Scholarlink

<p>The recycling and reuse of copper and aluminum from end-of-life vehicles (ELVs) present both significant economic and environmental opportunities. However, substantial differences in the amounts of copper and aluminum used in vehicles have led to an imbalance in the quantity of small scrap samples. Effectively detecting and sorting non-ferrous scraps from ELVs remains an urgent challenge. To address these challenges, this paper proposes an improved YOLOv8 detection algorithm, integrated with a small sample enhancement method for non-ferrous scrap. Sorting software was developed for conducting the experiments. The results demonstrate that efficient image fusion can be achieved by generating targeted images for underrepresented classes and carefully designing the fusion position, size, and method. The improved YOLOv8 detection model trained on the augmented dataset achieved optimal performance, with recognition accuracies of 98.5% for aluminum scrap and 98.1% for copper scrap, and a detection speed of 50 FPS, meeting real-time detection requirements. Furthermore, the developed system and software yielded sorting accuracy of 93.3% and purity of 92.6% during the scraps sorting experiments, both exceeding the industry standard of 90%. The proposed method of recovering these metals through sorting and recycling is not only environmentally friendly and energy-efficient but also offers substantial economic advantages.</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

Real-time high-accuracy sorting of imbalanced non-ferrous scraps in ELVs: a deep learning approach with small-sample optimization

  • 期刊论文

  • 发表日期：

    2025年10月29日

查看全文
  • Rui Wang,
  • Qiu Pang,
  • Zhili Hu,
  • Lin Hua

摘要

The recycling and reuse of copper and aluminum from end-of-life vehicles (ELVs) present both significant economic and environmental opportunities. However, substantial differences in the amounts of copper and aluminum used in vehicles have led to an imbalance in the quantity of small scrap samples. Effectively detecting and sorting non-ferrous scraps from ELVs remains an urgent challenge. To address these challenges, this paper proposes an improved YOLOv8 detection algorithm, integrated with a small sample enhancement method for non-ferrous scrap. Sorting software was developed for conducting the experiments. The results demonstrate that efficient image fusion can be achieved by generating targeted images for underrepresented classes and carefully designing the fusion position, size, and method. The improved YOLOv8 detection model trained on the augmented dataset achieved optimal performance, with recognition accuracies of 98.5% for aluminum scrap and 98.1% for copper scrap, and a detection speed of 50 FPS, meeting real-time detection requirements. Furthermore, the developed system and software yielded sorting accuracy of 93.3% and purity of 92.6% during the scraps sorting experiments, both exceeding the industry standard of 90%. The proposed method of recovering these metals through sorting and recycling is not only environmentally friendly and energy-efficient but also offers substantial economic advantages.