YOLOv7-SiamFF: Industrial defect detection algorithm based on improved YOLOv7

The task of accurately classifying defect types and pinpointing their locations in the domain of industrial product defect detection remains a formidable challenge. This paper introduces an advanced industrial defect detection framework, named YOLOv7-SiamFF, which utilizes the YOLOv7 as a feature ex...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computers & electrical engineering 2024-03, Vol.114, p.109090, Article 109090
Hauptverfasser: Yi, Feifan, Zhang, Haigang, Yang, Jinfeng, He, Liming, Mohamed, Ahmad Sufril Azlan, Gao, Shan
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The task of accurately classifying defect types and pinpointing their locations in the domain of industrial product defect detection remains a formidable challenge. This paper introduces an advanced industrial defect detection framework, named YOLOv7-SiamFF, which utilizes the YOLOv7 as a feature extraction and detection backbone with three feature reinforcement modules. Firstly, we employ a parallel Siamese network, facilitating differential feature extraction through dual-stream feature extraction channels, aimed at better highlighting defect features and suppressing background interference. Additionally, we introduce a depth information feature fusion module, which effectively integrates high and low-level features in the Siamese network, thus enhancing the model’s detection accuracy for small target defects. Finally, an attention mechanism is integrated into the feature extraction network, further enhancing the model’s precision in identifying defect-specific features. In the simulation experiment, a specialized visual dataset was created for object detection tasks focusing on industrial defects, dubbed the BC-DD dataset. Additionally, the effectiveness of the proposed model has been validated in this paper using the aforementioned dataset.
ISSN:0045-7906
1879-0755
DOI:10.1016/j.compeleceng.2024.109090