A 3D Point Cloud Classification Method Based on Adaptive Graph Convolution and Global Attention

Yaowei Yue,Xiaonan Li,Yun Peng

DOI: https://doi.org/10.3390/s24020617

IF: 3.9

2024-01-19

Sensors

Abstract:In recent years, there has been significant growth in the ubiquity and popularity of three-dimensional (3D) point clouds, with an increasing focus on the classification of 3D point clouds. To extract richer features from point clouds, many researchers have turned their attention to various point set regions and channels within irregular point clouds. However, this approach has limited capability in attending to crucial regions of interest in 3D point clouds and may overlook valuable information from neighboring features during feature aggregation. Therefore, this paper proposes a novel 3D point cloud classification method based on global attention and adaptive graph convolution (Att-AdaptNet). The method consists of two main branches: the first branch computes attention masks for each point, while the second branch employs adaptive graph convolution to extract global features from the point set. It dynamically learns features based on point interactions, generating adaptive kernels to effectively and precisely capture diverse relationships among points from different semantic parts. Experimental results demonstrate that the proposed model achieves 93.8% in overall accuracy and 90.8% in average accuracy on the ModeNet40 dataset.

engineering, electrical & electronic,chemistry, analytical,instruments & instrumentation

What problem does this paper attempt to address?

### Problems Addressed by the Paper

The paper aims to address key issues in 3D point cloud classification, particularly how to extract richer features from point clouds. Specifically, the researchers propose a new method based on global attention mechanism and adaptive graph convolution (Att-AdaptNet) to overcome the challenges faced by existing methods in classifying point clouds in complex environments.

#### Main Contributions
1. **Proposed a new 3D point cloud classification method**: The method named Att-AdaptNet combines global attention mechanism and adaptive graph convolution, which can directly handle raw point cloud data and focus on feature regions through global feature masks and adaptive graph convolution.
2. **Utilized adaptive graph convolution to extract global features**: This method dynamically generates convolution kernels to more accurately capture the relationships between different semantic parts.
3. **Experimental results show significant performance improvement**: On the widely used ModelNet40 benchmark dataset, Att-AdaptNet achieved a classification accuracy of 93.3%, showing significant improvement over other methods.

#### Core Issues
- Existing point cloud processing methods have limitations in handling irregular and sparse point cloud data, making it difficult to capture local geometric features and interactions between points.
- Conventional multi-layer perceptron (MLP) methods perform poorly in handling long-range dependencies.
- Traditional graph convolution methods use fixed convolution kernels, which cannot distinguish points in different semantic parts.

To address these issues, the researchers proposed Att-AdaptNet, a model that better captures features in point cloud data through adaptive convolution kernels and a global attention mechanism.