Scholarlink

<p>Aluminum castings can be prone to surface defects due to manufacturing process limitations and external factors, significantly restricting their performance and service life. To address the limitations of traditional detection algorithms in terms of accuracy and speed, this paper presents an efficient surface defect detection model for aluminum castings, named OHSW-YOLO, based on YOLOv8n. The model incorporates Omni-dimensional Dynamic Convolution(ODConv) to enhance performance, while reducing computational requirements. It integrates components of the HorNet architecture to improve feature extraction, introduces the SEAttention module to strengthen target feature representation, and employs a dynamic focus mechanism along with the Wise-IoU loss function to optimize object detection performance. Experimental results on a self-constructed aluminum casting defect dataset demonstrate that OHSW-YOLO achieves a mAP50 of 93.9%, representing a 1.5% improvement compared to the original model. Additionally, it achieves an inference speed of 115 FPS, an improvement of 24 FPS. This study highlights that OHSW-YOLO significantly enhances real-time performance, while maintaining high detection accuracy, providing an efficient solution for aluminum casting surface defect detection.</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

OHSW-YOLO: An Aluminum Casting Surface Defects Detection Model

  • 期刊论文

  • 发表日期：

    2025年4月25日

查看全文
  • Chengjun Wang,
  • Chengwei Ye

摘要

Aluminum castings can be prone to surface defects due to manufacturing process limitations and external factors, significantly restricting their performance and service life. To address the limitations of traditional detection algorithms in terms of accuracy and speed, this paper presents an efficient surface defect detection model for aluminum castings, named OHSW-YOLO, based on YOLOv8n. The model incorporates Omni-dimensional Dynamic Convolution(ODConv) to enhance performance, while reducing computational requirements. It integrates components of the HorNet architecture to improve feature extraction, introduces the SEAttention module to strengthen target feature representation, and employs a dynamic focus mechanism along with the Wise-IoU loss function to optimize object detection performance. Experimental results on a self-constructed aluminum casting defect dataset demonstrate that OHSW-YOLO achieves a mAP50 of 93.9%, representing a 1.5% improvement compared to the original model. Additionally, it achieves an inference speed of 115 FPS, an improvement of 24 FPS. This study highlights that OHSW-YOLO significantly enhances real-time performance, while maintaining high detection accuracy, providing an efficient solution for aluminum casting surface defect detection.