A forest characterization system and a digital planning and / or management platform for forestry applications

Abstract

The proposed technology relates to digital or computer-based systems and / or platforms and associated methods and / or procedures for forestry applications. There is provided a digital planning and / or management platform (300) for forestry applications, comprising a forest characterization system (100) and a forestry management system (200) configured to obtain forest characterization data from the forest characterization system (100) to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation. The forest characterization system (100) may be configured to perform at least one of segmentation / tree-level identification, vector embedding, similarity matching and attribute inheritance using global positioning data. The forest characterization system (100) may also be configured to operate based on data integration of aerial forestry data, harvester data and forest process industry data.

Description

[0001] TITLE

[0002] A forest characterization system and a digital planning and / or management platform for forestry applications.

[0003] FIELD OF THE INVENTION

[0004] The invention generally relates to digital or computer-based systems and / or platforms and associated methods and / or procedures for forestry applications. More particularly, the invention concerns a forest characterization system and a digital planning and / or management platform for forestry applications, a computer- implemented method for collecting data and building a database for forestry applications, a computer-implemented method for predictive forest characterization, a computer-implemented method for forestry management, a computer-implemented decision-support system or computer-implemented system for automated decisionmaking for forestry applications as well as associated databases, computer programs and computer-program products.

[0005] BACKGROUND OF THE INVENTION

[0006] In general, forestry management involves strategic planning, maintenance, and development of forested areas to meet various environmental, regulatory, qualitative and economic goals. It encompasses a wide range of activities aimed at sustaining healthy forest ecosystems while optimizing the benefits derived from the forests.

[0007] Examples of forestry management tasks and operations include forest inventory and monitoring, forest planning, planting and regeneration, forest restoration and planning and / or adapting harvesting operations.

[0008] It is well-known that forest characterization, i.e. , gathering information and processing data related to forested areas to extract useful information about a forest region, is of outmost importance for efficient forestry management. For this purpose, well- accepted techniques for gathering useful information about a forest region involve collecting aerial forestry data by imaging and / or mapping the region from an aerial vehicle such as a helicopter or drone. The state-of-the-art provides methods and systems related to various aspects of forestry management such as forestry machines, communication between such machines, optimization of harvesting operations, automated forest inventory mapping and forest stand attribute prediction.

[0009] EP 3 991 547 A1 relates to a processing head for a forestry machine. The processing head is intended to process a tree having a trunk with a longitudinal direction, branches extending from the trunk transversely to the longitudinal direction and knots extending into the trunk. The processing head comprises: a frame having a seat for receiving the trunk of the tree to be processed; a motorized device for moving the trunk relative to the seat, by advancing the trunk through the seat along the longitudinal direction of the trunk; one or more blades for cutting the branches from the trunk as the trunk advances; a detection system for detecting positions of the branches and / or of the knots on the trunk as the trunk advances. Information on the positions of the branches and / or of the knots is processed to determine an identification code that is based on the positions and that refers to the trunk or to a segment obtained from the trunk. The identification code is comparable against a code determined a posteriori for a specific trunk or for a specific segment of trunk, in order to establish whether the specific trunk or the specific segment of trunk corresponds to said trunk or to said segment of the trunk. The information on the positions of branches and / or of knots can also be used to determine, during a processing of the tree, one or more positions on the trunk in which to cut the trunk perpendicularly or transversely to the longitudinal direction, which is to say to optimize the truncation of the trunk.

[0010] SE 522 055 C2 relates to a method and a system for forestry management using a harvester comprising a vehicle and a harvester unit mounted thereon. The method is based on the fact that the harvester is equipped with at least one positioning device capable of determining the harvester's position with the help of externally, wirelessly receivable signals. The harvester is equipped with a marking device arranged to apply a marking on wood pieces obtained from trees. The harvester is equipped with a control unit, preferably a computer, arranged to control the marking device to mark the wood piece with the position information obtained from the positioning device or a code corresponding to this position information.

[0011] US 2023 / 0213943 A1 relates to a sensor system for synchronizing operational data for heavy equipment during tree felling operations. The system includes a first heavy equipment comprising a first winch and a second heavy equipment comprising a second winch. The system includes a first cable attached to the first winch and a fulcrum roller and a second cable attached to the second winch and the fulcrum roller. The system is such that the first heavy equipment communicates with the second heavy equipment by way of long-range radio signals.

[0012] US 2019 / 0090437 relates to a forestry management device, which includes processing circuitry configured to: a) receive an indication of a selected tree, b) receive selected tree location data associated with the selected tree, c) receive operator location data associated with an operator, and d) cause information associated with the selected tree to be displayed on a wearable computing device.

[0013] US 2006 / 0096667 A1 relates to a system for localizing and optimizing harvesting of trees or stands of trees. The system comprises a plurality of identification and localizing remote devices associated with individual trees or individual blocks of trees, each remote device including: a) a localization receiver for receiving GPS localization signals; b) a processor for processing the localization signals so as to provide location data; and c) a transmitter for communicating the location data to a remote server computer; and d) a plurality of users in electronic communication via a plurality of user interface devices with the server computer for selectively receiving and remotely further processing the location data for select trees or blocks of trees of the individual trees or said individual blocks of trees. The system further comprises a server computer including: a) a database for maintaining the identity, and location data and for associating each user with a particular, or particular group of, remote devices and for providing selective access of the users to the database; and b) a tracking and optimizing processor for tracking the location of the trees or group of trees before and after harvesting, and for optimized processing by the monitoring of the location data for the individual trees or blocks of trees to match the trees or blocks of trees to a transportation and comparison of a production schedule for a sawmill in order to timely fill a working order for lumber to be produced by the sawmill according to the production schedule, and for coordinating harvesting to accomplish the production schedule.

[0014] US 2023 / 0102406 A1 relates to a system and method for automated forest inventory mapping. The method includes receiving an image depicting an overhead view of a wooded area, the image comprising a plurality of pixels, receiving a set of climate data for a geographic region in which the wooded area is located, and receiving a point cloud of a digital surface model of the wooded area. The method further includes concatenating data corresponding to the plurality of pixels of the image, the set of climate data, and the point cloud into a feature vector, executing a machine learning model using the feature vector to generate timber data for each of the plurality of pixels of the image, and generating an interactive overlay from the timber data, the interactive overlay comprising the generated timber data for each of the plurality of pixels of the image.

[0015] US 11 ,074,447 B1 relates to a land-analysis system using drone-captured data. The system instructs a drone to fly along a flight path, capture images of the land below, and measure altitude data. The system processes the images using, for example, artificial intelligence, to identify locations at which plant material may be present. The system then further processes the images to identify the plant health of the plant material at the identified locations. The system further uses the altitude data to determine the strata of plants at the identified locations. Optionally, the system can further process the images to identify the soil moisture levels at the identified locations.

[0016] US 2022 / 0012820 A1 relates to forest stand target attribute prediction. A computer- implemented method includes obtaining direct indicator data about forest stands, the direct indicator data comprising at least one of forest inventory estimates, airborne laser scan data, field measurement data, optical, hyperspectral or radar satellite data, and aerial image data. The method involves obtaining indirect indicator data about the forest stands, the indirect indicator data providing data that helps to explain growth of trees in the forest stands, the indirect indicator data comprising at least one of silvicultural data, geographical data, geological data, historical weather and climate data. The method further involves obtaining empirical measurement data about the forest stands, the empirical measurement data comprising at least one of harvester machine data, X-ray data from saw mills, saw mill data, pulp mill data and integrated mills data. The method also involves dividing the forest stands into a grid composed of geographically non-overlapping cells, each cell being bound by geographic coordinates, the grid comprising a plurality of grid layers. Further, the method involves determining values of a forest stand target attribute for a first set of cells of a grid layer based on the empirical measurement data, wherein the forest stand target attribute refers to any attribute that is measurable for a forest stand and characterizes the forest stand, and determining values of a plurality of input variables for a second set of cells of the remaining grid layers based on the direct indicator data and the indirect indicator data so that cells of each remaining grid layer comprise values associated with the corresponding same input variable, the second set of cells geographically corresponding to the first set of cells. The method also involves converting the grid layers to grid-specific feature vectors so that each grid-specific feature vector corresponds to a single cell of the grid, and applying a supervised machine learning algorithm for the forest stand target attribute to generate a trained model for the forest stand target attribute based on the grid-specific feature vectors. Finally, the method involves determining values of the plurality of input variables for a given cell of the remaining grid layers based on the direct indicator data and the indirect indicator data, constructing an input feature vector for the given cell based on the values of the plurality of input variables for the given cell, and predicting the value of the forest stand target attribute for the given cell based on the input feature vector and the trained model for the forest stand target attribute.

[0017] US 2021 / 0019873 A1 relates to a computer implemented method for training a software infrastructure based on machine learning techniques and intended for analysis of data obtained from a three-dimensional tomographic inspection of objects of a predetermined type, such as logs, with the aim of determining information about internal characteristics of interest of the objects. Once a training set comprising a plurality of objects of the same predetermined type has been selected, for each object, the software infrastructure is supplied with training input data and corresponding training output data, which are processed by the software infrastructure for setting internal processing parameters of the software infrastructure which correlate the training input data with the training output data. The training input data comprises data obtained from a three-dimensional tomographic inspection of the object, and the training output data comprises information about internal characteristics of interest assessed at internal points of the object. The information about the internal characteristics of interest is at least partly assessed at real internal points of the object, previously made accessible by cutting or breaking the object.

[0018] JP 2023-158970 relates to forestry management. A low-altitude observation machine comprises: a ranging device for generating a distance map, and a positioning device for measuring the position of the low-altitude observation machine. The low-altitude observation machine moves in a region of a forest near the ground without branches or leaves. The low-altitude distance map generated by the ranging device and the position of the low-altitude observation machine are acquired by a data acquisition unit from the low-altitude observation machine. The position of a tree is specified by a position-specifying unit on the basis of the position of the low-altitude observation machine and the distance from the low-altitude observation machine to the tree specified on the basis of the low-altitude distance map. Identification information for identifying an individual tree on the basis of the position of the tree is acquired by an identification unit. Low-altitude tree information, which is information on the tree at a portion lower than branches and leaves of the individual measured on the basis of the low-altitude distance map, is recorded by a recording unit in a database in association with the identification information.

[0019] Although many advances have been made in the field of forest characterization and forestry management, there is still a general demand for improved procedures and systems for enhancing the performance of current forest characterization and corresponding forestry management.

[0020] BRIEF SUMMARY OF THE INVENTION

[0021] A general object of the proposed technology is to overcome at least part of the limitations of the prior art and to provide improvements with regard to enhanced forest characterization and forestry management.

[0022] It is a particular object to provide an improved forest characterization system.

[0023] It is also an object to provide an improved digital planning and / or management platform for forestry applications.

[0024] Another object is to provide a computer-implemented method for collecting data and building a database for forestry applications.

[0025] Yet another object is to provide a computer-implemented method for predictive forest characterization.

[0026] Still another object is to provide a computer-implemented method for forestry management.

[0027] It is also an object to provide a computer-implemented decision-support system or automated decision-making system for forestry applications.

[0028] Another object is to provide a forestry database implemented as a non-transitory computer-readable storage medium.

[0029] Yet another object is to provide a computer program comprising instructions, which, when executed by at least one processor, cause the at least one processor to perform methods and / or procedures for forest characterization and forestry management. Still another object is to provide a corresponding computer-program product.

[0030] These and other objects may be achieved by one or more embodiments of the proposed technology.

[0031] According to a first aspect of the invention, there is provided a forest characterization system. The forest characterization system is configured to obtain input data including at least aerial forestry data representing a forest region. The forest characterization system comprises: a segmentation module configured to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; a neural network architecture configured to perform segment-level vector embedding for generating a unique vector-based identification and / or characterization key per segment; and an attribute assignment module configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment.

[0032] For example, the segmentation module may be configured to perform tree-level identification when a segment represents an individual tree, and the attribute assignment module may be configured to perform tree-level attribute assignment to assign, for each segment representing an individual tree, a set of attributes representing the tree.

[0033] In a particular example, the attribute assignment module may be configured to associate each identification and / or characterization key, corresponding to an individual segment, with attribute data including harvester data and / or forest process industry data related to the segment to thereby provide for data integration of aerial forestry data, and harvester data and / or forest process industry data.

[0034] By way of example, the forest characterization system may be operated in a database building mode as a system for descriptive forest characterization, e.g., to build up a database of records with descriptive forest characterization data and / or in an inference mode as a system for predictive forest characterization.

[0035] According to a second aspect of the invention, there is provided a digital planning and / or management platform for forestry applications, comprising a forest characterization system according to the first aspect.

[0036] According to a third aspect of the invention, there is provided a computer- implemented method, performed by processing circuitry, for collecting data and building a database for forestry applications. The method comprises: obtaining input data including at least aerial forestry data representing a forest region; performing segmentation to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; performing segment-level vector embedding by using a neural network structure for generating a unique vector-based identification and / or characterization key per segment; performing segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment; and collecting the identification and / or characterization keys together with corresponding sets of attributes for structured storage as forest characterization data in the database. According to a fourth aspect of the invention, there is provided a computer- implemented method, performed by processing circuitry, for predictive forest characterization. The method comprises: obtaining input data including at least aerial forestry data representing a forest region under predictive analysis; performing segmentation to divide at least part of the aerial forestry data representing a forest region under predictive analysis into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; performing segment-level vector embedding by using a neural network structure for generating a unique vector-based identification and / or characterization key per segment; performing segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment, wherein each identification and / or characterization key, corresponding to an individual segment under predictive analysis, is associated with a respective set of attributes based on similarity matching and attribute inheritance in relation to previously generated identification and / or characterization keys and corresponding sets of attributes, related to at least one harvested forest region, stored in a database; and collecting the identification and / or characterization keys together with corresponding sets of attributes to provide a set of predictive forest characterization data.

[0037] According to a fifth aspect of the invention, there is provided a computer- implemented method, performed by processing circuitry, for forestry management. The method for forestry management comprises the steps of the method for collecting data and building a database for forestry applications according to the third aspect and / or the steps of the method for predictive forest characterization according to the fourth aspect, and the method for forestry management further comprises performing forestry management based on forest characterization data. According to a sixth aspect of the invention, there is provided a computer- implemented decision-support system or automated decision-making system for forestry applications. The decision-support system or automated decision-making system comprises: a forest characterization system configured to generate forest characterization data, and a forestry management system configured to obtain forest characterization data from said the characterization system to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

[0038] The forest characterization system, in turn, is configured to obtain input data including at least aerial forestry data representing a forest region, and to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree.

[0039] The forest characterization system is further configured to perform segmentlevel attribute assignment for associating each segment with attribute data including a respective set of attributes describing different aspects of the segment, wherein the attribute data includes harvester data and / or forest process industry data related to the segment to thereby provide for data integration of aerial forestry data, harvester data and forest process industry data.

[0040] According to a seventh aspect of the invention, there is provided a forestry database implemented as a non-transitory computer-readable storage medium. The database comprises records holding forest characterization data generated by the method according to the third aspect or the method according to the fourth aspect.

[0041] According to an eighth aspect of the invention, there is provided a computer program comprising instructions, which, when executed by at least one processor, cause the at least one processor to perform the method of any of the third, fourth and fifth aspects. According to a ninth aspect of the invention, there is provided a computer-program product comprising a non-transitory computer-readable storage medium carrying the computer program according to the eighth aspect.

[0042] The proposed technology provides a highly useful and efficient system for forest characterization.

[0043] The technology further opens up for enhanced possibilities for improved forestry management, e.g., based on better forest characterization data and / or efficient integration of such data from multiple sources at different stages of the overall production chain.

[0044] BRIEF DESCRIPTION OF DRAWINGS

[0045] The embodiments, together with further objects and advantages thereof, may best be understood by making reference to the following description taken together with the accompanying drawings, in which:

[0046] FIG. 1 is a schematic diagram illustrating an example of an overall digital planning and / or management platform, also referred to as a decision-support system, for forestry applications according to an embodiment of the invention.

[0047] FIG. 2 is a schematic diagram illustrating an example of a forest characterization system according to an embodiment of the invention.

[0048] FIG. 3 is a schematic diagram illustrating an example of a forestry management system according to an embodiment of the invention.

[0049] FIG. 4 is a schematic diagram illustrating an example of a forest characterization system according to an embodiment. FIG. 5 is a schematic diagram illustrating an example of a set-up for training a neural network architecture to perform segment-level vector embedding for forestry applications.

[0050] FIG. 6 is a schematic diagram illustrating an example of a system for descriptive forest characterization according to an embodiment of the invention.

[0051] FIG. 7 is a schematic diagram illustrating an example of a system for predictive forest characterization according to an embodiment of the invention.

[0052] FIG. 8 is a schematic diagram illustrating an example of a computer-implemented method for collecting forestry-related data and building a database for forestry applications according to an embodiment.

[0053] FIG. 9 is a schematic diagram illustrating an example of a computer-implemented method for predictive forest characterization according to an embodiment.

[0054] FIG. 10 is a schematic diagram illustrating an example of a computer-implemented method for forestry management according to an embodiment.

[0055] FIG. 11 A is a schematic diagram illustrating an example of various camera views of a forest region.

[0056] FIG. 11 B is a schematic diagram illustrating an example of a laser scan side view of an individual tree.

[0057] FIG. 12 is a schematic diagram illustrating an example of a pixelated image of a forest region showing an example of a specific delineated segment.

[0058] FIG. 13 is a schematic diagram illustrating an example of the relation between image pixels and global positioning coordinates related to a forest region. FIG. 14 is a schematic diagram illustrating an example of a multimodal Al-model for generating a specific vector embedding based on both image data and laser scan data of a particular segment representing a tree, as well as the corresponding creation of a database record comprising additional harvester data and sawmill data for storage in a database.

[0059] FIG. 15 is a schematic diagram illustrating an example of a segmented view of a forest region, and an example of a possible database record for a particular segment having a corresponding vector embedding key.

[0060] FIG. 16 is a schematic diagram illustrating another example of a forest characterization system according to an embodiment of the invention.

[0061] FIG. 17 is a schematic diagram illustrating an example of establishment of the relation between pixels and global positioning coordinates for storage in a database.

[0062] FIG. 18 is a schematic diagram illustrating an example of on-demand calculation of the relation between pixels and global positioning coordinates.

[0063] FIG. 19 is a schematic diagram illustrating an example of data integration of aerial forestry data, harvester data and sawmill data using a particular non-limiting combination model according to an embodiment of the invention.

[0064] FIG. 20 is a schematic diagram illustrating an example of a computer implementation according to an embodiment.

[0065] DETAILED DESCRIPTION

[0066] In the following, the present invention will be described with reference to non-limiting examples of the proposed technology.

[0067] As mentioned, forestry management involves strategic planning, maintenance, and development of forested areas to meet various environmental and economic goals. It encompasses a wide range of activities aimed at sustaining healthy forest ecosystems while optimizing the benefits derived from forests. Examples of forestry management tasks and operations include forest inventory and monitoring, forest planning, planting and regeneration, fertilization, forest restoration and planning and / or adapting harvesting operations.

[0068] It is well-known that forest characterization, i.e. , gathering information and processing data related to forested areas to extract useful information about a forest region, is of outmost importance for efficient forestry management. For example, it may be beneficial to provide a digital map of the trees in a forested area.

[0069] Aerial or Airborne Laser Scanning (ALS) is a powerful tool for capturing high- resolution topographic data from an airborne platform such as an aerial vehicle, e.g., a helicopter, drone or other aircraft. It has widespread applications in general mapping as well as forestry applications providing detailed and accurate 3D models of the terrain and surface features. ALS, also known as LIDAR (Light Detection and Ranging), is a remote sensing method that uses laser pulses emitted from an airborne platform to measure distances to the earth's surface. These measurements may then be used to create precise 3D models of the terrain and / or vegetation.

[0070] More particularly, a laser scanner aboard the aircraft emits rapid laser pulses toward the ground. These pulses reflect off the ground and objects (e.g., trees) and return to a sensor. The time taken for the pulses to return is measured, which is then used to calculate the distance between the scanner and the target. The scanner equipment typically collects millions of points per second, creating a dense point cloud of 3D coordinates (x, y, z) in local 3D space. Each point may be associated with RGB (color) data and Global Positioning System (GPS) data. By transforming point cloud coordinates to align with GPS coordinates or similar global positioning data, users can effectively integrate local 3D spatial data with global geographic information systems. The collected data may then be processed to remove noise and classify different surface types (e.g., ground and vegetation). This data can then be used to generate digital elevation models (DEMs), digital surface models (DSMs), digital terrain models (DTMs), canopy height models (CHMs) and other geographic information. In other words, high-resolution laser data may be utilized to generate map bases by creating various types of terrain and tree height models.

[0071] By way of example, for forestry management applications, an ALS system mounted on a helicopter or other aerial vehicle may fly over a forested area, collecting data that reveals tree heights, canopy structure and / or ground elevation. This information may be used to manage forest resources, plan harvesting operations and / or monitor forest health.

[0072] Aerial forestry data may alternatively, or complementary, include camera images from one or more cameras installed on an aerial vehicle for capturing different camera views of a forest region. By way of example, photogrammetric techniques such as Structure-from-Motion (SfM) may be used to estimate three-dimensional structures from two-dimensional image sequences, which may be coupled with local motion signals. When applied to camera motion, SfM normally involves analysing the movement of the camera through space to reconstruct the 3D geometry of the environment. The proposed technology may also utilize any well-accepted technologies to match aerial images to create ortho-photographic data like orthophotos. Orthophotos may be used in a geographic information system to provide a detailed, accurate visual representation of a forest region, which can be overlaid with other data for analysis and decision-making.

[0073] Considerable amounts of data, usually from several sources, need to be processed, and it is a challenge to process and efficiently extract relevant and useful information.

[0074] FIG. 1 is a schematic diagram illustrating an example of an overall digital planning and / or management platform, also referred to as a decision-support system or an automated decision-making system, for forestry applications according to an embodiment of the invention. The digital planning and / or management platform 300 for forestry applications may include a forest characterization system 100, and a forestry management system 200 configured to obtain forest characterization data from the forest characterization system 100 to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

[0075] In general, the overall platform may use aerial forestry data as input, and optionally also harvester data and / or process industry data to allow for full data integration across the production chain.

[0076] The digital platform may be regarded as a computer-implemented decision-support system or automated decision-making system for forestry applications. With exemplary reference to FIG.1 , such a decision-support system or automated decision-making system 300 comprises: a forest characterization system 100 configured to generate forest characterization data, and a forestry management system 200 configured to obtain forest characterization data from said the characterization system to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

[0077] The forest characterization system 100, in turn, is configured to obtain input data including at least aerial forestry data representing a forest region, and to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree. For example, segmentation may be performed by generating point clouds from the aerial forestry data (such as ALS data and / or image data) and performing object segmentation based on the generated point clouds and / or original aerial forestry data. By way of example, the open-source software Treeseg may be used to provide computer-based segmentation of trees. Treeseg may utilize any generic point cloud, e.g., to identify individual trees. Top view segmentation is also possible, e.g., to identify and segment individual trees or clusters of closely positioned trees from a camera top view (nadir). The forest characterization system 100 is further configured to perform segment-level attribute assignment for associating each segment with attribute data including a respective set of attributes describing different aspects of the segment.

[0078] By way of example, the attribute data may include harvester data and / or forest process industry data related to the segment to thereby provide for data integration of aerial forestry data and, harvester data and / or forest process industry data.

[0079] Preferably, although not necessarily, the attribute data includes both harvester data and forest process industry data to provide a richer data characterization.

[0080] A preferred example of process industry data is relevant data from a sawmill, as will be described in more detail. Other examples of process industry facilities that may provide useful forest process industry data include pulp and / or paper mills, biomass and bioenergy plants, board plants and pellet plants and so forth.

[0081] For example, the segment-level attribute assignment may be performed in different operation modes of the forest characterization system such as a database building mode and an inference mode for predictive analysis. In the latter mode of operation, the forest characterization system may be used, e.g., for predictive analysis of a new (i.e. not yet characterized) forest stand before harvesting using aerial data of the new forest stand as input data for analysis (e.g., see the dashed line denoted “INPUT DATA FOR ANALYSIS” in FIG. 1 ).

[0082] As indicated in FIG. 1 , the forest characterization system 100 may have various functionalities for segmentation / tree-level identification, vector embedding, similarity matching and / or attribute inheritance, depending on the operation mode of the forest characterization system. Typically, the above functionalities are bridged and / or interrelated by appropriate use of global positioning data, as will be explained in more detail later on.

[0083] FIG. 2 is a schematic diagram illustrating an example of a forest characterization system according to an embodiment of the invention. In various embodiments, the forest characterization system 100 may use vector embedding to capture complex relationships between multi-dimensional technical data sets, providing a structured descriptive representation of data. This involves transforming unstructured data, such as aerial forestry data, into a more structured form. Vector embedding leverages neural network architectures to learn continuous representations of this data in a high-dimensional vector space.

[0084] Herein, vector embedding is typically used to convert unstructured input data such as aerial input data (e.g., ALS data and / or image data) into structured numerical representations. The process begins by taking the input data such as aerial input data and feeding it into a neural network, such as a Convolutional Neural Network (CNN). The CNN processes the aerial input data through multiple layers, each designed to detect and extract different features such as edges, textures, and patterns. The output from the final layer of the CNN is a high-dimensional vector embedding. This vector embedding encapsulates the essential features of the aerial input data, transforming it into a format that can be easily processed and analyzed.

[0085] In this context, it should be understood that while ALS is not an imaging procedure in the same category as optical photography, it does create a form of "image" in the sense that it provides a visual representation. In this sense, ALS is a specialized remote sensing technique that complements traditional imaging methods by providing detailed 3D structural information.

[0086] It should also be understood that other forms or types of data may also be included in the input data used for vector embedding.

[0087] These embeddings are particularly crucial for similarity search tasks. By representing input data such as aerial input data as vectors in a high-dimensional space, the system can efficiently compare and measure the similarity between different data points or data sets. Similar data points or data sets will have embeddings that are close to each other in this space, while dissimilar data points or data sets will be further apart.

[0088] The proposed technology enables effective and accurate retrieval of relevant vector embeddings and associated payload in the form of attribute data.

[0089] In accordance with the invention, vector embedding thus denotes the technology of converting unstructured input data such as aerial input data by means of neural networks into structured numerical representations in the form of high-dimensional descriptive vectors, preferably such that similar data points or data sets are located near each other in the high-dimensional vector space.

[0090] As a baseline, the forest characterization system 100 use aerial forestry data such as laser scanning data and / or camera data as input. The system 100 may optionally use additional sets of input data originating from harvester machines, process industry facilities and optional user input. For example, harvester data may include global positioning data as well as data related to the outer dimensions of individual trees. Process industry data may for example include data from a sawmill or other process industry facility such as outer and / or inner characterization of logs or properties of wood and / or wood-derived material that preferably can be coupled to specific harvested trees.

[0091] As previously indicated, the forest characterization system may be configured to generate forest characterization data in various modes of operation by using segmentation / tree-level identification, vector embedding, similarity matching and / or attribute inheritance, depending on the operation mode of the system.

[0092] FIG. 3 is a schematic diagram illustrating an example of a forestry management system according to an embodiment of the invention.

[0093] In this example, the forestry management system 200 may be configured to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation based on forest characterization data obtained from the forest characterization system. This may manifest as decision-support output from the system and / or feedback for adaptation of forestry operations like harvesting. The system may also provide decision output so as to provide for automated process adaptation or other types of automated decision-making.

[0094] An automated decision-making system may be implemented as computer-based system that is configured to make decisions or assist in decision-making processes with no or at least limited human intervention. Such a system may be rule-based and / or employ Al-based technology. In the former case, the automated decisionmaking system may use predefined rules to make decisions or provide relevant estimations. In the latter case, machine learning and other Al procedures may be used to handle complex decision-making tasks.

[0095] Since the proposed technology provides highly relevant and improved forest characterization data, the forestry management system 200 may in turn provide improved forestry operations, enhanced operational efficiency, increased resource utilization and / or improved planning and inventory for forestry applications.

[0096] By way of example, the operations of one or more harvesters may be controlled and / or adapted based on suitable feedback from the forestry management system 200. As another example, it is feasible to provide an estimation / prediction of a possible production outcome of a set of individual trees in a forest region with standing trees based on harvester data and / or process industry data obtained from previous harvesting and processing of timber / logs from one or more harvested forest stands.

[0097] FIG. 4 is a schematic diagram illustrating an example of a forest characterization system according to an embodiment.

[0098] The forest characterization system 100 is configured to obtain input data including at least aerial forestry data representing a forest region. The forest characterization system comprises: a segmentation module 110 configured to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; a neural network architecture 120 configured to perform segment-level vector embedding for generating a unique vector-based identification and / or characterization key per segment; and an attribute assignment module 130 configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment.

[0099] In this way, a highly useful and efficient system for forest characterization is provided. Unlike traditional machine learning methods that require separate models to estimate each variable, this system links multiple variables to a single vector by connecting additional data to the embedding, enabling the description of numerous aspects simultaneously. With similarity matching, new segments may inherit attributes from the most similar existing segments, providing accurate and contextually rich characterizations. This approach offers improved efficiency, scalability, and a more comprehensive view of each segment compared to traditional methods.

[0100] Although the use of Artificial Intelligence (Al) and neural networks in particular is known in the forestry industry, especially for machine-learning-based prediction of timber data and for image processing and pattern recognition for visual inspection and assessment of trees, the use of segment-level vector-embedding in the manner described herein is not previously known.

[0101] For example, the segmentation module 110 may be configured to perform tree-level identification when a segment represents an individual tree, and the attribute assignment module may be configured to perform tree-level attribute assignment to assign, for each segment representing an individual tree, a set of attributes representing the tree. It should be understood that any well-accepted procedure may be used by the proposed technology to identify objects and provide segmentation. By way of example, point clouds may be generated based on image data and segmentation may then be performed based on the generated point clouds and / or the image data.

[0102] By way of example, the segment-level vector embedding to generate a unique vector-based identification and / or characterization key per segment may be performed based on 2D and / or 3D representations of the segment, i.e. , 2D and / or 3D representations of the identified tree of the delineated segment. For example, this may involve 2D projections, or more simply put side views, of the segmented tree. With sufficient processor capacity, large amount of segments may be processed using 3D representations of the segments as input for the vector embedding. An advantage of the forest characterization system according to the invention is that it is well-suited for handling 3D image representations and large amounts of image data. In a particular example, the attribute assignment module 130 may be configured to associate each identification and / or characterization key, corresponding to an individual segment, with attribute data including harvester data and / or forest process industry data related to the segment to thereby provide for data integration of aerial forestry data, on one hand and harvester data and / or forest process industry data on the other hand.

[0103] By way of example, the forest characterization system 100 may be operated in a database building mode as a system for descriptive forest characterization, e.g., to build up a database of records with descriptive forest characterization data and / or in an inference mode as a system for predictive forest characterization.

[0104] For example, the neural network architecture 120 may be configured to perform segment-level vector embedding for each segment based on aerial forestry data related to the segment and also harvester data and / or forest process industry data related to the segment. In a particular example, the harvester data includes, for each segment, information concerning the global positioning and external dimensions of an individual tree associated with the segment, wherein the harvester data is obtained from a forest harvester.

[0105] Typically, production reporting from harvesters is conducted for each individually produced log, facilitating a detailed analysis of the products manufactured in the forest. This reporting includes dimensions, length, and diameter measurements along the stem, quality assessments, and unique identification for each log. This comprehensive data supports forecasting of available forest resources and / or fuel and enables the calculation of properties such as density, heartwood content, and knot structure.

[0106] The reporting typically also includes GPS positioning, timestamps, and the operational status of each log. The harvester may collect data in compliance with the StanForD 2010 standard, or another forestry industry standard, ensuring that all information is structured and reported according to this globally recognized standard in the forestry industry.

[0107] The harvester data obtained from a forest harvester could therefore, for each segment, include further information in addition to information about global positioning and external dimensions, such as information about knot structure, knot positions, density, tree species, and / or heartwood content.

[0108] Optionally, the forest characterization system may be configured to match an individual tree handled by the forest harvester with a segment based on the information concerning global positioning of the individual tree obtained from the harvester and global positioning data associated with the segment as obtained from the aerial forestry data.

[0109] For example, the forest characterization system may be configured to match the information concerning global positioning of the individual tree obtained from the forest harvester to a corresponding delineated segment based on global positioning data associated with the aerial forestry data, where each delineated segment is associated with a respective set of global positioning data.

[0110] As an example, the forest process industry data may include, for each segment, information concerning outer dimensions and / or inner characterization of an individual tree related to the segment, wherein the forest process industry data is obtained from a process industry facility for timber processing.

[0111] The forest process industry data from a process industry facility for timber processing may be obtained from a scan of the log. The log scanning equipment may for example be based on 2D X-ray imaging, 3D X-ray imaging and / or laser technology. The scan of the log may provide a representation of the outer shape of the log and preferably also of the inner structure of the log. The scan of the log preferably provides a digital 3D representation of the log.

[0112] For example, in a sawmill equipped with a Computer Tomography (CT) scanner, a very detailed characterization of a log of timber can be obtained from a CT scan of the log. This may allow determination of not only the external dimensions of the log but also the “inner” quality of the log (by means of density, knots, wood damages, and so forth) and how different parts of the log can be cut and used.

[0113] A combination of different scanning techniques may be used and the log scanning equipment may be arranged at different positions in the industrial timber processing line, for example at an inflow station and / or a sorting station.

[0114] By way of example, the forest characterization system may be configured to identify, for each of at least a subset of trees harvested by a harvester, the same tree at the process industry facility and link information about the global position of the harvester, when the tree was harvested, to processed log(s) of the tree at the process industry facility to facilitate tracking of trees from the forest to the process industry facility. As a non-limiting example, the forest characterization system may be configured to match at least one log handled by a process industry facility to an individual tree related to a segment based on measurement matching and / or by physical marking of log(s).

[0115] For example, process industry data may be associated to a particular segment by i) matching measurement data related to the outer shape of a harvested tree by a harvester, on one hand, and the outer shape of corresponding logs in a process industry facility, on the other hand, and ii) using knowledge of the positioning of the tree at harvesting and thereby connecting the logs to the corresponding segment.

[0116] By way of example, each harvesting operation normally has a geographical boundary defined by a working order, e.g., a timber order. This working order is typically registered by the harvester and later also in the sawmill. Both the harvester and the sawmill measure the external shape of each tree, including the diameter along the length of the log and the total length. This information can be used to match the log at the sawmill with the harvested tree's position. By filtering the logs registered by the sawmill with the working order, the possible combinations of length and diameter are reduced, increasing the likelihood of accurately matching the log.

[0117] According to another embodiment of the invention, a camera in the harvester can capture a photo of the tree's end surface at the point of harvest. This photo can then be used to identify the same tree in the sawmill by comparing it with a photo of the log taken at the sawmill.

[0118] It may even be possible to use measurements of the DNA of the tree or log to match the log at the sawmill with the harvested tree.

[0119] Alternatively, it is possible to more directly mark each tree or log with a unique stamp or code that can later be identified at the sawmill (or another process industry facility for timber processing). The invention may utilize a marking system, for example integrated into the harvester head, which marks the tree in a unique way, such as printing a stamp, applying a unique code, or attaching an RFID tag. This mark is preferably connected to the timestamp when the tree was harvested by the harvester and linked to the record of the harvested stem and log. When the log reaches the sawmill the unique mark is recognized, for example by photographing the log (in the case with a printed stamp or code) or by a signal detector (in the case with the RFID tag). This method ensures a reliable link between the harvested tree and the processed log, facilitating accurate tracking from the forest to the sawmill.

[0120] In other words, to facilitate tracking of a tree from the forest to the process industry facility, the proposed technology provides a procedure and corresponding system components to identify, for each of at least a subset of harvested trees, the same tree in the process industry facility and link information about the global position of the harvester, when the tree was harvested, to the processed log(s) of the tree at the process industry facility. Preferably, this is performed on segment-level, where each matching tree pertains to a corresponding segment.

[0121] FIG. 5 is a schematic diagram illustrating an example of a set-up for training a neural network architecture to perform segment-level vector embedding for forestry applications.

[0122] By way of example, the neural network architecture 120 may be trained to perform the segment-level vector embedding based on i) segments obtained from input data including at least aerial forestry data related to a forest region before harvesting and ii) corresponding sets of attributes, at least part of which are post-harvesting attributes or attributes obtained during harvesting of the forest region.

[0123] For example, it is possible to use a deep neural network structure and train the network to optimize the neural network weights for the task at hand. In a sense, the attribute sets can be regarded as labels for “training” the neural network. The weights may, e.g., be optimized in a way that segments with similar sets of attributes are embedded closer compared to segments with more disparate sets of attributes. Once the neural network embedding model has been learned, it is possible to transform segments into corresponding vector embeddings, or more specifically characterization and / or identification keys, and store them, e.g., with a K-Nearest- Neighbor (KNN) index. Now, given a new “unseen” segment to be analyzed, such a new segment can be transformed into a vector embedding, or specifically a characterization and / or identification key, by using the trained / learned embedding model and retrieve one or more of its k-most similar vector embeddings to allow access to a corresponding set(s) of attributes. For similarity matching or search, it is required to calculate the distance between vectors to determine similarity. Vector embeddings can be used to calculate these distances to find the k-most similar vector(s).

[0124] In a particular example, the neural network architecture may be trained to classify segments into at least two different tree species (such as pine and spruce), and to perform the segment-level vector embedding within each of said species.

[0125] As indicated in the examples of FIG. 6 and FIG. 7, the forest characterization system may comprise a database configured for storage of identification and / or characterization keys together with corresponding sets of attributes.

[0126] For example, the forest characterization system may be configured to be operated in a database building mode as a system for descriptive forest characterization. As such, the system may be described as a system for collecting forestry-related data and building a database for forestry applications. Exemplary reference can be made to FIG. 6.

[0127] In this context, the database 140 may, e.g., be configured to store each identification and / or characterization key, corresponding to an individual segment, linkable to the respective set of attributes describing different aspects of the segment.

[0128] In a particular example, the forest characterization system may be configured to obtain input data including at least aerial forestry data related to a forest region before harvesting. The attribute assignment module 130 may be configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment, at least part of which are postharvesting attributes or attributes obtained during harvesting to thereby provide descriptive characterization of a harvested forest region.

[0129] In a sense, the attribute assignment module 130 may include a data coupling unit 135 configured to perform the segment-level attribute assignment by coupling or associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes.

[0130] Alternatively, or complementary, the forest characterization system may be configured to be operated in an inference mode as a system for predictive forest characterization. Exemplary reference can be made to FIG. 7.

[0131] For example, the forest characterization system may be configured to obtain input data including at least aerial forestry data representing a forest region under predictive analysis. The segmentation module 110 may be configured to divide at least part of the aerial forestry data representing the forest region under predictive analysis into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree. The attribute assignment module 130 may be configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment of the forest region under predictive analysis, with a respective set of attributes based on similarity matching and attribute inheritance in relation to identification and / or characterization keys and corresponding sets of attributes for at least one harvested forest region already stored in the database 140.

[0132] By way of example, the attribute assignment module 130 may include a similarity matching module 150 configured to compare each identification and / or characterization key, corresponding to an individual segment of the forest region under predictive analysis, with identification and / or characterization keys for at least one harvested forest region already stored in the database 140 to find a similar or matching identification and / or characterization key in the database 140 and use the matching identification and / or characterization key for retrieving the corresponding set of attributes from the database 140 that is associated with the matching key.

[0133] In a sense, this corresponds to attribute assignment by inheritance, where a segment of a new uncharacterized or unharvested forest stand to be analyzed takes on or inherits the attributes of an “old” segment already stored in the database 140.

[0134] In this way, segment by segment, predictive characterization of an uncharacterized or unharvested forest region is enabled. By way of example, the characterization may be performed to enable a proper assessment whether a forest stand or selected parts thereof should be harvested or not. It could be the case, that certain trees or parts of a forest stand should not be harvested, e.g., due to nature conservation and preservation purposes.

[0135] In other words, the attribute assignment module may be regarded as part of a system designed to assign specific characteristics (attributes) to individual parts (segments) of data. Each segment is identified or characterized by keys (or codes), and these keys are associated with attributes that describe various aspects of the segment. For example, a neural network may be configured to create a vector embedding (a numerical representation) for each segment. In inference mode, the vector embedding functionality may be used to find and match similar records from a vector database, ensuring that each segment is accurately described by the most relevant attributes.

[0136] In a particular example of the proposed technology, the forest characterization system is preferably configured to establish a mapping between image pixels of the aerial forestry data and global positioning coordinates to enable identification of a global position for each image pixel in a delineated segment and to enable determination of which image pixel that corresponds to a global position. In various example embodiments, the aerial forestry data includes map and / or image data obtained from aerial mapping and / or imaging of the forest region. As previously indicated, this may involve image data obtained from one or more cameras arranged on an aerial vehicle such as a drone or helicopter and / or Aerial Laser Scanning (ALS) data from a laser scanning system, which may also be arranged on an aerial vehicle.

[0137] Not all information can be collected via harvesters and measurements in process industry facilities. For example, trees with high natural values or trees with damages such as top fractures and stem damages may also be important to quantify. In the case of aspen trees or other tree species with high natural values, these are relatively rarely harvested. For such cases, the proposed system or platform may be configured with a user-based feedback loop where the user can describe an object, e.g., an aspen tree and delineate the object in a considered image. Then the coordinates of that object may be linked to a registered class. The Al model may be configured to recognize the shape, color and / or structure of the user-annotated class and with time and training the model progressively gets better at identifying the annotated objects.

[0138] According to a second aspect of the invention, there is provided a digital planning and / or management platform for forestry applications, comprising a forest characterization system according to the first aspect.

[0139] As previously discussed, e.g., with reference to FIG. 1 , the digital planning and / or management platform 300 for forestry applications may include a forestry management system 200 configured to obtain forest characterization data from the forest characterization system 100 to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

[0140] For example, the proposed technology offers an important possibility to determine forecasted outcomes from upcoming harvests. Traditionally, during planning, a rough estimate of the distribution of tree species and assortments is made. The quality of these estimates varies greatly. Forecasts for future deliveries to forest process industry are based on the law of large numbers, where experience with seasons and geographies plays a significant role. With the proposed technology it is possible to reach an entirely new level of precision. For each forest region under consideration, a tree database with detailed information about each tree can be obtained. Through the linkage between historical trees and detailed measurement data from both harvester and process industry facility, it is feasible to predict the production outcome of each individual tree. This is not merely a forecast of the expected outcome on average, but also a prediction of the quantities of finished products that can be expected and also within which parameters it is possible to control the outcome with adjusted bucking / crosscutting, as an example of forestry process adaptation.

[0141] In the category of new opportunities by means of forestry process adaptation, the example of customized bucking can be highlighted. By customized bucking, it is not only possible to provide a tailored approach for a given region, but rather an optimized bucking for each individual tree to match a desired product outcome. Procedures and / or system components of the proposed technology may thus be configured for performing customized bucking based on the determined forest characterization data, for a given forest region or for each of a number of individual trees in the forest region, to match a predetermined target product outcome.

[0142] By way of example, planning may be divided into three horizons: strategic, tactical, and operational. The proposed technology can contribute and provide data to strategic, tactical and operational planning. By providing better data about upcoming harvests, uncertainty in tactical plans can be reduced, which ultimately can decrease the need for large timber stockpiles. This in turn reduces the risk for timber quality degradation that otherwise may naturally occur during longer periods of stockpiling. It also saves costs since it is very expensive to keep large timber stockpiles.

[0143] For long-term strategic planning, relevant information about forest inventory including data about volumes, distribution of different species and / or overall health conditions are of outmost importance as the basis for decision-making. By using the proposed technology, detailed information about the inventory related to large, forested areas may be obtained by aerial supervision and using the aerial data as input for forest characterization.

[0144] As mentioned, forestry process adaptation may involve providing suitable feedback to the harvester(s) for enabling adapted and / or improved control of the timber harvesting operations. The computer-based control system in the harvester machine(s) may thus be provided with relevant feedback and / or control data for adapting, e.g., cutting operations. This adaptation and / or control may be very precise by using global positioning data and matching the global positioning of the harvester machines with the global positioning data associated with identified segments.

[0145] By way of example, cutting lengths of specific individual trees may be adapted for optimal resource utilization. This is sometimes more generally referred to as adaptive bucking, wherein cutting patterns that maximize the value of harvested trees are selected to optimize timber production.

[0146] Other examples of process industry facilities closely related to forestry include pulp and / or paper mills, biorefineries, and biomass and bioenergy plants. These facilities typically transform raw forest materials into valuable products through various chemical, physical, mechanical, thermal, and biological processes.

[0147] • Pulp and paper mills: These facilities convert lignocellulosic material from forests, such as wood chips, recycled paper, or other raw materials into pulp, which may be processed to produce paper and paper products such as tissue or paperboard products. Data generated in these mills which may be used as input data in the forest characterization system of the invention include information on wood properties, wood chips characteristics, and pulp properties. It may also include information on the efficiency or yield of pulp production, the quality of the final paper product, and various process parameters such as the composition of feedstock materials, process temperature and the concentrations of the chemicals used in the process. • Biorefineries: These facilities process lignocellulosic biomass, such as wood chips, sawdust, or other forest biomass, into a range of products such as biofuels, power, heat, and value-added chemicals. Data in biorefineries often include information on the feedstock material properties and composition, the yield of different products, the efficiency of conversion processes, and the concentrations of key biochemical components.

[0148] • Biomass and bioenergy plants: These plants convert forest residues, such as logging debris or dedicated energy crops, into energy through processes like combustion, anaerobic digestion, or gasification. Data from these facilities typically include energy output, efficiency metrics, and the composition of feedstock materials.

[0149] In all these facilities, process industry data is crucial for monitoring and optimizing operations. This data may involve measuring the presence and / or concentration of biochemical components such as cellulose, hemicellulose and lignin.

[0150] By monitoring these and other biochemical segments, as well as other parameters in the process, including temperatures, pressures, flows, consistencies, additives and machine-parameters, forestry-related facilities can optimize their processes to improve efficiency, product quality, and sustainability. For instance, adjusting these parameters can enhance the yield of desired products, reduce waste, and minimize environmental impact.

[0151] FIG. 8 is a schematic diagram illustrating an example of a computer-implemented method for collecting forestry-related data and building a database for forestry applications according to an embodiment.

[0152] When utilizing process industry data from a facility downstream of the sawmill, such as for example a pulp mill or a plant for production of bio-pellets, it may be preferred to match segments of trees, or groups of such segments, to batches of trees or tree- derived materials in the process.

[0153] According to a third aspect of the invention, there is provided a computer- implemented method, performed by processing circuitry, for collecting data and building a database for forestry applications. The method comprises the following steps:

[0154] S1 : obtaining input data including at least aerial forestry data representing a forest region;

[0155] S2: performing segmentation to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree;

[0156] S3: performing segment-level vector embedding by using a neural network structure for generating a unique vector-based identification and / or characterization key per segment;

[0157] S4: performing segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment; and

[0158] S5: collecting the identification and / or characterization keys together with corresponding sets of attributes for structured storage as forest characterization data in the database.

[0159] FIG. 9 is a schematic diagram illustrating an example of a computer-implemented method for predictive forest characterization according to an embodiment.

[0160] According to a fourth aspect of the invention, there is provided a computer- implemented method, performed by processing circuitry, for predictive forest characterization. The method comprises the following steps:

[0161] S11 : obtaining input data including at least aerial forestry data representing a forest region under predictive analysis;

[0162] S12: performing segmentation to divide at least part of the aerial forestry data representing a forest region under predictive analysis into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree;

[0163] S13: performing segment-level vector embedding by using a neural network structure for generating a unique vector-based identification and / or characterization key per segment;

[0164] S14: performing segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment, wherein each identification and / or characterization key, corresponding to an individual segment under predictive analysis, is associated with a respective set of attributes based on similarity matching and attribute inheritance in relation to previously generated identification and / or characterization keys and corresponding sets of attributes, related to at least one harvested forest region, stored in a database; and

[0165] S15: collecting the identification and / or characterization keys together with corresponding sets of attributes to provide a set of predictive forest characterization data.

[0166] FIG. 10 is a schematic diagram illustrating an example of a computer-implemented method for forestry management according to an embodiment.

[0167] According to a fifth aspect of the invention, there is provided a computer- implemented method, performed by processing circuitry, for forestry management. The method for forestry management comprises the steps of the method for collecting data and building a database according to the third aspect and / or the steps of the method for predictive forest characterization according to the fourth aspect, and the method for forestry management further comprises the step S21 of performing forestry management based on forest characterization data.

[0168] As previously discussed, the step of performing forestry management may involve performing forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation. By way of example, for any of the methods disclosed herein, the neural network structure may be trained to perform the segment-level vector embedding based on i) segments obtained from input data including at least aerial forestry data related to a forest region before harvesting and ii) corresponding sets of attributes, at least part of which are post-harvesting attributes or attributes obtained during harvesting of the forest region.

[0169] According to a sixth aspect of the invention, there is provided a computer- implemented decision-support system or automated decision-making system for forestry applications. With exemplary reference to FIG.1 , the decision-support system or automated decision-making system 300 comprises: a forest characterization system 100 configured to generate forest characterization data, and a forestry management system 200 configured to obtain forest characterization data from said the characterization system to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

[0170] The forest characterization system 100, in turn, is configured to obtain input data including at least aerial forestry data representing a forest region, and to divide at least part of the aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree.

[0171] The forest characterization system 100 is further configured to perform segment-level attribute assignment for associating each segment with attribute data including a respective set of attributes describing different aspects of the segment. The attribute data includes harvester data and forest process industry data related to the segment to thereby provide for data integration of aerial forestry data, harvester data and forest process industry data. According to a seventh aspect of the invention, there is provided a forestry database implemented as a non-transitory computer-readable storage medium. The database comprises records holding forest characterization data generated by the method according to the third aspect or the method according to the fourth aspect.

[0172] For a better understanding of the proposed technology, non-limiting examples of various aspects and embodiments will now be described, e.g., with complementary reference to FIGs. 11 to 19.

[0173] In a particular example, the methodology is at least partly tree-level based, starting from the identification of individual trees and digitally “following” them through the forest production chain to a process industry facility such as a sawmill.

[0174] By way of example, each of a number of trees is measured or captured by airborne sensors such as a laser scanner mounted on an aerial vehicle and / or one or more digital cameras.

[0175] FIG. 11 A is a schematic diagram illustrating an example of various camera views of a forest region. In this example, three different cameras are arranged on the aerial platform to capture images of a forested area in different view angles, including a view from the front, a top view (nadir imaging) and view from behind.

[0176] FIG. 11 B is a schematic diagram illustrating an example of a laser scan side view of an individual tree. The aerial laser scanning equipment may be used to provide several different side views, also referred to as 2D projections, of the tree as well as a top view of the crown of the tree (see also FIG. 14).

[0177] The aerial data serves as initial input for a digital twin of the physical tree. The airborne platform, equipped with a laser scanner and / or one or more digital cameras, flies over a forested area to generate aerial forestry data. Initially, or continuously, this is done in a database building phase. Later on, aerial forestry data may be captured for a forest region for which there is an impending management need, usually a logging operation.

[0178] The proposed technology may use image processing techniques such as Treeseg or similar systems to perform segmentation, e.g., by generating point clouds from the aerial forestry data (such as laser scanning data and / or image data) and performing object segmentation based on the generated point clouds and / or original aerial forestry data.

[0179] FIG. 12 is a schematic diagram illustrating an example of a pixelated image of a forest region showing an example of a specific delineated segment.

[0180] For example, since the trunk forms a centerline for the tree and is distinctly shaped, it creates an increased accumulation of points from the laser scanner and / or cameras, and in statistical terms creates a local maximum. This maximum may be used, e.g., as a starting point for the delineation of the tree’s crown. This results in a polygon where typically one tree is located, but it is not strictly assumed that one segment / polygon is only one tree, it could be one or many.

[0181] The same tree may later be harvested using a harvester, e.g., with a global navigation satellite system (GNSS) with real-time correction (RTK), combined with crane angle sensors, potentially achieving a position determination of the harvested tree within less than 50 cm.

[0182] The tree may then be measured in a traditional way using delimbing knives for diameter measurements and a measuring wheel for length measurement. By way of example, every 10 cm, a diameter measurement may be registered, creating a diameter vector along the trunk. The logs are processed / cut according to a predetermined processing instruction. Based on these measurements, the volume of the processed logs can be determined. Through the geographical position (GNSS + RTK), the data from harvester and the airborne sensors can be linked. Through a spatial linkage, the segments / polygons may thus be described based on measurements related to the harvested trees.

[0183] As an example, to maintain the digital chain, the harvested tree may also be individually and physically marked by a multi-digit code or ID. This code allows identification of the cut tree when the tree later reaches the process industry facility such as a sawmill, as the log end surfaces are photographed, and the tree’s ID can be recognized and linked back to a digital twin corresponding to the harvested segment / polygon.

[0184] In a particular example, the process industry facility may be equipped with a layer CT scanner. The logs may thus be scanned, and their internal properties measured, consequently adding process industry data such as sawmill data to the digital twin of the corresponding polygon / segment. Based on the log’s external properties, such as length and a diameter vector, a relationship between the sawmill’s length and diameter measurements and those of the harvester can be established. This, combined with each logging operation being delimited by a working order, also allows a more general relationship to be formed, which enables linkage of harvester data and process industry data, e.g., sawmill data, even if not every log is individually marked with a stamp.

[0185] FIG. 13 is a schematic diagram illustrating an example of the relation between image pixels and global positioning coordinates related to a forest region.

[0186] In this particular example, the image technology encompasses two main functionalities: Pixel-to-Global-Coordinates mapping and Global-Coordinates-to-Pixel mapping. By way of example, these functionalities serve to enhance the accuracy and utility of geospatial imaging by linking high-resolution image data directly to a global coordinate system and vice versa. This capability is highly useful in applications requiring detailed visual analysis and / or precise geographical placements, especially in the context of forestry management and environmental monitoring. For example, the Pixel-to-Global-Coordinates mapping may involve translating pixel coordinates (Px, Py) from high-resolution images into global spatial coordinates (X, Y, Z). This procedure is particularly advantageous in scenarios where raw imagery provides significantly more detail than typically available in processed orthophotos. For instance, forest planners and conservationists can detect subtle details related to vegetation patterns, tree health, and terrain features that are often missed in traditional imaging. The ability to correlate these details directly to global coordinates greatly enhances the effectiveness of forest management strategies and environmental conservation efforts.

[0187] Examples of advantages of the Pixel-to-Global-Coordinates mapping include:

[0188] • Enhanced Detail Visibility: The raw image data reveals finer details not visible in standard orthophotos, providing crucial visual information for detailed geographic assessments.

[0189] • Multi-directional Imaging Capability: Unlike traditional orthophotos that capture views directly from above (nadir view), this platform supports imaging from multiple angles. This capability allows forest professionals and image classification to view the sides of trees, providing a more comprehensive understanding of tree structure and health, which is vital for accurate forest assessment and management.

[0190] • Direct Global Localization: Objects of interest identified in high-resolution images can be precisely located on global maps, facilitating easier navigation and intervention.

[0191] • Human-Aided Training: The integration of human feedback into the image classification process enhances the accuracy and relevance of automated systems, tailoring Al behavior through iterative improvement. The Global-Coordinates-to-Pixel mapping allows for the reverse operation, where global spatial coordinates are converted back into image pixel coordinates. By way of example, this functionality enables users to identify an object in one image and subsequently pinpoint its presence across multiple images. By leveraging this procedure, it is possible to apply consistent, stable, and rapid classification across diverse datasets without the need for repetitive manual identification.

[0192] Examples of advantages of the Global-Coordinates-to-Pixel mapping include:

[0193] • Consistency Across Images: Once an object is identified in any image, it can be automatically located in all other images covering the object, due to the precise mapping from global coordinates back to pixel coordinates.

[0194] • Accelerated and Stable Classification: The ability to consistently classify objects across multiple images enhances both the speed and the reliability of image classification systems, benefiting a wide range of applications from automated surveillance to environmental monitoring.

[0195] These innovative mapping technologies provide a robust framework for integrating detailed image analysis with global geographic information systems, offering significant improvements over traditional methods in terms of detail, accuracy, and operational efficiency.

[0196] By way of example, the proposed technology may also involve an image viewing tool that enables the display of high-resolution images, e.g., in a web application. Users can interact with and add information to images, and exemplifying embodiments of the proposed technology may also provide map geometries, meaning that objects such as identified trees in the images are linked to the global coordinate system.

[0197] FIG. 14 is a schematic diagram illustrating an example of a multimodal Al-model for generating a specific vector embedding based on both image data and laser scan data of a particular segment representing a tree, as well as the corresponding creation of a database record comprising additional harvester data and sawmill data for storage in a database.

[0198] In this particular example, a multimodal Al-model is employed. It should though be understood that this is merely a non-limiting example, and that it is possible to use input data from a single modality such as laser scan data.

[0199] Based on the established connection between real trees and their digital twins, a neural network, e.g., based on a multimodal Al-model, can then be trained. The purpose of the training is typically not to create a classification but rather a numerical representation of the digital twin by means of vector embedding. By using vector embedding, the digital twins become searchable in a way that an image cannot be, a transformation from unstructured data to structured data.

[0200] Information from the harvester and / or process industry facility (such as a sawmill) can be saved in a vector database, developed for the specific purpose of quickly returning data based on vector embedding. Through this approach, a large amount of descriptive data in the form of descriptive attributes can be linked to the tree segment.

[0201] In inference mode, for example, for a forest region for which there is an impending management need, new aerial forestry data of that specific region may be used as input for segmentation of the aerial data into delineated segments, at least part of which represent individual trees. For example, segmentation may be performed by delineation of tree crowns, identifying the extent of tree crowns using accumulated laser scanning data, and optionally also camera images.

[0202] For each segment, a numerical Al-based representation of a digital twin may be created based on vector embedding. By way of example, laser and image data may be combined to form a complex numerical representation, i.e. a vector embedding key, for each segment / tree. The vector embedding key is also referred to as a vector- based identification and / or characterization key. Sometimes the vector embedding key is simply referred to as a vector embedding.

[0203] For inference, the previously generated vector database may be queried by comparing, for each segment under analysis, the vector embedding key of that segment with the vector embedding keys stored in the vector database to find a “matching” vector embedding key in the database, i.e. , a vector embedding that most closely matches the vector embedding key for the segment under analysis. Once such a match is found, the set of descriptive attributes that are stored together with the matching vector embedding key in the vector database are retrieved and linked to the segment under analysis. For example, the set of descriptive attributes corresponds to previously gathered information from harvester and / or process industry facility.

[0204] The retrieved data in the form of a set of descriptive attributes may define a comprehensive database entry for each segment / tree, enhancing both accuracy and traceability in forest management and timber processing.

[0205] FIG. 15 is a schematic diagram illustrating an example of a segmented view of a forest region, and an example of a possible database record for a particular segment having a corresponding vector embedding key.

[0206] In the database, each segment-level vector embedding key may be associated with a set of features and corresponding feature values. Examples of features may include coordinates, region indicator, height, volume, species, diameter data and so forth.

[0207] By querying the database, it is possible to retrieve relevant sets of data of previously harvested and processed trees from harvesters and / or sawmills. The retrieved sets of data can be linked to corresponding segments / trees representing a forest region for which there is an impending management need, thereby generating tree-level based forest characterization data for a forest region under analysis. The generated tree-level based forest characterization data may be stored in a dedicated part of the vector database or in a separate database.

[0208] FIG. 16 is a schematic diagram illustrating another example of a forest characterization system according to an embodiment of the invention. In this particular example, aerial forestry data including laser and image data are captured by individual Al models or possibly an integrated model to generate segment-level vector embedding keys, such that each segment is associated with its own vector embedding key(s). In the particular example of using individual Al models, each Al model may generate an individual vector embedding key. The generated individual vector embedding keys such as [0.1 , 0.2], [0.3, 0.4], [0.5, 0.6] may then be merged into an aggregated or combined vector embedding key such as [0.1 , 0.2, 0.3, 0.4, 0.5, 0.6], Alternatively, an integrated Al model is utilized for generating the vector embedding key.

[0209] For example, for each segment, the generated vector embedding key may then be coupled to corresponding harvester data and / or sawmill data (or similar process industry data) and / or other attribute data. The collective data for that segment may finally be stored in the vector database. In this way, each segment is associated with a vector embedding key and a relevant set of attribute data in the vector database.

[0210] In exemplifying embodiments, the proposed technology may be configured to optimize timber resource utilization and enhance operational efficiency. For example, an overall platform may be configured for integrating data from aerial platforms, harvesters and sawmills to provide comprehensive forest characterization and decision support. As an example, key features may involve individual tree identification, attribute estimation based on neural network architectures, and optionally a feedback loop for predictive modeling and customized processing. By maximizing data value and leveraging advanced technologies, the proposed technology may improve forestry operations, reduce losses and increase resource utilization. Traditionally, the forestry machinery lacks knowledge of the value of timber resources at the time of bucking, leading to sub-optimal operations and sometimes significant losses of valuable resources.

[0211] Enhanced tree-level forest characterization would, by way of example, facilitate improved bucking operations, e.g., by aligning process industry data such as sawmill requirements with standing forests prior to harvesting.

[0212] By way of example, the proposed technology may involve data collection before actual operational tasks, as well as integration and analysis of information from the entire value chain from forest to finished product or at least relevant parts thereof. For example, optimal value may be achieved by integrating and combining different data sources. Data collection for the digital platform may occur at multiple stages and different times, e.g., before harvesting with imaging and laser scanning, during harvesting by a harvester with high-resolution positioning functionality, and optionally also at the process industry with the use of advanced measurement equipment.

[0213] FIG. 17 is a schematic diagram illustrating an example of establishment of the relation between pixels and global positioning coordinates for storage in a database.

[0214] Input data may be aerial images, e.g., with information about position and orientation. The images may be matched and combined using photogrammetry to produce an output in the form of a 3D reconstruction. Based on the 3D reconstruction, a relation between each image pixel Px, Py and global coordinates X, Y, Z is extracted and stored in a database, for each image. This relation enables connecting image data with georeferenced information, i.e. , data available in global coordinates, such as ALS data and forest harvester data and other forest related information.

[0215] FIG. 18 is a schematic diagram illustrating an example of on-demand calculation of the relation between pixels and global positioning coordinates. Input data may be aerial images, e.g., together with camera calibration data, rotation matrices, position and orientation. In this example, the relation between each image pixel Px, Py and global coordinates X, Y, Z may be calculated on-demand by using matrix calculations.

[0216] FIG. 19 is a schematic diagram illustrating an example of data integration of aerial forestry data, harvester data and sawmill data using a particular non-limiting combination model according to an embodiment of the invention.

[0217] By way of example, in order to be able to identify individual trees, high-resolution aerial forestry data may be gathered and processed into a point cloud with sufficiently high point density (e.g., > 500 points / m2). Each point is associated with X, Y, Z coordinates. The result is a 3D reconstruction of the imaged forest area, which allows for segmentation down to the tree level. For example, the segmentation may be performed with a “top-down” procedure, which means that 3D data is transformed into a 2D map representation where a relatively high concentration of points in a location will be identified as a tree. This location will then be the basis to delineate the crown of the identified tree. The result is a vectorized surface area, i.e., a delineation of the tree crown in X and Y directions. When the tree is identified, the laser scan points that lie within the identified surface area may be extracted and / or visualized in a 2D view (e.g., see FIG. 11 B). The identified tree can be regarded as a digital twin.

[0218] More generally, at least part of the aerial forestry data representing a forest region is divided into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree.

[0219] By using an Al model, segment-level vector embedding is performed for generating a unique vector-based identification and / or characterization key per segment. These keys are sometimes simply referred to as vector embedding keys or vector embeddings. Optionally, additional input data to the Al model can be integrated per segment by using an optional combination process, taking harvester data and / or industry data into account.

[0220] When the harvester cuts a tree during harvesting operations, the harvested tree is matched to its digital twin based on a spatial correspondence between the registered global position of the harvester and the global position of the digital twin. For those cases that the global position of the harvester falls within the delineated segment corresponding to a digital twin as represented in global coordinates, the measurement data (attribute data) from the harvester may be associated to the segment, or more particularly to the vector embedding of that segment.

[0221] As an example, the delimbing knives of the harvester may be used to provide a diameter measurement along the tree trunk. Preferably, the measurements result in a diameter vector defining the diameter of the trunk every 10 cm or so along the entire trunk.

[0222] Once the tree arrives at a process industry facility such as a sawmill or pulp mill, further measurements performed by measuring equipment at the process industry facility may be collected and associated to the relevant segment, or more particularly to the vector embedding of that segment.

[0223] By way of example, at the process industry facility such as a sawmill, logs from the same tree may be logically merged based on detailed information about the outer and possibly inner form of the logs in order to identify a “complete” tree also at the process industry facility. By matching the information about the outer shape of the logically merged tree at the process industry facility and data about the outer shape of the tree obtained during harvesting, the positioning of the tree at harvesting and the corresponding delineated segment can be connected to the logically merged tree in the process industry facility. In effect, this means that process industry data may be associated to a particular tree or segment. The procedure may then be performed for each tree having a match between a) segmented tree, b) harvested tree and c) logically merged tree as measured in the process industry facility.

[0224] Alternatively, physical marking of logs during harvesting and subsequent scanning / recognition at the sawmill may be used for logically merging logs into individual trees.

[0225] To make this data integration over the production chain highly efficient and / or practically feasible, the inventors have realized that segment-level vector embedding combined with segment-level attribute assignment can be practically configured to associate each vector embedding key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment.

[0226] The harvester data and / or the process industry data related to a tree may here be regarded as reference data that can be loaded for subsequent storage together with the vector embedding key of the corresponding tree segment.

[0227] The resulting data can be seen as georeferenced (X, Y, Z) attribute data coupled to corresponding segments related to a set of tree in the forested region. This system for segment-level forest characterization may be used both in a database building mode as well as in inference mode for predictive modeling for planning, prognosis and / or adapted harvesting operations.

[0228] In inference mode, the database may be employed to let trees, that have not yet been harvested and are still standing in a forest area, inherent attributes and / or properties from similar trees in the database, e.g., based on similarity matching.

[0229] By way of example, similarity matching can be accomplished by transforming segment data into a form suitable for efficient comparison. The inventors have realized that 2D and / or 3D representations of any identified tree segment may be transformed by using neural network-based vector embedding to generate a unique vector embedding (a unique identification and / or characterization key or code) for each segment. Similarity matching can then be performed by comparing vector embeddings. In inference mode, for a “new” segment representing a tree that has not yet been harvested, a new vector embedding of a 2D and / or 3D representation of the identified tree can be generated, and the new segment may then inherit the attributes of a previously harvested and measured tree that has a vector embedding with the highest degree of similarity to the new vector embedding.

[0230] The similarity matching may for example be performed based on vector embeddings of side views and / or 2D projections of available 3D data for the corresponding segments that represent the trees under consideration in the comparison. It should however be understood that 3D-based similarity matching is also feasible, depending on the processing requirements and resources.

[0231] It will be appreciated that the systems, modules and sub-systems and / or arrangements described herein can be implemented, combined and re-arranged in a variety of ways.

[0232] For example, embodiments may be implemented in hardware, or at least partly in software for execution by suitable processing circuitry, or a combination thereof.

[0233] The steps, functions, procedures, and / or blocks described herein may be implemented in hardware using any conventional technology, such as discrete circuit or integrated circuit technology, including both general-purpose electronic circuitry and application-specific circuitry.

[0234] Alternatively, or as a complement, at least some of the steps, functions, procedures, and / or blocks described herein may be implemented in software such as a computer program for execution by suitable processing circuitry such as one or more processors or processing units. For example, the forest characterization system and / or the overall digital platform may be computer-implemented systems. Likewise, the methods and / or procedures described herein may be computer-implemented methods.

[0235] FIG. 20 is a schematic diagram illustrating an example of a computer implementation according to an embodiment. In this particular example, the system 400 comprises a processor 410 and a memory 420, the memory comprising instructions executable by the processor, whereby the processor is operative to perform the steps and / or actions described herein. The instructions are typically organized as a computer program 430, which may be preconfigured in the memory 420 or downloaded from an external memory device. Optionally, the system 400 comprises an input / output interface 440 that may be interconnected to the processor(s) 410 and / or the memory 420 to enable input and / or output of relevant data such as input parameter(s) and / or resulting output parameter(s).

[0236] In a particular example, the memory 420 comprises such a set of instructions executable by the processor, whereby the processor is operative to perform any of the methods and / or procedures described herein.

[0237] The term ‘processor’ should be interpreted in a general sense as any system or device capable of executing program code or computer program instructions to perform a particular processing, determining or computing task.

[0238] The processing circuitry including one or more processors is thus configured to perform, when executing the computer program, well-defined processing tasks such as those described herein.

[0239] The processing circuitry does not have to be dedicated to only execute the abovedescribed steps, functions, procedure and / or blocks, but may also execute other tasks. The proposed technology also provides a computer-program product comprising a computer-readable medium having stored thereon such a computer program.

[0240] By way of example, the software or computer program may be realized as a computer program product, which is normally carried or stored on a computer- readable medium, in particular a non-volatile medium. The computer-readable medium may include one or more removable or non-removable memory devices including, but not limited to a Read-Only Memory (ROM), a Random Access Memory (RAM), a Compact Disc (CD), a Digital Versatile Disc (DVD), a Blu-ray disc, a Universal Serial Bus (USB) memory, a Hard Disk Drive (HDD) storage device, a flash memory, a magnetic tape, or any other conventional memory device. The computer program may thus be loaded into the operating memory of a computer or equivalent processing device for execution by the processing circuitry thereof.

[0241] Method flows may be regarded as a computer action flows, when performed by one or more processors. A corresponding device, system and / or apparatus may be defined as a group of function modules, where each step performed by the processor corresponds to a function module. In this case, the function modules are implemented as a computer program running on the processor. Hence, the system and / or platform of the proposed technology may alternatively be defined as a group of function modules, where the function modules are implemented as a computer program running on at least one processor.

[0242] The computer program residing in memory may thus be organized as appropriate function modules configured to perform, when executed by the processor, at least part of the steps and / or tasks described herein.

[0243] Alternatively, it is possible to realize the modules predominantly by hardware modules, or alternatively by hardware. The extent of software versus hardware is purely implementation selection. The embodiments described above are merely given as examples, and it should be understood that the proposed technology is not limited thereto. It will be understood by those skilled in the art that various modifications, combinations and changes may be made to the embodiments without departing from the invention. In particular, different part solutions in the different embodiments can be combined in other configurations, where technically possible.

Claims

CLAIMS1. A forest characterization system (100; 100-A; 100-B), wherein said forest characterization system is configured to obtain input data including at least aerial forestry data representing a forest region, wherein said forest characterization system comprises: a segmentation module (110) configured to divide at least part of said aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; a neural network architecture (120) configured to perform segment-level vector embedding for generating a unique vector-based identification and / or characterization key per segment; and an attribute assignment module (130) configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment.

2. The forest characterization system of claim 1 , wherein said segmentation module (110) is configured to perform tree-level identification when a segment represents an individual tree, and said attribute assignment module (130) is configured to perform tree-level attribute assignment to assign, for each segment representing an individual tree, a set of attributes representing the tree.

3. The forest characterization system of claim 1 or 2, wherein said attribute assignment module (130) is configured to associate each identification and / or characterization key, corresponding to an individual segment, with attribute data including harvester data and / or forest process industry data related to the segment to thereby provide for data integration of aerial forestry data, and harvester data and / or forest process industry data.

4. The forest characterization system of any of the claims 1 to 3, wherein said neural network architecture (120) is configured to perform segment-level vector embedding for each segment based on aerial forestry data related to the segment and also harvester data and / or forest process industry data related to the segment.

5. The forest characterization system of claim 3 or 4, wherein said harvester data includes, for each segment, information concerning global positioning and outer dimensions of an individual tree related to the segment, wherein said harvester data is obtained from a forest harvester.

6. The forest characterization system of claim 5, wherein said forest characterization system (100; 100-A; 100-B) is configured to match an individual tree handled by the forest harvester with a segment based on the information concerning global positioning of the individual tree obtained from the harvester and global positioning data associated with the segment as obtained from the aerial forestry data.

7. The forest characterization system of claim 5 or 6, wherein said forest characterization system (100; 100-A; 100-B) is configured to match the information concerning global positioning of the individual tree obtained from the forest harvester to a corresponding delineated segment based on global positioning data associated with the aerial forestry data, where each delineated segment is associated with a respective set of global positioning data.

8. The forest characterization system of any of the claims 3 to 7, wherein said forest process industry data includes, for each segment, information concerning outer dimensions and / or inner characterization of an individual tree related to the segment, wherein said forest process industry data is obtained from a process industry facility for timber processing.

9. The forest characterization system of claim 8, wherein said forest characterization system (100; 100-A; 100-B) is configured to identify, for each of atleast a subset of trees harvested by a harvester, the same tree at the process industry facility and link information about the global position of the harvester, when the tree was harvested, to processed log(s) of the tree at the process industry facility to facilitate tracking of trees from the forest to the process industry facility.

10. The forest characterization system of claim 8 or 9, wherein said forest characterization system (100; 100-A; 100-B) is configured to match at least one log handled by a process industry facility to an individual tree related to a segment based on measurement matching and / or by physical marking of log(s).11 . The forest characterization system of any of the claims 1 to 10, wherein said neural network architecture (120) is trained to perform said segment-level vector embedding based on i) segments obtained from input data including at least aerial forestry data related to a forest region before harvesting and ii) corresponding sets of attributes, at least part of which are post-harvesting attributes or attributes obtained during harvesting of the forest region.

12. The forest characterization system of claim 11 , wherein said neural network architecture (120) is trained to classify segments into at least two different tree species, and to perform said segment-level vector embedding within each of said species.

13. The forest characterization system of any of the claims 1 to 12, wherein said forest characterization system (100; 100-A; 100-B) further comprises a database (140; 450) configured for storage of identification and / or characterization keys together with corresponding sets of attributes.

14. The forest characterization system of claim 13, wherein said forest characterization system (100; 100-A) is configured to be operated in a database building mode as a system for descriptive forest characterization.

15. The forest characterization system of claim 14, wherein said database (140;450) is configured to store each identification and / or characterization key, corresponding to an individual segment, linkable to the respective set of attributes describing different aspects of the segment.

16. The forest characterization system of claim 14 or 15, wherein said forest characterization system (100; 100-A) is configured to obtain input data including at least aerial forestry data related to a forest region before harvesting, wherein said attribute assignment module (130) is configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment, at least part of which are postharvesting attributes or attributes obtained during harvesting to thereby provide descriptive characterization of a harvested forest region.

17. The forest characterization system of claim 13, wherein said forest characterization system (100; 100-B) is configured to be operated in an inference mode as a system for predictive forest characterization.

18. The forest characterization system of claim 17, wherein said forest characterization system (100; 100-B) is configured to obtain input data including at least aerial forestry data representing a forest region under predictive analysis, wherein said segmentation module (110) is configured to divide at least part of said aerial forestry data representing said forest region under predictive analysis into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; and wherein said attribute assignment module (130) is configured to perform segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment of said forest region under predictive analysis, with a respective set of attributes based on similarity matching and attribute inheritance in relation to identification and / or characterizationkeys and corresponding sets of attributes for at least one harvested forest region already stored in the database (140; 450).

19. The forest characterization system of claim 18, wherein said attribute assignment module (130) includes a similarity matching module (150) configured to compare each identification and / or characterization key, corresponding to an individual segment of said forest region under predictive analysis, with identification and / or characterization keys for at least one harvested forest region already stored in the database (140; 450) to find a similar or matching identification and / or characterization key in the database (140; 450) and use the matching identification and / or characterization key for retrieving the corresponding set of attributes from the database (140; 450) that is associated with the matching key, to thereby enable predictive characterization of an uncharacterized or unharvested forest region.

20. The forest characterization system of any of the claims 1 to 19, wherein said forest characterization system (100; 100-A; 100-B) is configured to establish a mapping between image pixels of said aerial forestry data and global positioning coordinates to enable identification of a global position for each image pixel in a delineated segment and to enable determination of which image pixel that corresponds to a global position.21 . The forest characterization system of any of the claims 1 to 20, wherein said aerial forestry data includes map and / or image data obtained from aerial mapping and / or imaging of said forest region.

22. A digital planning and / or management platform (300) for forestry applications, comprising a forest characterization system (100; 100-A; 100-B) according to any of the claims 1 to 21 .

23. The digital planning and / or management platform (300) for forestry applications of claim 22, wherein said platform (300) comprises a forestry management system (200) configured to obtain forest characterization data from said forestcharacterization system (100; 100-A; 100-B) to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

24. A computer-implemented method, performed by processing circuitry, for collecting data and building a database for forestry applications, wherein said method comprises: obtaining (S1 ) input data including at least aerial forestry data representing a forest region; performing (S2) segmentation to divide at least part of said aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; performing (S3) segment-level vector embedding by using a neural network structure for generating a unique vector-based identification and / or characterization key per segment; performing (S4) segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment; and collecting (S5) the identification and / or characterization keys together with corresponding sets of attributes for structured storage as forest characterization data in said database.

25. A computer-implemented method, performed by processing circuitry, for predictive forest characterization, wherein said method comprises: obtaining (S11 ) input data including at least aerial forestry data representing a forest region under predictive analysis; performing (S12) segmentation to divide at least part of said aerial forestry data representing a forest region under predictive analysis into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree;performing (S13) segment-level vector embedding by using a neural network structure for generating a unique vector-based identification and / or characterization key per segment; performing (S14) segment-level attribute assignment for associating each identification and / or characterization key, corresponding to an individual segment, with a respective set of attributes describing different aspects of the segment, wherein each identification and / or characterization key, corresponding to an individual segment under predictive analysis, is associated with a respective set of attributes based on similarity matching and attribute inheritance in relation to previously generated identification and / or characterization keys and corresponding sets of attributes, related to at least one harvested forest region, stored in a database; and collecting (S15) the identification and / or characterization keys together with corresponding sets of attributes to provide a set of predictive forest characterization data.

26. A computer-implemented method, performed by processing circuitry, for forestry management, wherein said method for forestry management comprises the steps of the method for collecting data and building a database according to claim 24 and / or the steps of the method for predictive forest characterization according to claim 25, wherein said method for forestry management further comprises performing forestry management based on forest characterization data.

27. The computer-implemented method for forestry management according to claim 26, wherein said step of performing forestry management comprises performing forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation.

28. The computer-implemented method of any of the claims 24 to 27, wherein said neural network structure is trained to perform said segment-level vector embedding based on i) segments obtained from input data including at least aerial forestry data related to a forest region before harvesting and ii) corresponding sets of attributes, atleast part of which are post-harvesting attributes or attributes obtained during harvesting of the forest region.

29. A computer-implemented decision-support system or automated decisionmaking system (300) for forestry applications, comprising: a forest characterization system (100; 100-A; 100-B) configured to generate forest characterization data, and a forestry management system (200) configured to obtain forest characterization data from said forest characterization system (100; 100-A; 100-B) to perform forestry inventory, forestry prognosis, forestry planning and / or forestry process adaptation, wherein said forest characterization system (100; 100-A; 100-B) is configured to obtain input data including at least aerial forestry data representing a forest region, and to divide at least part of said aerial forestry data representing a forest region into a plurality of delineated segments, wherein each segment of at least a subset of the segments represents an individual tree; wherein said forest characterization system (100; 100-A; 100-B) is further configured to perform segment-level attribute assignment for associating each segment with attribute data including a respective set of attributes describing different aspects of the segment, wherein said attribute data includes harvester data and / or forest process industry data related to the segment to thereby provide for data integration of aerial forestry data, and harvester data and / or forest process industry data.

30. A forestry database (140; 450) implemented as a non-transitory computer- readable storage medium, said forestry database (140; 450) comprising records holding forest characterization data generated by the method according to claim 24 or claim 25.31 . A computer program comprising instructions, which, when executed by at least one processor, cause said at least one processor to perform the method of any of the claims 24 to 28.

32. A computer-program product comprising a non-transitory computer-readable storage medium carrying the computer program of claim 31 .

Images