Scholarlink

<p>Surface defect detection on aluminum profiles (APs) is an important task in the APs production process. However, existing detection methods for surface defects on APs still have room for improvement in detecting defects with low contrast to the background and small defects. This paper proposes an effective model, named DSLE-YOLO, based on YOLOv8 to detect surface defects on APs. Firstly, we introduce the Deformable Convolution-C2f (DC-C2f) module in the backbone, strengthening multi-scale feature extraction ability of the model. Secondly, the ScConv Spatial Pyramid Pooling Fast (SCSPPF) module is proposed to replace the original SPPF module in the backbone of YOLOv8. The SCSPPF module boosts local feature extraction and preserves spatial details, significantly improving the detection accuracy for defects with low contrast to the background. Finally, we design the Fusion Attention Mechanism (FAM) module to capture fine-grained information of small defects and multi-scale spatial information of surface defects with large scale variation. We validate our model on two public APs surface defects datasets. Experimental results show that the DSLE-YOLO has superior detection performance compared with other comparison models.</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

Dsle-yolo: an effective method for detecting surface defects on aluminum profiles

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Meilin Wang,
  • Haichao Wang,
  • Xiaojing Zhu

摘要

Surface defect detection on aluminum profiles (APs) is an important task in the APs production process. However, existing detection methods for surface defects on APs still have room for improvement in detecting defects with low contrast to the background and small defects. This paper proposes an effective model, named DSLE-YOLO, based on YOLOv8 to detect surface defects on APs. Firstly, we introduce the Deformable Convolution-C2f (DC-C2f) module in the backbone, strengthening multi-scale feature extraction ability of the model. Secondly, the ScConv Spatial Pyramid Pooling Fast (SCSPPF) module is proposed to replace the original SPPF module in the backbone of YOLOv8. The SCSPPF module boosts local feature extraction and preserves spatial details, significantly improving the detection accuracy for defects with low contrast to the background. Finally, we design the Fusion Attention Mechanism (FAM) module to capture fine-grained information of small defects and multi-scale spatial information of surface defects with large scale variation. We validate our model on two public APs surface defects datasets. Experimental results show that the DSLE-YOLO has superior detection performance compared with other comparison models.