A tree image recognition system based on laser point cloud technology
By integrating data from drones and handheld devices and combining it with AI technology, the tree image recognition system achieves efficient fusion of air and ground data and matching of real-world images with point cloud data. This solves the problems of low data fusion and low processing efficiency in existing technologies, and improves the accuracy and intuitiveness of the recognition system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-30
AI Technical Summary
Existing tree image recognition systems suffer from low spatial-ground data fusion during data collection, low processing efficiency, insufficient intelligence, and a lack of intuitive integration between real-world images and point cloud data.
By integrating data from drone radar and handheld devices, and utilizing AI technology for efficient classification and recognition, the system achieves the fusion of air and ground data, and matches real-world images and point cloud data under the same three-dimensional geographic coordinates.
This improved the completeness and accuracy of the data, enabling efficient and accurate classification and identification of trees, and enhancing the objectivity and intuitiveness of the assessment process.
Smart Images
Figure CN122313261A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image recognition technology, and in particular relates to a tree image recognition system based on laser point cloud technology. Background Technology
[0002] Laser point cloud technology is a technique that uses a lidar system to collect three-dimensional spatial information about the surface of an object. Its core principle is to emit laser pulses towards the target and record the time interval between the pulse's emission and return. Combined with precise navigation and positioning data, the three-dimensional coordinates of a large number of discrete points on the target surface are calculated, thus forming a set of points, i.e., a "laser point cloud." In agriculture and forestry, combining laser point cloud technology with tree image recognition systems can obtain information such as the number and type of trees, and is commonly used for forest resource surveys, biomass estimation, and vegetation cover analysis.
[0003] Current tree image recognition systems include a data acquisition system and an analysis system. The data acquisition system is used to collect tree data in the area to be measured. The tree data includes three-dimensional point cloud data and image data. The three-dimensional point cloud data is used to acquire point cloud data. The image data is used to perform image recognition to obtain tree information. The analysis system is used to obtain first measurement data based on the point cloud data and second measurement data based on the tree information.
[0004] Existing tree image recognition systems rely on drones equipped with radar to scan the tree canopy from the air during on-site scanning, while the trunk is scanned by handheld radar. This results in low data fusion between the two scans. Furthermore, current point cloud data processing largely relies on various tools, such as single-tree segmentation tools, which have low levels of intelligence, leading to low processing efficiency. Additionally, current laser point cloud tree image recognition systems only allow users to view the point cloud data without combining it with on-site photographs, resulting in low intuitiveness during viewing. Summary of the Invention
[0005] To address the aforementioned technical issues, this invention provides a tree image recognition system based on laser point cloud technology. By integrating data captured by drone radar and handheld devices, it fuses aerial and ground data, improving data integrity and accuracy. AI technology is used for efficient tree classification and recognition, achieving objectivity, efficiency, and accuracy in the evaluation process. Matching real-world images with point cloud data allows for a more intuitive view of the real-world image and the original 3D point cloud data under the same 3D geographic coordinates.
[0006] A tree image recognition system based on laser point cloud technology includes a data acquisition module, a data backup and processing module, and an image recognition module;
[0007] The data acquisition module includes an image capture module, a laser scanning module, and a data storage and transmission module.
[0008] The image recognition module includes a coordinate matching module and a coordinate-based image retrieval and display module;
[0009] The image capturing module includes an airborne camera and a handheld camera. The airborne camera is installed on a drone carrying a LiDAR laser scanner, and the handheld camera is installed on a handheld laser scanner. The image capturing module captures images of a certain forest area to obtain real-world images with geographic location coordinate information.
[0010] The laser scanning module includes a drone equipped with a LiDAR laser scanner and a handheld laser scanner. The drone flies along a preset route and uses the LiDAR laser scanner to collect three-dimensional data of the tree crowns of a certain area to form the first three-dimensional point cloud raw data. The handheld laser scanner is used to collect three-dimensional data of the tree trunks of a certain area to form the second three-dimensional point cloud raw data.
[0011] The data storage and transmission module includes a data storage module and a data transmission module. The first three-dimensional point cloud raw data is stored through the data storage module in the LiDAR laser scanner, and the second three-dimensional point cloud raw data is stored through the data storage module in the handheld laser scanner. After the scanning is completed, the first three-dimensional point cloud raw data and the second three-dimensional point cloud raw data are transmitted to the data backup and processing module through the data transmission module.
[0012] Preferably, the data backup and processing module includes a data backup module, a data verification module, and a data processing module. The first 3D point cloud original data and the second 3D point cloud original data are transmitted to the data backup module for backup. The first 3D point cloud original data and the second 3D point cloud original data are transmitted to the data processing module for processing. After the transmission of the first 3D point cloud original data and the second 3D point cloud original data is completed, the data verification module compares the verification values of the source file and the target file of the 3D point cloud original data to confirm that there is no loss or damage in the data transmission.
[0013] Preferably, the data processing module includes a data integration module, a point cloud preprocessing module, a ground segmentation module, a single tree segmentation module, a single tree feature extraction module, a tree species identification module, and a forest stand index statistics module;
[0014] The data integration module integrates the first and second three-dimensional point cloud raw data, and combines the crown and trunk of each tree to form complete three-dimensional point cloud raw data.
[0015] The point cloud preprocessing module has a built-in AI screening model. Through the processing of the AI screening model, the point cloud preprocessing module removes non-target noise points from the point cloud and retains the point cloud data of the ground and trees. The point cloud preprocessing module includes a coarse denoising module, a fine denoising module, and a point cloud resampling module. The coarse denoising module removes invalid noise points from laser reflections, the fine denoising module removes non-tree noise points from the ground, and the point cloud resampling module thins out overly dense point clouds and adds points to overly sparse areas to ensure uniform point cloud density. The preprocessed point cloud data is obtained through the point cloud preprocessing module.
[0016] Preferably, the ground segmentation module has a built-in AI segmentation model. The ground segmentation module uses the AI segmentation model to segment the point clouds of the ground and trees, eliminating the error of ground undulation on tree height. The ground segmentation module is equipped with a ground segmentation tool, which separates the ground and tree point clouds after preprocessing the point cloud data. Using the ground points as a reference, the elevation of the tree point cloud is set to zero, thereby realizing the segmentation of the ground and tree point clouds and forming three-dimensional point cloud data of trees.
[0017] Preferably, the single-tree segmentation module is equipped with an AI single-tree segmentation tool, which automatically splits the trees and generates a unique number for each tree, and simultaneously counts the total number of trees to form single-tree segmentation data.
[0018] Preferably, the single-tree feature extraction module is equipped with a tree feature extraction tool to extract single-tree indicators with centimeter-level precision for each tree and generate a single-tree ledger. Through the tree feature extraction tool, users select the required indicators, such as tree height, diameter at breast height, branching point height, etc. The tree feature extraction tool automatically extracts indicators for each tree and generates a single-tree indicator ledger that also includes a number, indicator, and geographical coordinates.
[0019] Preferably, the tree species identification module is equipped with an AI tree species identification tool. After loading the tree species model, the AI tree species identification tool automatically labels the tree species for each tree. The AI tree species identification tool integrates deep learning, 3D point cloud processing and multimodal data, and realizes intelligent and automated identification and classification of tree species by automatically extracting the 3D structural features and spectral features of trees.
[0020] Preferably, the forest stand index statistics module includes a forest stand statistics tool, a 3D GIS extension module, and the forest stand statistics tool automatically summarizes individual tree indicators and calculates indicators such as forest stand density, canopy closure, and total volume. The 3D GIS extension module combines the statistical results with geographic coordinates to generate a spatial distribution map of trees and outputs a forest stand statistics report.
[0021] Preferably, the coordinate matching module is used to unify the real-scene image with geographic location coordinate information and the complete three-dimensional point cloud raw data into the same three-dimensional geographic coordinate system, so that the real-scene image with geographic location coordinate information and the complete three-dimensional point cloud raw data are spatially aligned.
[0022] The same coordinate map retrieval and display module can retrieve real-world images and raw 3D point cloud data under the same 3D geographic coordinates and display the retrieved information.
[0023] Beneficial effects
[0024] Compared with the prior art, the present invention has the following beneficial effects:
[0025] 1. By integrating data from drone radar and handheld devices, air-to-ground data is fused, improving the completeness and accuracy of the data.
[0026] 2. Use AI technology to efficiently classify and identify trees, achieving objectivity, efficiency, and accuracy in the assessment process.
[0027] 3. Matching real-world images with point cloud data allows for a more intuitive view of the real-world images and the original 3D point cloud data under the same 3D geographic coordinates. Attached Figure Description
[0028] 1. Data Acquisition Module; 2. Data Backup and Processing Module; 3. Coordinate Matching and Display Module; 11. Laser Scanning Module; 12. Data Storage and Transmission Module; 21. Data Backup Module; 22. Data Verification Module; 23. Data Processing Module; 230. Data Integration Module; 231. Point Cloud Preprocessing Module; 232. Ground Segmentation Module; 233. Individual Tree Segmentation Module; 234. Individual Tree Feature Extraction Module; 235. Tree Species Identification Module; 236. Stand Index Statistics Module; 31. Coordinate Matching Module; 32. Same Coordinate Map Retrieval and Display Module.
[0029] Figure 1 This is a structural diagram of a tree image recognition system based on laser point cloud technology.
[0030] Figure 2 This is a structural diagram of the laser scanning module.
[0031] Figure 3 This is a structural diagram of the data backup and processing module.
[0032] Figure 4 This is a structural diagram of the data processing module.
[0033] Figure 5 This is a structural diagram of the coordinate matching and display module. Detailed Implementation
[0034] The present invention will be further described below with reference to the accompanying drawings:
[0035] In the picture:
[0036] As attached Figure 1 and attached Figure 2 As shown:
[0037] A tree image recognition system based on laser point cloud technology specifically includes: a data acquisition module 1, a data backup and processing module 2, and an image recognition module 3;
[0038] The data acquisition module 1 includes an image capturing module 10, a laser scanning module 11, and a data storage and transmission module 12.
[0039] The image recognition module 3 includes a coordinate matching module 31 and a same coordinate number image retrieval and display module 32;
[0040] The image capturing module 10 includes an airborne camera and a handheld camera. The airborne camera is installed on a drone carrying a LiDAR laser scanner, and the handheld camera is installed on a handheld laser scanner. The image capturing module 10 captures images of a certain forest area to obtain real-scene images with geographic location coordinate information.
[0041] The laser scanning module 11 includes a drone equipped with a LiDAR laser scanner and a handheld laser scanner. The drone flies along a preset route and uses the LiDAR laser scanner to collect three-dimensional data of the tree crowns of a certain area to form the first three-dimensional point cloud raw data. The handheld laser scanner is used to collect three-dimensional data of the tree trunks of a certain area to form the second three-dimensional point cloud raw data.
[0042] The data storage and transmission module 12 includes a data storage module 121 and a data transmission module 122. The first three-dimensional point cloud raw data is stored through the data storage module 121 in the LiDAR laser scanner, and the second three-dimensional point cloud raw data is stored through the data storage module 121 in the handheld laser scanner. After the scanning is completed, the first three-dimensional point cloud raw data and the second three-dimensional point cloud raw data are transmitted to the data backup and processing module 2 through the data transmission module 122.
[0043] In this implementation plan, in conjunction with the appendix Figure 3As shown, the data backup and processing module 2 includes a data backup module 21, a data verification module 22, and a data processing module 23. The first 3D point cloud original data and the second 3D point cloud original data are transmitted to the data backup module 21 for backup. The first 3D point cloud original data and the second 3D point cloud original data are transmitted to the data processing module 23 for processing. After the first 3D point cloud original data and the second 3D point cloud original data are transmitted, the data verification module 22 compares the verification values of the source file and the target file of the 3D point cloud original data to confirm that there is no loss or damage in the data transmission.
[0044] In this implementation plan, in conjunction with the appendix Figure 4 As shown, the data processing module 23 includes a data integration module 230, a point cloud preprocessing module 231, a ground segmentation module 232, a single tree segmentation module 233, a single tree feature extraction module 234, a tree species identification module 235, and a forest stand index statistics module 236.
[0045] The data integration module 230 integrates the first three-dimensional point cloud raw data and the second three-dimensional point cloud raw data, and splices the crown and trunk of each tree to form complete three-dimensional point cloud raw data.
[0046] The point cloud preprocessing module 231 has a built-in AI screening model. The point cloud preprocessing module 231 extracts non-target noise points from the point cloud through the processing of the AI screening model, and retains the point cloud of the ground and trees. The point cloud preprocessing module 231 includes a coarse denoising module, a fine denoising module and a point cloud resampling module. The coarse denoising module removes invalid noise points of laser reflection, the fine denoising module removes non-tree noise points on the ground, and the point cloud resampling module thins out the overly dense point cloud and adds points to the overly sparse areas to ensure uniform point cloud density. The point cloud preprocessing module 231 obtains the preprocessed point cloud data.
[0047] In this embodiment, specifically, the ground segmentation module 232 has a built-in AI segmentation model. The ground segmentation module 232 uses the AI segmentation model to segment the point clouds of the ground and trees, eliminating the error of ground undulation on tree height. The ground segmentation module is equipped with a ground segmentation tool, which separates the ground and tree point clouds after preprocessing the point cloud data. Using the ground points as a reference, the elevation of the tree point cloud is set to zero, thereby realizing the segmentation of the ground and tree point clouds and forming three-dimensional point cloud data of trees.
[0048] In this implementation scheme, specifically, the single tree segmentation module 233 is equipped with an AI single tree segmentation tool. The AI single tree segmentation tool automatically splits the trees and generates a unique number for each tree, and simultaneously counts the total number of trees to form single tree segmentation data.
[0049] In this implementation scheme, specifically, the single tree feature extraction module 234 is equipped with a tree feature extraction tool to extract single tree indicators with centimeter-level precision for each tree and generate a single tree ledger. Through the tree feature extraction tool, the user selects the required indicators, such as tree height, diameter at breast height, branching point height, etc. The tree feature extraction tool automatically extracts indicators for each tree and generates a single tree indicator ledger that also includes a number, indicator, and geographical coordinates.
[0050] In this implementation scheme, specifically, the tree species identification module 235 is equipped with an AI tree species identification tool. After loading the tree species model, the AI tree species identification tool automatically labels the tree species for each tree. The AI tree species identification tool integrates deep learning, 3D point cloud processing, and multimodal data, and achieves intelligent and automated identification and classification of tree species by automatically extracting the 3D structural features and spectral features of trees. The AI tree species identification tool utilizes AI... The AI-powered tree species identification tool automatically identifies tree species using deep learning, point cloud models, and image recognition. This high-efficiency method enables batch identification of trees in large forest areas, significantly reducing manual labor and improving the efficiency of species identification. The tool's automatic identification accuracy is stable, leveraging the generalization ability of deep learning models and the precision of 3D point cloud features to avoid misjudgments caused by experience differences and visual fatigue in manual identification. This ensures the objectivity of the identification results, reduces the misjudgment rate, and allows for continuous improvement in accuracy through iterative training with samples. Furthermore, it enhances adaptability to different regions, being compatible with various types of laser point cloud data, including airborne and ground-based systems. This allows it to adapt to different climate zones, forest stand structures, and tree species identification needs, and is compatible with different regions and tree species databases.
[0051] In this implementation scheme, specifically, the forest stand index statistics module 236 is equipped with a forest stand statistics tool, a 3D GIS extension module, and the forest stand statistics tool automatically summarizes individual tree indicators and calculates indicators such as forest stand density, canopy closure, and total volume. The 3D GIS extension module combines the statistical results with geographic coordinates to generate a tree spatial distribution map and outputs a forest stand statistics report.
[0052] In this implementation plan, in conjunction with the appendix Figure 5 As shown, the coordinate matching module 31 is used to unify the real-scene image with geographic location coordinate information and the complete three-dimensional point cloud raw data into the same three-dimensional geographic coordinate system, so that the real-scene image with geographic location coordinate information and the complete three-dimensional point cloud raw data are aligned in spatial position.
[0053] The same coordinate map retrieval and display module 32 can retrieve real-world images and original 3D point cloud data under the same three-dimensional geographic coordinates, and display the retrieved information, so that users can intuitively see real-world images and original 3D point cloud data under the same three-dimensional geographic coordinates.
[0054] Any technical solution that achieves the above-mentioned technical effects by utilizing the technical solutions described in this invention, or by designing similar technical solutions by those skilled in the art under the inspiration of the technical solutions described in this invention, falls within the protection scope of this invention.
Claims
1. A tree image recognition system based on laser point cloud technology, characterized in that, include: The data acquisition module (1), the data backup and processing module (2), and the image recognition module (3) are included. The data acquisition module (1) includes an image capturing module (10), a laser scanning module (11), and a data storage and transmission module (12). The image recognition module (3) includes a coordinate matching module (31) and a coordinate number image retrieval and display module (32); The image capturing module (10) includes an airborne camera and a handheld camera. The airborne camera is installed on a drone carrying a LiDAR laser scanner, and the handheld camera is installed on a handheld laser scanner. The image capturing module (10) captures images of a forest area to obtain real-world images with geographic location coordinate information. The laser scanning module (11) includes a drone equipped with a LiDAR laser scanner and a handheld laser scanner. The drone flies along a preset route and uses the LiDAR laser scanner to collect three-dimensional data of the tree crowns of a certain area to form the first three-dimensional point cloud raw data. The handheld laser scanner is used to collect three-dimensional data of the tree trunks of a certain area to form the second three-dimensional point cloud raw data. The data storage and transmission module (12) includes a data storage module (121) and a data transmission module (122). The first three-dimensional point cloud raw data is stored through the data storage module (121) in the LiDAR laser scanner, and the second three-dimensional point cloud raw data is stored through the data storage module (121) in the handheld laser scanner. After the scanning is completed, the first three-dimensional point cloud raw data and the second three-dimensional point cloud raw data are transmitted to the data backup and processing module (2) through the data transmission module (122).
2. The tree image recognition system based on laser point cloud technology as described in claim 1, characterized in that, The data backup and processing module (2) includes a data backup module (21), a data verification module (22), and a data processing module (23). The original data of the first three-dimensional point cloud and the original data of the second three-dimensional point cloud are transmitted to the data backup module (21) for backup. The original data of the first three-dimensional point cloud and the original data of the second three-dimensional point cloud are transmitted to the data processing module (23) for processing. After the original data of the first three-dimensional point cloud and the original data of the second three-dimensional point cloud are transmitted, the data verification module (22) compares the verification values of the source file and the target file of the original three-dimensional point cloud to confirm that the data transmission is without loss or damage.
3. The tree image recognition system based on laser point cloud technology as described in claim 2, characterized in that, The data processing module (23) includes a data integration module (230), a point cloud preprocessing module (231), a ground segmentation module (232), a single tree segmentation module (233), a single tree feature extraction module (234), a tree species identification module (235), and a forest stand index statistics module (236). The data integration module (230) integrates the first three-dimensional point cloud raw data and the second three-dimensional point cloud raw data, and splices the crown and trunk of each tree to form complete three-dimensional point cloud raw data. The point cloud preprocessing module (231) has an AI screening model built in. The point cloud preprocessing module (231) extracts non-target noise points in the point cloud through the processing of the AI screening model, and retains the point cloud of the ground and trees. The point cloud preprocessing module (231) includes a coarse denoising module, a fine denoising module and a point cloud resampling module. The coarse denoising module removes invalid noise points of laser reflection, the fine denoising module removes non-tree noise points on the ground, and the point cloud resampling module thins out the overly dense point cloud and fills in the overly sparse areas to ensure uniform point cloud density. The preprocessed point cloud data is obtained through the point cloud preprocessing module (231).
4. The tree image recognition system based on laser point cloud technology as described in claim 3, characterized in that, The ground segmentation module (232) has an AI segmentation model built in. The ground segmentation module (232) segments the point cloud of the ground and trees through the AI segmentation model, eliminating the error of ground undulation on tree height. The ground segmentation module is equipped with a ground segmentation tool. The preprocessed point cloud data is separated into ground and tree point clouds through the ground segmentation tool in the ground segmentation module (232). The elevation of the tree point cloud is set to zero based on the ground point, thereby realizing the segmentation of the ground and tree point clouds and forming three-dimensional point cloud data of trees.
5. The tree image recognition system based on laser point cloud technology as described in claim 3, characterized in that, The single tree segmentation module (233) is equipped with an AI single tree segmentation tool. The AI single tree segmentation tool automatically splits the trees and generates a unique number for each tree, and simultaneously counts the total number of trees to form single tree segmentation data.
6. The tree image recognition system based on laser point cloud technology as described in claim 3, characterized in that, The single tree feature extraction module (234) is equipped with a tree feature extraction tool, which extracts single tree indicators with centimeter-level precision for each tree and generates a single tree ledger. Through the tree feature extraction tool, the user selects the required indicators, such as tree height, diameter at breast height, branching point height, etc. The tree feature extraction tool automatically extracts indicators for each tree and generates a single tree indicator ledger that also includes a number, indicator and geographical coordinates.
7. The tree image recognition system based on laser point cloud technology as described in claim 3, characterized in that, The tree species identification module (235) is equipped with an AI tree species identification tool. After loading the tree species model, the AI tree species identification tool automatically labels the tree species for each tree. The AI tree species identification tool integrates deep learning, three-dimensional point cloud processing and multimodal data, and realizes intelligent and automated identification and classification of tree species by automatically extracting the three-dimensional structural features and spectral features of trees.
8. The tree image recognition system based on laser point cloud technology as described in claim 3, characterized in that, The forest stand index statistics module (236) includes a forest stand statistics tool, a three-dimensional GIS extension module, and a forest stand statistics tool that automatically summarizes individual tree indicators and calculates indicators such as forest stand density, canopy closure, and total volume. The three-dimensional GIS extension module combines the statistical results with geographic coordinates to generate a tree spatial distribution map and outputs a forest stand statistics report.
9. The tree image recognition system based on laser point cloud technology as described in claim 1, characterized in that, The coordinate matching module (31) is used to unify the real-scene image with geographic location coordinate information and the complete three-dimensional point cloud raw data into the same three-dimensional geographic coordinate system, so that the real-scene image with geographic location coordinate information and the complete three-dimensional point cloud raw data are aligned in spatial position. The same coordinate map retrieval and display module (32) can retrieve real-world images and original three-dimensional point cloud data under the same three-dimensional geographic coordinates and display the retrieved information.