Luyang Wang

Luyang Wang

Auto-generated profile, based on publicly available information

Luyang Wang

Robot and Intelligent Equipment Center, Guangzhou Institute of Advanced Technology, China and College of Mechanical and Electrical Engineering, Shaanxi University of Science and Technology, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 August 2024
Verified
Conference Paper
Open access
LWLuyang Wang
Luyang Wang
GZGongxue Zhang
Gongxue Zhang
WWWeijun Wang
Weijun Wang
ZHZucheng Huang
Zucheng Huang
JWJian Wang
Jian Wang
Due to imperfect manufacturing processes and external factors, aluminum profiles can exhibit various surface defects, which significantly affect their service life. To address the issues of low accuracy and high computational load in current aluminum profile surface defect detection methods, this paper proposes a high-precision and lightweight aluminum profile surface defect detection model, YOLOv8-ARL. The model first replaces some Conv modules in the backbone and neck with an Adown down-sampling module, which reduces the number of parameters while retaining more image information. Next, a lightweight module, RGCSPELAN (RepGhostCSPELAN), is introduced to replace the C2f module in the network, enhancing gradient flow information while improving computational resource utilization. Finally, a lightweight detection head is constructed using the concept of shared convolution, replacing the BN (Batch Normalization) layers in the Conv modules to ensure that lightweight design does not compromise accuracy. The improved YOLOv8-ARL algorithm achieves an average mean precision (mAP50%) of 88.6% on the APDDD aluminum profile dataset, a 2.2% increase over the original YOLOv8s model, with a parameter count of 4.55M, reduced by 59% compared to the original model, and a computational load reduced by 48%. The detection time per image is 2.2ms. Experimental results demonstrate that the improved YOLOv8-ARL model maintains excellent detection accuracy and speed while being lightweight.