Scholarlink

<p>Strip steel is widely used as an important material in various fields, but it is easy to cause surface damage due to various factors in production. Although the existing model takes into account the accuracy and parameters, the number of parameters is still huge. From the perspective of reference quantity, this paper proposes an ultra-lightweight model called Edge-FastestDet for monitoring strip surface defects. Based on ShuffleNetV2 architecture, the model significantly improves feature fusion capability and defect detection accuracy by introducing SiLU activation function and “split-transform-splice" structure. In addition, a lightweight spatial pyramid pool module (LSPP) and dynamic enhanced detection head are proposed to further reduce the number of parameters and improve feature discrimination. The results show that the mAP50 of Edge-FastestDet on the NEU-DET and GC10 datasets is 49.0% and 33.4% respectively, the FPS is 148.2 and 163.2 respectively, and the number of model parameters is only 0.29M. The model successfully detects defects in a simulated industrial environment. This model has very low computing cost and high real-time performance, which can provide a reference for resource-limited industrial scenarios. The code can be found at <a href="https://github.com/01WineCool/Strong-FastestDet">https://github.com/01WineCool/Strong-FastestDet</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

Edge-FastestDet: a real-time and ultra-lightweight model for detection

  • 期刊论文

  • 发表日期：

    2025年7月24日

查看全文
  • Meishun Wu,
  • Jinmin Peng,
  • Xinyi Yu,
  • Heng Xu,
  • Liulu Zhang,
  • Chaoqi Jiang,
  • Wenkai Dong,
  • Liangshen Chen

摘要

Strip steel is widely used as an important material in various fields, but it is easy to cause surface damage due to various factors in production. Although the existing model takes into account the accuracy and parameters, the number of parameters is still huge. From the perspective of reference quantity, this paper proposes an ultra-lightweight model called Edge-FastestDet for monitoring strip surface defects. Based on ShuffleNetV2 architecture, the model significantly improves feature fusion capability and defect detection accuracy by introducing SiLU activation function and “split-transform-splice" structure. In addition, a lightweight spatial pyramid pool module (LSPP) and dynamic enhanced detection head are proposed to further reduce the number of parameters and improve feature discrimination. The results show that the mAP50 of Edge-FastestDet on the NEU-DET and GC10 datasets is 49.0% and 33.4% respectively, the FPS is 148.2 and 163.2 respectively, and the number of model parameters is only 0.29M. The model successfully detects defects in a simulated industrial environment. This model has very low computing cost and high real-time performance, which can provide a reference for resource-limited industrial scenarios. The code can be found at https://github.com/01WineCool/Strong-FastestDet.