Title: Real-time detector on SMT chip component defect detection

Authors: Wangyou Gui; Zhenyi Xu; Yubo Zhao; Lijun Zhao

Addresses: Jianghuai Advance Technology Center, Hefei, 230000, China ' Key Laboratory of System Control and Information Processing, Ministry of Education, Shanghai, 200240, China; Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, 230088, China ' Department of Automation, University of Science and Technology of China, Hefei, 230026, China ' Yangtze River Delta Hart robot industry Technology Research Institute, Wuhu, 241060, China

Abstract: Defect detection in surface mount technology (SMT) is critical for electronics quality control. Traditional automated optical inspection (AOI) faces slow speed and high misdetection, while existing deep learning methods trade accuracy (one-stage) for real-time performance (two-stage). We propose SMT-DETR, a lightweight real-time algorithm for SMT defect detection. First, an adaptive collaborative feature extraction module (ACBlock) enhances sensitivity to component deformations and defects. Second, a novel IMIoU loss improves tiny object detection and accelerates convergence by optimising IoU computation. We also construct the SMT chip component dataset (SMTC). Experiments show SMT-DETR outperforms state-of-the-art methods on SMTC and achieves competitive results on public datasets, balancing accuracy and real-time efficiency for industrial applications.

Keywords: SMT defect detection; IMIoU Loss; adaptive collaborative feature.

DOI: 10.1504/IJSCIP.2025.150923

International Journal of System Control and Information Processing, 2025 Vol.4 No.4, pp.323 - 339

Received: 27 Dec 2023
Accepted: 13 Aug 2024
Published online: 05 Jan 2026 *

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