An automatic forest land identification and classification method, device and storage device

Abstract

The present invention discloses a method, device and storage device for automatic identification and classification of forest land. The method is as follows: Obtain true orthophoto images and high-density LiDAR point cloud data, and perform preprocessing to obtain preprocessed data; Select point clouds from the preprocessed data to obtain two-dimensional canopy point cloud data; Calculate the canopy density of the two-dimensional canopy point cloud data, and automatically identify forest land according to the canopy density; Perform attribute annotation on the forest land; Complete topological inspection and data correction. The beneficial effects of the present invention are: The present invention comprehensively utilizes the spatial proximity features of remote sensing images and high-density LiDAR point cloud data, filters out the point cloud data of non-target land types, realizes the rapid, accurate and intelligent identification and extraction of forest land, effectively saves the information extraction time, and improves work efficiency; The implementation steps of the present invention have the characteristics of simplicity, operability and low manual participation, providing new technical support for the development and protection, resource monitoring and resource management of forest land resources.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a method, device and storage device for automatically identifying and classifying forest land. Background Art

[0002] As a new technical means for obtaining three-dimensional spatial data, airborne LiDAR technology has the advantages of high precision, high density, all-weather, short operation cycle, etc., and is widely used in fields such as urban planning, emergency support, resource monitoring, etc. At present, the research on airborne LiDAR point cloud data mainly focuses on aspects such as ground filtering, scene segmentation, spatial clustering, and feature extraction.

[0003] With the continuous advancement of China's ecological civilization construction and the rapid development of remote sensing technology, it is becoming increasingly important to study the automated classification method of forest land LiDAR point clouds in forest land resource investigation, monitoring, protection, planning, ecological restoration, and optimizing the pattern of national land space development.

[0004] Most of the existing methods for identifying forest land types are based on spectral features and require manual extraction and intervention, resulting in low efficiency. Based on this, there is an urgent need to provide a method and system for identifying forest land based on remote sensing technology and LiDAR point cloud data to solve the problems in the prior art. Summary of the Invention

[0005] In order to solve the technical problems of low efficiency and non-automation in identifying forest land types, the present invention provides a method, device and storage device for automatically identifying and classifying forest land, and the method specifically includes the following steps:

[0006] S1. Obtain a true orthophoto image and high-density LiDAR point cloud data, and perform preprocessing to obtain preprocessed data;

[0007] S2. Select point clouds from the preprocessed data to obtain two-dimensional point cloud data of the forest canopy layer;

[0008] S3. Calculate the canopy density of the two-dimensional point cloud data of the forest canopy layer, and automatically identify forest land according to the canopy density;

[0009] S4. Perform attribute annotation on the forest land;

[0010] S5. Complete topological inspection and data correction.

[0011] A storage device stores instructions and data for implementing a method for automatically identifying and classifying forest land.

[0012] A device for automatically identifying and classifying forest land includes: a processor and a storage device; the processor loads and executes the instructions and data in the storage device for implementing a method for automatically identifying and classifying forest land.

[0013] The beneficial effects provided by the present invention are as follows: By comprehensively utilizing the spatial proximity features of remote sensing images and high-density LiDAR point cloud data, the point cloud data of non-target land types is filtered, enabling the rapid, accurate, and intelligent identification and extraction of forest land, effectively saving the time for information extraction and improving work efficiency; The implementation steps of the present invention are characterized by simplicity, operability, and low manual participation, providing new technical support for the development and protection, resource monitoring, and resource management of forest land resources. Description of the Drawings

[0014] Figure 1 is the flowchart of the method of the present invention;

[0015] Figure 2 is the working diagram of the hardware device of the present invention. Detailed Embodiments

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0017] Please refer to Figure 1 , Figure 1 which is the flowchart of the method of the present invention.

[0018] The present invention provides a method, device, and storage device for automatic identification and classification of forest land. Specifically, the method includes the following steps:

[0019] S1. Obtain an orthorectified image and high-density LiDAR point cloud data, and perform preprocessing to obtain preprocessed data;

[0020] Step S1 is specifically as follows:

[0021] S11. Obtain an orthorectified image (TDOM) and high-density LiDAR point cloud data with a resolution greater than a first preset value, and perform data denoising;

[0022] S12. Input the denoised data into the reclassification unit of the spatial analysis tool to extract the polygon of the vegetation coverage area;

[0023] S13. Create a new layer, and add the field attributes of the forest land type LDLX and the field attributes of the average height PJGD of the forest land.

[0024] As an embodiment, in the present invention, an orthorectified image (TDOM) and high-density LiDAR point cloud data with a resolution better than 5CM are prepared, and data preprocessing is performed, including point cloud data denoising, etc.;

[0025] Import the true orthogonal image (TDOM) into ArcGIS software and use the "Reclassify" function in the "Spatial Analyst Tools" to extract the polygons of vegetation coverage;

[0026] Create a new layer in ArcGIS software, import polygons into the layer, open the attribute table, and add the forest type field "LDLX" and the forest average height field "PJGD".

[0027] S2, performing point cloud selection on the preprocessed data to obtain two-dimensional point cloud data of the canopy layer;

[0028] The specific process of point cloud selection in step S2 is as follows:

[0029] S21. Use the function of classifying ground points to separate ground points and non-ground points from the denoised high-density LiDAR point cloud data;

[0030] S22, preset a two-dimensional grid G of a specified size of m1*m1 within the polygon of the vegetation coverage area i , i is the grid number, i = 1, 2, ..., n; calculate the difference between the highest point height value and the lowest point height value of the point cloud data in the vertical space of each grid h1, h2, ..., h n Calculate the average height of the height difference H1 = (h1 + h2 + ... + h n ) / n, if H1 is greater than or equal to H max , the vegetation range polygon is automatically classified as the initially screened candidate forest land range, and the candidate forest land is marked in the field LDLX, otherwise it is classified as non-forest land; where H max is the default value;

[0031] S23: Within the vertical space of the candidate forest land, a designated space is delineated; and within the designated space, a two-dimensional grid G is screened according to the point cloud quantity, point cloud height, and point cloud uniformity that meet the multi-dimensional standards at the same time. j Reserved, j is the serial number of the two-dimensional grid that is retained; the two-dimensional grid G j The size is divided into m2*m2, m2 <m1;

[0032] S24, choose any one of the remaining two-dimensional grids G j , with G j As the center, R1 is the radius, and search whether there are other preserved two-dimensional grids within the radius. If so, merge the two-dimensional grids within the range into one group as the trunk plane range;

[0033] S25. Set up a two-dimensional grid G of m3 * m3 within the range of the candidate forest land k , search for the point cloud data with a height greater than the preset value within the range of R2 radius around each tree trunk, and project it vertically onto the two-dimensional grid G k , to obtain the two-dimensional point cloud data of the forest canopy layer; where m3 < m2.

[0034] As an embodiment, the selection of the point cloud includes the following parts:

[0035] 1. Separate ground points and non-ground points.

[0036] Import the denoised high-density LiDAR point cloud data into the LiDAR360 software, and use the "ground point classification" function in "classification" to separate the point cloud data of ground points and non-ground points;

[0037] 2. Preliminary classification of forest land by range method. Set up a two-dimensional grid G1 of m1 * m1 (the default constraint value of m1 is set to 2m) within the vegetation range polygon, calculate the extreme height difference of the point cloud data within the vertical space of each grid, that is, the difference between the highest point height value and the lowest point height value of the point cloud data. Calculate the extreme height differences h1, h2,..., h of all grid point clouds by the extreme value method n , calculate the average height H1 of the height differences = (h1 + h2 +... + h n ) / n. If H1 is greater than or equal to H max (H max The default constraint value is set to 3m), then automatically classify the vegetation range polygon as the candidate forest land range initially screened out, and mark "candidate forest land" in the field "LDLX", otherwise classify it as non-forest land;

[0038] 3. Select the point cloud with tree trunk characteristics. In the vertical space where the above candidate forest land range is located, select the interval [H2, H3] (the default constraint value of H2 is set to 2m, and the default constraint value of H3 is set to 3m) above the lowest point cloud height value (ground point), and divide it into S1 - S5 segments (the default constraint value is set to 5 segments) evenly by height. Divide the candidate forest land range into a two-dimensional grid G2 of m2 * m2 (the default constraint value of m2 is set to 5cm). Set that if the point cloud data in the grid G2 simultaneously meets the following conditions, then select and retain this grid:

[0039] (1) When the point cloud data is projected vertically into G2, if the number of point clouds is greater than or equal to n2 (the default constraint value is set to 20), then retain this G2 grid;

[0040] (2) Retain the G2 grid where the point cloud height values are distributed in each of the S1 - S5 height segments;

[0041] (3) Calculate and check the uniformity of the point cloud height within each grid. Set the L k value as the standard value for measuring the uniformity of the point cloud height, and the calculation rule of the L k value is as follows:

[0042] ① Let the height values of N point clouds within the G2 grid be L1, L2, …, L n , and the average height value be M0; the L k value divides the height values of these valid point clouds into two categories: C L and C H (low value and high value); C L represents the point clouds with height values [L1, L2, …, L k , and C H represents the point clouds with height values [L k+1 , …, L n . The total numbers of these two types of point clouds are N L and N H respectively, and the average heights are M L and M H .

[0043] Then, the proportion, average height value, and total height average value of each category are given by the following formulas respectively:

[0044] P L = N L / N (1)

[0045] P H = 1 - P L = N H / N (2)

[0046]

[0047]

[0048]

[0049] Formulas (1) to (5) can all be calculated by the program.

[0050] ② The degree of difference δ 2 = P L (M L - M0) 2 + P H (M H - M0) 2 (6)

[0051] ③ Traverse all the height values of L1, L2, …, L n , and the program finds the height value L k that maximizes formula (6).If L is satisfied k Located in the middle section S3 of the height section (the default constraint value is set to the middle section), that is, the G2 grid is retained.

[0052] 4. Grid merging and grouping. After the above screening, randomly select a G2 grid, and sequentially query the G2 grids within the range of radius R1 (the default value of R1 is set to 1m) clockwise in a loop. Merge the grids that meet the conditions into a group in sequence. If no grid is found within this range, the program automatically terminates. The range of the G2 grids after merging and grouping is regarded as the trunk plane range.

[0053] 5. Canopy point cloud extraction. Set a two-dimensional grid G3 of m3*m3 (the default constraint value of m3 is set to 1cm) within the range of the candidate forest land. Search for the point clouds with a height greater than the H3 height value (the default constraint value of H3 is set to 3m) set in "Step 3 of Step 2" within the range of radius R2 (the default value of R2 is set to 1m) around each trunk, and project them vertically onto the grid G3 to obtain the two-dimensional canopy point cloud data.

[0054] S3. Calculate the canopy density of the two-dimensional canopy point cloud data, and automatically identify forest land according to the canopy density;

[0055] Specifically, step S3 is as follows:

[0056] S31. According to the proportion β of the two-dimensional canopy point cloud data in each two-dimensional grid G k in, assign a value to the two-dimensional grid G k ; if β is greater than or equal to α, then the two-dimensional grid G k is assigned a value of 1, otherwise 0; where α is a preset value;

[0057] S32. Calculate the canopy density according to the formula , where CC represents the canopy density of the forest land to be calculated, N C1 is the area of the grid assigned a value of 1, and S is the area of the candidate forest land range;

[0058] S33. Identify forest land according to the value of CC. If CC is greater than or equal to the preset threshold, it is automatically classified as the forest land range, otherwise it is the non-forest land range.

[0059] Canopy density refers to the degree to which the tree crowns in the forest cover the ground, and is often expressed as the ratio of the vertical projection area of the canopy to the forest land area. According to the "Classification of Current Land Use" (GB / T 21010-2017), forest land refers to tree land with a canopy density greater than or equal to 0.2, including mangrove land and bamboo forest land.

[0060] As an example, the canopy density is calculated as follows:

[0061] 1. Grid network assignment. According to the proportion of the two-dimensional point cloud data of the forest canopy in each grid network G3, assign "1" or "0" to each grid; where "1" represents that the proportion of the two-dimensional point cloud in this grid is greater than or equal to α, and "0" represents that the proportion of the two-dimensional point cloud in this grid <α (the default value of α is set to 50%);

[0062] 2. Crown density calculation. Calculate the crown density of this forest land according to the following formula In the formula, CC represents the crown density of the forest land to be calculated, and N C1 is the area of the grid network assigned "1", and S is the area of the range of forest land to be candidate;

[0063] 3. Forest land identification and classification. According to the calculation result of the crown density, if CC is greater than or equal to 0.2, then automatically classify the polygon of the range of forest land to be candidate as the polygon of the range of forest land; if CC <0.2, classify it as a polygon of the range of non-forest land and delete it.

[0064] S4. Perform attribute annotation on the forest land; the specific process is as follows:

[0065] 1. Annotate the attribute of "forest land" in the attribute field "LDLX" of the polygon of the range of forest land obtained in the above steps;

[0066] 2. Calculate the maximum value and the minimum value of all the point clouds in the vertical space where the polygon of the range of forest land is located. Calculate the heights t1, t2,..., tn of all the trees by the extreme value method. According to the estimated number of trees N_total obtained by the number of merged groups in "Step 4 of Step 2", calculate the average height H_forest of the polygon of the range of forest land = (t1 + t2 +... + tn) / N_total, and annotate it in the attribute field "PJGD".

[0067] S5. Complete topological inspection and data correction.

[0068] Under the directory folder of the ArcGIS software, create a "personal GeoDatabase", create a new dataset "dataset", import the polygon elements of the range of forest land into the dataset "inport--feature class single", create a new topology polygon, add topological processing rules, and perform topological analysis. Check each topological error one by one until the data is corrected without error.

[0069] Please refer to Figure 2 , Figure 2 is a schematic diagram of the working of the hardware device of the embodiment of the present invention. The hardware device specifically includes: a forest land automatic identification and classification device 401, a processor 402, and a storage device 403.

[0070] An automatic forest land identification and classification device 401: The automatic forest land identification and classification device 401 implements the automatic forest land identification and classification method.

[0071] Processor 402: The processor 402 loads and executes the instructions and data in the storage device 403 to implement the automatic forest land identification and classification method.

[0072] Storage device 403: The storage device 403 stores instructions and data; the storage device 403 is used to implement the automatic forest land identification and classification method.

[0073] The beneficial effects of the present invention are as follows: By comprehensively utilizing the spatial proximity features of remote sensing images and high-density LiDAR point cloud data, the point cloud data of non-target land types is filtered, realizing the rapid, accurate, and intelligent identification and extraction of forest land, effectively saving the information extraction time and improving work efficiency; The execution steps of the present invention are characterized by simplicity, operability, and low human participation, providing new technical support for the development and protection, resource monitoring, and resource management of forest land resources.

[0074] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An automatic recognition and classification method for forest land, characterized in that: Including the following steps: S1. Obtain true orthophoto images and high-density LiDAR point cloud data, and perform preprocessing to obtain preprocessed data; The specific steps of step S1 are as follows: S11. Obtain true orthophoto images and high-density LiDAR point cloud data with a resolution greater than the first preset value, and perform data denoising; S12. Input the denoised data into the reclassification unit of the spatial analysis tool to extract the polygons of the vegetation coverage area; S13. Create a new layer, and add the field attributes of the forest land type LDLX and the field attributes of the average height PJGD of the forest land; S2. Select point clouds from the preprocessed data to obtain the two-dimensional point cloud data of the forest canopy layer; The specific process of point cloud selection in step S2 is as follows: S21. Use the function of classifying ground points to separate ground points and non-ground points from the denoised high-density LiDAR point cloud data; S22. Preset a two-dimensional grid G with a specified size of m1*m1 within the polygon of the vegetation coverage range i , where i is the serial number of the grid, i = 1, 2,..., n; calculate the difference h1, h2,..., h between the highest point height value and the lowest point height value of the point cloud data within the vertical space of each grid n ; calculate the average height H1 = (h1 + h2 +... + h n ) / n. If H1 is greater than or equal to H max , then automatically classify the polygon of this vegetation range as the candidate forest land range initially screened out, and mark the candidate forest land in the field LDLX. Otherwise, classify it as non-forest land; where H max is a preset value; S23: Define a specified space within the vertical space range of the candidate forest land area; and within the specified space, filter the two-dimensional grid G within the specified space according to the multi-dimensional simultaneous satisfaction criteria of the point cloud quantity, point cloud height, and point cloud uniformity j Keep it, where j is the serial number of the remaining two-dimensional grid; among them, the size of the two-dimensional grid G j is divided into m2*m2, where m2 < m1; S24. Optionally select one of the remaining two-dimensional grid networks G j , and with G j as the center and R1 as the radius, search whether there are other remaining two-dimensional grid networks within its radius. If so, merge the two-dimensional grid networks within the range into a group as the trunk plane range; S25. Set up a two-dimensional grid G of m3 * m3 within the candidate forest land area k , search for the point cloud data with a height greater than the preset value within the range of radius R2 around each tree trunk, and vertically project it onto the two-dimensional grid G k to obtain the two-dimensional point cloud data of the forest canopy layer; where m3 < m2; S3. Calculate the canopy density of the two-dimensional point cloud data of the forest canopy layer, and automatically identify forest land according to the canopy density; S4. Perform attribute annotation on the forest land; S5. Complete topological inspection and data correction.

2. The automatic forest land identification and classification method according to claim 1, characterized in that: The specific content of step S3 is: S31, based on the two-dimensional point cloud data of the canopy layer, in each two-dimensional grid G k The proportion β in the two-dimensional grid G k Assign a value; if β is greater than or equal to α, then the two-dimensional grid G k The value is assigned as 1, otherwise it is assigned as 0; where α is the preset value; S32. Calculate the canopy density according to the formula where CC represents the canopy density of the forest land to be calculated, N C1 is the area of the grid assigned a value of 1, and S is the area of the candidate forest land range; S33. Identify forest land according to the value of CC. If CC is greater than or equal to the preset threshold, it is automatically classified as the forest land range; otherwise, it is the non-forest land range.

3. The automatic forest land identification and classification method according to claim 1, characterized in that: In step S23, according to the point cloud quantity standard, the two-dimensional grid G within the specified space is screened j The condition is: when the point cloud data is vertically projected onto G j if the number of projected point clouds is greater than or equal to n2, then G j is retained.

4. The automatic forest land identification and classification method according to claim 1, wherein: In step S23, according to the point cloud height standard, the two-dimensional grid G within the specified space is screened j The condition is that if the point cloud height value is within the preset range, then G j is retained.

5. The automatic recognition and classification method of forest land according to claim 1, characterized in that: In step S23, according to the point cloud uniformity standard, the two-dimensional grid G within the specified space is screened j The condition is: set L k The value is the standard value for measuring the height uniformity of the point cloud. According to L k Set a threshold. If the threshold condition is met, then G j Is retained.

6. A storage device, characterized in that: The storage device stores instructions and data for implementing a method for automatically identifying and classifying forest land according to any one of claims 1 to 5.

7. An automatic forest land identification and classification device, characterized in that: Including: A processor and a storage device; the processor loads and executes the instructions and data in the storage device for implementing a method for automatically identifying and classifying forest land according to any one of claims 1 to 5.

Images