Image Classification using Parallel CPU and GPU Computing
Prathamesh Borhade1, Rajvardhan Deshmukh2, Samridhi Murarka3, Rishav Agarwal4
1Prathamesh Borhade*, Computer science and engineering, Vellore Institute of Technology, Vellore, India.
2Rajvardhan Deshmukh, Computer science and engineering, Vellore Institute of Technology, Vellore, India.
3Rishav Agarwal, Computer science and engineering, Vellore Institute of Technology, Vellore, India.
4Samridhi Murarka, Computer science and engineering, Vellore Institute of Technology, Vellore, India.
Manuscript received on March 29, 2020. | Revised Manuscript received on April 25, 2020. | Manuscript published on April 30, 2020. | PP: 839-843 | Volume-9 Issue-4, April 2020. | Retrieval Number: D7870049420/2020©BEIESP | DOI: 10.35940/ijeat.D7870.049420
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Image classification algorithms such as Convolutional Neural Network used for classifying huge image datasets takes a lot of time to perform convolution operations, thus increasing the computational demand of image processing. Compared to CPU, Graphics Processing Unit (GPU) is a good way to accelerate the processing of the images. Parallelizing multiple CPU cores is also another way to process the images faster. Increasing the system memory (RAM) can also decrease the computational time of image processing. Comparing the architecture of CPU and GPU, the former consists of a few cores optimized for sequential processing whereas the later has thousands of relatively simple cores clocked at approx. 1Ghz. The aim of this project is to compare the performance of parallelized CPUs and a GPU. Python’s Ray library is being used to parallelize multicore CPUs. The benchmark image classification algorithm used in this project is Convolutional Neural Network. The dataset used in this project is Plant Disease Image Dataset. Our results show that the GPU implementation achieves 80% speedup compared to the CPU implementation.
Keywords: Convolutional Neural Network, parallel computing, speedup.