Scholarlink

The paper addresses the issue of tool segmentation and classification in an automated manufacturing environment by employing the One-Shot Learning technique. A novel architecture is proposed that integrates Siamese neural networks with a pretrained ConvNeXt model for feature extraction, enabling efficient classification tasks with minimal data input. The distinctive aspect of this approach is the employment of the LRASPP semantic segmentation technique, which enhances the precision of object identification in images. A comparative analysis of traditional and contemporary One-Shot Learning approaches is conducted, and the efficacy of the suggested architecture is validated through testing on a custom dataset comprising four distinct tool classes. Through experimentation, a classification accuracy rate of 96.87% is achieved, demonstrating the high efficacy and dependability of the proposed methodology. The findings demonstrate that the proposed system has the potential to enhance the efficiency and precision of operator monitoring in the production process, thereby reducing the likelihood of errors and defects.

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

A Combined Approach to the Classification and Semantic Segmentation of Production Tools Through the Use of the One-Shot Learning Method

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • D. M. Grabar,
  • S. V. Zhiganov,
  • Y. S. Ivanov

摘要

The paper addresses the issue of tool segmentation and classification in an automated manufacturing environment by employing the One-Shot Learning technique. A novel architecture is proposed that integrates Siamese neural networks with a pretrained ConvNeXt model for feature extraction, enabling efficient classification tasks with minimal data input. The distinctive aspect of this approach is the employment of the LRASPP semantic segmentation technique, which enhances the precision of object identification in images. A comparative analysis of traditional and contemporary One-Shot Learning approaches is conducted, and the efficacy of the suggested architecture is validated through testing on a custom dataset comprising four distinct tool classes. Through experimentation, a classification accuracy rate of 96.87% is achieved, demonstrating the high efficacy and dependability of the proposed methodology. The findings demonstrate that the proposed system has the potential to enhance the efficiency and precision of operator monitoring in the production process, thereby reducing the likelihood of errors and defects.