Method and system for Forest Hight Mapping

Abstract

The present invention relates to a method and system for creating a forest tree height map. A method for creating a forest tree height map according to one embodiment of the present invention is a method for creating a forest tree height map by a forest tree height map creation system, comprising: a data preparation step of collecting Sentinel-2 data, which is optical image data including dense spatial sampling, and collecting GEDI (Global Ecosystem Dynamics Investigation) Level-2 data including sparse spatial sampling; a model training step of inputting the Sentinel-2 data into a U-Net model to train the U-Net model to predict a forest tree height map; and a tree height map generation step of inputting the Sentinel-2 data into the U-Net model to generate a forest tree height map.

Description

Technology Field

[0001] The present invention relates to a method and system for creating a forest tree height map, and more specifically, to a method and system for creating a high-resolution forest tree height map on a wide scale. Background Technology

[0003] Currently, the calculation of forest carbon storage reported in Korea's greenhouse gas inventory is carried out by multiplying the carbon storage based on field observations at sample points evenly distributed across national forests by the forest area.

[0004] As this method is based on the premise that sample points can represent the entire national forest, it is essential to develop verification methods for carbon storage calculated using this approach, and furthermore, to develop a more accurate method for estimating forest carbon storage, in order to advance Korea's greenhouse gas inventory.

[0005] However, existing methods for mapping forest heights relied on attaching LiDAR sensors to aircraft or drones, which presented limitations in terms of cost (aerial surveying) and scale (drone surveying) for creating a map of forest heights across the entire country of Korea.

[0006] Accordingly, GEDI (Global Ecosystem Dynamics Investigation), a satellite-based lidar sensor operated by NASA, is an excellent alternative to aircraft and drones as it repeatedly measures tree heights on a global scale; however, it has the problem that it cannot produce a tree height map of the entire forest as with the sampling point method because the distance between measurement points is considerable (sparse spatial sampling).

[0007] Therefore, technology is needed to create spatially dense forest height maps on a wide scale. Prior art literature

[0009] Registered Patent No. 10-1080985 (2011.11.09) The problem to be solved

[0010] The present invention was devised to solve the aforementioned problem, and the present invention aims to produce a forest height map by fusing an optical image including dense spatial sampling and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data including sparse spatial sampling.

[0011] Through this, the present invention provides the effect of upscaling data containing sparse spatial sampling into data containing dense spatial sampling, and thus can be used to upscale not only GEDI outputs but also various other field observation data (soil carbon content, leaf area indices), thereby enabling diverse applications in all fields of remote sensing.

[0012] In addition, according to the present invention, it is possible to create a forest tree height map at a resolution of 10m on a wide-area scale. Through this, the carbon storage amount of the forest can be calculated through the forest tree height, and by creating a forest tree height map of the entire Korea at a resolution of 10m, it aims to provide a more accurate calculation compared to the sample point method. means of solving the problem

[0014] A method for creating a forest tree height map according to an embodiment of the present invention for solving the aforementioned problem is a method for creating a forest tree height map by a forest tree height map creation system, comprising: a data preparation step of collecting Sentinel-2 data, which is optical image data including dense spatial sampling, and collecting GEDI (Global Ecosystem Dynamics Investigation) Level-2 data including sparse spatial sampling; a model training step of inputting the Sentinel-2 data into a U-Net model to train the prediction of a forest tree height map of the U-Net model; and a tree height map generation step of inputting the Sentinel-2 data into the U-Net model to generate a forest tree height map.

[0015] According to another embodiment of the present invention, the data preparation step may be configured to include the step of collecting median value data for a certain period from Sentinel-2 data, which is optical image data including dense spatial sampling, to construct Sentinel-2 model input data.

[0016] According to another embodiment of the present invention, the data preparation step may be configured to include the step of generating GEDI Level-2 model reference data by correcting the location information of the GEDI Level-2 data using a Digital Terrain Model (DTM).

[0017] According to another embodiment of the present invention, the model training step may be configured to include: a step of training a U-Net model using the Sentinel-2 model input data; a step of calculating a model loss function value by comparing the predicted value of the forest height map of the U-Net model with the GEDI Level-2 model reference data; a step of correcting the model loss function value using the GEDI Level-2 data distribution value derived from the GEDI Level-2 model reference data; and a step of training the U-Net model using the Sentinel-2 model input data by applying the corrected model loss function value.

[0018] A forest tree height map generation system according to one embodiment of the present invention comprises: a data collection unit that collects Sentinel-2 data, which is optical image data including dense spatial sampling, and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data including sparse spatial sampling; a model training unit that inputs the Sentinel-2 data into a U-Net model to train the U-Net model to predict a forest tree height map; and a tree height map generation unit that inputs the Sentinel-2 data into the U-Net model to generate a forest tree height map.

[0019] According to another embodiment of the present invention, the data collection unit may be configured to include an input data composition unit that collects median value data for a certain period from Sentinel-2 data, which is optical image data including dense spatial sampling, to form Sentinel-2 model input data.

[0020] The above data collection unit may be configured to include a reference data generation unit that generates GEDI Level-2 model reference data by correcting the location information of the GEDI Level-2 data using a Digital Terrain Model (DTM).

[0021] According to another embodiment of the present invention, the model training unit trains a U-Net model using the Sentinel-2 model input data, calculates a model loss function value by comparing the predicted value of the forest height map of the U-Net model with the GEDI Level-2 model reference data, corrects the model loss function value using the GEDI Level-2 data distribution value derived from the GEDI Level-2 model reference data, and trains the U-Net model using the Sentinel-2 model input data by applying the corrected model loss function value. Effects of the invention

[0023] The present invention can produce a forest height map by fusing optical images including dense spatial sampling and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data including sparse spatial sampling.

[0024] Through this, the present invention provides the effect of upscaling data containing sparse spatial sampling into data containing dense spatial sampling, and since it can be used to upscale not only GEDI outputs but also various other field observation data (soil carbon content, leaf area indices, etc.), it can be utilized in various applications in all fields of remote sensing.

[0025] In addition, according to the present invention, it is possible to create a forest tree height map at a resolution of 10m on a wide-area scale. Through this, the carbon storage amount of the forest can be calculated through the forest tree height, and by creating a forest tree height map of the entire Korea at a resolution of 10m, a more accurate calculation is possible compared to the sample point method. Brief explanation of the drawing

[0027] FIG. 1 is a drawing for explaining a method for creating a forest tree height map according to an embodiment of the present invention. FIGS. 2 to 4 are drawings for explaining in more detail a method for creating a forest tree height map according to an embodiment of the present invention. FIG. 5 is a diagram illustrating a method for correcting location information of GEDI Level-2 data using a Digital Terrain Model (DTM) according to an embodiment of the present invention. Figure 6 is a graph showing the results of the output of a model according to the prior art. FIG. 7 is a drawing illustrating a forest tree height map calculated by a method for creating a forest tree height map according to an embodiment of the present invention. FIG. 8 is a configuration diagram of a forest tree height map creation system according to an embodiment of the present invention. Specific details for implementing the invention

[0028] The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the description of the invention. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention.

[0029] However, in describing the embodiments, if it is determined that a detailed description of related known functions or configurations could unnecessarily obscure the essence of the invention, such detailed description is omitted. Furthermore, the sizes of each component in the drawings may be exaggerated for illustrative purposes and do not represent the actual sizes applied.

[0030] Furthermore, throughout the specification, when a component is referred to as being "connected" or "joined" with another component, it should be understood that the component may be directly connected or joined to the other component, but unless specifically stated otherwise, it may also be connected or joined through an intermediate component. Additionally, throughout the specification, when a part is described as "including" a component, unless specifically stated otherwise, this means that it may include additional components rather than excluding other components.

[0031] FIG. 1 is a drawing for explaining a method for creating a forest tree height map according to an embodiment of the present invention.

[0032] A method for creating a forest tree height map according to an embodiment of the present invention is performed by a corn map creation system. The forest tree height map creation system according to an embodiment of the present invention may be composed of a computer terminal, a server, or a dedicated device, or each component providing each function may be composed of a computer terminal, a server, or a dedicated device. Additionally, each component providing each function in the corn map creation system according to an embodiment of the present invention may be composed of hardware or software.

[0033] In one embodiment of the present invention, the U-Net model, which shows overall good performance in the field of computer vision, was used as the deep learning model used for creating a forest tree height map.

[0034] Sentinel-2 optical satellite images with a spatial resolution of 10 m were used as input data for the U-Net model, and GEDI (Global Ecosystem Dynamics Investigation) Level-2 forest height data were used for training the U-Net model.

[0035] In addition, the positional information of NASA's GEDI L2A forest height products used in model training was corrected.

[0036] It is known that there is an uncertainty of about 10 m in the location information of the above NASA Level-2 forest height data. To correct this, the present invention uses a Digital Terrain Model (DTM) to correct the location information of the GEDI output.

[0037] According to the present invention, a method for correcting the model loss function value was used to mitigate overestimation and underestimation that may occur when utilizing the U-Net model.

[0038] It is widely known that when a U-Net model is used to derive specific values ​​rather than classification, the model output tends to be skewed toward the mode. Given that most real-world information follows a normal distribution, it is common for low values ​​to be overestimated and high values ​​to be underestimated. Therefore, to mitigate this phenomenon, a forest height map was produced by applying a method to correct the model's loss function.

[0039] FIGS. 2 to 4 are drawings for explaining in more detail a method for creating a forest tree height map according to an embodiment of the present invention.

[0040] In addition, FIG. 5 is a diagram illustrating a method for correcting location information of GEDI Level-2 data using a Digital Terrain Model (DTM) according to an embodiment of the present invention, FIG. 6 is a graph showing the result of an output of a model according to the prior art, and FIG. 7 is a diagram showing a forest height map calculated by a method for creating a forest height map according to an embodiment of the present invention.

[0041] From now on, a method for creating a forest tree height map according to an embodiment of the present invention will be described in more detail with reference to FIGS. 2 to 7.

[0042] According to the method for creating a forest tree height map according to one embodiment of the present invention, data is first prepared (S110).

[0043] More specifically, Sentinel-2 data, which is optical image data including dense spatial sampling, and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data, which includes sparse spatial sampling, are collected.

[0044] At this time, median data for a certain period (e.g., April to September) can be collected from Sentinel-2 data, which is optical image data including dense spatial sampling, to form input data for the Sentinel-2 model.

[0045] In addition, GEDI Level-2 model reference data can be generated by correcting the location information of the above GEDI Level-2 data using a Digital Terrain Model (DTM).

[0046] Referring to Fig. 5, the actual location of the measurement value for the Given GEDI Location in the GEDI Level-2 data is one of the Potential GEDI Locations. Therefore, this is corrected using a high-resolution DTM.

[0047] In addition, referring to Fig. 6, it can be seen that the output of a conventional model that does not use the loss function correction technique according to the present invention shows an underestimation phenomenon at high heights and an overestimation phenomenon at low heights.

[0048] Afterwards, as shown in Fig. 3, the Sentinel-2 data is input into the U-Net model to train the prediction of the forest tree height map of the U-Net model (S120).

[0049] To explain in more detail, first, a U-Net model is trained using the input data of the Sentinel-2 model. The predicted value of the forest tree height map of the U-Net model is compared with the reference data of the GEDI Level-2 model to calculate the model loss function value, and the model loss function value is corrected using the GEDI Level-2 data distribution value derived from the reference data of the GEDI Level-2 model. In addition, the U-Net model is trained using the input data of the Sentinel-2 model by applying the corrected model loss function value.

[0050] Afterwards, as shown in FIG. 4, the Sentinel-2 data is input into the U-Net model to produce a forest height map (S130).

[0051] Figure 7 illustrates a forest tree height map produced by the forest tree height map creation method as described above.

[0052] In this way, the present invention can produce a forest height map by fusing an optical image containing dense spatial sampling and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data containing sparse spatial sampling.

[0053] FIG. 8 is a configuration diagram of a forest tree height map creation system according to an embodiment of the present invention.

[0054] From now on, the configuration of a forest tree height map creation system according to an embodiment of the present invention will be described with reference to FIG. 8.

[0055] A forest tree height map creation system according to one embodiment of the present invention may be composed of a computer terminal, a server, or a dedicated device, or each component providing each function may be composed of a computer terminal, a server, or a dedicated device. In addition, a corn map creation system according to one embodiment of the present invention may have each component providing each function composed of hardware or software.

[0056] More specifically, a forest tree height map generation system (100) according to one embodiment of the present invention comprises a data collection unit (110), a model training unit (120), and a tree height map generation unit (130).

[0057] The data collection unit (110) collects Sentinel-2 data, which is optical image data including dense spatial sampling, and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data, which includes sparse spatial sampling. At this time, the data collection unit (110) can collect Sentinel-2 data and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data from the database (200).

[0058] More specifically, the data collection unit (110) may be configured to include an input data configuration unit (111) and a reference data generation unit (112).

[0059] At this time, the input data composition unit (111) can compose the Sentinel-2 model input data by collecting median value data for a certain period from Sentinel-2 data, which is optical image data including dense spatial sampling.

[0060] In addition, the reference data generation unit (112) can generate GEDI Level-2 model reference data by correcting the location information of the GEDI Level-2 data using a Digital Terrain Model (DTM).

[0061] The above model training unit (120) inputs the above Sentinel-2 data into the U-Net model to train the prediction of the forest height map of the U-Net model.

[0062] To explain in more detail, the model training unit (120) can train the U-Net model using the Sentinel-2 model input data, calculate the model loss function value by comparing the predicted value of the forest tree height map of the U-Net model with the GEDI Level-2 model reference data, correct the model loss function value using the GEDI Level-2 data distribution value derived from the GEDI Level-2 model reference data, and train the U-Net model using the Sentinel-2 model input data by applying the corrected model loss function value.

[0063] Accordingly, the above-mentioned tree height map output unit (130) can output a forest tree height map by inputting the above-mentioned Sentinel-2 data into the above-mentioned trained U-Net model.

[0064] Through this, the present invention provides the effect of upscaling data containing sparse spatial sampling into data containing dense spatial sampling, and since it can be used to upscale not only GEDI outputs but also various other field observation data (soil carbon content, leaf area indices, etc.), it can be utilized in various applications in all fields of remote sensing.

[0065] In addition, according to the present invention, it is possible to create a forest tree height map at a resolution of 10m on a wide-area scale. Through this, the carbon storage amount of the forest can be calculated through the forest tree height, and by creating a forest tree height map of the entire Korea at a resolution of 10m, a more accurate calculation is possible compared to the sample point method.

[0066] In the detailed description of the present invention as described above, specific embodiments have been described. However, various modifications are possible within the scope of the present invention. The technical concept of the present invention should not be limited to the aforementioned embodiments, but should be defined by the claims as well as equivalents thereof. Explanation of the symbols

[0068] 100: Forest Tree Height Mapping System 110: Data Collection Department 111: Input data composition section 112: Reference material generation section 120: Model Training Department 130: Labor Map Output Section 200: Database

Claims

Claim 1 A method for generating a forest tree height map using a forest tree height map generation system comprises: a data preparation step of collecting Sentinel-2 data, which is optical image data including dense spatial sampling, and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data, which is sparse spatial sampling, and collecting median value data for a certain period from the Sentinel-2 data, which is optical image data including dense spatial sampling, to constitute Sentinel-2 model input data; a step of generating GEDI Level-2 model reference data by correcting location information using a Digital Terrain Model (DTM) for the GEDI Level-2 data; training a U-Net-based deep learning model with the Sentinel-2 data as input, calculating a model loss function value by comparing the predicted value of the forest tree height map predicted by the deep learning model with the GEDI Level-2 model reference data, and from the GEDI Level-2 model reference data A method for creating a forest tree height map comprising: a model training step of correcting the loss function using derived tree height GEDI data distribution values ​​and then training the U-Net-based deep learning model by applying the corrected loss function; and a tree height map generation step of generating a forest tree height map by inputting the Sentinel-2 data into the trained U-Net-based deep learning model. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 A data collection unit that collects Sentinel-2 data, which is optical image data including dense spatial sampling, and GEDI (Global Ecosystem Dynamics Investigation) Level-2 data including sparse spatial sampling; a reference data generation unit that generates GEDI Level-2 model reference data by correcting location information using a Digital Terrain Model (DTM) for the GEDI Level-2 data; and a model training unit that, for a U-Net-based deep learning model using the Sentinel-2 data as input, calculates a loss function by comparing predicted forest height values ​​with the GEDI Level-2 model reference data, corrects the loss function using height data distribution values ​​derived from the GEDI Level-2 model reference data, and trains the deep learning model by applying the corrected loss function. A forest tree height map generation system characterized by including: a tree height map generation unit that generates a forest tree height map with high-density spatial resolution using the above-mentioned learned deep learning model. Claim 6 A forest height map creation system according to claim 5, wherein the data collection unit comprises an input data composition unit that generates model input data through median value calculation from multiple time points of Sentinel-2 images collected over a certain period. Claim 7 delete Claim 8 delete Claim 9 A computer-readable recording medium containing a program for performing the method of creating a forest tree height map according to claim 1.

Images