CN-122023686-A - Land high-precision three-dimensional modeling optimization method and system based on unmanned aerial vehicle remote sensing

CN122023686ACN 122023686 ACN122023686 ACN 122023686ACN-122023686-A

Abstract

The invention discloses a land high-precision three-dimensional modeling optimization method and system based on unmanned aerial vehicle remote sensing, which relate to the technical field of remote sensing mapping and comprise the steps of carrying out local multi-view geometric solution and incremental optimization on a multi-source view block sequence to obtain a view block geometric parameter set, and carrying out online training on a nerve elevation coding network by using the view block geometric parameter set to generate a gridding result; the method comprises the steps of carrying out semantic projection and grid subdivision on a gridding result, clustering to obtain a plurality of renovation operation blocks, carrying out graph construction on each renovation operation block by using preprocessed topographic data to obtain a land block instance scene graph, carrying out visual interaction on the land block instance scene graph to generate a renovation index, and updating the land block instance scene graph to obtain an updated land block instance scene graph, wherein the problem of weak land block renovation semantic association is solved by constructing the nerve height Cheng Chang as the land block instance scene graph.

Inventors

WEI YANFEI
CAO LIWEI
TONG XINHUA
HUANG LONGCHENG
ZHANG YUAN

Assignees

南宁师范大学
广西致达远土地规划设计有限公司

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (10)

1. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing is characterized by comprising the following steps of, Collecting topographic data, remediation planning data and unmanned aerial vehicle remote sensing data of a target area, preprocessing the topographic data, the remediation planning data and the unmanned aerial vehicle remote sensing data, and constructing a multi-source view block sequence; carrying out local multi-view geometric solution and incremental optimization on the multi-source view block sequence to obtain a view block geometric parameter set, and carrying out online training on the nerve elevation coding network by using the view block geometric parameter set to generate a gridding result; carrying out semantic projection and grid subdivision on the gridding result, clustering to obtain a plurality of renovation operation blocks, and carrying out graph construction on each renovation operation block by using the preprocessed topographic data to obtain a land block instance scene graph; and performing visual interaction on the land parcel instance scene graph to generate a remediation index, and updating the land parcel instance scene graph to obtain an updated land parcel instance scene graph.
2. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as set forth in claim 1, wherein the constructing the multi-source view block sequence using the preprocessed topographic data, the remediation planning data and the unmanned aerial vehicle remote sensing data comprises the following steps, Calculating the land mass importance of each land mass by using the preprocessed remediation planning data to obtain land mass importance distribution; planning a route of the unmanned aerial vehicle according to the land mass importance distribution, the preprocessed topographic data and the unmanned aerial vehicle remote sensing data to obtain unmanned aerial vehicle tracks and photo-taking point sets; And controlling the unmanned aerial vehicle, performing aerial photographing operation according to the flight path and the photographing point set of the unmanned aerial vehicle to obtain an aerial photographing image sequence, and aggregating the aerial photographing image sequence to obtain a multi-source view block sequence.
3. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 1, wherein the local multi-view geometric solution and incremental optimization are carried out on the multi-source view block sequence to obtain a view block geometric parameter set, and the specific steps are as follows, Carrying out local feature extraction and matching on the multi-source view block sequence, and carrying out local multi-view geometric solution to obtain an absolute camera external azimuth parameter and a pixel ray direction; and performing incremental optimization on the external azimuth parameters of the absolute camera and the pixel ray directions to obtain a view block geometric parameter set.
4. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 1, wherein the method is characterized in that the method uses a view block geometric parameter set to train a neural elevation coding network on line to generate a gridding result, and comprises the following specific steps of, Performing online counter propagation training on a local area of the nerve elevation coding network by using the view block geometric parameter set to obtain a nerve elevation field; and performing gridding sampling on the nerve Gao Chengchang to obtain a gridding result.
5. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 1, wherein the semantic projection and grid subdivision are carried out on the gridding result, the specific steps are as follows, Carrying out semantic projection on the gridding result by using the multi-source view block sequence to obtain a gridding result with semantics; and expanding the semantic gridding result into a three-dimensional space by adopting the nerve height Cheng Chang to obtain a three-dimensional earth surface grid.
6. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 1, wherein the construction of the map is carried out on each renovation work block by using the preprocessed topographic data to obtain a land block instance scene map, and the specific steps are as follows, Determining mapping information of the land block and the renovation operation block according to the preprocessed topographic data and the renovation planning data; abstracting the land parcels and the remediated operation blocks into nodes by using mapping information, and establishing side relations among the nodes to obtain a land parcels relation graph; the plot relationship graph is expanded into a plot instance scene graph using the nerve height Cheng Chang and the semantic gridding result.
7. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 1, wherein the visual interaction is carried out on the land parcel instance scene graph to generate a remediation index, and the specific steps are as follows, Loading the land parcel instance scene graph to a three-dimensional visual platform, and responding to drawing operation of a user in a visual interface to obtain a geometric shape drawn by the user; Binding the geometric shapes drawn by the user to the corresponding nodes, and updating the nodes to obtain the treatment index.
8. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 1, wherein the land high-precision three-dimensional modeling optimization method is characterized in that the land instance scene graph is updated to obtain an updated land instance scene graph, and comprises the following specific steps of, Local constraint adjustment is carried out on the nerve elevation field by using the adjustment index, so as to obtain a target topography; and adjusting the land parcel instance scene graph by adopting the target terrain to obtain an updated land parcel instance scene graph.
9. The land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing as claimed in claim 8, wherein the land instance scene graph is adjusted by adopting target terrain, and comprises the following specific steps of, Generating a local gridding result according to the target topography and the treatment index, and carrying out local replacement on the three-dimensional earth surface grid to obtain an updated three-dimensional earth surface grid; and (3) adjusting the plot relation diagram by using the updated three-dimensional earth surface grid, and re-expanding to obtain an updated plot instance scene diagram.
10. The land high-precision three-dimensional modeling optimization system based on unmanned aerial vehicle remote sensing is characterized by comprising the following components according to any one of the claims 1-9, The construction module is used for collecting the topographic data, the renovation planning data and the unmanned aerial vehicle remote sensing data of the target area and preprocessing the topographic data, the renovation planning data and the unmanned aerial vehicle remote sensing data to construct a multi-source view block sequence; the training module performs local multi-view geometric solution and incremental optimization on the multi-source view block sequence to obtain a view block geometric parameter set, and performs online training on the nerve elevation coding network by using the view block geometric parameter set to generate a gridding result; The example module performs semantic projection and grid subdivision on the gridding result, clusters the semantic projection and the grid subdivision to obtain a plurality of renovation operation blocks, and performs image construction on each renovation operation block by using the preprocessed topographic data to obtain a land block example scene image; And the updating module is used for carrying out visual interaction on the land parcel instance scene graph, generating a remediation index, and updating the land parcel instance scene graph to obtain an updated land parcel instance scene graph.

Description

Land high-precision three-dimensional modeling optimization method and system based on unmanned aerial vehicle remote sensing Technical Field The invention relates to the technical field of remote sensing mapping, in particular to a land high-precision three-dimensional modeling optimization method and system based on unmanned aerial vehicle remote sensing. Background Three-dimensional modeling technology based on unmanned aerial vehicle remote sensing has become a mainstream method for acquiring high-precision topographic data in land remediation projects. Most of the existing three-dimensional modeling methods rely on structured light, laser radar or stereoscopic image matching, a digital surface model is generated through multi-view geometry of aerial images, and then topographic mapping and ground object identification are realized by combining digital orthographic images and digital surface models. In order to improve the precision and timeliness of the model, researchers propose a multi-view three-dimensional reconstruction algorithm, a deep learning semantic segmentation model and a point cloud fusion technology, so that an unmanned aerial vehicle can realize high-precision earth surface modeling under a large range and complex terrain. The existing method can be improved, and firstly, the existing method generally adopts an off-line reconstruction and step-by-step optimization mode, so that the multi-source data fusion accuracy is low. Second, the lack of space-time constraints and semantic associations between multi-source data makes it difficult to form interactive and updatable digital terrain representation structures. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing solves the problems of low multi-source data fusion precision and weak land block renovation semantic association. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the invention provides a land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing, which comprises the following steps of, Collecting topographic data, remediation planning data and unmanned aerial vehicle remote sensing data of a target area, preprocessing the topographic data, the remediation planning data and the unmanned aerial vehicle remote sensing data, and constructing a multi-source view block sequence; carrying out local multi-view geometric solution and incremental optimization on the multi-source view block sequence to obtain a view block geometric parameter set, and carrying out online training on the nerve elevation coding network by using the view block geometric parameter set to generate a gridding result; carrying out semantic projection and grid subdivision on the gridding result, clustering to obtain a plurality of renovation operation blocks, and carrying out graph construction on each renovation operation block by using the preprocessed topographic data to obtain a land block instance scene graph; and performing visual interaction on the land parcel instance scene graph to generate a remediation index, and updating the land parcel instance scene graph to obtain an updated land parcel instance scene graph. As a preferable scheme of the land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing, the invention constructs a multi-source view block sequence by using preprocessed topographic data, remediation planning data and unmanned aerial vehicle remote sensing data, and comprises the following specific steps, Calculating the land mass importance of each land mass by using the preprocessed remediation planning data to obtain land mass importance distribution; planning a route of the unmanned aerial vehicle according to the land mass importance distribution, the preprocessed topographic data and the unmanned aerial vehicle remote sensing data to obtain unmanned aerial vehicle tracks and photo-taking point sets; And controlling the unmanned aerial vehicle, performing aerial photographing operation according to the flight path and the photographing point set of the unmanned aerial vehicle to obtain an aerial photographing image sequence, and aggregating the aerial photographing image sequence to obtain a multi-source view block sequence. As a preferable scheme of the land high-precision three-dimensional modeling optimization method based on unmanned aerial vehicle remote sensing, the invention comprises the steps of carrying out local multi-view geometric solution and incremental optimization on a multi-source view block sequence to obtain a view block geometric parameter set, specifically comprising the following steps of, Carrying out local feature extraction and matchin