Method for acquiring key attributes of canopy trees based on oblique photography of unmanned aerial vehicle

Abstract

A method for obtaining key attributes of canopy trees based on unmanned aerial vehicle oblique photography relates to the field of image data processing, and comprises the following steps: S1, unmanned aerial vehicle oblique photography, S2, DSM data splicing, S3, DEM data obtaining, S4, data processing, and S5, forest key factor extraction. According to the method, a large area can be covered in a short time, the image data of the forest trees can be rapidly collected, indoor data analysis is carried out, key attribute elements of the trees are obtained, the cost is only one tenth of that of traditional manual investigation, and almost no personnel safety risk exists. And simultaneously. Position signals are obtained in the air, the positioning precision is high, the tree tip position is recognized through an algorithm, the geographic coordinates and the height of the tree are obtained, and the data precision is high; image data are collected through different angles such as inclination, the tree height and crown breadth are identified through an algorithm, manual intervention is greatly reduced, and the investigation efficiency and precision are improved.

Description

Technical Field

[0001] The present invention relates to the field of image data processing, and in particular to a method for obtaining key attributes of canopy trees based on unmanned aerial vehicle (UAV) oblique photography. Background Art

[0002] Tree attributes such as geographical coordinates, tree height, and crown width are key elements necessary for forestry production. The traditional method of obtaining the above-mentioned tree attributes can only be done through manual single-tree surveys, but the manual single-tree survey method has the following defects: First, the traditional manual collection of tree geographical coordinates, tree height, crown width and other attributes requires deep forest area to collect each tree one by one, which has high time cost, labor cost and safety risk; Second, due to the obstruction of vegetation in the forest and poor positioning signals of GPS and other equipment, the manual positioning of single-tree geographical coordinates is slow and accurate, and can only be located around the tree body. The coordinates are offset, which is not conducive to the precise management needs of forestry; Third, due to the closure of the forest, manual surveys are often difficult to accurately and quickly identify the highest point of tree growth and the edge of the crown. The speed of collecting tree height and crown width attributes is slow and the accuracy is low. The efficiency and accuracy of data surveys are significantly related to the experience of investigators, and there is a lot of manual intervention and the intensity of survey work is high. Based on this, under the premise of ensuring personnel safety, in order to greatly improve the accuracy and efficiency of the investigation of forest attributes and complete the collection of forest information, the applicant has conducted intensive research and thus came up with this case. Summary of the invention

[0003] The present invention provides a method for obtaining key attributes of canopy trees based on oblique photography by unmanned aerial vehicles. The main purpose of the method is to overcome the technical problems mentioned above caused by the manual collection of existing tree attributes such as geographical coordinates, tree height, and canopy width.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A method for obtaining key attributes of canopy trees based on drone oblique photography, comprising the following steps:

[0006] S1. Drone oblique photography: Plan the route for the survey area, set the gimbal pitch angle to 20-45°, set the route according to forward / rearward or left / right vision, and conduct drone oblique photography of the survey area;

[0007] S2, stitching DSM data: the images taken by the drone are processed into DSM data through oblique photography processing software;

[0008] S3. Obtain DEM data: interpolate the DSM data generated by drone oblique photography at locations with obvious lack of vegetation (such as forest windows, forest roads, etc.) to generate DEM data;

[0009] S4, data processing: adjust the resolution of DSM and DEM data to 1 / 10-3 / 10 of the average tree crown width to reduce the error rate of tree identification and improve calculation efficiency, and subtract DSM from DEM data to obtain vegetation net height data;

[0010] S5. Extract key factors of trees: Generate contour lines according to the vegetation net height data. The highest point in the contour line area is regarded as the top of the tree, that is, the geographical coordinates of the tree, and the value is the tree height. The lowest point in the contour line area is regarded as the edge of the tree crown, and the lowest point around the tree top is the tree crown width.

[0011] Furthermore, in step S2, the process of splicing DSM data is as follows:

[0012] S21. Check the data taken by the drone to ensure that the geographical location information taken is accurate;

[0013] S22, importing the checked shooting data into a general 3D modeling software;

[0014] S23, point cloud registration, aligning the point cloud data in photos taken at different angles to form a complete point cloud model;

[0015] S24. Generate DSM data using point cloud data through TIN triangulation model.

[0016] Furthermore, in step S3, the process of obtaining DEM data is as follows:

[0017] S31. Determine the vegetation-free area: The areas with obvious vegetation-free areas (such as forest windows, forest roads, etc.) or the areas with clear vegetation height in the DSM data generated by the drone oblique photography minus the vegetation height are regarded as the vegetation-free areas in the DSM. Assume that we define a threshold h, that is, the area with height h is considered to be the vegetation-free area;

[0018] Use a binary mask M to represent the vegetation-free area:

[0019]

[0020] Where (x, y) represents the coordinate position in the DSM data;

[0021] S32, the height of the vegetation-free area is regarded as the ground height: for the area where the mask M(x, y) = 1, the height value of the DSM is directly regarded as the height value of the DEM;

[0022] Generate initial DEM data DEMinit(x,y):

[0023] DEMinit(x,y)=M(x,y)·DSM(x,y);

[0024] S33. Interpolation generates complete DEM data: For the area where M(x,y)=0, the DEM height value is estimated by the interpolation algorithm:

[0025]

[0026] Among them, Interp(DEM init (x′, y′)) represents the interpolation estimation of the adjacent DEM values, specifically: find the nearest point (x', y') where M(x', y') = 1, and interpolate based on the height values ​​of these points.

[0027] Furthermore, in step S4, the data processing process is as follows:

[0028] S41. Define input and output grids:

[0029] Input raster data: Raster input (i,j), where (i,j) represents the pixel position of the input raster data,

[0030] Output raster data: Raster output (m,n), where (m,n) represents the pixel position of the output raster data,

[0031] Resolution of input raster: R input ,

[0032] Output raster resolution: R output =M·R input ,

[0033] M is the target resolution for adjustment;

[0034] S42, interpolation adjustment to target resolution:

[0035] The output grid position (m,n) corresponds to the input grid position (i,j):

[0036]

[0037]

[0038] The data at position (m,n) is interpolated using the nearest neighbor method according to the position value at (i,j):

[0039] Raster output (m, n) = Raster output (i, j);

[0040] S43, obtaining vegetation net height data DSM0: subtracting the DSM and DEM raster data adjusted to the same resolution to obtain vegetation net height data DSM0:

[0041] DSM0=DSM-DEM.

[0042] Furthermore, in step S5, the data processing process is as follows: the DEM0 data generated in S43 is used as a two-dimensional matrix DEM (i, j), where (i, j) represents the coordinates in the grid, h (i, j) represents the elevation value of the point, the neighborhood is defined as a 3×3 window, and the treetop position (tree geographic coordinates), tree height elevation value (tree height), crown edge position and crown area (crown width) are marked;

[0043] S51. Find the tree height:

[0044] 1) Initialize the highest point position set:

[0045]

[0046] 2) Traverse each point in the DEM data:

[0047] For each (i, j) in DEM;

[0048] 3) Get the neighborhood of point (i, j):

[0049] Neighbors(i,j)={(i′,j′)|i-1≤i′≤i+1, j-1≤j′≤j+1, (i′,j′)≠(i,j)}

[0050] 4) Determine whether it is the tree top position:

[0051] if h(i,j)>h(i′,j′)for all(i′,j′)∈Neighbors(i,j)then

[0052] Peaks = Peaks∪{(i, j)};

[0053] S52. Find the edge of the tree crown:

[0054] 1) Initialize the lowest point position set:

[0055]

[0056] 2) Traverse each point in the DEM data:

[0057] For each (i, j) in DEM;

[0058] 3) Get the neighborhood of point (i, j):

[0059] Neighbors(i,j)={(i′,j′)|i-1≤i′≤i+1, j-1≤j′≤j+1, (i′,j′)≠(i,j)};

[0060] 4) Determine whether it is the tree top position:

[0061] if h(i,j)× <h(i′,j′)for all(i′,j′)∈Neighbors(i,j)then

[0062] Valleys=Valleys∪{(i,j)}

[0063] S53. Mark tree height, crown edge, and crown width:

[0064] 1) Initialize the tag set:

[0065]

[0066] 2) Traverse each point in the highest point position:

[0067] For each (i, j) in Peaks;

[0068] 3) Find the nearest lowest point in the X direction:

[0069]

[0070]

[0071] 4) Find the nearest lowest point in the Y direction:

[0072]

[0073]

[0074] 5) Mark the highest point and the edge of the crown:

[0075] MarkedPeaks=MarkedPeaks∪{i,j,h(i,j),Valley left , Valley right , Valley up , Valley down)};

[0076] 6) Calculate the crown area:

[0077] Long axis:

[0078] a=distance(Valley left , Valley right );

[0079] Short Axis:

[0080] b=distance(Valley up , Valley down )

[0081] Area=0.25πa b.

[0082] Furthermore, in step S4, the resolution of DSM and DEM data is adjusted to 1 / 5 of the average tree crown width.

[0083] It can be seen from the above description of the present invention that, compared with the prior art, the present invention has the following advantages:

[0084] 1. The present invention can cover a large area in a short time, quickly collect forest image data, conduct indoor data analysis, and obtain key attribute elements of trees. The cost is only a few tenths of traditional manual surveys, and there is almost no risk to personnel safety.

[0085] 2. The present invention obtains position signals in the air with accurate positioning accuracy, identifies the top positions of trees through algorithms, obtains the geographical coordinates and tree heights of trees with high data accuracy; collects image data at different angles such as inclination, identifies tree heights and crown widths through algorithms, greatly reduces manual intervention and improves the efficiency and accuracy of the survey. DETAILED DESCRIPTION

[0086] A method for obtaining key attributes of canopy trees based on drone oblique photography, comprising the following steps:

[0087] S1. Drone oblique photography: Plan the route for the survey area, set the gimbal pitch angle to 20-45°, set the route according to forward / rearward or left / right vision, and conduct drone oblique photography of the survey area;

[0088] S2, stitching DSM (digital surface model) data: the images taken by the drone are processed into DSM data through oblique photography processing software;

[0089] S3. Obtain DEM (digital elevation model) data: interpolate the DSM data generated by drone oblique photography at locations with obvious lack of vegetation (such as forest windows, forest roads, etc.) to generate DEM data;

[0090] S4, data processing: adjust the resolution of DSM and DEM data to 1 / 10-3 / 10 of the average tree crown width to reduce the error rate of tree identification and improve calculation efficiency, and subtract DSM from DEM data to obtain vegetation net height data;

[0091] S5. Extract key factors of trees: Generate contour lines according to the vegetation net height data. The highest point in the contour line area is regarded as the top of the tree, that is, the geographical coordinates of the tree, and the value is the tree height. The lowest point in the contour line area is regarded as the edge of the tree crown, and the lowest point around the tree top is the tree crown width.

[0092] In step S1, when the drone plans the route for oblique photography, it only sets the route according to the forward / rearward view or the left / right view, and determines that the pitch angle of the gimbal is within the range of 20 to 45 degrees. In this way, image data can be collected at different angles, and the workload of data collection and post-processing is only 40% of the workload of traditional oblique photography. While meeting the data acquisition accuracy, the work efficiency is greatly improved.

[0093] Specifically, in step S2, the process of splicing DSM data is as follows:

[0094] S21. Check the data taken by the drone to ensure that the geographical location information taken is accurate;

[0095] S22. Import the checked shooting data into common 3D modeling software (such as Pix4D, ContextCapture, Agisoft Metashape, DJI Zhitu, etc.);

[0096] S23, point cloud registration, aligning the point cloud data in photos taken at different angles to form a complete point cloud model;

[0097] S24. Generate DSM data using point cloud data through TIN triangulation model.

[0098] Specifically, in step S3, the process of obtaining DEM data is as follows:

[0099] S31. Determine the vegetation-free area: The areas with obvious vegetation-free areas (such as forest windows, forest roads, etc.) or the areas with clear vegetation height in the DSM data generated by the drone oblique photography minus the vegetation height are regarded as the vegetation-free areas in the DSM. Assume that we define a threshold h, that is, the area with height h is considered to be the vegetation-free area;

[0100] Use a binary mask M to represent the vegetation-free area:

[0101]

[0102] Where (x, y) represents the coordinate position in the DSM data;

[0103] S32, the height of the vegetation-free area is regarded as the ground height: for the area where the mask M(x, y) = 1, the height value of the DSM is directly regarded as the height value of the DEM;

[0104] Generate initial DEM data DEMinit(x, y):

[0105] DEMinit(x,y)=M(x,y)·DSM(x,y);

[0106] S33, interpolation to generate complete DEM data: For the area where M(x, y) = 0, interpolation is performed

[0107] The algorithm estimates the DEM height value:

[0108]

[0109] Among them, Interp(DEM init (x', y')) represents the interpolation estimation of the adjacent DEM values, specifically: find the nearest point (x', y') where M(x', y') = 1, and interpolate according to the height values ​​of these points.

[0110] Specifically, in step S4, the data processing process is as follows:

[0111] S41. Define input and output grids:

[0112] Input raster data: Raster input (i, j), where (i, j) represents the pixel position of the input raster data,

[0113] Output raster data: Raster output (m, n), where (m, n) represents the pixel position of the output raster data,

[0114] Resolution of input raster: R input ,

[0115] Output raster resolution: R output =M·R input ,

[0116] M is the target resolution for adjustment;

[0117] S42, interpolation adjustment to target resolution:

[0118] Output grid position (m, n) corresponds to input grid position (i, j):

[0119]

[0120]

[0121] The data at position (m, n) is interpolated using the nearest neighbor method according to the position value at (i, j):

[0122] Rasteroutput (m, n) = Raster output (i, j);

[0123] S43, obtaining vegetation net height data DSM0: subtracting the DSM and DEM raster data adjusted to the same resolution to obtain vegetation net height data DSM0:

[0124] DSM0=DSM-DEM.

[0125] When processing data, the DEM and DSM data were preprocessed, and the grid resolution was adjusted to 1 / 5 of the average canopy width of the trees in the survey area. Adjusting the DEM and DSM data to the same resolution can greatly reduce the workload of data calculation and improve data processing efficiency. Adjusting the resolution to 1 / 5 of the average canopy width can significantly reduce the misidentification rate of evergreen broad-leaved forests.

[0126] Specifically, in step S5, the data processing process is as follows: the DEM0 data generated in S43 is used as a two-dimensional matrix DEM (i, j), where (i, j) represents the coordinates in the grid, h (i, j) represents the elevation value of the point, the neighborhood is defined as a 3×3 window, and the treetop position (tree geographic coordinates), tree height elevation value (tree height), crown edge position and crown area (crown width) are marked;

[0127] S51. Find the tree height:

[0128] 1) Initialize the highest point position set:

[0129]

[0130] 2) Traverse each point in the DEM data:

[0131] For each (i, j) in DEM;

[0132] 3) Get the neighborhood of point (i, j):

[0133] Neighbors(i,j)={(i′,j′)|i-1≤i′≤i+1, j-1≤j′≤j+1, (i′,j′)≠(i,j))

[0134] 4) Determine whether it is the tree top position:

[0135] if h(i,j)>h(i′,j′)for all(i′,j′)∈Neighbors(i,j)then

[0136] Peaks = Peaks∪{(i, j)};

[0137] S52. Find the edge of the tree crown:

[0138] 1) Initialize the lowest point position set:

[0139]

[0140] 2) Traverse each point in the DEM data:

[0141] For each (i, j) in DEM;

[0142] 3) Get the neighborhood of point (i, j):

[0143] Neighbors(i,j)={(i′,j′)|i-1≤i′≤i+1, j-1≤j′≤j+1, (i′,j′)≠(i,j)};

[0144] 4) Determine whether it is the tree top position:

[0145] if h(i,j)x<h(i′,j′)for all(i′,j′)∈Neighbors(i,j)then

[0146] Valleys = Valleys∪{i,j)}

[0147] S53. Mark tree height, crown edge, and crown width:

[0148] 1) Initialize the tag set:

[0149]

[0150] 2) Traverse each point in the highest point position:

[0151] For each (i, j) in Peaks;

[0152] 3) Find the nearest lowest point in the X direction:

[0153]

[0154]

[0155] 4) Find the nearest lowest point in the Y direction:

[0156]

[0157]

[0158] 5) Mark the highest point and the edge of the crown:

[0159] MarkedPeaks=MarkedPeaks∪{(ij,h(i,j),Valley left , Valley right , Valley up , Valley down)};

[0160] 6) Calculate the crown area:

[0161] Long axis:

[0162] a=distance(Valley left , Valley right );

[0163] Short Axis:

[0164] b=distance(Valley up , Valley down )

[0165] Area=0.25πa b.

[0166] The above algorithm can obtain the axial attributes of trees such as tree height by generating a three-dimensional model through oblique photography. Since the forest canopy is relatively complex and it is difficult to segment individual trees, the radial attributes of trees cannot be obtained. The above algorithm can obtain the key radial attribute of trees - crown width.

[0167] The present invention can cover a large area in a short time, quickly collect image data of forests, conduct indoor data analysis, and obtain key attribute elements of trees. The cost is only a few tenths of that of traditional manual surveys, and there is almost no risk to personnel safety. At the same time. The present invention obtains position signals in the air with accurate positioning accuracy, identifies the top positions of trees through algorithms, obtains the geographical coordinates and tree heights of trees, and has high data accuracy; collects image data at different angles such as tilt, and identifies tree heights and crown widths through algorithms, which greatly reduces manual intervention and improves the efficiency and accuracy of surveys.

[0168] The above is only a specific implementation of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A method for obtaining key attributes of canopy trees based on drone oblique photography, characterized in that: The following steps are involved: S1. Drone oblique photography: Plan the route for the survey area, set the gimbal pitch angle to 20-45°, set the route according to forward / rearward or left / right vision, and conduct drone oblique photography of the survey area; S2, stitching DSM data: the images taken by the drone are processed into DSM data through oblique photography processing software; S3. Obtain DEM data: interpolate the DSM data generated by drone oblique photography at locations where there is obviously no vegetation to generate DEM data; S4, data processing: adjust the resolution of DSM and DEM data to 1 / 10-3 / 10 of the average tree crown width to reduce the error rate of tree identification and improve calculation efficiency, and subtract DSM from DEM data to obtain vegetation net height data; S5. Extract key factors of trees: Generate contour lines according to the vegetation net height data. The highest point in the contour line area is regarded as the top of the tree, that is, the geographical coordinates of the tree, and the value is the tree height. The lowest point in the contour line area is regarded as the edge of the tree crown, and the lowest point around the tree top is the tree crown width.

2. The method for obtaining key attributes of canopy trees based on drone oblique photography as claimed in claim 1, characterized in that: In step S2, the process of splicing DSM data is as follows: S21. Check the data taken by the drone to ensure that the geographical location information taken is accurate; S22, importing the checked shooting data into a general 3D modeling software; S23, point cloud registration, aligning the point cloud data in photos taken at different angles to form a complete point cloud model; S24. Generate DSM data using point cloud data through TIN triangulation model.

3. The method for obtaining key attributes of canopy trees based on drone oblique photography as claimed in claim 1, characterized in that: In step S3, the process of obtaining DEM data is as follows: S31. Determine the vegetation-free area: The area with obvious vegetation-free area or the area with clear vegetation height in the DSM data generated by the UAV oblique photography minus the vegetation height is regarded as the vegetation-free area in the DSM. Assume that we define a threshold h, that is, the area with height h is considered to be the vegetation-free area; Use a binary mask M to represent the vegetation-free area: Where (x, y) represents the coordinate position in the DSM data; S32, the height of the vegetation-free area is regarded as the ground height: for the area where the mask M(x, y) = 1, the height value of the DSM is directly regarded as the height value of the DEM; Generate initial DEM data DEMinit(x,y): DEMinit(x,y)=M(x,y)·DSM(x,y); S33. Interpolation generates complete DEM data: For the area where M(x,y)=0, the DEM height value is estimated by the interpolation algorithm: Among them, Interp(DEM init (x′, y′)) represents the interpolation estimation of the adjacent DEM values, specifically: find the nearest point (x', y') where M(x', y') = 1, and interpolate based on the height values ​​of these points.

4. The method for obtaining key attributes of canopy trees based on drone oblique photography as claimed in claim 1, characterized in that: In step S4, the data processing process is as follows: S41. Define input and output grids: Input raster data: Raster input (i,j), where (i,j) represents the pixel position of the input raster data, Output raster data: Raster output (m,n), where (m,n) represents the pixel position of the output raster data, Resolution of input raster: R input , Output raster resolution: R output =M·R input , M is the target resolution for adjustment; S42, interpolation adjustment to target resolution: The output grid position (m,n) corresponds to the input grid position (i,j): The data at position (m,n) is interpolated using the nearest neighbor method according to the position value at (i,j): Raster output (m,n)=Raster output (i,j); S43, obtaining vegetation net height data DSM0: subtracting the DSM and DEM raster data adjusted to the same resolution to obtain vegetation net height data DSM0: DSM0=DSM-DEM.

5. The method for obtaining key attributes of canopy trees based on drone oblique photography as claimed in claim 4, characterized in that: In step S5, the data processing process is as follows: the DEM0 data generated in S43 is used as a two-dimensional matrix DEM (i, j), where (i, j) represents the coordinates in the grid, h (i, j) represents the elevation value of the point, the neighborhood is defined as a 3×3 window, and the treetop position, tree height elevation value, crown edge position and crown area are marked; S51. Find the tree height: 1) Initialize the highest point position set: 2) Traverse each point in the DEM data: For each (i, j) in DEM; 3) Get the neighborhood of point (i, j): Neighbors(i,j)={(i′,j′)|i-1≤i′≤i+1, j-1≤j′≤j+1, (i′,j′)≠(i,j)} 4) Determine whether it is the tree top position: if h(i,j)>h(i′,j′)for all(i′,j′)∈Neighbors(i,j)then Peaks = Peakss∪{(i, j)}; S52. Find the edge of the tree crown: 1) Initialize the lowest point position set: 2) Traverse each point in the DEM data: For each (i, j) in DEM; 3) Get the neighborhood of point (i, j): Neighbors(i,j)={(i′,j′)|i-1≤i′≤i+1, j-1≤j′≤j+1, (i′,j′)≠(i,j)}; 4) Determine whether it is the tree top position: if h(i,j)x<h(i′,j′)for all(i′,j′)∈Neighbors(i,j)then Valleys=Valleys∪{(i,j)} S53. Mark tree height, crown edge, and crown width: 1) Initialize the tag set: 2) Traverse each point in the highest point position: For each (i, j) in Peaks; 3) Find the nearest lowest point in the X direction: 4) Find the nearest lowest point in the Y direction: 5) Mark the highest point and the edge of the crown: MarkedPeaks=MarkedPeaks∪{(i,j,h(i,j),Valley left ,Valley right ,Valley up ,Valley down )}; 6) Calculate the crown area: Long axis: a=distance(Vallev left ,Valley right ); Short Axis: b=distance(Valley up ,Valley down ) Area=0.25πa b.

6. The method for obtaining key attributes of canopy trees based on drone oblique photography as claimed in claim 1, characterized in that: In step S4, the resolution of DSM and DEM data is adjusted to 1 / 5 of the average tree crown width.

Images