A method for rapidly determining forest stock volume based on ground-based lidar data

Abstract

This invention discloses a method for rapidly determining forest stock volume based on ground-based lidar data, comprising the following steps: S1, acquiring raw point cloud data; S2, processing the raw point cloud data; S3, data preprocessing; S4, point cloud data extraction; S5, segmentation and recognition of tree trunk-like objects; S6, fitting cylindrical objects to traverse the data; S7, identification and extraction of arc-like objects; S8, extraction of forest stock volume calculation factors; and S9, calculation of forest stock volume. This invention solves the technical problems of low fieldwork efficiency, long processing cycles, high workload, and large discrepancies in survey results in existing forest stock volume measurement technologies. It improves work efficiency, reduces labor intensity, and provides high-quality data with high accuracy.

Description

Technical Field

[0001] This invention belongs to the field of forestry remote sensing technology, specifically relating to a method for rapidly determining forest stock volume based on ground-based lidar data. Background Technology

[0002] Forest stock volume is an important indicator of the abundance of forest resources and the quality of the forest ecological environment. It is also a significant marker of forest carbon sequestration capacity and a crucial component of forest aboveground biomass. Therefore, stock volume is a vital factor in forest stand surveys, providing essential information for forest harvesting, utilization, and management. With the increasing application and maturation of remote sensing technology in forest resource monitoring, numerous domestic and international scholars have conducted research on remote sensing estimation of forest stock volume.

[0003] Methods for estimating forest stock volume based on remote sensing data can be mainly divided into two categories: traditional mathematical models and machine learning. For example, multivariate regression analysis uses algorithms such as least squares, partial least squares, stepwise regression, and ridge estimation to obtain parameters, as well as algorithms such as radial basis function neural networks, artificial neural networks, k-nearest neighbors, random forests, and support vector machines.

[0004] Based on the source of remote sensing data, remote sensing can be divided into two main categories: passive remote sensing and active remote sensing. Passive remote sensing, also known as passive remote sensing system, is a detection system that does not have a radiation source itself. During remote sensing, the detection instrument acquires and records electromagnetic wave information emitted or reflected by the target object itself from natural radiation sources, such as the sun. Examples include aerial and aerospace photography systems and infrared scanning systems. The advantage of active remote sensing is that it does not rely on solar radiation, can work day and night, and can actively select the wavelength and emission mode of electromagnetic waves according to different detection purposes. The electromagnetic waves generally used are microwave bands and lasers, often using pulse signals, but some also use continuous beams. Ordinary radar, side-looking radar, synthetic aperture radar, infrared radar, and lidar all belong to active remote sensing systems.

[0005] LiDAR is an active remote sensing technology that uses laser light emitted by a sensor to determine the distance between a sensor and a target. Depending on the platform on which it is mounted, it can be divided into spaceborne LiDAR (Geoscience Laser Altimeter System, GLAS), airborne LiDAR (Airborne Laser Scanner, ALS), mobile LiDAR (Mobile Laser Scanner, MLS), and ground-based LiDAR (Terrestrial Laser Scanner, TLS). It can effectively penetrate forests and directly acquire the three-dimensional spatial coordinates of ground features, offering unparalleled advantages over other remote sensing technologies in measuring tree height and obtaining information on the vertical structure of forest stands.

[0006] Existing methods for surveying forest stock volume include manual field surveys and lidar system surveys. Manual field surveys are further divided into whole-forest measurement methods, standard plot survey methods, and angle gauge control measurement methods. These methods require technicians to reach the heart of the forest and use relatively primitive measuring instruments such as altimeters, diameter gauges, tape measures, and compasses to obtain forest stand factors. The survey techniques are relatively backward and extensive, not only time-consuming and labor-intensive but also only able to obtain information on a small area of ​​the forest. The survey scope is incomplete, the labor intensity is high, and there are unpredictable dangers involved.

[0007] LiDAR systems can be categorized into ALS estimation, TLS measurement, and backpack-mounted MLS estimation based on the sensor platform. ALS transmits laser beams from an aircraft platform to the ground, enabling the acquisition of large-scale forest canopy structure information. However, due to the lack of laser penetration, the beam is often obstructed by tree crowns and foliage when the canopy closure is high and the number of trees is large, making it difficult to accurately acquire lower-layer information. TLS is primarily used for high-precision extraction of lower-layer parameters in small-scale forests, and is more suitable for the accurate extraction of individual tree factors and the reconstruction of 3D forest scenes. Its working range is extremely limited, and it struggles to fully acquire canopy top structure information under closed canopy conditions. When forest scenes are complex, multi-station scanning is often required. However, with too many stations, issues such as target movement or loss can easily generate invalid data, resulting in very low data acquisition and post-assembly efficiency. Backpack-mounted MLS, while eliminating the need for multiple station scans and avoiding extensive post-assembly work, offers a more comprehensive solution. However, its applicability to forest scenarios is very limited. During use, it must be carried by the survey personnel. In order to ensure data accuracy, the equipment must be moved at a relatively uniform speed and walk steadily. The forest area is often complex, with many steep slopes and potholes, making it very easy to fall and slip, and it is impossible to move forward in an ideal state. Therefore, the data obtained by scanning is prone to layering and mismatch, resulting in large errors in the final results, which cannot meet the actual needs. Summary of the Invention

[0008] The purpose of this invention is to provide a method for rapidly determining forest stock volume based on ground-based lidar data. This method can solve the technical problems of low fieldwork efficiency, long processing time, high workload, and large discrepancies in survey results in existing forest stock volume measurement technologies. It can improve work efficiency, reduce labor intensity, and provide high-quality data with high measurement accuracy.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A method for rapidly determining forest stock volume based on ground-based lidar data includes the following steps:

[0011] S1. Obtain raw point cloud data;

[0012] S2, Solve the original point cloud data;

[0013] S3, Data Preprocessing;

[0014] S4, Point Cloud Data Extraction;

[0015] S5. Tree-like object segmentation and recognition;

[0016] S6. Traverse the object to fit a cylinder;

[0017] S7. Recognition and extraction of curved surface objects;

[0018] S8. Extraction of factors for forest stock volume measurement;

[0019] S9. Calculation of forest stock volume.

[0020] Furthermore, the method for step S1 is as follows: select the area to be measured, use a ground-based lidar scanning device to perform a fixed-point panoramic scan of the forest, and obtain raw point cloud data.

[0021] Furthermore, the method in step S2 is as follows: export the original point cloud data from the lidar scanning device, and use post-processing software to perform data calculation and enhancement processing to obtain the point cloud result data of the target area.

[0022] Furthermore, in step S3, LiDAR 360 is used to preprocess LiDAR point cloud data.

[0023] Furthermore, the method for step S4 is as follows: extract a subset of the preprocessed point cloud data from a distance of 1.3m ± 0.5m as input data.

[0024] Furthermore, in step S5, mathematical statistics methods are used, and an algorithm is written in Python to achieve point cloud object segmentation.

[0025] Furthermore, the method in step S6 is as follows: fit a cylinder based on the denoised point cloud data to obtain the cylinder's axial direction, axis, cylinder radius, and fitting error data, and calculate the point cloud fitting effect evaluation index based on the distance from the point cloud data to the cylinder's axis and the cylinder's radius. The index includes goodness of fit, root mean square error, and estimation accuracy.

[0026] Furthermore, the method for step S7 is as follows: First, a batch of training data is created using a manual judgment method. Each data point contains information on whether it is a curved surface object and information on the evaluation index of the cylinder fitting effect. A random forest classifier is trained using the data. Then, the goodness of fit, root mean square error, and estimation accuracy obtained after fitting the cylinder for each object in S6 are used as independent variables and input into the trained random forest classifier to determine whether the object is a curved surface object. If it is, it is retained; otherwise, it is deleted, resulting in a curved surface object file.

[0027] Furthermore, the method for step S8 is as follows: based on the arc-shaped object obtained in S7, and based on the cylinder fitted in S6, calculate the diameter at breast height (DBH) and cross-sectional area of ​​the sample tree corresponding to the object; based on the DBH of the sample tree and the number of arc-shaped objects, extract the number of trees per unit area of ​​the forest stand; based on the relative position of the station and the point cloud object and the DBH of the sample tree, calculate the horizontal distance from the station to the DBH position of the sample tree.

[0028] Furthermore, the method for step S9 is as follows: using the factors calculated in S8, the forest stock volume is calculated through dendrometric theory.

[0029] The advantages of this invention compared to the prior art are as follows:

[0030] This invention utilizes the characteristics of ground-based lidar, such as fast scanning speed and high data accuracy, to ensure estimation accuracy while maintaining high work efficiency. It can directly extract forest volume measurement factors from the structural information inside the forest stand, without the need for a large amount of sample data for modeling and inversion. The field survey process is simple, the indoor processing time is short, the processing speed is fast, and the measurement results are highly accurate. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the present invention;

[0032] Figure 2 This is the point cloud result data map obtained in step S2 of embodiment 1 of the present invention;

[0033] Figure 3 This is the point cloud data image before denoising in step S3 of embodiment 1 of the present invention;

[0034] Figure 4 This is the point cloud data image after denoising in step S3 of embodiment 1 of the present invention;

[0035] Figure 5 This is the data map obtained from ground point classification in step S3 of embodiment 1 of the present invention;

[0036] Figure 6 This is a data graph before normalization processing in step S3 of embodiment 1 of the present invention;

[0037] Figure 7 This is a data graph after normalization processing in step S3 of embodiment 1 of the present invention;

[0038] Figure 8 This is the data image before data cropping in step S3 of embodiment 1 of the present invention;

[0039] Figure 9 This is a data image after data cropping in step S3 of embodiment 1 of the present invention;

[0040] Figure 10 This is a top-view data diagram of the results captured in step S4 of embodiment 1 of the present invention;

[0041] Figure 11 This is a partial data image of the result extracted in step S4 of embodiment 1 of the present invention.

[0042] Figure 12 This is the data image before filtering and denoising in step S5 of embodiment 1 of the present invention;

[0043] Figure 13 This is a data image after filtering and denoising in step S5 of embodiment 1 of the present invention;

[0044] Figure 14 This is a data graph of 3D density clustering in step S5 of embodiment 1 of the present invention;

[0045] Figure 15 This is a data diagram before the extraction of the tree trunk object in step S5 of embodiment 1 of the present invention;

[0046] Figure 16 This is a data diagram after the tree trunk object is extracted in step S5 of embodiment 1 of the present invention;

[0047] Figure 17 This is the pre-projection data image of 2D projection denoising in step S5 of embodiment 1 of the present invention;

[0048] Figure 18 This is the projected data image after 2D projection denoising in step S5 of embodiment 1 of the present invention;

[0049] Figure 19 This is the data image before denoising in step S5 of embodiment 1 of the present invention;

[0050] Figure 20 This is the denoised data image of 2D projection denoising in step S5 of embodiment 1 of the present invention;

[0051] Figure 21 This is a top view of the fitting effect in step S6 of embodiment 1 of the present invention;

[0052] Figure 22 This is a side view of the fitting effect in step S6 of embodiment 1 of the present invention. Detailed Implementation

[0053] Before further describing specific embodiments of the present invention, it should be understood that the scope of protection of the present invention is not limited to the specific embodiments described below; it should also be understood that the terminology used in the embodiments of the present invention is for describing specific embodiments and not for limiting the scope of protection of the present invention.

[0054] 1. A method for rapidly determining forest stock volume based on ground-based lidar data, such as... Figure 1 As shown, it includes the following steps:

[0055] S1. Obtain raw point cloud data: Select the area to be measured, and use a ground-based lidar scanning device to perform a fixed-point panoramic scan of the forest to obtain raw point cloud data.

[0056] S2. Solve the raw point cloud data: Export the raw point cloud data from the lidar scanning device, and use post-processing software to solve and enhance the data to obtain the point cloud result data of the target area.

[0057] S3. Data Preprocessing: LiDAR point cloud data preprocessing is performed using LiDAR 360;

[0058] S4. Point cloud data extraction: Extract a subset of preprocessed point cloud data from a distance of 1.3m ± 0.5m as input data;

[0059] S5. Tree-like object segmentation and recognition: Using mathematical statistics methods and writing algorithms in Python, point cloud object segmentation is achieved.

[0060] S6. Traversing the object to fit a cylinder: Fit a cylinder based on the denoised point cloud data to obtain the cylinder's axis, center point, cylinder radius, and fitting error data. Calculate the point cloud fitting effect evaluation index based on the distance from the point cloud data to the cylinder's axis and the cylinder radius. The index includes goodness of fit, root mean square error, and estimation accuracy.

[0061] S7. Recognition and extraction of curved surface objects: First, a batch of training data is created through manual judgment. Each data point contains information on whether it is a curved surface object and evaluation index information on the cylinder fitting effect. A random forest classifier is trained using the data. Then, the goodness of fit, root mean square error, and estimation accuracy obtained after fitting the cylinder for each object in S6 are used as independent variables and input into the trained random forest classifier to determine whether the object is a curved surface object. If it is, it is kept; otherwise, it is deleted, resulting in a curved surface object file.

[0062] S8. Extraction of forest stock measurement factors: Based on the arc-shaped object obtained in S7, and the cylinder fitted in S6, calculate the diameter at breast height (DBH) and basal area of ​​the corresponding sample tree. Then, based on the DBH of the sample tree and the number of arc-shaped objects, extract the number of trees per unit area of ​​the forest stand. Based on the relative position of the station and the point cloud object and the DBH of the sample tree, calculate the horizontal distance from the station to the DBH position of the sample tree.

[0063] S9. Forest stock volume calculation: Using the factors obtained in S8, forest stock volume is calculated through dendrometric theory.

[0064] The following description uses more specific examples.

[0065] Example 1

[0066] S1. Obtain raw point cloud data:

[0067] In Ganwei Village, Ganwei Town, Wuming District, Nanning City, a PENTAX S-3180TLS scanning system was set up to conduct a fixed-point panoramic scan of the forest and obtain raw three-dimensional laser point cloud data.

[0068] S2: Solve the original point cloud data;

[0069] Raw data is exported from the device, and post-processing software Z+FLaserControl provided by the hardware manufacturer is used for data processing and enhancement to obtain standard LAS format 3D laser point cloud data of the target area, such as... Figure 2 As shown.

[0070] S3: Data preprocessing;

[0071] LiDAR point cloud data preprocessing was performed using LiDAR 360. The preprocessing included the following steps:

[0072] ① Data denoising:

[0073] Data denoising was performed using five standard deviations. Before denoising, the dataset contained 17,074,150 data points; after denoising, it contained 16,998,143 data points. Figure 3-4 As shown:

[0074] ② Ground point classification: Based on the actual conditions of the data collection area, ground points are classified using a gentle slope scenario, such as... Figure 5 As shown;

[0075] ③ Normalization: Based on the ground point classification results, the point cloud data, which originally had terrain undulations, is normalized to the same horizontal plane, such as... Figure 6-7 As shown;

[0076] ④ Data cropping: Using the station location as the center, crop a circular area with a radius greater than or equal to 14.57m, such as... Figure 8-9 As shown;

[0077] S4: Point cloud data extraction: Extract a subset of preprocessed point cloud data from a distance of 1.3m ± 0.5m as input data, such as... Figure 10-11 As shown.

[0078] S5: Tree-like object segmentation and recognition: Using mathematical statistics methods and writing algorithms in Python, point cloud object segmentation is achieved, including the following steps:

[0079] ① Filtering and noise reduction: Select the 20 nearest points to a given point, calculate the average Euclidean distance from each point to the other points, and use the 3-standard-deviation principle to remove outliers, such as... Figure 12-13 As shown;

[0080] ② Clustering and segmentation:

[0081] A 3D density clustering method, utilizing the DBSCAN algorithm, is employed to segment point cloud data into objects. Through neighborhood search, points with a neighborhood distance of less than 0.02m and a neighborhood of ≥10 points are clustered into one class and stored separately. For example... Figure 14 As shown in the figure, the point cloud enclosed by each cube is classified as a separate category, dividing a single point cloud data into N objects;

[0082] ③ Trunk-like object extraction: Obtain the AABB and OBB bounding boxes of the point cloud object. If the bounding box attributes simultaneously meet the following three conditions, it is identified as a trunk-like object; otherwise, it is discarded. Figure 15-16 As shown.

[0083] ①The AABB box is a vertical cuboid;

[0084] ② The volume of the OBB box is > 0.0005m³ 3 ;

[0085] ③ The height of the OOB box is greater than 0.3m.

[0086] ④ Projection Denoising: Project the point cloud object onto a plane perpendicular to the central axis of the OOB bounding box, such as... Figure 17-18 As shown; the point cloud distribution and two-dimensional centroid coordinates (xc, yc) are calculated. The Mahalanobis distance from each point to the distribution center is calculated, and filtering and noise reduction are performed based on three times the standard deviation. A circular neighborhood with a radius of 0.05 cm is used for point-by-point traversal. Center points with fewer than 10 points within the neighborhood are removed as noise to reduce the point cloud noise caused by fallen bark and branches. Figures 19-20 As shown.

[0087] S6: Traverse the object to fit a cylinder;

[0088] A cylinder is fitted based on the denoised point cloud data, such as Figure 21-22 As shown, the axial direction, center point, cylinder radius, and fitting error data of the cylinder are obtained. The point cloud fitting effect evaluation index is calculated based on the distance from the point cloud data to the cylinder axis and the cylinder radius. The index includes goodness of fit, root mean square error, and estimation accuracy.

[0089] S7: Recognition and extraction of curved surface objects;

[0090] First, a batch of training data is created using a manual judgment method. Each data point includes information on whether it is a curved surface object and evaluation index information on the cylinder fitting effect. A random forest classifier is trained using the data. Then, the goodness of fit, root mean square error, and estimation accuracy obtained after fitting the cylinder for each object in S6 are used as independent variables and input into the trained random forest classifier to determine whether the object is a curved surface object. If it is, it is kept; otherwise, it is deleted, resulting in a curved surface object file.

[0091] S8: Factor extraction calculation;

[0092] By traversing the arc-shaped object files in S7 and the fitted cylinders in S6, the diameter at breast height (DBH) and cross-sectional area (g) of the corresponding sample trees are calculated. Based on the DBH and the number of arc-shaped objects, the number of trees per unit area of ​​the stand can be extracted. Based on the relative position of the station and the point cloud objects and the DBH of the sample trees, the horizontal distance (Dist) from the station to the DBH position of the sample trees can be calculated, as shown in Table 1.

[0093] Table 1. Data Table for Arc-shaped Surface Objects

[0094] file name Dist(m) DBH(cm) 1 / g Cluster_Dbscan_1.xyz 13.8922 7.3 238.93 Cluster_Dbscan_10.xyz 11.0372 17.4 42.05 Cluster_Dbscan_100.xyz 7.3952 18.8 36.02 Cluster_Dbscan_101.xyz 7.6407 8.8 164.42 …… … … …

[0095] S9: Forest Stock Volume Measurement:

[0096] Using the factors obtained from S8, the volume per unit area of ​​the forest stand was calculated using dendrometric theory, as shown in Table 2.

[0097] Table 2 Data Table for Accumulation Measurement

[0098]

[0099]

[0100] Measurements showed that the measured volume per hectare in the artificially surveyed sample plots in this area was 203.44 m³. 3 Using this method to extract forest stock measurement factors and perform calculations, the forest stand stock per hectare was found to be 210.78 m³. 3 The error relative to the measured value is 3.61%. This demonstrates that the forest stock volume calculated using this method is accurate, simple to operate, requires minimal workload, and has excellent application prospects.

[0101] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for rapidly determining forest stock volume based on ground-based lidar data, characterized in that, Includes the following steps: S1. Obtain raw point cloud data; S2, Solve the original point cloud data; S3, Data Preprocessing; S4, Point Cloud Data Extraction; S5. Tree-like object segmentation and recognition; S6. Traverse the object to fit a cylinder; S7. Recognition and extraction of curved surface objects; S8. Extraction of factors for forest stock volume measurement; S9. Forest stock volume calculation; The method for step S7 is as follows: First, a batch of training data is created by manual judgment. Each data point contains information on whether it is a curved surface object and information on the evaluation index of cylinder fitting effect. The random forest classifier is trained using the data. Then, the goodness of fit, root mean square error, and estimation accuracy obtained after fitting the cylinder for each object in S6 are used as independent variables and input into the trained random forest classifier to determine whether the object is a curved surface object. If it is, it is kept; otherwise, it is deleted, and a curved surface object file is obtained. The method for step S8 is as follows: Based on the arc-shaped object obtained in S7, and the cylinder fitted in S6, calculate the diameter at breast height (DBH) and cross-sectional area of ​​the sample tree corresponding to the object. Then, based on the DBH of the sample tree and the number of arc-shaped objects, extract the number of trees per unit area of ​​the forest stand. Based on the relative position of the station and the point cloud object and the DBH of the sample tree, calculate the horizontal distance from the station to the DBH position of the sample tree.

2. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 1, characterized in that, The method for step S1 is as follows: Select the area to be measured, and use a ground-based lidar scanning device to perform a fixed-point panoramic scan of the forest to obtain raw point cloud data.

3. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 2, characterized in that, Step S2 involves exporting raw point cloud data from the lidar scanning device, performing data calculation and enhancement processing using post-processing software, and obtaining point cloud result data for the target area.

4. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 3, characterized in that, In step S3, LiDAR 360 is used to preprocess LiDAR point cloud data.

5. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 4, characterized in that, The method for step S4 is as follows: extract a subset of the preprocessed point cloud data from a distance of 1.3m ± 0.5m and use it as input data.

6. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 5, characterized in that: In step S5, mathematical statistics methods are used, and the algorithm is written in Python to achieve point cloud object segmentation.

7. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 6, characterized in that, Step S6 is as follows: Fit a cylinder based on the denoised point cloud data to obtain the cylinder's axial direction, center point, cylinder radius, and fitting error data. Calculate the point cloud fitting effect evaluation index based on the distance from the point cloud data to the cylinder's axis and the cylinder's radius. The index includes goodness of fit, root mean square error, and estimation accuracy.

8. The method for rapidly determining forest stock volume based on ground-based lidar data according to claim 1, characterized in that, The method for step S9 is as follows: using the factors obtained in S8, calculate the forest stock volume through dendrometric theory.

Images