Scholarlink

<p>Gamma radiography (GR) is widely used across various industries to assess the quality of welded components. However, identifying and pinpointing flaws in Gamma Radiography Images (GRIs) remains challenging due to issues with contrast, noise, and visibility. This study seeks to enhance the identification and localization of welding defects in GRIs by developing a Deep Learning (DL) model named SKBOT3-YOLOv5. The study’s objectives include: first, applying preprocessing techniques to reduce noise and enhance contrast in GRIs; second, implementing data augmentation methods to expand the limited dataset and improve generalization; and third, enhancing the model by integrating the YOLOv5 architecture with advanced attention mechanisms, i.e., the Bottleneck Transformer (BOT3) and selective kernel (SK). The SKBOT3-YOLOv5 model was validated through comparative experiments against the original YOLOv5, YOLOv7, and the more recent YOLOv8. The model’s efficacy was further confirmed by testing various backbones and attention mechanisms within the YOLOv5 architecture. An ablation study was performed to emphasize the importance of critical parameters within the proposed network. The experimental findings demonstrate that the SKBOT3-YOLOv5 model surpasses other models, attaining a recall of 97.7%, precision of 94.1%, and detection accuracy of 97.6%. Consequently, our method significantly enhances automatic weld fault localization and detection in GRIs. The proposed SKBOT3-YOLOv5 model is available at: <a href="https://github.com/zeinabelsharkawy/SKBOT3-YOLOv5-for-weld-flaws-localization">https://github.com/zeinabelsharkawy/SKBOT3-YOLOv5-for-weld-flaws-localization</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Developing YOLOv5 for weld flaws detection and localization in gamma radiography images based on attention mechanisms

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Zeinab F. Elsharkawy,
  • H. Kasban,
  • Mohammed Y. Abbass

摘要

Gamma radiography (GR) is widely used across various industries to assess the quality of welded components. However, identifying and pinpointing flaws in Gamma Radiography Images (GRIs) remains challenging due to issues with contrast, noise, and visibility. This study seeks to enhance the identification and localization of welding defects in GRIs by developing a Deep Learning (DL) model named SKBOT3-YOLOv5. The study’s objectives include: first, applying preprocessing techniques to reduce noise and enhance contrast in GRIs; second, implementing data augmentation methods to expand the limited dataset and improve generalization; and third, enhancing the model by integrating the YOLOv5 architecture with advanced attention mechanisms, i.e., the Bottleneck Transformer (BOT3) and selective kernel (SK). The SKBOT3-YOLOv5 model was validated through comparative experiments against the original YOLOv5, YOLOv7, and the more recent YOLOv8. The model’s efficacy was further confirmed by testing various backbones and attention mechanisms within the YOLOv5 architecture. An ablation study was performed to emphasize the importance of critical parameters within the proposed network. The experimental findings demonstrate that the SKBOT3-YOLOv5 model surpasses other models, attaining a recall of 97.7%, precision of 94.1%, and detection accuracy of 97.6%. Consequently, our method significantly enhances automatic weld fault localization and detection in GRIs. The proposed SKBOT3-YOLOv5 model is available at: https://github.com/zeinabelsharkawy/SKBOT3-YOLOv5-for-weld-flaws-localization.