Point cloud classification segmentation network based on point cloud multi-scale parallel feature extraction and attention mechanism

The invention discloses a point cloud classification segmentation network Paraller-Net based on point cloud parallel multi-scale feature extraction and an attention mechanism in order to solve the defects in the prior art, particularly relates to a point cloud classification segmentation network bas...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: CHEN JINYAN, WANG CHENGXI, HU WENJIE, LIU ZHENZE, SUN JI, ZANG YIFAN
Format: Patent
Sprache:chi ; eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The invention discloses a point cloud classification segmentation network Paraller-Net based on point cloud parallel multi-scale feature extraction and an attention mechanism in order to solve the defects in the prior art, particularly relates to a point cloud classification segmentation network based on point cloud parallel multi-scale feature extraction and an attention mechanism, and aims to improve the accuracy. The method adopts a parallel multi-scale feature extraction and cross attention mechanism, and comprises the following steps: 1) inputting point cloud data, and carrying out feature learning; 2) performing feature extraction and sampling through a downsampling algorithm; 3) introducing a self-attention mechanism to calculate position correlation; 4) using cross attention to process a plurality of point cloud features after down-sampling; and 5) transmitting feature information by applying an up-sampling algorithm. According to the method, 3D point cloud features can be effectively extracted, and t