An Algorithm for Point Cloud Object Classification Combining Bidirectional Attention Mechanism and Edge Convolution

Abstract

Aiming at the problem that existing deep learning point cloud object classification algorithms do not adequately mine the global contextual information resulting in low classification accuracy, this paper proposes a point cloud object classification algorithm that combines the bidirectional attention mechanism and edge convolution. First, the contextual information and local information of the point cloud are extracted separately using a single layer of bidirectional attention mechanism and edge convolution; then, the two parts of the information are fused and then passed to the next layer for feature extraction, and the features extracted from each layer are integrated into the global features, thus enhancing the capture of contextual information. With the help of the dataset ModelNet40, the overall classification accuracy of this paper's algorithm reaches 92.5% and the mean accuracy reaches 89.8%. Experimental results show that the algorithm in this paper performs better than other point cloud object classification algorithms in terms of classification performance and is more robust.