Journal Article

A novel efficient method for welding spots detection

Li, Eric, Ma, Yadong, Wang, Zhenxing, He, Zhicheng

2022

Overview

The traditional welding spots detection methods are very sensitive to light pollution, rust and oil stain. In this work, a novel method that combines light-weight deep learning model and circle detection algorithm for welding spots detection is proposed. First, a light-weight model called Modified You Only Look Once version 2 (M-YOLOv2) which substitutes MobileNetv2 for Darknet-19 convolution structure in YOLOv2 and incorporates fine-grained features (FGF) is developed. The proposed M-YOLOv2 is very effective and efficient to capture the position of welding spot compared with YOLOv2, YOLOv3 and Tiny-YOLOv2. Secondly, a novel right-triangle circle detection (RTCD) is developed to refine the shapes of the welding spots. The novel algorithm called M-YOLOv2-RTCD that combines M-YOLOv2 and RTCD is able to accurately and fast detect welding spots. The theoretical analysis and experimental results demonstrate that the proposed M-YOLOv2-RTCD outperforms the methods only using M-YOLOv2 or RTCD, in terms of accuracy and real-time performance.

• Publisher: Springer US,Springer Nature B.V
• Subject: Accuracy / Algorithms / Computer Communication Networks / Computer Science / Data Structures and Information Theory / Datasets / Deep learning / Efficiency / Light pollution / Machine learning / Methods / Multimedia / Multimedia Information Systems / Neural networks / Optimization / Special Purpose and Application-Based Systems / Spots / Triangles / Welding