21 / 2023-03-27 08:49:16

Vegetation extraction from tilt photogrammetry point clouds based on Transformer

Tilt Photogrammetric point cloud; Transformer; self-attention, Vegetation Extraction; Semantic Segmentation

Abstract: The application of deep learning, which has self-feature learning advantages, to point cloud classification can currently be mainly divided into three categories: multi-view, volex-based, and 3D point cloud classification methods. However, in the field of oblique photogrammetry, due to different research ideas and a lack of publicly available oblique photogrammetric point cloud datasets, research on oblique photogrammetric point cloud classification has not received much attention. In this paper, we propose an oblique photogrammetric point cloud dataset that satisfies the semantic segmentation tasks, and use deep learning methods to perform point cloud semantic segmentation tasks. Additionally, inspired by the significant achievements of self-attention mechanism in natural language processing and the promsing progress in visual fields such as object detection, we investigate the application of self-attention networks in oblique photogrammetric point cloud classification tasks. We develop a point cloud classification method for oblique photogrammetric point clouds using self-attention networks, and design an RDpoint Transformer model based on self-attention networks that can be used to perform oblique photogrammetric point cloud semantic segmentation tasks. Through extensive comparative experiments, we demonstrate the effectiveness of the self-attention network point cloud classification method for oblique photogrammetric point clouds.