Title: An end-to-end instance segmentation method based on improved ConvNeXt V2

Authors: Wenlu Wang; Ying Sun; Manman Xu; Dongxu Bai; Li Huang; Chunlong Zou; Baojia Chen; Dalai Tang

Addresses: Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan, Hubei, China; Research Centre for Biomimetic Robot and Intelligent Measurement and Control, Wuhan University of Science and Technology, Wuhan, Hubei China ' Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan, Hubei, China; Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan University of Science and Technology, Wuhan, Hubei, China ' Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan, Hubei, China; Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan University of Science and Technology, Wuhan, Hubei, China ' Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan, Hubei, China; Research Centre for Biomimetic Robot and Intelligent Measurement and Control, Wuhan University of Science and Technology, Wuhan, Hubei China ' College of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan, Hubei, China; Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System, Wuhan University of Science and Technology, Wuhan, Hubei, China ' College of Mechanical Engineering, Hubei University of Automotive Technology, Shiyan, Hubei, China ' Hubei Key Laboratory of Hydroelectric Machinery Design & Maintenance, China Three Gorges University, Yichang, Hubei, China ' School of Computer Information Management, Inner Mongolia University of Finance and Economics, Hohhot, China

Abstract: In order to improve the efficiency of indoor mobile robots in locating and segmenting environmental instances, an instance segmentation method based on RTMDet is proposed. Firstly, the more powerful ConvNeXt V2 is used as the backbone network of the model, which improves the performance of the existing RTMDet model. Subsequently, NAS-FPN is used as a converged network to detect the network at any given time. Finally, AdamW is used as the model optimiser, which effectively solves the problem of excessive memory occupation when updating parameters and improves the performance of the model. The proposed method obtains the highest evaluation index in all cases. After training and testing on the Cityscapes data set, the recognition accuracy reaches 65.0%.

Keywords: instance segmentation; ConvNeXt V2; deep learning; optimiser; feature fusion.

DOI: 10.1504/IJWMC.2024.138852

International Journal of Wireless and Mobile Computing, 2024 Vol.26 No.4, pp.384 - 392

Received: 20 Sep 2023
Accepted: 31 Dec 2023
Published online: 31 May 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article