Year: 2025
Type: conference-paper
Abstract: This paper investigates the development of a lightweight convolutional neural network (LCNN) based on group convolution, with the objective of addressing the issue of excessive parameters and computational demands in traditional convolutional neural network(CNN), which render them unsuitable for mobile and embedded device applications. The study begins by presenting an overview of the fundamental architecture of CNN and conducting an in-depth analysis of conventional convolution operation and pooling layers. It then delves into the enhanced method of group convolution, applying this technique to MobileNetV3 through experimentation. The findings demonstrate that under identical conditions, the proposed LCNN effectively reduces both parameter count and computational load while preserving high accuracy.
Authors: Song Wang, Qiong Wang
Institution: Wuhan Donghu University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4413845301