Scholarlink

The study presents the development of an algorithm for detecting the blades of a foaming flotation machine using a neural network. Based on the carried-out analysis of the current object detection approaches such as R-CNN, Faster R-CNN and YOLO, taking into account their feasibility to the task of real-time video stream processing, a configuration of YOLO-11s was chosen to provide a balance between speed and detection rate. Training was performed on a dataset marked up consisting of 410 images for training and 88 images for testing, using data augmentation and CVAT annotation instruments. The quality metrics of the mAP50 and mAP50-95 models were equal to 0.96 and 0.75, respectively that confirmed advantageous object identification. The developed algorithm is integrated into video stream processing software for flotation machines. A comparative evaluation with the pre-existing algorithm was performed and showed 5.1% reduced number of runs of the foaming machine to skip.

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

Using Neural Networks to Improve Quality of Flotation Parameter Detection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Andrey Zatonskiy,
  • Stanislav Kuznetsov,
  • Kristina Salomatova,
  • Tatiana Gorbunova,
  • Ruslan Bazhenov

摘要

The study presents the development of an algorithm for detecting the blades of a foaming flotation machine using a neural network. Based on the carried-out analysis of the current object detection approaches such as R-CNN, Faster R-CNN and YOLO, taking into account their feasibility to the task of real-time video stream processing, a configuration of YOLO-11s was chosen to provide a balance between speed and detection rate. Training was performed on a dataset marked up consisting of 410 images for training and 88 images for testing, using data augmentation and CVAT annotation instruments. The quality metrics of the mAP50 and mAP50-95 models were equal to 0.96 and 0.75, respectively that confirmed advantageous object identification. The developed algorithm is integrated into video stream processing software for flotation machines. A comparative evaluation with the pre-existing algorithm was performed and showed 5.1% reduced number of runs of the foaming machine to skip.