance, mobile robot vision, image search engine in database, etc. Besides, the rapid growth of CNN has shown that the performance of CNN now surpasses of the other type of visual recognition algorithms, and even beyond the human accuracy on certain conditions. In this work, the FPGA platform is used to implement the CNNs of different application. This is because FPGA has good performance, high energy efficiency, fast development round, and capability of reconfiguration. We also set up two different platforms which are CPU-only (Intel i5-4200M) and GPU-only (NVIDIA GTX-750Ti) to run the YOLOv2 so that the data can be obtained and compared with the FPGAs. The results show that the YOLOv2 and ResNet50 on FPGA have achieved low power consumption and high-power efficiency in this project. Besides, the accuracy of CNN on ZedBoard for digits recognition is satisfactory and the power consumption is very low. For the BNN, the time taken for hardware implementation to classify an image is faster than the software implementation when using the PYNQ-Z2.

Item Type: Final Year Project (Project Report)
Uncontrolled Keywords: Convolutional Neural Network, FPGA platform, YOLOv2, ResNet50, Power efficiency
Subjects: T Technology > T Technology (General)
T Technology > TA Engineering (General). Civil engineering (General)
Divisions: Library > Final Year Project > FKEKK
Depositing User: Norfaradilla Idayu Ab. Ghafar
Date Deposited: 03 Apr 2025 03:55
Last Modified: 03 Apr 2025 03:55
URI: http://digitalcollection.utem.edu.my/id/eprint/35274

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year