Scholarlink

Abstract <p>&#xa0;In recent times, the creation of 3D content has exponentially increased and the amount of information in the form of 3D models available to the public is also increasing. In this paper, we propose a hybrid deep convolutional neural network (DCNN) and support vector machines (SVM) to classify 3D objects. First, we discretize the 3D object model as binary voxel data to improve the learning phase of 3D complex features. Second, we employ a deep neural network architecture consisting of several layers to generate a feature map. The obtained feature map plays a significant role in preserving the overall geometrical structure of the object. As a feature learning model, we employ a nonlinear support vector machine (SVM) with a radial basis function (RBF) kernel to enhance the learning efficiency of the 3D deep feature map, thereby improving prediction performance. We use the ModelNet10 standard large-scale 3D CAD dataset to evaluate the proposed approach. The experimental evaluation shows that the proposed approach has higher performance compared to other approaches in terms of accuracy and precision.</p> Graphic Abstract <p></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

A novel hybrid DCNN–SVM method for 3D object classification

  • 期刊论文

  • 发表日期：

    2025年7月16日

查看全文
  • Mohamed Othmani,
  • Brahim Issaoui,
  • Salim El Khediri,
  • Rehan Ullah Khan

摘要

Abstract

 In recent times, the creation of 3D content has exponentially increased and the amount of information in the form of 3D models available to the public is also increasing. In this paper, we propose a hybrid deep convolutional neural network (DCNN) and support vector machines (SVM) to classify 3D objects. First, we discretize the 3D object model as binary voxel data to improve the learning phase of 3D complex features. Second, we employ a deep neural network architecture consisting of several layers to generate a feature map. The obtained feature map plays a significant role in preserving the overall geometrical structure of the object. As a feature learning model, we employ a nonlinear support vector machine (SVM) with a radial basis function (RBF) kernel to enhance the learning efficiency of the 3D deep feature map, thereby improving prediction performance. We use the ModelNet10 standard large-scale 3D CAD dataset to evaluate the proposed approach. The experimental evaluation shows that the proposed approach has higher performance compared to other approaches in terms of accuracy and precision.

Graphic Abstract