EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3030. Ensemble of YOLO Networks for Welding Detection

Invited abstract in session MD-4: Hybrid Appraches in Deep Learning and Machine Learning, stream Recent Advancements in AI .

Monday, 14:30-16:00
Room: 1001 (building: 202)

Authors (first author is the speaker)

1. Melisa Caliskan-Demir
Industrial Engineering, Istanbul Aydin University
2. Alev Taskin
Industrial Engineering, Yildiz Technical University

Abstract

The YOLO series of neural networks is widely recognized within the domain of target detection due to its robust feature extraction capabilities, streamlined network architecture, and efficient detection speed, thus rendering it a cornerstone in contemporary research and practical applications. Ensemble learning, a technique that combines multiple learning algorithms, is employed in the literature to increase predictive accuracy. In this study, ensemble learning is judiciously employed with state-of-the-art object detection algorithms, including YOLOv5, YOLOv7, YOLOv8 and MASK R-CNN. The overarching objective is to refine welding detection outcomes by leveraging an original post-welding dataset comprising images captured from seed-throwing machinery utilized in agricultural contexts. The principal aim is to discern any potential faults within the welding processes applied to the seeding legs of these instruments. The findings of this investigation underscore the superior efficacy of the proposed ensemble approach are compared with individual methodologies documented in extant literature. Although the YOLOv5 model manifests the highest individual performance metrics, the ensemble learning paradigm transcends these benchmarks, culminating in the most promising outcomes relative to the established performance criteria.

Keywords

Status: accepted

Back to the list of papers