Scholarlink

<p>Noisy point cloud data pose a major obstacle to the accurate inspection and modeling of high-speed railway track fasteners, often leading to degraded structural interpretation and unreliable automation. To address this issue, we propose an efficient PointNet-Based Multifaceted Autoencoder (PMAE) specifically designed for denoising such data. Our approach integrates neural network innovations, including a multimodal fusion module and a coordinate augmentation block, to enhance denoising effectiveness. Extensive experiments on a proprietary point cloud dataset reveal that PMAE achieves a 15% improvement in Chamfer Distance (CD) and a 20% reduction in Mean Squared Error (MSE) compared to traditional baselines. Additionally, this model preserves over 90% of critical structural features, improves feature localization by 18%, and reduces overfitting by 12%, thanks to architectural components like Coordinate Convolution Block and Orthogonal Regularizer. These results demonstrate that PMAE offers a robust and practical solution for point cloud denoising in railway infrastructure inspection and 3D modeling applications.</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

An efficient PointNet-based multifaceted autoencoder (PMAE) for denoising rail track fastener point clouds

  • 期刊论文

  • 发表日期：

    2025年9月25日

查看全文
  • Qasim Zaheer,
  • Jin Wang,
  • S. Muhammad Ahmed Hassan Shah,
  • Zunaira Atta,
  • Weidong Wang,
  • Momina Malik,
  • Chengbo Ai,
  • Shi Qiu

摘要

Noisy point cloud data pose a major obstacle to the accurate inspection and modeling of high-speed railway track fasteners, often leading to degraded structural interpretation and unreliable automation. To address this issue, we propose an efficient PointNet-Based Multifaceted Autoencoder (PMAE) specifically designed for denoising such data. Our approach integrates neural network innovations, including a multimodal fusion module and a coordinate augmentation block, to enhance denoising effectiveness. Extensive experiments on a proprietary point cloud dataset reveal that PMAE achieves a 15% improvement in Chamfer Distance (CD) and a 20% reduction in Mean Squared Error (MSE) compared to traditional baselines. Additionally, this model preserves over 90% of critical structural features, improves feature localization by 18%, and reduces overfitting by 12%, thanks to architectural components like Coordinate Convolution Block and Orthogonal Regularizer. These results demonstrate that PMAE offers a robust and practical solution for point cloud denoising in railway infrastructure inspection and 3D modeling applications.