Scholarlink

<p>To tackle challenges posed by large metal sheet areas, diverse defect types, and small, hard-to-identify defects, we propose an improved detection algorithm for lightweight metal-like surface based on YOLOv8. Firstly, we preprocess the original dataset to segment it into sizes suitable for YOLOv8 input dimensions, ensuring defect information integrity. Secondly, the LWBNet (Lightweight Backbone Networks) was designed to replace YOLOv8’s backbone network, achieving model lightweighting and enhancing its portability. Additionally, the ECSA (Efficient Channel Spatial Attention) module and the I-BiFPN (Improved Bidirectional Feature Pyramid Network) network structure were introduced to effectively extract and integrate defect features. Furthermore, the Wise-IoU loss function was adopted to enhance detection performance. Experimental results demonstrate that the model reduces parameters by 39.8%, computational load by 36.6%, and increases FPS by 6 frames per second compared to YOLOv8. Precision, Recall, mAP50 and mAP50-95 improve by 4.3, 2.5, 2.9 and 0.3%, respectively, validating its effectiveness for real-time industrial inspection.</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

An improved lightweight metal sheets surface defect detection algorithm based on YOLOv8

  • 期刊论文

  • 发表日期：

    2025年8月30日

查看全文
  • Yuechao Bian,
  • Haorong Wu,
  • Fuchun Sun,
  • Qiurong Lv,
  • Xiaoxiao Li

摘要

To tackle challenges posed by large metal sheet areas, diverse defect types, and small, hard-to-identify defects, we propose an improved detection algorithm for lightweight metal-like surface based on YOLOv8. Firstly, we preprocess the original dataset to segment it into sizes suitable for YOLOv8 input dimensions, ensuring defect information integrity. Secondly, the LWBNet (Lightweight Backbone Networks) was designed to replace YOLOv8’s backbone network, achieving model lightweighting and enhancing its portability. Additionally, the ECSA (Efficient Channel Spatial Attention) module and the I-BiFPN (Improved Bidirectional Feature Pyramid Network) network structure were introduced to effectively extract and integrate defect features. Furthermore, the Wise-IoU loss function was adopted to enhance detection performance. Experimental results demonstrate that the model reduces parameters by 39.8%, computational load by 36.6%, and increases FPS by 6 frames per second compared to YOLOv8. Precision, Recall, mAP50 and mAP50-95 improve by 4.3, 2.5, 2.9 and 0.3%, respectively, validating its effectiveness for real-time industrial inspection.