Oblique photography point cloud classification method based on multi-feature integration deep learning model

Abstract

The invention discloses an oblique photographic point cloud classification method based on a multi-feature integrated deep learning model in the technical field of photogrammetry data processing. First, by studying the point cloud classification optimization considering the visual attention mechanism, the point cloud stereoscopic visual attention feature analysis method is realized. , and a deep learning-based stereo target visual attention evaluation method, and evaluate and sort the recognition target attention strength; secondly, use the stereo visual attention mechanism to filter the point cloud of the oblique photographic point cloud to be recognized, and carry out point cloud Primary feature description and self-learning sub-model research; finally, the point cloud filtered by the visual attention mechanism of the point cloud scene relative to the target to be recognized is used as the point cloud to be recognized; the present invention obtains oblique photography with practical value Real point cloud classification technology, in order to effectively promote the development of oblique photography applications from "visualization" to "computable".

Description

technical field [0001] The invention relates to the technical field of photogrammetry data processing, in particular to a method for classifying oblique photographic point clouds based on a multi-feature integrated deep learning model. Background technique [0002] Oblique photography technology has been widely used in large-scale topographic map surveying and mapping, urban spatial data infrastructure construction, and urban fast true 3D modeling, etc. The degree of automation of oblique photography information extraction directly affects large-scale data production, a wide range of industry applications, and the full application of data. At present, through the steps of "image preprocessing, automatic joint aerial three-dimensional calculation, dense image matching, 3D point cloud generation, triangulation network construction, texture mapping" and other steps, the field data of oblique photography can quickly, efficiently and fully automatically obtain the real area. It ...

AI Technical Summary

Problems solved by technology

[0004] Under the current technical conditions, the classification of oblique photographic point clouds based on deep learning mainly faces the following challenges: 1) "Phenomena such as diverse targets, complex morphological structures, target occlusion and overlapping, and differences in spatial density are the key to automatic fine classification of 3D point clouds. Compared with laser point cloud, photogrammetry point cloud has more noise and more uneven distribution, which requires a more robust algorithm; 2) For a long time, the extraction of oblique photography information has mostly been carried out around the "manual stereo mapping" mode, Most of its production and application processes lack the link of oblique photographic point cloud classification, resulting in the lack of attention to the research on oblique photographic point cloud classification. There are not many related studies, it is difficult to obtain training samples for deep learning, and there is a lack of public sample data sets; 3) Although the current research It presents the technical development direction of the fusion of artificial feature prior knowledge and deep learning model, and proposes several point cloud feature learning strategies and model construction methods, but there is still a lack of research on the use of human stereo vision experience; 4) Practical application The deep learning model is aimed at replacing the actual manual operation, oblique photography artificial stereo mapping, even if there are obvious data missing, data occlusion and data overlap, shape or texture distortion, shadow interference, flying spots or dirty spots and other data abnormalities Under normal circumstances, human vision can easily and clearly identify vegetation, waters, ground, buildings, and finer types of ground features by comprehensively perceiving the three-dimensional shape, color, texture and other characteristics of ground objects.

Images