Forest information analysis device, forest information analysis method, and program

Abstract

Using optical images and / or LiDAR point cloud data acquired from the air, this system enables highly accurate estimation of tree crown regions, tree species identification, and estimation of tree height, diameter at breast height (DBH), trunk volume / biomass, and CO2 fixation at the individual tree level, regardless of acquisition conditions (presence or absence of point clouds, etc.), regional differences, or seasonal factors such as pollen attachment period and autumn foliage period. The results can then be saved and utilized in a GIS. [Solution] Optical images or optical images and LiDAR point cloud data are acquired, and orthomosaic images and crown height models (CHMs) are generated. Depending on the presence or absence of point cloud data, a DSM is created using SfM, and the CHM is obtained by the difference with an external DEM, or a DSM / DTM is generated from the point cloud, and the CHM is obtained by the difference. Crown polygons are generated based on the CHM or orthomosaic images, and the corresponding images are identified by a trained model. The model / coefficients are switched according to the location information, and the color tone is corrected by hue correction and color distribution conversion before identification and estimation, and saved as attributed GIS data.

Description

[[Technical Field]]

[0001] The present invention relates to a technology for analyzing forest-related information, and particularly to a forest information analysis apparatus, a forest information analysis method, and a program for executing the same, which analyze forest information in individual tree units using optical images acquired from the air and / or LiDAR point cloud data. [[Background Art]]

[0002] Forest resources have various functions not limited to timber production, such as carbon dioxide absorption and fixation, biodiversity conservation, and water conservation. In order to appropriately evaluate these functions and achieve sustainable management, it is required to accurately grasp the state of forests. In particular, information such as the number of trees, tree species, tree height, diameter at breast height (DBH), and stem volume is important basic information for evaluating forest resource volume and formulating management plans.

[0003] Conventionally, such forest information has been acquired through manual surveys such as full timber surveys and sample plot surveys conducted on site. However, these survey methods require significant labor and time, and have the problem that survey costs increase when targeting vast forest areas. In addition, in steep terrain or areas that are difficult to access, the survey itself is often difficult from the viewpoint of safety.

[0004] In recent years, technologies for analyzing forest information using aerial images and LiDAR point cloud data acquired using aircraft, unmanned aerial vehicles (drones), etc. have been proposed to address such issues. For example, technologies are known that use orthoimages generated from aerial images, digital surface models, or digital terrain models generated from LiDAR point cloud data to grasp the tree height distribution and stand structure of forests.

[0005] However, many of these conventional techniques are limited to analysis at the stand or mesh level, and techniques for consistently estimating tree height, diameter at breast height, trunk volume, biomass, and carbon dioxide sequestering at the individual tree level, i.e., single tree level, have not been sufficiently established. Furthermore, it has not been easy to flexibly switch analysis methods depending on the acquisition conditions and availability of image data and LiDAR point cloud data, or to cope with the decrease in tree species identification accuracy caused by regional and seasonal differences.

[0006] Furthermore, the mechanisms for saving analysis results in a format usable by Geographic Information Systems (GIS) and for aggregating and utilizing them by region or management area were far from adequately developed. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2023-3767 [Overview of the project] [Problems that the invention aims to solve]

[0008] As mentioned above, technologies have been proposed to analyze forest information using image data and LiDAR point cloud data acquired from the air, with the aim of improving the understanding and management of forest resources. For example, Patent Document 1 discloses a technology for analyzing the position and height of trees using point cloud data.

[0009] However, prior art, including Patent Document 1, relies primarily on analysis based on point cloud data, and configurations that flexibly switch analysis methods depending on the acquisition conditions and availability of optical images and LiDAR point cloud data have not been sufficiently considered. Furthermore, it is difficult to say that sufficient accuracy and versatility have been ensured in stably separating the canopy region at the individual tree level, identifying the tree species for each canopy region, and consistently estimating multiple forest information such as tree height, diameter at breast height, trunk volume or biomass, and carbon dioxide sequestering amount.

[0010] Furthermore, there was a problem in that the accuracy of tree species identification decreased when the color tone and appearance of images changed depending on the type of forest, regional characteristics, and seasonal factors (e.g., pollen attachment period or autumn foliage period), and sufficient effective measures to address this had not been taken. In addition, despite the fact that the calculation formulas for trunk volume or biomass and the calculation parameters for carbon dioxide sequestration differed depending on the region or country being analyzed, there was no mechanism in place to automatically switch and apply these.

[0011] Furthermore, there was a problem in that sufficient consideration had not been given to the structure for saving the analysis results as geographic information data and outputting and aggregating them in a format that is easy to use for forest management and resource assessment.

[0012] This invention has been made in view of the problems of the prior art described above, and aims to provide a forest information analysis device, a forest information analysis method, and a program that can perform tree canopy region separation, tree species identification, and forest information estimation at the individual tree level with high accuracy and efficiency, while appropriately applying analysis methods according to acquisition conditions and regional characteristics, using optical images and / or LiDAR point cloud data acquired from the air. [Means for solving the problem]

[0013] A forest information analysis device according to the first embodiment of the present invention is a forest information analysis device that analyzes forest information at the individual tree level using data on forests acquired from the air, A data acquisition unit that acquires either an optical image obtained by photographing a forest, or both the optical image and LiDAR point cloud data, A forest analysis data creation unit that generates orthomosaic images and a crown height model (CHM) based on the acquired data, A forest analysis unit that separates the canopy region for each tree and generates canopy polygons based on the orthomosaic image and the canopy height model, identifies the tree species for each canopy polygon, and estimates at least one of the tree height, diameter at breast height, trunk volume or biomass, and carbon dioxide sequestering amount for each canopy polygon, The system is characterized by comprising a storage unit that stores geographic information data associated with identified or estimated forest information to the tree canopy polygon.

[0014] A forest information analysis device according to a second embodiment of the present invention is a forest information analysis device according to a first embodiment, wherein the forest analysis data creation unit performs the following actions depending on the presence or absence of LiDAR point cloud data: (a) A process to generate a numerical surface model (DSM) by performing Structure from Motion (SfM) processing on the optical image, and generate the canopy height model by the difference between it and an externally provided numerical elevation model (DEM), or (b) A process to generate a numerical terrain model (DTM) and a DSM from the LiDAR point cloud data, and generate the canopy height model from the difference between them. It is characterized by selectively executing the following.

[0015] A third embodiment of the present invention is a forest information analysis device according to the first or second embodiment, characterized in that the digital elevation model is obtained by an API call to an external public DEM service or as a user-specified DEM file.

[0016] A forest information analysis device according to a fourth embodiment of the present invention is the forest information analysis device according to any one of the first to third embodiments, wherein the forest analysis data creation unit calculates an approximate plane or an approximate curved surface based on height values of ground corresponding regions included in the crown height model, and performs correction based on the approximate plane or the approximate curved surface on the crown height model.

[0017] A forest information analysis device according to a fifth embodiment of the present invention is the forest information analysis device according to any one of the first to fourth embodiments, wherein the forest analysis unit (a) a first crown separation method using a local maximum filter and watershed segmentation for the crown height model, (b) a second crown separation method based on object detection or region extraction using a trained model for the orthoimage, and (c) a third crown separation method based on object detection or region extraction that does not use a trained model for the orthoimage or CHM, separates a crown region using at least one of the foregoing.

[0018] A forest information analysis device according to a sixth embodiment of the present invention is the forest information analysis device according to any one of the first to fifth embodiments, wherein the forest analysis unit extracts an image region corresponding to the crown polygon, inputs the image region to a trained model for tree species identification to identify a tree species, is characterized in that the tree species identification model to be used is selected based on position information, or tree species candidates to be identified are restricted.

[0019] A forest information analysis device according to a seventh embodiment of the present invention is the forest information analysis device according to any one of the first to sixth embodiments, wherein the forest analysis unit is characterized in that, in order to correct color tone changes in a pollen adhesion period or an autumn leaf discoloration period, tree species identification is performed after performing hue correction or color distribution conversion on the image region.

[0020] A forest information analysis apparatus according to an eighth embodiment of the present invention is the forest information analysis apparatus according to any one of the first to seventh embodiments, wherein the forest analysis unit calculates a maximum value of the canopy height model within the canopy polygon as a tree height, and estimates a diameter at breast height using a formula for each tree species based on the tree height and a canopy area.

[0021] A forest information analysis apparatus according to a ninth embodiment of the present invention is the forest information analysis apparatus according to any one of the first to eighth embodiments, wherein the forest analysis unit specifies a region or a country based on position information, and selects a calculation formula or a parameter for stem volume, biomass, or carbon dioxide fixation amount according to the region or the country.

[0022] A forest information analysis method according to a tenth embodiment of the present invention is a forest information analysis method for analyzing forest information in individual tree units using data related to a forest acquired from the air, acquiring at least one of an optical image obtained by photographing a forest, and LiDAR point cloud data acquired simultaneously with or at a different time from the optical image; generating an orthoimage and a canopy height model (CHM) based on the acquired data; separating a canopy region for each tree to generate a canopy polygon based on the orthoimage and the canopy height model; identifying a tree species for each of the canopy polygons; estimating at least one of tree height, diameter at breast height, stem volume or biomass, and carbon dioxide fixation amount for each of the canopy polygons; saving the forest information identified or estimated for the canopy polygons as associated geographic information data; comprising:

[0023] A forest information analysis method according to an eleventh embodiment of the present invention is the forest information analysis method according to the tenth embodiment, wherein the step of generating the canopy height model comprises, depending on the presence or absence of LiDAR point cloud data, (a) A process to generate a numerical surface model (DSM) by performing Structure from Motion (SfM) processing on the optical image, and generate the tree canopy height model by the difference between it and an externally provided numerical elevation model (DEM), or (b) A process of generating a numerical terrain model (DTM) and a DSM from the LiDAR point cloud data, and generating the canopy height model from the difference between them. It is characterized by selectively executing the following.

[0024] A forest information analysis method according to the twelfth embodiment of the present invention is a forest information analysis method according to the tenth or eleventh embodiment, characterized in that the digital elevation model is obtained by making an API call to an external public DEM service or as a user-specified DEM file.

[0025] A forest information analysis method according to the 13th embodiment of the present invention is a forest information analysis method according to any of the 10th to 12th embodiments, characterized in that the separation of the canopy region is performed by at least one of a method based on a canopy height model or a method based on orthophotos.

[0026] A forest information analysis method according to the 14th embodiment of the present invention is a forest information analysis method according to any of the 10th to 13th embodiments, wherein the step of separating the tree canopy region is: (a) A first canopy separation method using a local maximum filter and watershed segmentation for the canopy height model, (b) A second tree canopy separation method using object detection or region extraction with a trained model on the orthophoto, and (c) A third tree canopy separation method using object detection or region extraction without using a pre-trained model for the orthoimage or CHM, The method is characterized by including a process of separating the canopy region using at least one of the following.

[0027] A forest information analysis method according to the 15th embodiment of the present invention is a forest information analysis method according to any of the 10th to 14th embodiments, wherein the step of identifying the tree species is to extract an image region corresponding to the tree canopy polygon, input the image region into a trained model for tree species identification to identify the tree species, This method is characterized by including a process to select a tree species identification model to be used based on location information, or to limit the candidate tree species to be identified.

[0028] A forest information analysis method according to the 16th embodiment of the present invention is a forest information analysis method according to any of the 10th to 15th embodiments, characterized in that the step of identifying the tree species is performed on the image region in order to correct for color changes during the pollen attachment period or the autumn foliage period, and then the tree species is identified.

[0029] The forest information analysis method according to the 17th embodiment of the present invention is a forest information analysis method according to any of the 10th to 16th embodiments, characterized in that the diameter at breast height is calculated by an estimation formula for each tree species based on tree height and crown area.

[0030] The forest information analysis method according to the 18th embodiment of the present invention is a forest information analysis method according to any of the 10th to 17th embodiments, characterized in that it outputs statistical information obtained by aggregating the estimation results by region.

[0031] The forest information analysis program according to the 19th embodiment of the present invention is characterized in that it causes a computer to function as a forest information analysis device according to any of the first to eighth embodiments.

[0032] The forest information analysis program according to the 20th embodiment of the present invention is characterized by causing a computer to execute a forest information analysis method according to any of the 10th to 18th embodiments. [Effects of the Invention]

[0033] According to the present invention, forest information can be analyzed with high accuracy and efficiency at the individual tree level using optical images and / or LiDAR point cloud data acquired from the air. In other words, the forest information analysis device, forest information analysis method, and program according to the present invention can flexibly switch between analysis based on optical images and analysis based on LiDAR point cloud data depending on the type of data that can be acquired to generate a canopy height model, thus enabling stable forest analysis even when the shooting conditions and sensor configurations differ.

[0034] Furthermore, by applying ground height correction to the crown height model, the impact of errors between data obtained from different data sources for DSM and DEM can be reduced, thereby improving the accuracy of tree height calculations. In addition, by using a crown separation method based on the crown height model and a crown separation method based on orthomosaic images in combination or selectively, it becomes possible to separate appropriate crown regions according to the type and structure of the forest, enabling stable crown extraction at the individual tree level.

[0035] Furthermore, by using image regions corresponding to tree canopy polygons for tree species identification, selecting a pre-trained model to use based on location information, or restricting candidate tree species, highly accurate tree species identification that takes regional characteristics into account can be achieved. In addition, by performing color correction for color changes during pollen attachment and autumn foliage seasons, it is possible to suppress the decrease in identification accuracy caused by seasonal variations.

[0036] In addition, by estimating the diameter at breast height using tree height and crown area calculated based on the crown height model, and further estimating the trunk volume, biomass, or carbon dioxide sequestration amount using calculation formulas or parameters appropriate to the region or country, it is possible to quantitatively grasp the amount of forest resources and environmental value at the individual tree level. Furthermore, since the analysis results can be saved as geographic information data linked to the crown polygon, and statistical information aggregated by region or management area can be output, it becomes possible to provide highly practical information for forest management, resource assessment, and environmental policy planning.

[0037] As described above, the present invention offers the remarkable advantage of enabling highly accurate and efficient consistent forest information analysis at the individual tree level using aerial data, which was difficult with conventional techniques. [Brief explanation of the drawing]

[0038] [Figure 1] This is a system configuration diagram schematically showing the overall configuration of a forest information analysis system equipped with a forest information analysis device according to one embodiment of the present invention. [Figure 2] This is a block diagram showing an example of the hardware configuration of a forest information analysis device according to one embodiment of the present invention. [Figure 3] This is a block diagram showing an example of the functional configuration of a forest information analysis device according to one embodiment of the present invention. [Figure 4] This flowchart shows an example of the overall processing procedure for a forest information analysis method according to one embodiment of the present invention. [Figure 5] This flowchart shows an example of a process that switches the tree canopy height model generation process depending on whether or not LiDAR point cloud data is available. [Figure 6] This is a schematic diagram illustrating the relationship between the numerical surface model (DSM), the numerical terrain model (DTM), and the canopy height model (CHM). [Figure 7] This is an explanatory diagram showing an example of input data used in one embodiment of the present invention: an orthophoto, a digital surface model (DSM), and a digital terrain model (DTM). [Figure 8] This is an explanatory diagram showing an example of the screen for selecting the ground portion and the correction method (planar / curved surface) in the ground correction of the tree canopy height model (CHM). [Figure 9] This flowchart shows an example of a forest analysis processing procedure that includes generating a canopy map, a tree species map, a tree height / diameter at breast height map, a trunk volume map, and a CO2 sequester map. [Figure 10]Figure 9 is a flowchart detailing the process for generating a tree canopy map, illustrating an example of a procedure for separating tree canopies at the individual level using orthomosaic images and / or a tree canopy height model (CHM) as input. [Figure 11] This is an explanatory diagram showing an example of the results of tree apex detection using a local maximum filter and tree crown separation using watershed segmentation. [Figure 12] Figure 9 is a flowchart showing an example of the detailed process for generating a tree species map (dead tree detection, tree species identification, candidate narrowing by region, etc.). [Figure 13] This is an explanatory diagram showing an example of displaying tree species identification results at the canopy polygon level using color coding. [Figure 14] Figure 9 is a flowchart showing an example of the detailed process for generating the tree height / diameter at breast height map. [Figure 15] Figure 9 is a flowchart showing an example of the detailed process for generating the trunk volume map. [Figure 16] Figure 9 is a flowchart showing an example of the detailed process for generating the CO2 fixation map. [Figure 17] This is an explanatory diagram showing an example of how to display the estimated results for tree height and diameter at breast height (DBH). [Figure 18] This is an explanatory diagram showing an example of how to display the estimated results of trunk volume and CO2 fixation. [Figure 19] This is an explanatory diagram showing examples of how geographic information data from the analysis results and statistical information for the selected area are displayed. [Figure 20] This is an explanatory diagram showing an example of GIS data in which analysis results such as tree species, tree height, diameter at breast height (DBH), trunk volume, and CO2 fixation amount are linked to a tree canopy polygon and saved. [Figure 21] This is an explanatory diagram illustrating an example of color correction performed by hue substitution in the HSV color space. [Figure 22] This is an explanatory diagram illustrating an example of performing color correction using a 1D LUT (One-Dimensional Lookup Table). [Modes for carrying out the invention]

[0039] One embodiment of the present invention will be described below with reference to the drawings. The individual embodiments of the present invention are not independent but can be combined and implemented as appropriate. This embodiment is not limited to the present invention, and various modifications are possible without departing from the spirit of the invention.

[0040] Figure 1 is a schematic system configuration diagram showing the overall configuration of a forest information analysis system equipped with a forest information analysis device according to one embodiment of the present invention. In this embodiment, the forest information analysis system is configured to include a computer 100 that functions as a forest information analysis device and an unmanned aerial vehicle (e.g., a drone) 200 that serves as a camera for aerial photography of a forest 300. The computer 100 is an information processing device equipped with a processing unit, a storage device, an input device, a display device, etc., and functions as a forest information analysis device according to the present invention by executing a forest information analysis program. The computer 100 may be, for example, a notebook computer or a desktop computer, or a server device on a cloud computing environment.

[0041] The unmanned aerial vehicle (UAV) 200 is a photographic device that flies over the forest 300 and photographs the forest. The UAV 200 is equipped with an optical camera for photographing the forest and can acquire visible light images. Note that the optical camera is not limited to a visible light camera and may include a near-infrared camera, etc. In addition, the UAV 200 may be equipped with a LiDAR sensor in addition to, or in place of, the optical camera, and may be configured to acquire LiDAR point cloud data related to the forest 300.

[0042] Optical images and / or LiDAR point cloud data acquired by the unmanned aerial vehicle 200 are transmitted to the computer 100 via wireless communication or the like. The data may be transmitted in real time, or it may be transmitted as post-processing via a recording medium.

[0043] The computer 100 generates orthomosaic images and a crown height model (CHM) based on optical images and / or LiDAR point cloud data acquired from the unmanned aerial vehicle 200, and uses the generated orthomosaic images and crown height model to analyze the trees that make up the forest 300 on a tree-by-tree basis.

[0044] Specifically, the computer 100 separates the canopy region of each tree to generate canopy polygons, identifies the tree species for each canopy polygon, and estimates at least one of the following: tree height, diameter at breast height, trunk volume or biomass, and carbon dioxide sequestering amount. These analysis results are stored as geographic information data associated with the canopy polygons and used for forest management, resource assessment, or environmental assessment.

[0045] In this embodiment, an example using an unmanned aerial vehicle (e.g., a drone) 200 as the imaging device is shown, but the present invention is not limited thereto, and any imaging device capable of photographing forests from the air, such as a manned aircraft or a satellite, can be used. As described above, according to this embodiment, it is possible to analyze the trees constituting the forest 300 with high accuracy at the individual tree level using data on the forest acquired from the air.

[0046] Figure 2 is a block diagram showing an example of the hardware configuration of a computer that functions as a forest information analysis device according to one embodiment of the present invention. The computer 100 is configured to include a central processing unit (CPU) 11, memory 12, bus 13, input / output interface 14, input unit 15, output unit 16, storage unit 17, communication unit 18, and image processing unit (GPU) 19.

[0047] The CPU 11 is a processing unit that controls the operation of the entire computer (forest information analysis device 100) and executes various instructions included in the forest information analysis program, which will be described later. The CPU 11 performs various calculations necessary for forest information analysis, such as image processing, point cloud processing, model inference, and calculation of various parameters.

[0048] Memory 12 is the main memory used by the CPU 11 when executing processes, and is composed of, for example, RAM. The running program, data in progress, calculation results, etc., are temporarily stored in Memory 12.

[0049] Bus 13 is a communication path that interconnects various components such as the CPU 11, memory 12, and input / output interface 14, and transmits data and control signals. The input / output interface 14 is an interface for inputting and outputting data between the computer 100 and external devices. The input unit 15, output unit 16, storage unit 17, and communication unit 18 are connected to bus 13 via the input / output interface 14.

[0050] The GPU 19 is a computing device that works in cooperation with the CPU 11 to perform processing suitable for parallel computing. In this embodiment, the separation of the tree canopy region and the identification of tree species include inference processing by a trained model that takes an orthophoto or an image region corresponding to the tree canopy polygon as input, and the GPU 19 can be used to speed up the calculations in this inference processing. For example, the inference processing of tree canopy separation (object detection, instance segmentation, or region extraction) and tree species identification by a trained model may be performed on the GPU 19. Note that the GPU 19 may be replaced by an accelerator such as an NPU, TPU, or FPGA.

[0051] The input unit 15 is a device for inputting operation instructions and setting information from the user, and includes, for example, a keyboard, mouse, or touch panel. Through the input unit 15, the target area for analysis is specified, and analysis conditions are set. The output unit 16 is a device for outputting analysis results and processing status, and includes, for example, a display or printer. The output unit 16 displays or outputs tree canopy polygons, tree species identification results, forest information estimation results, etc.

[0052] The memory unit 17 is a storage device for storing forest information analysis programs and various data, and is composed of, for example, a hard disk drive, a semiconductor storage device, etc. The memory unit 17 stores optical images acquired by drones, LiDAR point cloud data, generated orthomosaic images, tree crown height models, and geographic information data as analysis results. The memory unit 17 also stores trained models for tree crown separation and tree species identification, estimation formulas and coefficient sets for each region or country, candidate tree species lists, and conversion parameters or lookup tables used for hue correction or color distribution conversion, which are read out as needed by the CPU 11 or GPU 19.

[0053] The memory unit 17 may store correspondence information for selecting a trained model for tree species identification, a formula for estimating diameter at breast height, a formula or coefficient set for calculating trunk volume or biomass, and a formula or coefficient set for calculating carbon dioxide fixation, according to the regional classification and tree species (or tree species class). The correspondence information may be configured, for example, as a correspondence table (hereinafter referred to as the model selection table) that associates the applicable model ID or coefficient set ID with the regional classification ID and tree species class ID as keys.

[0054] The coefficient sets or calculation formulas may be managed by country, prefecture, or a predetermined regional classification. The storage unit 17 may store, for each coefficient set or calculation formula, at least an identifier (ID), applicable region, applicable tree species (or tree species class), version number, and application start date. When a coefficient set or calculation formula is updated, the storage unit 17 adds the new version while retaining the old version, and the forest analysis unit 113 may select the coefficient set or calculation formula with the version number to be applied according to the shooting time associated with the data to be analyzed, or according to user specification. This makes it possible to switch coefficient systems according to region or country while ensuring the reproducibility of the analysis results.

[0055] The communication unit 18 is a communication device for communicating with external devices, and includes, for example, a wired or wireless communication interface. Through the communication unit 18, data acquired from the unmanned aerial vehicle 200 is received, data is sent and received with a cloud server, or external digital elevation model provision services are accessed. The communication unit 18 is also used for communication to acquire digital elevation models by making API calls to external public DEM services.

[0056] The hardware configuration shown in Figure 2 is merely an example, and the present invention is not limited thereto. For example, the configuration may involve a dedicated computing device such as a GPU executing some or all of the functions of the CPU 11, or it may be a distributed processing configuration using multiple computers or a configuration implemented on a cloud computing environment. The forest information analysis device according to the present invention is realized by a computer 100 configured as described above.

[0057] Figure 3 is a block diagram showing an example of the functional configuration of a forest information analysis device 100 according to one embodiment of the present invention. The forest information analysis device 100 is implemented by a forest information analysis program executed on a computer and mainly comprises a CPU 11, a storage unit 17, and a communication unit 18.

[0058] The CPU 11 is the core processing unit of the forest information analysis device 100, and functionally operates as a data acquisition unit 111, a data creation unit 112, and a forest analysis unit 113. These functional units do not need to be hardware-independent; they are logically realized when the CPU 11 executes the forest information analysis program.

[0059] The data acquisition unit 111 is a functional unit that acquires data for analysis related to forests. Specifically, the data acquisition unit 111 acquires optical images and / or LiDAR point cloud data transmitted from imaging devices such as the unmanned aerial vehicle 200 shown in Figure 1 via the communication unit 18. The data acquisition unit 111 can also acquire external data such as digital elevation models (DEMs) via an external network N such as the Internet. Furthermore, the data acquisition unit 111 can also acquire metadata such as location information, shooting time, and shooting conditions associated with the acquired data.

[0060] The location information acquired by the data acquisition unit 111 is, for example, the shooting location (latitude and longitude) and altitude assigned by the shooting device 200, or location information included in the shooting log associated with the shooting data. Based on this location information, the forest analysis unit 113 identifies the area to be analyzed at a predetermined level of granularity. Specifically, the forest analysis unit 113 may be configured to identify the area in the following order of priority: (1) country, (2) administrative divisions such as prefectures, and (3) predetermined mesh divisions or management areas. For example, if centroid coordinates associated with tree canopy polygons can be acquired, the area may be identified based on these centroid coordinates. If location information at the tree canopy polygon level cannot be acquired, the area may be identified based on a representative point of the shooting range (such as the center point of the shooting range) or the flight log of the shooting device 200.

[0061] The forest analysis unit 113 may perform a fallback process to continue subsequent processing without identifying the region if location information is missing or the accuracy of the location information is below a predetermined threshold. For example, the forest analysis unit 113 may perform at least one of the following: (a) adopt a region specified by the user; (b) adopt a default coefficient set or estimation model that is applicable when a country or region cannot be identified; or (c) replace the tree species with a broad classification (coniferous trees / broad-leaved trees, etc.) and select an estimation model. This allows the estimation process of trunk volume or biomass and carbon dioxide sequestration to be performed without interruption, even if there are deficiencies in location information.

[0062] The data creation unit 112 (forest analysis data creation unit) is a functional unit that generates analytical data for use in forest analysis based on the data acquired by the data acquisition unit 111. Specifically, the data creation unit 112 generates orthomosaic images from optical images and also generates a crown height model (CHM) based on optical images and / or LiDAR point cloud data. The crown height model generation process in the data creation unit 112 is switched depending on whether or not LiDAR point cloud data is available, and the detailed processing procedure will be described later.

[0063] The forest analysis unit 113 is a functional unit that analyzes forest information using orthomosaic images and canopy height models generated by the data creation unit 112. Specifically, the forest analysis unit 113 separates the canopy region of each tree to generate canopy polygons and identifies the tree species for each canopy polygon. Furthermore, based on the canopy polygons and canopy height models, the forest analysis unit 113 estimates at least one of the following: tree height, diameter at breast height, trunk volume or biomass, and carbon dioxide sequestering amount.

[0064] The storage unit 17 is a storage area for storing various data used by the forest information analysis device 100, and a database 171 is provided within the storage unit 17. The database 171 stores optical images acquired by the data acquisition unit 111, LiDAR point cloud data, external data, orthomosaic images and canopy height models generated by the data creation unit 112, and canopy polygons and forest information generated by the forest analysis unit 113.

[0065] The communication unit 18 is a functional unit for communicating between the forest information analysis device 100 and the external network N. Through the communication unit 18, data is transmitted and received with the imaging device, communication with external databases and cloud services is performed, and analysis results are transmitted to external systems. With the forest information analysis device 100 configured as described above, it becomes possible to consistently analyze forest information at the individual tree level using optical images and / or LiDAR point cloud data acquired from the air.

[0066] Figure 4 is a flowchart showing the overall processing procedure of a forest information analysis method according to one embodiment of the present invention. Note that the processing procedure shown in Figure 4 is an example, and some processing may be omitted or modified depending on the state of the forest to be analyzed, the type of data that can be obtained, or user settings. The processing procedure in Figure 4 will be described below, showing the correspondence with each functional part of the forest information analysis device 100 shown in Figure 3. The forest information analysis method is executed on a computer that functions as a forest information analysis device 100 by a forest information analysis program.

[0067] First, when the forest information analysis process is started (start), in step S101, the data acquisition unit 111 acquires data relating to the forest to be analyzed. Specifically, the data acquisition unit 111 acquires optical images obtained by photographing the forest, and, if necessary, acquires LiDAR point cloud data acquired simultaneously with or separately from the optical images (step S101). The data acquisition unit 111 may acquire at least one of the optical images obtained by photographing the forest and the LiDAR point cloud data acquired simultaneously with or separately from the optical images. This data may be acquired from a photographing device such as a drone via the communication unit 18, or via a storage medium or the like.

[0068] Next, in step S102, the data creation unit 112 generates an orthophoto image based on the optical image acquired in step S101. The orthophoto image is an orthographic image corrected for the effects of the camera's orientation and terrain at the time of shooting, and is used in the subsequent tree canopy analysis process.

[0069] Next, in step S103, the data creation unit 112 generates a canopy height model (CHM). The canopy height model is generated based on optical images and / or LiDAR point cloud data, and is calculated, for example, as the difference between a digital surface model (DSM) and a digital terrain model (DTM). The canopy height model generation process is switched depending on whether or not LiDAR point cloud data is available, but the details of this will be described later.

[0070] Next, in step S104, the forest analysis unit 113 separates the canopy region for each tree based on the generated canopy height model and generates a canopy map. The canopy map is information indicating the canopy region corresponding to each tree and is represented as a canopy polygon. Then, in step S105, the forest analysis unit 113 identifies the tree species based on the canopy map and orthomosaic image and generates a tree species map. Tree species identification is performed by extracting the image region corresponding to the canopy polygon and using a trained model. The generation of the canopy map in S104 and the generation of the tree species map in S105 include inference processing using a trained model, and these processes may be executed at high speed using a GPU 19 suitable for parallel computing. This makes it possible to efficiently perform canopy separation and tree species identification at the individual tree level, even when targeting a wide area of ​​forest.

[0071] Next, in step S106, the forest analysis unit 113 estimates tree height and diameter at breast height based on the canopy height model and canopy map, and generates a tree height / diameter at breast height map. For example, tree height is calculated as the maximum value of the canopy height model within the canopy polygon, and diameter at breast height is calculated using an estimation formula based on tree height and canopy area. Then, in step S107, the forest analysis unit 113 calculates trunk volume or biomass based on the estimated tree height and diameter at breast height, as well as tree species information, and generates a trunk volume or biomass map. A calculation formula or parameters appropriate to the region or country are used to calculate trunk volume or biomass. The tree height / diameter at breast height map, trunk volume or biomass map, and CO2 fixation map generated in the processing from S106 onwards are intermediate products that are sequentially generated using the canopy map and tree species map generated in S104 and S105 as input, and the output results of each step are used in the next stage of processing.

[0072] Furthermore, in step S108, the forest analysis unit 113 calculates the amount of carbon dioxide sequestered based on the estimated trunk volume or biomass and generates a CO2 sequester map. Once the above processes are completed, the forest information analysis process ends (end). These maps and analysis results are stored in the database 171 of the storage unit 17 and output in a format usable as geographic information data. This geographic information data is used for forest management, resource assessment, environmental value assessment, etc.

[0073] Figure 5 is a flowchart showing an example of a processing procedure for generating a tree canopy height model (CHM) according to one embodiment of the present invention. This process is performed by the data acquisition unit 111 and the data creation unit 112 in the forest information analysis device 100 shown in Figure 3.

[0074] First, when processing begins (start), in step S201, the data acquisition unit 111 acquires optical images obtained by photographing the forest to be analyzed. The optical images are acquired by a photographing device such as a drone and may include visible light images or near-infrared images.

[0075] Next, in step S202, the data creation unit 112 performs Structure from Motion (SfM) processing on the optical image acquired in step S201 to generate an orthomosaic image. Through SfM processing, the shooting orientation and three-dimensional structure are estimated from multiple optical images, and an orthomosaic image with terrain distortion corrected is generated. Subsequently, in step S203, the data creation unit 112 determines whether or not LiDAR point cloud data has been acquired.

[0076] If LiDAR point cloud data has been acquired (YES), the process proceeds to step S204; otherwise, the process proceeds to step S208. If LiDAR point cloud data has been acquired, in step S204, the data creation unit 112 generates a digital surface model (DSM) based on the LiDAR point cloud data. Next, in step S205, ground detection is performed based on the generated DSM or LiDAR point cloud data, and points or regions corresponding to the ground are extracted.

[0077] Next, in step S206, a digital terrain model (DTM) is generated based on the extracted ground-corresponding points. Then, in step S207, a tree canopy height model (CHM) is generated by calculating the difference between the generated DSM and DTM. In this way, when LiDAR point cloud data is available, a tree canopy height model that reflects the ground shape with high accuracy can be generated.

[0078] On the other hand, if LiDAR point cloud data has not been acquired, in step S208, the data creation unit 112 generates a digital surface model (DSM) based on the three-dimensional structure generated by the SfM process. Next, in step S209, a digital elevation model (DEM) provided by an external organization such as the Geospatial Information Authority of Japan is acquired. The DEM may be acquired by making an API call to an externally published service, or it may be acquired as a pre-prepared DEM file.

[0079] Next, in step S210, the difference between the generated DSM and the acquired DEM is calculated. Then, in step S211, a correction is performed so that the height of the ground portion becomes zero, and a canopy height model (CHM) is generated. Once the above processing is completed, the canopy height model generation process is terminated (end). In this way, according to this embodiment, the canopy height model generation method can be switched depending on the presence or absence of LiDAR point cloud data, and it is possible to generate a canopy height model flexibly and practically according to the conditions of the data that can be acquired.

[0080] Figure 6 is a schematic diagram illustrating the relationship between the Digital Surface Model (DSM), Digital Topography Model (DTM), and Canopy Height Model (CHM) used in one embodiment of the present invention. As shown in Figure 6, the Digital Surface Model (DSM) is a model that represents the height distribution including objects on the ground surface, and includes height information of tree canopies, buildings, etc., in addition to the ground. The DSM is generated, for example, based on LiDAR point cloud data or based on a three-dimensional structure generated by Structure from Motion (SfM) processing.

[0081] Furthermore, a Digital Terrain Model (DTM) is a model that represents the height distribution of the ground itself, excluding features such as trees and buildings from the Earth's surface. DTMs are generated by detecting ground-corresponding points from LiDAR point cloud data and are used as a reference plane that reflects the topography. In Figure 6, the DTM is represented as a curve indicating the ground shape.

[0082] Furthermore, the Canopy Height Model (CHM) is a model calculated as the difference between the Distance Measurement Model (DSM) and the Distance Measurement Model (DTM), and represents the tree canopy height as the relative height from the ground. In other words, the CHM is a model that shows tree height information relative to ground level at each location.

[0083] In the example shown in Figure 6, the dotted curve represents the DSM, and the solid curve represents the DTM. The height obtained as the difference between DSM and DTM is the CHM, and for the central tree, the height from the ground (DTM) to the top of the canopy (DSM) is indicated by an arrow. The height indicated by this arrow is the CHM value corresponding to the tree's height. Thus, DSM is a height model that includes all structures on the ground surface, DTM is a height model that represents only the ground, and CHM is a tree height model obtained as the difference between these two.

[0084] In the present invention, if LiDAR point cloud data can be acquired, the DSM and DTM are directly generated and the CHM is calculated using the processing procedure shown in Figure 5. If LiDAR point cloud data cannot be acquired, the CHM is generated from the difference between the DSM generated by SfM processing and the externally acquired DEM (equivalent to DTM). By using the CHM defined as described above, the present invention makes it possible to perform stable tree height estimation and tree crown analysis on a single tree basis without being affected by the topography.

[0085] Figure 7 is an explanatory diagram showing a specific example of input data used in one embodiment of the present invention, and shows (a) an orthophoto, (b) a digital surface model (DSM), and (c) a digital terrain model (DTM) in parallel. The orthophoto shown in Figure 7(a) is generated by applying orthorectification correction to multiple optical images acquired from the air, and represents the shape, color tone, and arrangement of the tree canopy in correspondence with its planar position. The orthophoto is used as input data to provide appearance information of the tree canopy in the tree canopy separation process and tree species identification process described later.

[0086] The DSM shown in Figure 7(b) is a numerical surface model generated based on Structure from Motion (SfM) processing of optical images or LiDAR point cloud data, and shows the height distribution including tree canopies, topography, structures, etc., present on the ground surface. The DSM retains height information, including the position of the top of the trees.

[0087] On the other hand, the DTM shown in Figure 7(c) is a numerical terrain model generated by extracting ground-corresponding points from LiDAR point cloud data, and shows the height distribution of the ground surface itself with features such as trees and structures removed. In this invention, the tree crown height model (CHM) shown in Figure 6 is generated by calculating the difference between the DSM shown in Figure 7(b) and the DTM shown in Figure 7(c). In this way, by using orthomosaic images, DSM, and DTM as input data, it becomes possible to accurately grasp the height and canopy structure of trees while eliminating the influence of terrain relief.

[0088] Figure 8 is an explanatory diagram showing an example of ground correction (ground height correction) processing for a tree canopy height model (CHM) in one embodiment of the present invention. In this embodiment, the data creation unit 112 (forest analysis data creation unit) calculates an approximate plane or approximate surface based on the height values ​​of the ground-corresponding area included in the generated CHM, and applies a correction to the CHM based on the approximate plane or approximate surface.

[0089] CHM is generated from the difference between DSM and DTM (or DEM), but due to terrain relief, errors in external DEMs, reconstruction errors derived from SfM, etc., the area corresponding to the ground may not necessarily be at zero height, and a planar offset or gentle slope (drift) may remain. Therefore, the data creation unit 112 corrects the CHM based on the ground-corresponding area and adjusts it so that the ground height becomes a reference value (e.g., 0).

[0090] Figure 8(a) shows an example of a screen for selecting the ground area (ground-corresponding region). The data creation unit 112 acquires multiple points or regions that are determined to correspond to the ground on the CHM through user operation or automatic extraction. In Figure 8(a), the selected ground points are shown as orange points. Ground points may be selected from areas without vegetation, such as forest roads, bare ground, or logging sites, and it is desirable to select multiple points to ensure the stability of the approximate surface. Ground points may be newly selected, or they may be acquired as a saved ground point file and reused / edited.

[0091] Next, Figure 8(b) shows an example of a screen for selecting a correction method. The data creation unit 112 calculates an approximate surface to be used as a correction standard based on the acquired ground points (or height values ​​of the ground-corresponding area). As correction methods, for example, (i) plane correction, which fits an approximate plane to the ground points using the least squares method, or (ii) surface correction, which fits an approximate surface (e.g., a polynomial surface, a spline surface, etc.) considering the curvature of the terrain can be used. Plane correction is effective when the ground offset is approximately constant or appears as a slope in one direction, while surface correction is effective when the error is distributed over a wide area in undulating terrain.

[0092] The data creation unit 112 corrects the height value of each pixel (or mesh) of the CHM based on the calculated approximate plane or approximate surface. For example, by subtracting the height value of the approximate surface from the CHM, the height corresponding to the selected ground point is corrected to zero, and the corrected CHM (corrected CHM data) is generated. This reduces the effects of topographic residuals and reference plane displacement, and improves the accuracy of subsequent processes such as tree crown separation, tree height estimation, and breast height diameter estimation.

[0093] Figure 9 is a flowchart showing an example of a processing procedure in a forest information analysis device 100 according to one embodiment of the present invention, which uses orthomosaic images and a crown height model (CHM) to separate the crown region, identify tree species, estimate tree height and diameter at breast height, and estimate trunk volume or biomass and carbon dioxide sequestering amount.

[0094] This process is primarily performed by the forest analysis unit 113 shown in Figure 3. First, in step S301, the forest analysis unit 113 performs canopy separation processing based on orthomosaic images and CHM to extract the canopy region for each tree. Canopy separation is performed, for example, by using a local maximum filter and watershed segmentation on the CHM, or by region extraction processing using a trained model on the orthomosaic image. Furthermore, canopy separation can also be performed by object detection or region extraction without using a trained model on the orthomosaic image or CHM. This process generates a canopy map in which the canopy region corresponding to each tree is represented as a polygon.

[0095] Next, in step S302, the forest analysis unit 113 performs tree species identification and dead tree detection based on the generated canopy map and orthomosaic image. Specifically, it extracts image regions corresponding to canopy polygons and identifies tree species by inputting these image regions into a trained model for tree species identification. It can also detect dead trees based on the color tone and shape of the canopy. This generates a tree species map.

[0096] Next, in step S303, the forest analysis unit 113 estimates the height of each tree based on the tree crown height model (CHM) and the tree species map. For example, the forest analysis unit 113 calculates the maximum value from the CHM height values ​​within the area corresponding to each tree crown polygon as the tree height.

[0097] In step S304, the forest analysis unit 113 estimates the diameter at breast height for each tree based on the tree height and crown area estimated in step S303 and the tree species estimated in step S302. The diameter at breast height may be calculated using a pre-set estimation formula or a trained model for each tree species or tree type. The forest analysis unit 113 also records the estimated tree height results obtained in step S303 and the estimated diameter at breast height results obtained in step S304, associating them with the crown polygons, and generates a tree height / diameter at breast height map. The tree height / diameter at breast height map is configured, for example, as geographic information data in which the tree height and diameter at breast height are assigned to each crown polygon.

[0098] Next, in step S305, the forest analysis unit 113 estimates the trunk volume (or biomass) for each tree based on the tree height and breast height diameter estimated in step S304 and the tree species estimated in step S302. Specifically, the forest analysis unit 113 uses the tree height, breast height diameter, and tree species information associated with each tree crown polygon to calculate the trunk volume of the tree using a predetermined trunk volume calculation formula or a trained model. For example, empirical formulas or regression formulas defined for each tree species can be used as trunk volume calculation formulas, and in these formulas, tree height and breast height diameter are used as explanatory variables.

[0099] Furthermore, the forest analysis unit 113 can identify the region or country to which the trees belong based on the location information of the trees to be analyzed, and can select the trunk volume calculation formula or calculation parameters to be used according to the region or country. The forest analysis unit 113 also automatically identifies the region (country, prefecture, etc.) to which the trees belong by referring to the location information (e.g., GPS information, latitude and longitude) associated with the data to be analyzed (photographed data or tree canopy polygons). Then, based on the identified region and the tree species identified in the tree species map, the forest analysis unit 113 automatically selects and sets the model to be applied to each tree from the tree species and region-specific trunk volume estimation models (volume formulas, coefficient sets, etc.) stored in the memory unit 17.

[0100] The forest analysis unit 113 refers to the model selection table stored in the memory unit 17 and determines the model ID or coefficient set ID corresponding to the identified regional division and tree species (or tree species class). Based on the determined model ID or coefficient set ID, the forest analysis unit 113 may automatically set the trunk volume estimation model and the calculation parameters for carbon dioxide sequestration.

[0101] For example, by specifying input data (e.g., polygon data with tree height and breast height diameter linked to tree crown polygons), the system can acquire a region (e.g., prefecture) based on GPS information associated with the data, and then automatically assign a trunk volume estimation model (e.g., region-specific, tree species-specific model) associated with that region for each tree species (e.g., cedar, cypress, broadleaf trees, etc.). Users may also check or manually change the model assigned to each tree species as needed.

[0102] This makes it possible to perform more realistic estimations of trunk volume, taking into account regional characteristics and differences in forest management conditions. Furthermore, the forest analysis unit 113 records the estimated trunk volume in association with tree canopy polygons and generates a trunk volume map. The trunk volume map is composed of geographic information data in which, for example, estimated trunk volume values ​​are assigned to each tree canopy polygon, and is used for forest management, resource assessment, or logging planning.

[0103] Next, in step S306, the forest analysis unit 113 estimates the amount of carbon dioxide fixed by each tree based on the trunk volume map generated in step S305. Specifically, the forest analysis unit 113 uses the estimated trunk volume associated with each tree canopy polygon and calculates the amount of carbon dioxide fixed by the tree by applying parameters such as the expansion factor, underground part ratio, volume density, carbon content, or carbon dioxide conversion factor defined for each tree species. These parameters may be set by tree species, region, or country, and appropriate parameters are selected based on the location information of the trees to be analyzed.

[0104] Furthermore, the forest analysis unit 113 records the estimated carbon dioxide fixation amount in association with the canopy polygons and generates a CO2 fixation amount map. The CO2 fixation amount map is composed of geographic information data in which, for example, the estimated carbon dioxide fixation amount is assigned to each canopy polygon.

[0105] Furthermore, the forest analysis unit 113 can aggregate the amount of carbon dioxide sequestration in the entire analysis area or in any arbitrary area based on the generated CO2 sequestration map and output it as statistical information. This statistical information is used for evaluating the carbon storage capacity of forests, calculating carbon credits, or measuring the effectiveness of environmental policies. The various maps generated in this way are stored in the storage unit 17 as geographic information data associated with tree canopy polygons and are used for forest management, resource assessment, or environmental assessment.

[0106] Figure 10 is a flowchart showing the details of the process for generating a canopy map (step S301) in the forest information analysis process shown in Figure 9. It illustrates a specific procedure for separating the canopy region of each tree based on orthomosaic images and / or canopy height models (CHMs) and generating a canopy map represented as canopy polygons. Figure 10 shows an example of a processing procedure in the forest information analysis device 100 according to one embodiment of the present invention, where the canopy separation method is switched according to the forest conditions to be analyzed when performing canopy separation (step S301 in Figure 9). This process is mainly performed by the forest analysis unit 113.

[0107] First, once processing begins, in step S401, the forest analysis unit 113 determines whether or not it wants to detect tree canopies in planted forests or the like based on specific parameters. For example, the forest analysis unit 113 may determine whether to apply a parameter-adjustable canopy separation method or a canopy separation method using a trained model, based on user settings, attribute information of the area to be analyzed, or the estimated forest type.

[0108] (Parameter-adjustable: CHM-based canopy separation) If "YES" is determined in step S401, the forest analysis unit 113 performs canopy separation processing using the canopy height model (CHM) and orthomosaic images. Specifically, first, in step S402, the forest analysis unit 113 generates a mask (vegetation area mask) for extracting vegetation areas. The vegetation area mask may be generated, for example, by extracting areas where the height value of the CHM is greater than or equal to a predetermined threshold as vegetation areas, or it may be generated using the pixel values ​​(hue, brightness, vegetation index, etc.) of the orthomosaic image. Furthermore, morphological processing such as expansion and contraction may be applied for noise reduction and area correction.

[0109] Next, in step S403, the forest analysis unit 113 applies a local maximum filter to the CHM to detect tree vertices (points corresponding to the tops of trees). When applying the local maximum filter, tree vertices may be extracted while suppressing over-detection or under-detection by using predetermined parameters such as window size, smoothing conditions, and minimum tree height threshold. These parameters can be selected or changed according to, for example, tree species, forest age, planting density, or regional characteristics. Next, in step S404, the forest analysis unit 113 performs canopy separation using watershed segmentation. For example, the forest analysis unit 113 performs watershed segmentation using CHM (or its gradient) and estimates the canopy boundaries between adjacent trees by using the tree vertices detected in step S403 as markers. Furthermore, to improve the accuracy of the canopy boundaries, the boundaries may be corrected by using edge information and texture information from orthomosaic images in combination. In this way, the canopy region of each tree is separated, and a canopy map is generated, which is represented as canopy polygons.

[0110] (Pre-trained model: orthomosaic-based tree canopy separation) On the other hand, if the result in step S401 is "NO", the process proceeds to step S405 to determine whether or not to use the trained model. If the result in step S405 is "YES", the forest analysis unit 113 takes the orthomosaic image as input and performs canopy separation using the trained model (AI) (step S406). The trained model may be, for example, a model that performs object detection, instance segmentation, or region extraction, and it extracts canopy regions on the orthomosaic image on a tree-by-tree basis to generate canopy polygons. With this method, canopy regions can be separated based on the appearance features of the orthomosaic image, even when the canopy shape is diverse and the boundaries are complex.

[0111] If the determination in step S405 is "NO", the forest analysis unit 113 takes the orthomosaic image or CHM as input and performs canopy separation by object detection or region extraction (step S407). The object detection or region extraction here may be rule-based methods such as edge detection, region growth, clustering, or thresholding. The selection of orthomosaic image or CHM as input data can be made according to the data quality of the target area, the degree of canopy overlap, the presence or absence of known parameters, etc., and both may be used in combination.

[0112] The forest analysis unit 113 may assign a crown ID to the crown area obtained in step S404, S406, or S407, and manage it by associating it with characteristic quantities such as crown area, crown diameter, and tree height. This makes it possible to aggregate individual values ​​in subsequent stages such as tree species map generation, tree height / breast height diameter map generation, trunk volume or biomass map generation, and CO2 fixation amount map generation.

[0113] If the input data is wide-area or high-resolution, the input may be divided into tiles of a predetermined size, inference or division processing may be performed on a tile-by-tile basis, and then the overlapping regions of the tile boundaries may be aligned and integrated. During integration, post-processing such as suppression of overlap detection (estimation of overlaps within the same tree canopy), boundary smoothing, and removal of small regions may be applied.

[0114] As described above, according to this embodiment, when stable detection is possible in artificial forests, etc., by adjusting parameters, a CHM-based canopy separation method can be applied, and in other cases, a canopy separation method using a trained model can be applied, allowing for the selection of an appropriate canopy separation method according to the conditions of the analysis target. This enables stable extraction of canopy regions on a tree-by-tree basis, even when the acquired data and forest conditions differ, contributing to improved accuracy in subsequent tree species identification and forest information estimation.

[0115] Figure 11 is an explanatory diagram showing an example of displaying the results of tree apex detection and tree crown separation in one embodiment of the present invention. In this embodiment, the forest analysis unit 113 detects tree apex using a local maximum filter based on a tree crown height model (CHM) in the tree crown separation process shown in Figure 10 (for example, steps S403 and S404), and generates tree crown polygons by separating the tree crown region for each tree by performing watershed segmentation using the tree apex as a marker.

[0116] In Figure 11, the circular dots indicate tree vertices (points corresponding to the tops of trees) obtained by the local maximum filter. The forest analysis unit 113 searches for local maximums in the CHM with a predetermined window size and extracts these local maximums as tree vertex candidates. Parameters such as window size, smoothing conditions, and minimum tree height threshold may be set or changed according to the tree species, forest age, planting density, or type of forest being analyzed.

[0117] Furthermore, in Figure 11, the thick lines indicate the canopy boundaries estimated by watershed segmentation, and the area enclosed by these boundaries corresponds to the canopy polygon (canopy region). The forest analysis unit 113 performs watershed segmentation using, for example, CHM or its inverted image, gradient image, etc., and estimates the boundaries between adjacent trees by using the tree vertices obtained in step S403 as markers. This makes it possible to stably extract the canopy region on a single tree basis, even when the canopies are in contact, such as in densely planted forests.

[0118] Furthermore, as shown in step S405 of Figure 10, the forest analysis unit 113 can also perform canopy separation using a trained model (AI) that takes orthomosaic images as input. In this case, tree apex detection is not mandatory, and only lines corresponding to the canopy boundaries (canopy polygons corresponding to the thick lines shown in Figure 11) may be generated. As described above, according to this embodiment, tree apex detection based on CHM, canopy separation by watershed segmentation, and canopy separation by AI can be selected or used in combination depending on the conditions of the analysis target, and the generation of canopy maps can be achieved with high accuracy at the individual tree level.

[0119] Figure 12 is a flowchart illustrating the details of the tree species map generation process (step S302) in the forest information analysis process shown in Figure 9. It illustrates a specific procedure for identifying tree species on a per-canopy polygon basis based on orthomosaic images and canopy maps (canopy polygon sets), detecting dead trees as needed, and generating a tree species map by associating the results with the canopy map. This process is mainly performed by the forest analysis unit 113.

[0120] First, in step S501, the forest analysis unit 113 refers to the orthomosaic image and the canopy map and extracts image regions (patches) corresponding to each canopy polygon. The image regions extracted here may be, for example, the bounding rectangle region of the canopy polygon, or the region masked by the canopy polygon. The extracted image regions are used as input data for the subsequent identification process by the AI ​​model.

[0121] Next, in step S502, the forest analysis unit 113 changes the AI ​​model used for tree species identification according to the country or region to be analyzed. For example, considering that the distribution of training data and the composition of tree species differ from country to country, a pre-trained model optimized for each country may be selected and used. Subsequently, in step S503, the forest analysis unit 113 determines whether or not to perform dead tree detection (whether or not to select dead tree detection). If dead tree detection is to be performed (YES), the process proceeds to step S504; if not (NO), the process proceeds to step S508.

[0122] (When detecting dead trees) When performing dead tree detection, in step S504, the forest analysis unit 113 sets a trained model for dead tree detection (dead tree detection AI model). Next, in step S505, the set dead tree detection AI model identifies dead trees in the image region corresponding to each canopy polygon. Then, in step S506, the identification results of whether a tree is dead or healthy are merged (associated) with the canopy map based on the identifier (ID) assigned to the canopy polygon.

[0123] Furthermore, in step S507, post-processing may be performed as needed, such as deleting polygons that do not correspond to dead trees (for example, healthy tree polygons). This makes it possible to obtain geographic information data (dead tree map) in which only dead trees are extracted.

[0124] (When identifying tree species) On the other hand, if dead tree detection is not performed in step S503 (NO), the forest analysis unit 113 proceeds to tree species identification processing. Specifically, in step S508, the forest analysis unit 113 determines whether or not it wants to focus on identifying a specific tree species (or a limited set of tree species) in the planted forest, etc. If it wants to identify a specific tree species (YES), the process proceeds to step S509; otherwise (NO), the process proceeds to step S512.

[0125] (When identifying a specific tree species) When identifying a specific tree species, in step S509, the forest analysis unit 113 sets a trained model (specific tree species identification AI model) suitable for identifying a specific tree species. Next, in step S510, tree species identification is performed using this model. Then, in step S511, the tree species identification results are merged (associated) with the canopy map based on the ID of each canopy polygon. This generates a tree species map in which a tree species is assigned to each canopy polygon.

[0126] (When identifying multiple tree species: This includes narrowing down candidates by region.) If the system is to identify a variety of tree species rather than limiting itself to a specific species, in step S512, the forest analysis unit 113 sets up a trained model (multi-tree species identification AI model) suitable for identifying other tree species. Next, in step S513, the forest analysis unit 113 determines whether or not to narrow down the target tree species based on the region. If narrowing down is performed (YES), the process proceeds to step S514; if narrowing down is not performed (NO), the process proceeds to step S520. If narrowing down by region is performed, in step S514, the forest analysis unit 113 acquires location information corresponding to the data to be analyzed (each tree canopy polygon or image data).

[0127] Next, in step S515, the region or administrative division (e.g., prefecture) corresponding to the data is obtained based on the location information. Then, in step S516, candidate tree species to be identified (e.g., a list of tree species with a high probability of occurrence) are obtained in the corresponding region or administrative division. Furthermore, in step S517, the tree species to be identified are narrowed down based on the obtained candidates, and a set of candidate tree species to be used for the identification process is set. After that, in step S518, tree species identification is performed considering the narrowed-down set of candidate tree species. Subsequently, in step S519, the tree species identification results are merged (associated) with the canopy map based on the ID, and a tree species map is generated.

[0128] On the other hand, if regional filtering is not performed in step S513 (NO), in step S520, tree species identification is performed using a multi-tree species identification AI model, and in step S521, the tree species identification results are merged (associated) with the canopy map based on the ID. As described above, according to this embodiment, tree species identification can be performed on image regions extracted at the canopy polygon level while selecting a trained model and narrowing down tree species candidates according to the country, region, etc. This makes it possible to generate a tree species map that is consistent with the canopy map while suppressing misidentification caused by regional characteristics.

[0129] Figure 13 is an explanatory diagram showing an example of displaying tree species identification results in one embodiment of the present invention. In this embodiment, the forest analysis unit 113 identifies tree species on a per-canopy polygon basis based on the canopy map (a collection of canopy polygons) and orthomosaic images, and generates the identification results as a tree species map (for example, step S302 in Figure 9 and the processing shown in Figure 12).

[0130] Specifically, the forest analysis unit 113 extracts image regions of orthomosaic images corresponding to each canopy polygon (for example, image patches cut out or masked by the canopy polygons), inputs these image regions into a trained model (AI model) for tree species identification, and estimates the tree species class. The AI ​​model for tree species identification classifies tree species or tree species groups (conifers, broad-leaved trees, or cedar, cypress, etc.) based on the appearance features of each instance (color tone, leaf texture, canopy shape, shading pattern, etc.).

[0131] The forest analysis unit 113 assigns the estimated tree species (or tree species class) as attribute information, associating it with the identifier (polygonID, etc.) of the canopy polygon. This generates a tree species map (for example, GIS data such as a shapefile) in which tree species information is linked to each canopy polygon. The tree species map is used as input information for selecting estimation formulas or coefficients for each tree species in subsequent tree height / breast height diameter estimation, trunk volume estimation, CO2 fixation amount estimation, etc.

[0132] In the example shown in Figure 13, tree canopy polygons are drawn with thick borders, and each polygon is displayed in varying shades (color coding) according to the tree species identification result. For example, as shown in the legend, the trees are classified into multiple classes such as cedar, cypress, and others, and the class (classID) or tree species name is assigned to the tree canopy polygon. Users can switch the display / hide of layers (tree species map, tree canopy map, orthomosaic image, etc.) on the screen to visualize or extract data for each tree species class.

[0133] Furthermore, the forest analysis unit 113 may be configured to switch the tree species identification model used based on location information, or to restrict the candidate tree species to be identified according to the region. This allows for the stable generation of a tree species map consistent with the canopy map while suppressing misidentification caused by regional differences.

[0134] Figure 14 is a flowchart illustrating the details of the process for generating tree height and breast height diameter maps (steps S303 and S304) in the forest information analysis process shown in Figure 9. It illustrates the specific procedure for estimating the tree height of each tree based on the tree species map and the crown height model (CHM), and further estimating the breast height diameter based on the crown area and estimation formula. This process is mainly performed by the forest analysis unit 113.

[0135] First, in step S601, the forest analysis unit 113 extracts the height value of the CHM within each canopy polygon included in the canopy map. In order to suppress the inclusion of contaminants near the canopy boundary (adjacent trees, ground, gaps, etc.), the extraction target may be an area in which the canopy polygon has been reduced inward by a predetermined amount (inset).

[0136] Next, in step S602, the forest analysis unit 113 obtains a representative value of the extracted CHM height and stores it as tree height. As the representative value, the maximum value may be used, but from the viewpoint of suppressing outliers (noise points), for example, (i) the maximum value after median filtering or smoothing, (ii) the upper percentile value (e.g., the 95th or 99th percentile), (iii) the maximum value after excluding values ​​exceeding the mean ± predetermined range, etc., may be used. This makes it possible to stably calculate tree height values ​​on a per-canopy polygon basis while reducing the influence of spike-like noise contained in the CHM and outliers caused by canopy separation errors.

[0137] Next, in step S603, the forest analysis unit 113 calculates the area of ​​each tree canopy polygon (tree canopy area). The tree canopy area is calculated, for example, as an area based on a geographic coordinate system. In addition, to suppress underestimation or overestimation of the tree canopy area, post-processing such as smoothing the polygon shape, removing minute areas, or correcting for overlap with adjacent polygons may be performed.

[0138] Next, in step S604, the forest analysis unit 113 determines whether or not there are actual measured values ​​(training data) of the diameter at breast height measured in the area to be analyzed. For example, this determination may be made based on the input field survey data or the presence or absence of actual measured data registered in the database 171 of the storage unit 17.

[0139] (A) If field measurement data is available: Calibrate and apply the estimation formula for each tree species (S605~S606) If a breast height diameter measured on-site exists (YES), in step S605, the forest analysis unit 113 creates a breast height diameter estimation formula (or estimation model) for each tree species, using the crown area and tree height as explanatory variables. The estimation formula may be created, for example, by regression analysis or machine learning, and coefficients or parameters are estimated for each tree species based on the tree species information provided by the tree species map.

[0140] For example, any one or a combination of the following can be used as a concrete example of an estimation formula. Example 1: DBH = a · A b · H c (A: crown area, H: tree height, a, b, c: species coefficients) Example 2: DBH = a · H + b · √A + c Example 3: A trained model (regression tree, neural network, etc.) that outputs DBH = f(H, A) Note that the abc parameters (tree species coefficients) differ depending on the tree species and tree type (broadleaf tree / coniferous tree).

[0141] Furthermore, when creating the estimation formula, in order to suppress the influence of outliers, outlier removal of measured DBH, robust regression, or generalization performance evaluation based on cross-validation may be performed. Next, in step S606, the forest analysis unit 113 estimates the diameter at breast height for each tree species based on the crown area and tree height using the estimation formula created in step S605.

[0142] (B) If there is no on-site measurement data: Select and apply the default estimation formula (S607~S608) On the other hand, if no breast height diameter has been measured on-site (NO), the forest analysis unit 113 estimates the breast height diameter using a pre-set estimation formula or externally provided estimation parameters. Specifically, in step S607, the crown area is calculated (or reused), and in step S608, the breast height diameter is estimated for each tree species from the estimation formula based on the crown area and tree height. The estimation formula or parameters used in this case may be selected, for example, in the following order of priority: (1) Calibration formulas (regional calibration parameters) previously created for the same region and tree species (2) Default formulas for tree species (regional default parameters) established on a regional or national basis. (3) Generalized formulas (general parameters) specific to tree species (4) If the above is not available, apply the estimation formula after substituting the tree species with a broad classification (coniferous trees / broad-leaved trees, etc.).

[0143] Furthermore, to ensure the validity of the estimation results, constraints such as lower and upper limits for breast height diameter, or consistency conditions with tree height (e.g., excluding DBH that is unnaturally large relative to tree height) may be imposed, and re-estimation or flagging (confidence rating) may be performed as necessary.

[0144] Finally, the forest analysis unit 113 associates the estimated results of tree height (step S602) and diameter at breast height (step S606 or S608) with the ID of each canopy polygon and assigns them as attribute information to generate a tree height / diameter at breast height map. The tree height / diameter at breast height map is configured as geographic information data that holds, for example, tree height values ​​and diameter at breast height values ​​(and, if necessary, estimation method type, coefficient group, confidence level, etc.) for each canopy polygon, and is used in subsequent estimation of trunk volume (step S305) and carbon dioxide sequestration (step S306).

[0145] Figure 15 is a flowchart illustrating the details of the process for generating a trunk volume map (step S305) in the forest information analysis process shown in Figure 9. It illustrates a specific procedure for estimating trunk volume on a per-canopy polygon basis based on tree height / breast height diameter maps and tree species maps, and generating the results as a trunk volume map. This process is not limited to estimating trunk volume, but also includes estimating above-ground biomass (hereinafter simply referred to as "biomass") or a similar quantity as an alternative indicator to trunk volume. This process is mainly performed by the forest analysis unit 113.

[0146] First, once processing begins, in step S701, the forest analysis unit 113 acquires the relevant region of the data to be analyzed. The relevant region may be identified, for example, based on location information (latitude and longitude, etc.) associated with each tree canopy polygon or image data, and may be acquired as an administrative division such as a country or prefecture, or as a predetermined regional division (forest management area, vegetation zone division, etc.).

[0147] Next, in step S702, the forest analysis unit 113 selects a calculation formula (calculation formula) for calculating the trunk volume or biomass according to the acquired region. For example, since the applicable volume formula, biomass conversion formula, coefficient system, or forest stand characteristics may differ depending on the region or country, the forest analysis unit 113 can select an appropriate one from among several calculation formulas or coefficient sets prepared for each region.

[0148] The selection of the calculation formula may be made, for example, by referring to a regional table stored in the memory unit 17, or by using coefficient information obtained from an external database or service. Subsequently, in step S703, the forest analysis unit 113 determines the calculation formula or parameters to be used based on the tree species and region, and estimates the trunk volume or biomass in terms of canopy polygons based on the tree height and breast height diameter included in the tree height / breast height diameter map.

[0149] Specifically, the forest analysis unit 113 obtains the tree species, tree height, and breast height diameter associated with the ID of each canopy polygon, and substitutes these values ​​into a volume formula or biomass estimation formula set for each tree species (for example, a regression equation with tree height and breast height diameter as explanatory variables) to calculate an estimated value of trunk volume or biomass. The estimated value is assigned as attribute information to the canopy polygon and generated as a trunk volume map or biomass map.

[0150] Furthermore, the calculation formula used to estimate trunk volume or biomass may have different coefficients for each tree species, and even for the same tree species, the coefficients may differ depending on the region. In addition, if the tree species cannot be identified, the estimation may be continued by applying a calculation formula based on broad classifications such as coniferous trees / broad-leaved trees. As described above, according to this embodiment, it is possible to estimate trunk volume or biomass by selecting a calculation formula according to the region and tree species, and to generate a trunk volume map (or biomass map) at the individual tree level.

[0151] Figure 16 is a flowchart showing the details of the process for generating a CO2 fixation map (step S306) in the forest information analysis process shown in Figure 9. It illustrates a specific procedure for estimating the amount of carbon dioxide fixed at the canopy polygon level based on a trunk volume map or biomass map, etc., and generating the result as a CO2 fixation map. This process is mainly performed by the forest analysis unit 113.

[0152] First, once processing begins, in step S801, the forest analysis unit 113 acquires the relevant region of the data to be analyzed. The relevant region may be identified, for example, based on location information (latitude and longitude, etc.) associated with each tree canopy polygon or image data, and may be acquired as an administrative division such as a country or prefecture, or as a predetermined regional division.

[0153] Next, in step S802, the forest analysis unit 113 selects a calculation formula (calculation formula) for calculating the amount of carbon dioxide sequestration according to the acquired region. For example, since the handling of carbon content, timber density, expansion factor, conversion factor, etc. may differ depending on the region or country, the forest analysis unit 113 can select an appropriate one from multiple calculation formulas or coefficient sets prepared for each region.

[0154] Next, in step S803, the forest analysis unit 113 determines the calculation formula or parameters to be used based on the tree species and region, and estimates the amount of carbon dioxide fixed based on the trunk volume or biomass. Specifically, the forest analysis unit 113 obtains the tree species and trunk volume (or biomass) associated with the ID of each tree canopy polygon, and calculates the amount of carbon dioxide fixed by applying a conversion factor set for each tree species and / or region. For example, biomass may be calculated from the trunk volume, the amount of carbon may be obtained by multiplying the biomass by the carbon content rate, and then converted to the amount of CO2 using the carbon dioxide conversion factor. Alternatively, if the biomass has already been estimated, the amount of CO2 fixed may be calculated directly from the biomass.

[0155] The forest analysis unit 113 assigns the calculated carbon dioxide fixation amount as attribute information to the canopy polygons and generates a CO2 fixation amount map. The CO2 fixation amount map is composed of geographic information data in which, for example, an estimated value of CO2 fixation amount is assigned to each canopy polygon. As described above, according to this embodiment, it is possible to estimate the amount of carbon dioxide fixation by selecting a calculation formula according to the region and tree species, and to generate a CO2 fixation amount map on a tree-by-tree basis.

[0156] Figure 17 is an explanatory diagram showing an example of displaying (a) the estimated tree height and (b) the diameter at breast height (DBH) in one embodiment of the present invention. In this embodiment, the forest analysis unit 113 calculates the tree height and diameter at breast height on a per-canopy polygon basis based on the canopy map (set of canopy polygons), the tree species map, and the canopy height model (CHM), and generates it as a tree height / diameter at breast height map (for example, steps S303 and S304 in Figure 9, and the processing shown in Figure 15).

[0157] For tree height estimation, the forest analysis unit 113 extracts the CHM height value within each tree canopy polygon and determines the representative value of that height (e.g., maximum value, upper percentile value, or maximum value after outlier removal) as the tree height of that tree. This makes it less susceptible to the influence of topographic relief and ground elevation, and allows for stable estimation of the tree height for each individual tree.

[0158] For estimating the diameter at breast height, the forest analysis unit 113 calculates the diameter at breast height on a per-canopy polygon basis, using the canopy area calculated from the canopy polygons, the estimated tree height, and the tree species or tree type (coniferous / broadleaf, etc.) assigned by the tree species map. As a formula for calculating the diameter at breast height, for example, the empirical formula (regression formula) shown in the following equation may be used. DBH = a × (tree crown area) b × (Tree height) c Here, a, b, and c are parameters, and are set to different values ​​for each tree species or tree type. The forest analysis unit 113 identifies the target tree species or tree type based on the tree species map, and selects the corresponding parameters (a, b, c) according to the identification result and applies the above formula.

[0159] The forest analysis unit 113 associates the calculated tree height and diameter at breast height with the identifier (polygonID, etc.) of the canopy polygon and assigns it as attribute information, saving it as a tree height / diameter at breast height map. The tree height / diameter at breast height map is structured as GIS data, such as a shapefile, and the tree height value and diameter at breast height value are linked to each canopy polygon.

[0160] In the example display in Figure 17, (a) shows the estimated tree height and (b) shows the estimated diameter at breast height (DBH), both of which are displayed using shading (color coding) categorized according to the estimated values ​​in canopy polygon units. Through these displays, users can visually grasp the distribution of tree height and diameter at breast height in the target area and use this information for subsequent estimations of trunk volume and CO2 fixation.

[0161] Figure 18 is an explanatory diagram showing an example of displaying (a) the estimated trunk volume and (b) the estimated carbon dioxide sequester amount (CO2 sequester amount) in one embodiment of the present invention. In this embodiment, the forest analysis unit 113 estimates the trunk volume in units of canopy polygons based on the tree height / breast height diameter map (including the estimated results shown in Figure 17) and the tree species map (see, for example, step S305 in Figure 9 and Figure 15), and further estimates the CO2 sequester amount based on the estimated trunk volume (or biomass) (see, for example, step S306 in Figure 9 and Figure 16).

[0162] In estimating trunk volume, the forest analysis unit 113 identifies the region or country (or administrative division such as a prefecture, if necessary) to which the tree belongs, based on location information associated with each canopy polygon (e.g., photographic data or GPS information held as the centroid coordinates of the canopy polygon). Then, based on this identification result and the tree species (or tree type such as conifer / broadleaf tree) assigned by the tree species map, it automatically selects from multiple candidates a calculation formula, coefficient system, or trained model (hereinafter collectively referred to as the "trunk volume estimation model") to be used for calculating trunk volume. This makes it possible to perform estimations that reflect regional differences (differences in the volume formula system, differences in forest stand conditions, etc.) and tree species differences, even with the same estimation input (e.g., tree height and diameter at breast height).

[0163] Specifically, the forest analysis unit 113 acquires the tree height and diameter at breast height (DBH) associated with the identifier (polygonID, etc.) of each canopy polygon, inputs them into the selected trunk volume estimation model, and calculates the trunk volume. The calculated trunk volume is assigned as attribute information to the canopy polygon and generated as a trunk volume map. The trunk volume map is structured as GIS data, such as a shapefile, and the estimated trunk volume is associated with each canopy polygon.

[0164] Next, in estimating the amount of CO2 sequester, the forest analysis unit 113 applies conversion factors (e.g., wood density, expansion factor, carbon content, carbon dioxide conversion factor, etc.) set for each tree species and / or region to the estimated trunk volume (or biomass) and calculates the amount of CO2 sequester per tree canopy polygon. Here again, the set of coefficients or calculation formula to be applied may be automatically selected based on the location information (GPS information) and tree species information. The calculation results are assigned as attribute information to the tree canopy polygons and generated as a CO2 sequester map.

[0165] In the example display in Figure 18, (a) shows the estimated results of trunk volume and (b) shows the estimated results of CO2 sequestering, both of which are color-coded according to the estimated values ​​in terms of tree canopy polygons. Through these GIS displays, users can visually grasp the distribution of trunk volume and CO2 sequestering in the target area. The trunk volume map and CO2 sequestering map are stored in the storage unit 17 as geographic information data associated with tree canopy polygons and are used for aggregation by region or management area, forest management planning, environmental value assessment, etc.

[0166] Figure 19 is an explanatory diagram showing an example in one embodiment of the present invention in which analysis results obtained by the forest analysis unit 113 are stored as geographic information data, and displayed, referenced, and aggregated. The analysis results are managed using tree crown polygons, which represent the canopy area of ​​each tree, as units. Each tree crown polygon is associated with attribute information such as tree species, tree height, diameter at breast height (DBH), trunk volume, and carbon dioxide fixation amount (CO2 fixation amount). This geographic information data is output in a GIS format such as a shapefile (shp) or GeoJSON, and stored in the storage unit 17 (database 171).

[0167] Figure 19(a) shows an example of the forest analysis data viewer display, where canopy polygons are superimposed on the orthomosaic image, and the attribute table at the bottom stores and displays records such as the canopy polygon identifier (polygonID, etc.), tree vertex coordinates (treetop, etc.), canopy area, tree species, tree height, DBH, trunk volume, and CO2 fixation amount. When the user selects any canopy polygon on the map, that polygon is highlighted, and the corresponding record in the attribute table is identified, allowing the user to check the estimated values ​​for individual trees. As a result, the analysis results are not just an image output, but can be used as a database linked to the estimated results for each canopy polygon.

[0168] Figure 19(b) shows an example of the area statistics viewer displaying statistical information for an arbitrary area (for example, a range specified by the user on a map, a management area, or a selected set of polygons). The forest analysis unit 113 extracts the canopy polygon group included in the selected area and, based on the attribute information assigned to the polygon group, can calculate and display statistics such as the number of trees, density per unit area (number of trees / ha, etc.), average tree height, average DBH, average trunk volume, total trunk volume, and total CO2 fixation. This makes it possible to aggregate the estimation results at the individual tree level by region or management area and utilize them for forest management, resource assessment, forest management planning, or environmental value assessment.

[0169] Figure 20 is an explanatory diagram showing an example of the data content when the analysis results generated in this embodiment are saved as GIS data (vector data) such as a shapefile. In other words, in this embodiment, each canopy polygon (polygon on a single tree basis) obtained by canopy separation is held as a feature, and attribute information such as tree species, tree height, diameter at breast height (DBH), trunk volume, and CO2 fixation amount, which are the analysis results, are associated with the polygon and saved.

[0170] In the example shown in Figure 20, each record corresponds to one tree (one canopy polygon), and for example, using the polygon identifier (polygonID) as a key, the treetop position (treetop), canopy area, identification class (classID), tree species, tree height, diameter at breast height (DBH), trunk volume, and CO2 fixation amount are stored as columns (fields). This allows the geometric shape (location and shape) of the canopy polygon and the estimated forest information (attributes) to be managed together and used for visualization, searching, and area aggregation (e.g., number of trees in an arbitrary polygon, average tree height, total trunk volume, total CO2 fixation amount) on a GIS. Note that this GIS data is not limited to shapefiles and may be saved in other formats such as GeoJSON. In addition, auxiliary information such as the regional division used for estimation, the type of estimation formula applied, the coefficient set, and the estimation confidence level may be stored as additional fields as needed.

[0171] Figures 21 and 22 are explanatory diagrams illustrating a specific example of performing color correction (hue correction or color distribution conversion) before tree species identification to address the problem of reduced tree species identification accuracy due to changes in the color tone of the tree canopy during the pollen attachment period or the autumn foliage period. For example, Japanese cedar and cypress exhibit reddish to brownish hues during the pollen accumulation period, and tree species identification models based on training data from normal periods are prone to misidentification. Therefore, the forest analysis unit 113 of this embodiment generates a corrected image by correcting the reddish to brownish hues to greenish tones for the image region (patch) or the entire orthomosaic image extracted in correspondence with the tree canopy polygon, and inputs this corrected image into a trained model for tree species identification, thereby ensuring tree species identification accuracy even under seasonal changes in appearance. This correction process can also be applied to color changes (changes from yellow to red) in deciduous trees during the autumn foliage period.

[0172] Figure 21 shows an example of color correction performed by Hue (color) substitution in the HSV color space. The forest analysis unit 113 converts the input image to the HSV color space and, for pixels whose hue component H belongs to a predetermined range (for example, the range corresponding to red to brown), replaces or shifts said H to a green hue value. On the other hand, by maintaining the saturation S and brightness V or adjusting them with predetermined correction coefficients, the appearance of the tree canopy can be normalized to a color tone close to that of the normal period without excessively damaging shape features such as shadows and textures. The range of hue substitution, the target hue value for substitution, and the adjustment conditions for S and V may be switched according to the tree species (e.g., cedar / cypress), region, shooting conditions, and season (pollen attachment period / autumn foliage period, etc.). Furthermore, the correction target may be limited to pixels inside the tree canopy polygon, thereby suppressing unnecessary correction to pixels in the background or ground surface.

[0173] Figure 22 shows an example of performing color correction using a 1D LUT (one-dimensional lookup table). The forest analysis unit 113 can perform color conversion at high speed by determining the output value by referring to a predefined conversion table for each pixel value of the input image (e.g., each RGB channel or each HSV component). The 1D LUT may be created, for example, based on the correspondence between "canopy images taken during the pollen attachment period" and "a training image corrected to a color tone equivalent to that of the normal period," or it may be generated to satisfy the matching conditions of the color histogram with respect to a reference image (a representative image of the normal period). This makes it possible to map the color tone biased towards red to brown to a green tone while maintaining the brightness distribution and contrast of the canopy, and converting it to an appearance suitable for tree species identification. Furthermore, multiple LUTs may be prepared for each region or season, and a configuration may be implemented to automatically select one according to location information and shooting time.

[0174] The methods for achieving color correction (hue correction or color distribution transformation) are not limited to those shown in Figures 21 and 22. For example, image processing techniques such as hue rotation, color statistics transfer (matching of statistical quantities such as mean and variance), color distribution transformation based on optimal transport, and histogram matching can be applied. In any of these cases, inputting the corrected image into the tree species identification model reduces the influence of appearance variations caused by seasonal factors, thereby stabilizing and improving the accuracy of tree species identification. [Industrial applicability]

[0175] The forest information analysis device, forest information analysis method, and program according to the present invention use optical images and / or LiDAR point cloud data acquired from the air to estimate forest information such as canopy region separation, tree species identification, tree height, diameter at breast height, trunk volume, biomass, and carbon dioxide sequestration amount on a tree-by-tree basis, and can store and aggregate this information as geographic information data. Therefore, the present invention can be used for forest management (management planning, resource assessment, logging and reforestation planning, etc.), forest resource assessment, forestry management support, protected forest management, disaster risk assessment, dead tree monitoring, etc. Furthermore, it can be applied to carbon storage amount assessment based on the estimated carbon dioxide sequestration amount, carbon credit calculation, greenhouse gas emission management by companies and local governments, and measurement of the effectiveness of environmental policies, etc. In addition, since the analysis results can be output in shapefile or GIS format such as GeoJSON, it can be practically used in conjunction with various systems in government, research institutions, forestry businesses, consulting firms, insurance and finance sectors, etc. [Explanation of Symbols]

[0176] 11: CPU 12: Memory 13: Bus 14: Input / Output Interface 15: Input section 16: Output section 17: Storage section 18: Communications Department 19: GPU 100: Computers, forest information analysis equipment 111: Data acquisition unit 112: Data Creation Department 113:Forest Analysis Department 171: Database 200: Drones, unmanned aerial vehicles 300: Forest N: External network

Claims

1. A forest information analysis device that analyzes forest information at the individual tree level using forest data acquired from the air, A data acquisition unit that acquires either an optical image obtained by photographing a forest, or both the optical image and LiDAR point cloud data, A forest analysis data creation unit that generates orthomosaic images and a crown height model (CHM) based on the acquired data, A forest analysis unit that, based on the orthomosaic image and the crown height model, separates the crown region for each tree and generates crown polygons, identifies the tree species for each crown polygon, generates a tree species map associating the tree species information with the crown polygons, calculates the maximum value of the crown height model within the crown polygon as the tree height, calculates the crown area of ​​the crown polygon, selects an estimation formula for each tree species based on the tree species associated with the crown polygon using the tree species map, and estimates the diameter at breast height by applying the tree height and the crown area to the selected estimation formula, A forest information analysis device comprising a storage unit that stores geographic information data associated with identified or estimated forest information to the aforementioned tree canopy polygons.

2. The aforementioned forest analysis data creation unit, depending on the presence or absence of LiDAR point cloud data, (a) A process to generate a numerical surface model (DSM) by performing Structure from Motion (SfM) processing on the optical image, and generate the canopy height model by the difference between it and an externally provided numerical elevation model (DEM), or (b) A process of generating a numerical terrain model (DTM) and a DSM from the LiDAR point cloud data, and generating the canopy height model from the difference between them. A forest information analysis device according to claim 1, which selectively performs the following:

3. The forest information analysis device according to claim 2, wherein the digital elevation model is obtained by making an API call to an external public DEM service, or as a user-specified DEM file.

4. The forest information analysis device according to claim 2, wherein the forest analysis data creation unit calculates an approximate plane or approximate surface based on the height values ​​of the ground-corresponding area included in the tree canopy height model, and applies a correction to the tree canopy height model based on the approximate plane or approximate surface.

5. The aforementioned forest analysis unit, (a) A first canopy separation method using a local maximum filter and watershed segmentation for the canopy height model, (b) A second tree canopy separation method using object detection or region extraction with a trained model on the orthophoto, and (c) A third tree canopy separation method by object detection or region extraction without using a pre-trained model for the orthoimage or CHM, The forest information analysis device according to claim 1, which separates the canopy region using at least one of the following.

6. The forest analysis unit extracts the image region corresponding to the tree canopy polygon, inputs the image region into a trained model for tree species identification, and identifies the tree species. A forest information analysis device according to claim 1, which selects a tree species identification model to be used based on location information, or limits the candidate tree species to be identified.

7. The forest information analysis device according to claim 6, wherein the forest analysis unit performs hue correction or color distribution conversion on the image region in order to correct for color changes during the pollen attachment period or the autumn foliage period, and then performs tree species identification.

8. The forest information analysis device according to claim 1, wherein the estimation formula for each tree species is DBH = a × (crown area) b × (tree height) c, and a, b, and c are set to different values ​​for each tree species or tree type.

9. The forest information analysis device according to claim 1, wherein the forest analysis unit identifies a region or country based on location information and selects a formula or parameter for calculating trunk volume, biomass, or carbon dioxide sequestration amount according to the region or country.

10. A forest information analysis method that analyzes forest information at the individual tree level using forest data acquired from the air, The steps include acquiring either an optical image obtained by photographing a forest, or LiDAR point cloud data acquired simultaneously with or separately from the optical image, The steps include generating orthomosaic images and a crown height model (CHM) based on the acquired data, The steps include: generating canopy polygons by separating the canopy region for each tree based on the orthomosaic image and the canopy height model; The steps include: identifying the tree species for each of the aforementioned canopy polygons and generating a tree species map that associates the tree species information with the aforementioned canopy polygons; The steps include: calculating the maximum value of the crown height model within the crown polygon as the tree height; calculating the crown area of ​​the crown polygon; selecting an estimation formula for each tree species based on the tree species associated with the crown polygon by the tree species map; and estimating the diameter at breast height by applying the tree height and the crown area to the selected estimation formula; The steps include: storing the identified or estimated forest information associated with the aforementioned tree canopy polygon as geographic information data; A method for analyzing forest information, including the analysis of forest information.

11. The step of generating the aforementioned canopy height model depends on whether or not LiDAR point cloud data is available. (a) A process to generate a numerical surface model (DSM) by performing Structure from Motion (SfM) processing on the optical image, and generate the tree canopy height model by calculating the difference between it and an externally provided numerical elevation model (DEM), or (b) A process of generating a numerical terrain model (DTM) and a DSM from the LiDAR point cloud data, and generating the canopy height model from the difference between them. A forest information analysis method according to claim 10, wherein the following is selectively performed.

12. The forest information analysis method according to claim 11, wherein the digital elevation model is obtained by making an API call to an external public DEM service, or as a user-specified DEM file.

13. The forest information analysis method according to claim 10, wherein the separation of the canopy region is performed by at least one of a method based on a canopy height model or a method based on orthophotos.

14. The step of separating the canopy region is, (a) A first canopy separation method using a local maximum filter and watershed segmentation for the canopy height model, (b) A second tree canopy separation method using object detection or region extraction with a trained model on the orthophoto, and (c) A third tree canopy separation method by object detection or region extraction without using a pre-trained model for the orthoimage or CHM, The forest information analysis method according to claim 10, comprising a process of separating the canopy region using at least one of the following.

15. The step of identifying the tree species involves extracting an image region corresponding to the tree canopy polygon, inputting the image region into a trained model for tree species identification, and identifying the tree species. The forest information analysis method according to claim 10, comprising a process of selecting a tree species identification model to be used based on location information, or limiting the candidate tree species to be identified.

16. The forest information analysis method according to claim 15, wherein the step of identifying the tree species is to perform hue correction or color distribution conversion on the image region in order to correct for color changes during the pollen attachment period or the autumn foliage period, and then identify the tree species.

17. The forest information analysis method according to claim 10, wherein the estimation formula for each tree species is DBH = a × (crown area) b × (tree height) c, and a, b, and c are set to different values ​​for each tree species or tree type.

18. The forest information analysis method according to claim 10, which outputs statistical information that aggregates the estimation results by region.

19. A program for causing a computer to function as a forest information analysis device according to any one of claims 1 to 9.

20. A program for causing a computer to perform the forest information analysis method described in any one of claims 10 to 18.

Images