Title: Identifying assembly features within threaded components using intelligent detection with YOLOv5s

Authors: Zhuonan Yu; Tao Liu; Zhaofeng Chen; Ziyan Zhang; Chunlin Tian

Addresses: College of Mechanical and Electrical Engineering, Changchun University of Science and Technology, Changchun, 130000, China ' College of Mechanical and Electrical Engineering, Changchun University of Science and Technology, Changchun, 130000, China ' College of Mechanical and Electrical Engineering, Changchun University of Science and Technology, Changchun, 130000, China ' College of Mechanical and Electrical Engineering, Changchun University of Science and Technology, Changchun, 130000, China ' College of Mechanical and Electrical Engineering, Changchun University of Science and Technology, Changchun, 130000, China

Abstract: Addressing the challenge of localising threaded components during the assembly process, we propose a threaded component detection model called YOLO-DH, based on an enhanced version of YOLOv5s. To achieve lightweighting and optimise efficiency, we introduce the Ghost module in the YOLOv5s backbone network and integrate the CBAM attention mechanism for enhanced feature extraction. This improved model serves as a student model for knowledge distillation, boosting target detection accuracy. Rigorous experiments on the same dataset show notable improvements: 2.3% increase in detection accuracy (mAP), 19.4% reduction in the number of parameters (Params), 35.2% reduction in the number of floating-point operations (Flops), and 0.2 ms decrease in average detection time compared to Yolov5s. These enhancements maintain higher detection accuracy while achieving model lightweighting. The resulting model is suitable for intelligent recognition in the assembly features of threaded components.

Keywords: YOLOv5; threaded component; knowledge distillation; lightweight model; CBAM attention mechanism; deep learning; object detection.

DOI: 10.1504/IJNM.2024.144315

International Journal of Nanomanufacturing, 2024 Vol.19 No.1, pp.23 - 38

Received: 05 Feb 2024
Accepted: 26 Mar 2024

Published online: 06 Feb 2025 *

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