Scholarlink

The transition from traditional industrial production to intelligent manufacturing has become a prevailing trend. Artificial Intelligence (AI) serves as a crucial foundation for achieving intelligent manufacturing. One significant challenge that still persists in the industrial sector is the detection and counting of densely packed workpieces. Machine vision algorithms are essential for applying AI to industrial inspection. However, most current mainstream algorithms are designed for general scenarios, and their performance significantly declines when applied in industrial environments. This decline is due to the strong interference from the background in industrial settings, which is a common characteristic of factories due to their complexity. To address this issue, we propose constructing a workpiece counting sketch dataset (WCSD) designed to reduce background noise. This dataset aims to minimize the interference caused by the background during the training of vision models, thereby enhancing the semantic features of the workpieces. The dataset offers a vast array of usable class templates and consistent visual annotations. It comprises 838 images and 91,724 objects across 277 categories.

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

Performance Evaluation of Workpiece Counting Neural Network with a Novel Sketch Dataset for Reducing Background Noise

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chengxuan Wang,
  • Chaojun Dong,
  • Ye Li,
  • Xiankun Liu,
  • Jianhong Zhou,
  • Yikui Zhai

摘要

The transition from traditional industrial production to intelligent manufacturing has become a prevailing trend. Artificial Intelligence (AI) serves as a crucial foundation for achieving intelligent manufacturing. One significant challenge that still persists in the industrial sector is the detection and counting of densely packed workpieces. Machine vision algorithms are essential for applying AI to industrial inspection. However, most current mainstream algorithms are designed for general scenarios, and their performance significantly declines when applied in industrial environments. This decline is due to the strong interference from the background in industrial settings, which is a common characteristic of factories due to their complexity. To address this issue, we propose constructing a workpiece counting sketch dataset (WCSD) designed to reduce background noise. This dataset aims to minimize the interference caused by the background during the training of vision models, thereby enhancing the semantic features of the workpieces. The dataset offers a vast array of usable class templates and consistent visual annotations. It comprises 838 images and 91,724 objects across 277 categories.