Scholarlink

The rapid growth of 3D object databases in mechanical engineering leads to the need for efficient shape-based clustering and retrieval methods. Accurate and fast clustering of 3D mechanical objects is vital for design reuse, similarity-based search, and optimizing manufacturing processes. This paper introduces a Multi-View Multi-Slice Convolutional Neural Network (MVMS-CNN) with semi-supervised learning for efficient 3D object clustering and shape matching. Our method automatically extracts complex features from multiple views and slices, enhancing accuracy and efficiency in shape retrieval for large databases. The MVMS-CNN captures both external and internal features of 3D objects. When combined with K-Means clustering, the MVMS-CNN demonstrates better purity and clustering accuracy than existing methods. The semisupervised learning approach also addresses the challenge of varied and unknown class numbers, making it well-suited for real-world scenarios involving large and diverse datasets. This combination allows for precise clustering, benefiting numerous applications in mechanical engineering and manufacturing.

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

3D Mechanical Object Clustering for Shape Matching Using Multi-view Multi-slice CNN Architecture

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Dipesh Shrestha,
  • Arpan Man Sainju,
  • Abigail Kelly

摘要

The rapid growth of 3D object databases in mechanical engineering leads to the need for efficient shape-based clustering and retrieval methods. Accurate and fast clustering of 3D mechanical objects is vital for design reuse, similarity-based search, and optimizing manufacturing processes. This paper introduces a Multi-View Multi-Slice Convolutional Neural Network (MVMS-CNN) with semi-supervised learning for efficient 3D object clustering and shape matching. Our method automatically extracts complex features from multiple views and slices, enhancing accuracy and efficiency in shape retrieval for large databases. The MVMS-CNN captures both external and internal features of 3D objects. When combined with K-Means clustering, the MVMS-CNN demonstrates better purity and clustering accuracy than existing methods. The semisupervised learning approach also addresses the challenge of varied and unknown class numbers, making it well-suited for real-world scenarios involving large and diverse datasets. This combination allows for precise clustering, benefiting numerous applications in mechanical engineering and manufacturing.