Device and method for ai adaptation of farmland reporting data and farm map integration

Abstract

The present invention relates to an AI-adapted farmland automatic reporting data and a device and method for linking with a farm map. An electronic device according to one embodiment of the present invention includes a memory and a processor connected to the memory, and the processor receives metadata including an image of farmland captured from an unmanned aerial vehicle and environmental information at the time of capture, generates a dataset by normalizing and mutually mapping the farmland image and the metadata according to a preset standard schema, derives a quality index and an uncertainty index for the dataset, determines the suitability of the dataset for artificial intelligence learning based on the quality index and the uncertainty index, and if the suitability for artificial intelligence learning is determined to be suitable, derives a preset object from the farmland image included in the dataset through an artificial intelligence module, performs labeling on the object based on a preset domain ontology, and transmits the labeled dataset to a farmland information management system.

Description

Technology Field

[0001] The present invention relates to an AI-adapted farmland automatic reporting data and a device and method for linking with a farm map. Background Technology

[0003] Unless otherwise indicated in this specification, the contents described in this section are not prior art for the claims of this application, and are not to be recognized as prior art simply because they are included in this section.

[0004] In the modern agricultural sector, drone-based surveys of farmland use are being actively adopted to prevent land speculation and ensure efficient land management. While traditional farmland surveys involved surveyors visiting sites in person and visually inspecting the land, the use of drones to photograph large areas of farmland and analyze the data is becoming a common practice.

[0005] However, conventional drone-based farmland survey methods involve the cumbersome process of requiring photographers to manually organize images and write result reports after acquiring footage. During this process, quality variations in image resolution, altitude, and coordinate precision occur depending on the photographer's skill level and the drone model used. Furthermore, the lengthy time required for humans to manually inspect and report on vast amounts of data—amounting to tens of thousands of plots—causes bottlenecks in administrative processing.

[0006] In addition, although there have been recent attempts to automatically identify crop types or cultivation conditions using artificial intelligence (AI) technology, there is a limitation in that existing accumulated drone footage is managed only for simple administrative reporting or verification purposes, making it unsuitable for AI learning. Specifically, existing data lacks environmental information (metadata) at the time of filming, such as weather conditions, drone battery status, and GPS reception sensitivity (RTK), or lacks unified data labeling standards, which causes a decrease in the reliability of AI models.

[0007] Above all, conventional data does not include quality indicators or uncertainty information regarding how reliable the AI ​​model is of the data. Consequently, problems arise where the AI ​​learns errors in the training data as is or produces incorrect inference results based on uncertain data. Furthermore, since the captured data is not linked in real-time with government-managed farmland information systems (e.g., Farm Map) and requires a separate manual input process, it is difficult to guarantee data integrity, leading to reduced work efficiency.

[0008] Therefore, technical means are required to normalize raw data collected by drones into standardized specifications so that it can be immediately utilized for AI learning and inference, to ensure reliability by quantitatively calculating data quality and uncertainty, and to automatically link this data to farm maps and the like through electronic systems. Prior art literature

[0010] Korean Registered Patent No. 10-2507230 (March 2, 2023) The problem to be solved

[0011] One embodiment of the present invention provides an AI-adapted farmland automatic reporting data and a farm map linkage device and method.

[0012] The technical problems to be solved by the present invention are not limited to those mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below. means of solving the problem

[0014] To achieve the above-mentioned purpose, an electronic device according to one embodiment of the present invention includes a memory and a processor connected to the memory, and the processor receives metadata including an image of farmland captured from an unmanned aerial vehicle and environmental information at the time of capture, generates a dataset by normalizing and mutually mapping the farmland image and the metadata according to a preset standard schema, derives a quality index and an uncertainty index for the dataset, determines the suitability of the dataset for artificial intelligence learning based on the quality index and the uncertainty index, and if the suitability for artificial intelligence learning is determined to be suitable, derives a preset object from the farmland image included in the dataset through an artificial intelligence module, performs labeling on the object based on a preset domain ontology, and transmits the labeled dataset to a farmland information management system.

[0015] At this time, the metadata may include latitude and longitude coordinate information indicating the shooting location of the farmland image, flight information including the shooting altitude of the unmanned aerial vehicle, GNSS log information including PDOP and horizontal position accuracy indicating the position positioning precision of the unmanned aerial vehicle, and optical information including the focal length of the camera and the camera sensor size for calculating the ground sample distance of the farmland image.

[0016] At this time, the processor may apply at least one indexing method among H3 or QuadKey to the dataset to assign a spatial tile index, unify it into a preset standard coordinate system and divide it into tile units to generate one or more packets, and store the generated one or more packets in the memory as input data for the artificial intelligence module.

[0017] At this time, the processor may perform the labeling by referring to a label dictionary constructed by combining a preset rule and inference based on a large language model (LLM), thereby performing the labeling for the object.

[0018] At this time, the processor may generate and grant a machine-readable Rights Token (RT) for the dataset, which includes data license information, scope of consent, retention period, and access rights information.

[0019] At this time, the processor can form an untamper-proof lineage chain by recording a hash value, a parent hash value, processor information, and an applied model version for the output of each processing step in order to track the creation and modification history of the dataset.

[0020] At this time, the processor compares the calculated quality index and uncertainty index with the preset quality threshold and uncertainty threshold, and in the first case where the quality index is greater than or equal to the quality threshold and the uncertainty index is less than or equal to the uncertainty threshold, determines the AI ​​learning suitability of the dataset as suitable; in the second case where the quality index is greater than or equal to the quality threshold and the uncertainty index exceeds the uncertainty threshold, determines the AI ​​learning suitability of the dataset as insufficient, determines the dataset as data requiring inspection, moves it to an Active Learning queue, and transmits a labeling inspection request signal to the administrator terminal; and in the third case where the quality index is less than the quality threshold, determines the AI ​​learning suitability of the dataset as substandard, discards or isolates the dataset separately, and transmits a re-photography request signal for the area corresponding to the dataset to the administrator terminal.

[0021] At this time, the processor can calculate the quality index based on a sharpness measurement value representing the clarity of the farmland image, a Real Time Kinematic (RTK) reliability value representing the precision of location information included in the metadata, an intersection rate between the object area derived by the artificial intelligence module and the stored cadastral map area, and a GSD stability index calculated based on the difference between the preset target ground sample distance (GSD) and the actual captured GSD.

[0022] At this time, the above quality index is derived by the following mathematical formula,

[0023]

[0024] Q represents the quality index for the corresponding dataset, S_meas represents the sharpness measurement value for the corresponding dataset, S_ref represents the preset maximum sharpness reference value, P_meas represents the position error value measured as the RTK reliability value for the corresponding dataset, P_ref represents the preset position error allowable threshold value, I_IoU represents the crossover rate, G_tgt represents the target ground sample distance, and G_act may represent the actual captured GSD.

[0025] At this time, the processor can extract a PDOP (Position Dilution of Precision) value or a Horizontal Accuracy value from the GNSS (Global Navigation Satellite System) log data included in the metadata and set it as the position error value (P_meas), call the legal parcel boundary (Cadastral Boundary) corresponding to the location from the cadastral map database stored based on the coordinate information of the farmland image, and overlay the object area inferred by the artificial intelligence module with the legal parcel boundary, and calculate the ratio of the intersection area to the union area of ​​the two areas (Intersection over Union) to derive the intersection rate (I_IoU).

[0026] At this time, the processor can calculate the uncertainty index based on the ensemble variance calculated through the variance between the inferred results obtained by applying a plurality of different artificial intelligence models to the dataset, the error propagation value calculated according to the error propagation law regarding the degree of influence of input noise on the result value according to the sensor specifications of the unmanned aerial vehicle, and the confidence interval width calculated through the difference between the upper limit and the lower limit of the estimate distribution derived by repeatedly performing bootstrap sampling on the dataset.

[0027] At this time, the above uncertainty index is derived by the following mathematical formula,

[0028]

[0029] U represents the uncertainty index for the dataset, V_ens represents the ensemble variance value, V_ref represents a preset maximum allowable variance threshold, E_prop represents the error propagation value, E_ref represents a preset allowable error propagation threshold, W_CI represents the width of the confidence interval, and W_ref represents a preset maximum allowable width of the confidence interval. Effects of the invention

[0031] As such, according to one embodiment of the present invention, an AI-adapted farmland automatic reporting data and a farm map linkage device and method can be provided.

[0032] The effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description below. Brief explanation of the drawing

[0034] Other aspects, features, and benefits of specific preferred embodiments of the present invention, as described above, will become more apparent from the following description in conjunction with the accompanying drawings. FIG. 1 is a conceptual diagram of an AI adaptation and farm map linkage device for farmland automatic reporting data according to one embodiment of the present invention. FIG. 2 is a block diagram of an electronic device according to one embodiment of the present invention. FIG. 3 is a diagram showing tile unit packet generation according to one embodiment of the present invention. FIG. 4 is a drawing showing labeling according to one embodiment of the present invention. FIG. 5 is a diagram showing the chaining of rights tokens and history according to one embodiment of the present invention. FIG. 6 is a diagram showing the artificial intelligence learning suitability according to one embodiment of the present invention. FIG. 7 is a flowchart of a method for AI adaptation of farmland automatic reporting data and farm map linkage according to one embodiment of the present invention. It should be noted that in the drawings above, similar reference numbers are used to illustrate identical or similar elements, features, and structures. Specific details for implementing the invention

[0035] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings.

[0036] In describing the embodiments, technical details that are well known in the art to which the present invention belongs and are not directly related to the present invention are omitted. This is intended to convey the essence of the present invention more clearly without obscuring it by omitting unnecessary explanations.

[0037] For the same reason, some components in the attached drawings have been exaggerated, omitted, or schematically depicted. Additionally, the size of each component does not entirely reflect its actual dimensions. Identical or corresponding components in each drawing have been assigned the same reference numbers.

[0038] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0039] At this point, it will be understood that each block of the process flow diagrams and combinations of the flow diagrams can be executed by computer program instructions. Since these computer program instructions can be loaded into the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing equipment, the instructions executed through the processor of the computer or other programmable data processing equipment create means to perform the functions described in the flow diagram block(s). Since these computer program instructions can also be stored in computer-available or computer-readable memory that can be directed toward the computer or other programmable data processing equipment to implement the function in a specific way, the instructions stored in computer-available or computer-readable memory can also produce a manufactured item containing instruction means to perform the function described in the flow diagram block(s). Since computer program instructions can be loaded onto a computer or other programmable data processing equipment, instructions that perform a series of operation steps on the computer or other programmable data processing equipment to create a process executed by the computer can also provide steps for executing the functions described in the flowchart block(s).

[0040] Additionally, each block may represent a module, segment, or part of code containing one or more executable instructions for executing a specified logical function(s). It should also be noted that in some alternative execution examples, the functions mentioned in the blocks may occur out of order. For instance, two blocks described in succession may actually be executed substantially simultaneously, or the blocks may be executed in reverse order according to their corresponding functions.

[0041] In this embodiment, the term "part" refers to a software or hardware component such as a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC), and the "part" performs certain roles. However, the meaning of "part" is not limited to software or hardware. The "part" may be configured to reside in an addressable storage medium or configured to run one or more processors. Accordingly, as an example, the "part" includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functions provided within the components and "parts" may be combined into a smaller number of components and "parts" or further separated into additional components and "parts." In addition, the components and '~parts' may be implemented to play one or more CPUs within the device or secure multimedia card.

[0042] In describing the embodiments of the present invention in detail, the primary focus will be on examples of specific systems, but the main point claimed in this specification is applicable to other communication systems and services having a similar technical background without significantly departing from the scope disclosed in this specification, and this will be possible at the judgment of a person with skilled technical knowledge in the relevant technical field.

[0043] In addition, the administrator terminal described below may include a communication-enabled desktop computer, laptop computer, notebook, smartphone, tablet PC, mobile phone, smart watch, smart glass, e-book reader, PMP (portable multimedia player), portable game console, navigation device, digital camera, DMB (digital multimedia broadcasting) player, digital audio recorder, digital audio player, digital video recorder, digital video player, PDA (Personal Digital Assistant), etc.

[0045] FIG. 1 is a conceptual diagram of an AI-adapted farmland automatic reporting data and farm map linkage device according to one embodiment of the present invention, and FIG. 2 is a block diagram of an electronic device (100) according to one embodiment of the present invention.

[0046] An electronic device (100) according to one embodiment includes a processor (110) and a memory (120). The processor (110) can perform at least one of the methods described above. The memory (120) can store information related to the method described above or store a program in which the method described above is implemented. The memory (120) may be volatile memory or non-volatile memory. The memory (120) may be referred to as a 'database', 'storage unit', etc.

[0047] The processor (110) can execute a program and control the electronic device (100). The code of the program executed by the processor (110) can be stored in memory (120). The device (100) can be connected to an external device (e.g., a personal computer or a network) through an input / output device (not shown) and exchange data.

[0048] At this time, the processor (110) can receive metadata including an image of farmland captured from an unmanned aerial vehicle and environmental information at the time of capture.

[0049] At this time, the metadata may include latitude and longitude coordinate information indicating the shooting location of the farmland image, flight information including the shooting altitude of the unmanned aerial vehicle, GNSS log information including PDOP and horizontal position accuracy indicating the position positioning precision of the unmanned aerial vehicle, and optical information including the focal length of the camera and the camera sensor size for calculating the ground sample distance of the farmland image.

[0050] Through this, the quality index and uncertainty index for the dataset can be derived, as described below. More details will be provided later.

[0051] In addition, the processor can generate a dataset by normalizing and mutually mapping the farmland image and the metadata according to a preset standard schema.

[0052] In this case, the aforementioned pre-configured Standard Schema refers to a pre-defined data structure designed to manage heterogeneous data collected from unmanned aerial vehicles of different manufacturers or models according to a unified standard and to optimize it for artificial intelligence processing.

[0053] According to one embodiment, the standard schema may be formed into a hierarchical structure including a spatial information group, a temporal information group, and a sensor information group.

[0054] Specifically, the spatial information group may be defined to include a coordinate system (CRS), tile identifier (Tile ID), geometry information, and ground sample distance (GSD) statistics, the time information group may include a shot time and mission identifier (Mission ID), and the sensor information group may be defined to include a camera identifier, lens information, and RTK status information.

[0055] Furthermore, the above normalization refers to the process of converting the received raw metadata into conformance to the specifications of the standard schema.

[0056] For example, the processor can perform processing to convert latitude and longitude information collected in different coordinate systems (e.g., WGS84, Bessel, etc.) into a preset standard coordinate system (e.g., EPSG:5179, etc.) or to unify time information of different formats into a standard time zone (UTC, etc.).

[0057] In addition, the above normalization process may further include a process of resizing or converting the resolution or format of the image to match the input specifications of the artificial intelligence model.

[0058] Furthermore, the aforementioned mutual mapping refers to the process of logically combining the normalized metadata and the farmland image to link them into a single entity.

[0059] At this time, the processor is configured to store environmental information, location information, and sensor information at the time the image was captured in a metadata field in a 1:1 ratio using the unique identifier (Image ID) of the farmland image as a key, so that the image and attribute information are not separated but are always processed together.

[0060] The dataset generated accordingly refers not to a simple collection of image files, but to a collection of information in which the aforementioned farmland images and corresponding normalized metadata are structurally combined.

[0061] In other words, the aforementioned dataset refers to an output processed into an AI-Ready state, allowing an artificial intelligence model to immediately load and learn or infer data without a separate preprocessing step, and serves as the basic unit for calculating the quality index and uncertainty index described later.

[0062] In addition, the processor may derive a quality index and an uncertainty index for the dataset and determine the suitability of the dataset for artificial intelligence learning based on the quality index and the uncertainty index. This will be described in more detail later.

[0063] In addition, if the judgment result indicates that the artificial intelligence learning suitability is suitable, the processor can derive a preset object from the farmland images included in the dataset through the artificial intelligence module.

[0064] At this time, the AI ​​module may be implemented by including various algorithms based on machine learning or deep learning to extract and learn visual features from the input farmland image.

[0065] According to one embodiment, the artificial intelligence module may be built based on at least one algorithm among a Convolutional Neural Network (CNN), a Region-based CNN (R-CNN) family model (Fast R-CNN, Faster R-CNN, etc.), YOLO (You Only Look Once), U-Net, Mask R-CNN, or Vision Transformer (ViT) to detect specific objects within an image or to segment regions.

[0066] These algorithms are configured to precisely determine which object a pixel belongs to when a new farmland image is input, by pre-training on crop color, texture, leaf shape, and planting pattern using large amounts of farmland data.

[0067] In addition, the aforementioned pre-configured object refers to a target that the artificial intelligence module must identify for the purpose of investigating and managing the actual use of farmland. Specifically, the object includes actual crops being cultivated on the farmland (e.g., rice, garlic, onions, soybeans, fruit trees, etc.), and may include the actual cultivation boundary distinct from the legal boundary on the cadastral map, agricultural facilities such as greenhouses or farm sheds, or weed colonies on fallow or abandoned land.

[0068] In addition, according to the embodiment, the object may further include identification mark or sign data physically installed around the crop to increase the recognition rate of the artificial intelligence or to identify specific points. Through this, the processor can calculate not only the type of crop but also the actual cultivated area, and automatically determine whether there is illegal conversion or fallow status.

[0069] At this time, the processor can perform labeling based on a pre-configured domain ontology for the object and transmit the labeled dataset to the farmland information management system.

[0070] In this context, the aforementioned Domain Ontology refers to a knowledge model that is hierarchically structured so that a computer can understand the concepts, terms, and relationships between them used in the fields of agriculture and farmland management.

[0071] For example, regarding the general term 'field,' the aforementioned domain ontology defines it by subdividing it into 'fields,' 'orchards,' and 'specialty crop cultivation areas' under the superclass 'cultivation land.' It is configured to manage attribute information regarding the growth stages of each crop (sowing, growing, harvesting) or the types of facilities (vinyl greenhouses, glass greenhouses, livestock barns) using a standardized ID system. This is intended to unify terms such as 'paddy field,' 'rice paddy,' and 'rice cultivation area,' which are used interchangeably depending on the photographer or inputter, into a single standardized classification system.

[0072] In this case, performing labeling based on the domain ontology means that the artificial intelligence module does not simply attach text tags to objects derived from images, but rather maps attribute information with unique identifiers (Ontology IDs) defined in the ontology system. According to one embodiment, the processor combines pre-configured rule-based mapping with the contextual inference capabilities of a Large Language Model (LLM) to determine which class on the ontology a detected object belongs to, and automatically enters the standard name and attribute value of the corresponding class into the label field of the metadata. Through this, data consistency and interoperability can be ensured.

[0073] In addition, the aforementioned farmland information management system refers to a server or database system operated by government agencies, local governments, or public institutions related to agriculture to manage the ownership and usage status of farmland, and examples include the management system for the national electronic map of farmland, 'Farm Map,' or the agricultural business entity registration system.

[0074] At this time, the reason the processor transmits the labeled dataset to the farmland information management system is to eliminate the inefficiency of the conventional manual input method and to implement the automation of administrative processing.

[0075] Previously, investigators had to return to the office to compare photos taken at the field one by one on a PC, write a report, and upload it to the system, but according to the present invention, a dataset with verified quality and uncertainty is automatically transmitted to the management system in real time or periodically and is immediately registered.

[0076] Through this, officials or managers in charge can immediately identify whether farmland is actually being cultivated, whether it has been illegally converted, and whether it is eligible for direct payments, and make decisions without separate data processing.

[0078] FIG. 3 is a diagram showing tile unit packet generation according to one embodiment of the present invention.

[0079] Referring to FIG. 3, the processor may apply at least one indexing method among H3 or QuadKey to the dataset to assign a spatial tile index, unify it into a preset standard coordinate system and divide it into tile units to generate one or more packets, and store the generated one or more packets in the memory as input data for the artificial intelligence module.

[0080] In this context, the aforementioned H3 indexing method refers to the hexagonal hierarchical grid system developed by Uber. This method divides the Earth's surface into hexagonal cells of uniform size and assigns unique indices to them. Because the hexagonal structure ensures equal distances between all adjacent cells, it minimizes boundary distortion when analyzing continuous terrain, such as farmland, and offers advantages for movement path optimization and radius searches.

[0081] In addition, the above QuadKey indexing method is a rectangular hierarchical grid system used in Bing Maps, etc., which recursively divides the map into four parts (Quadtree). This is similar to the pixel structure of a general raster image, making it intuitive for image processing and having the advantage of facilitating data aggregation based on the zoom level.

[0082] At this time, the spatial tile index refers to a unique identifier (ID) assigned to each grid area (Tile) divided according to the indexing method described above. Instead of managing the vast total farmland data using only latitude / longitude coordinates, the processor can manage "specific zones" by replacing them with simple strings or numbers through this spatial tile index, thereby dramatically improving database search speed and reducing spatial computation load.

[0083] Furthermore, the aforementioned Standard Coordinate System refers to a unified Coordinate Reference System (CRS) defined to guarantee the positional accuracy of collected data and maintain input consistency for artificial intelligence models. For example, while drone GPS typically uses the WGS84 (EPSG:4326) coordinate system for latitude and longitude, national farmland systems such as FarmMap may use the GRS80 (EPSG:5179, etc.), a planar rectangular coordinate system.

[0084] The present invention prevents positional misalignment during subsequent data overlay by converting and unifying these heterogeneous coordinate systems into a single standard (e.g., based on FarmMap).

[0085] Furthermore, the packets generated by dividing into tile units refer to individual computational units formed by cutting the entire dataset into small sections corresponding to the spatial tile indices. Each packet is structured to include a 'cropped farmland image fragment' corresponding to the tile area and the 'local metadata' of that area. The reason for dividing data into packets and storing them in memory in this way is for parallel processing. Inputting a massive, high-resolution entire map into an AI model at once can cause out-of-memory errors or degrade computation speed.

[0086] Accordingly, the present invention allows data to be split into independent packets so that an artificial intelligence module can process them at high speed sequentially or in parallel, and since only the packet needs to be reprocessed when an error occurs in a specific tile (packet), it has the effect of maximizing the stability and efficiency of the system.

[0088] FIG. 4 is a drawing showing labeling according to one embodiment of the present invention.

[0089] Referring to FIG. 4, the processor can perform the labeling by referring to a label dictionary constructed by combining a preset rule and inference based on a large language model (LLM), thereby labeling the object.

[0090] In this case, the above-mentioned Label Dictionary refers to a mapping database for converting the classification code (Class ID) or basic name of an object visually detected by the artificial intelligence module from the farmland image into standardized administrative terms on a pre-established domain ontology.

[0091] In this case, the aforementioned preset rule refers to a 1:1 mapping logic applied when the visual reading result of the artificial intelligence module is clear. For example, if the artificial intelligence module identifies an object in an image as a 'vinyl greenhouse' with a 99% probability, the processor immediately labels it as 'agricultural facility', which is a standard code on the ontology, without separate inference. This is intended to quickly process objects with distinct visual features.

[0092] On the other hand, the above-mentioned inference based on the Large Language Model (LLM) is utilized when it is difficult to determine the exact attributes of an object based solely on visual information (Ambiguity) or when surrounding environmental information (Context) must be comprehensively considered. For example, when an AI module derives primary visual information such as "blue crop," the Large Language Model performs inference by combining it with metadata received together (time of shooting: October, location: plain area). If it is "a blue crop found in a plain area in October," the Large Language Model contextually determines that it is highly likely to be "barley for double cropping" or "feed crop" rather than "rice before harvest," and recommends or determines an ontology label that corresponds to this.

[0093] In other words, the fact that the above processor refers to a label dictionary constructed by combining these means that it combines the 'vision of image AI' and the 'background knowledge of LLM' to perform high-level labeling that goes beyond simply stating "there is a crop," such as "this is classified as [Ontology ID: A-01-05 (Barley)] when considering the environmental context of a specific time and location."

[0095] FIG. 5 is a diagram showing the chaining of rights tokens and history according to one embodiment of the present invention.

[0096] Referring to FIG. 5, the processor can generate and grant a machine-readable Rights Token (RT) for the dataset, which includes license information, consent scope, retention period, and access rights information for the data.

[0097] In this case, the above Rights Token (RT) refers to a digital metadata container structured to enable the software system to independently interpret and enforce data ownership, the scope of permission to use, and privacy policies, going beyond simple written contracts or terms of use. According to one embodiment, the Rights Token is included in an encrypted format within the header of a dataset or combined into a separate sidecar file and propagated along the data's path.

[0098] In addition, the information included in the above rights token is linked to the system's access control policy (Policy-based Access Control). For example, if "limited to administrative reporting purposes" is specified in the 'Consent Scope' field, the processor can automatically detect and block attempts to export the dataset for 'artificial intelligence training' or share it with a 'third-party company'.

[0099] Additionally, when the expiration date set in the 'Retention' field arrives, the system can be triggered to automatically delete or de-identify the corresponding dataset.

[0100] By granting machine-readable rights tokens in this manner, the present invention can technically guarantee data sovereignty and compliance with personal information protection laws throughout the entire lifecycle, from data collection to utilization and disposal. Furthermore, even after the data is subsequently linked to external systems such as Farm Map, the rights information of the original creator (landowner or photographer) is clearly preserved without being lost, thereby effectively preventing copyright disputes or unauthorized misuse at the source.

[0102] In addition, the processor can form an untamper-proof lineage chain by recording a hash value, parent hash value, processor information, and applied model version for the output of each processing step in order to track the creation and modification history of the dataset.

[0103] At this time, the lineage chain formed above can be managed by linking with an external blockchain network to maximize data transparency.

[0104] At this time, considering data processing speed and storage efficiency, the processor may perform an anchoring process in which the original large-capacity farmland image and all metadata are stored in a separate database (off-chain), and only core integrity information including the generated hash value (Root Hash), parent hash value, and timestamp is extracted and recorded in a distributed ledger on the blockchain.

[0105] Specifically, since the aforementioned blockchain network has a structure in which multiple nodes verify and store data through a consensus algorithm, the genealogy information once recorded on the blockchain possesses immutability, meaning it cannot be arbitrarily modified or deleted by specific administrators or hackers.

[0106] If suspicions of tampering with a specific dataset are raised in the future, the processor can immediately verify whether the data has been manipulated by comparing the hash value recalculated from the current dataset with the original hash value recorded on the blockchain.

[0107] Consequently, this blockchain-based lineage management serves as a means to mathematically and cryptographically guarantee data integrity and trust in administrative processes where interests are sharp, such as applications for direct payments for farmland, calculation of disaster compensation, or crackdowns on illegal conversion of farmland.

[0109] FIG. 6 is a diagram showing the artificial intelligence learning suitability according to one embodiment of the present invention.

[0110] Referring to FIG. 6, the processor compares the calculated quality index and uncertainty index with the preset quality threshold and uncertainty threshold, and in the first case where the quality index is greater than or equal to the quality threshold and the uncertainty index is less than or equal to the uncertainty threshold, the processor can determine the artificial intelligence learning suitability of the dataset as suitable and transmit the dataset determined to be suitable as labeling and farm map.

[0111] In addition, in the second case where the quality index is greater than or equal to the quality threshold and the uncertainty index exceeds the uncertainty threshold, the processor may determine that the AI ​​learning suitability of the dataset is insufficient, determine that the dataset is data requiring inspection, move it to an Active Learning queue, and transmit a labeling inspection request signal to an administrator terminal.

[0112] In this context, the aforementioned Active Learning refers to a machine learning strategy in which, instead of the artificial intelligence model randomly learning all collected data, the model selects data that is most difficult to judge (Hard Examples) or rich in information in order to maximize learning efficiency, and transmits this data to a human (Oracle) to request a correct answer (Label). In this invention, by selecting only datasets that are difficult to machine read due to high uncertainty and loading them into the Active Learning queue, the system is configured to efficiently improve the performance of the artificial intelligence model with minimal intervention, without the need for an administrator to inspect the entire dataset.

[0113] In addition, the labeling inspection request signal is a signal that triggers the control software or web interface installed on the administrator terminal to output an inspection screen. Specifically, the signal is transmitted in the form of a packet that includes not only a simple alarm, but also the farmland image to be inspected, the predicted label primarily inferred by the artificial intelligence model, and the basis information for the reason for the uncertainty of the judgment (e.g., discrepancy between the results of Model A and Model B, variance of the probability distribution, etc.).

[0114] Upon receiving this, the administrator terminal visualizes and displays the corresponding image and inference results on the screen, and provides an interface that allows the administrator to accept or correct them. If the administrator inputs a modified label, the processor confirms the data as high-reliability training data (Ground Truth), stores it in the database, and prioritizes its reflection in the subsequent retraining or fine-tuning process of the artificial intelligence model, thereby forming a feedback loop structure that gradually reduces uncertainty regarding the same type of data.

[0115] In addition, the processor may determine that the artificial intelligence learning suitability of the dataset is substandard in the third case where the quality index is less than the quality threshold, discard the dataset or isolate it separately, and transmit a signal requesting re-shooting of the area corresponding to the dataset to the administrator terminal.

[0116] In this case, the third case above refers to a state in which correction is impossible due to severe physical defects in the data, such as when motion blur occurs due to strong winds, when the clarity is significantly low because foreign matter is stuck to the lens, or when the positional accuracy is outside the acceptable range because the GNSS signal is unstable at the time of shooting.

[0117] If such low-quality data is used as is for training an artificial intelligence model, it causes a degradation in the model's inference performance (Garbage In, Garbage Out) and poses a risk of incorrect information being listed on the farm map. Therefore, the processor fundamentally prevents the high-quality training data pool from being contaminated by discarding the data or isolating it as a separate 'Defect Dataset'.

[0118] Furthermore, the configuration for transmitting the re-shooting request signal enables the drone operator at the site to immediately recognize and address data defects before withdrawing. Conventionally, there was an inefficiency where data defects were identified only after returning to the office, requiring a revisit to the site; however, the present invention induces immediate re-shooting at the site through real-time quality assessment, thereby significantly reducing the time and cost required for re-investigation.

[0119] At this time, the quality threshold and uncertainty threshold may be arbitrarily set by the manager of the present invention for each effect, and may also be set as the average of the quality index and uncertainty index of the datasets selected by the manager as the standard dataset.

[0121] In addition, the processor can calculate the quality index based on a sharpness measurement value representing the clarity of the farmland image, a Real Time Kinematic (RTK) reliability value representing the precision of location information included in the metadata, an intersection rate between the object area derived by the artificial intelligence module and the stored cadastral map area, and a GSD stability index calculated based on the difference between the preset target ground sample distance (GSD) and the actual captured GSD.

[0122] Looking at it in more detail, the above quality index can be derived by the following mathematical formula 1.

[0123] [Mathematical Formula 1]

[0124]

[0125] In this case, Q represents the quality index for the corresponding dataset, S_meas represents the sharpness measurement value for the corresponding dataset, S_ref represents the preset maximum sharpness reference value, P_meas represents the position error value measured as the RTK reliability value for the corresponding dataset, P_ref represents the preset position error allowable threshold value, I_IoU represents the crossover rate, G_tgt represents the target ground sample distance, and G_act may represent the actual captured GSD.

[0126] At this time, the processor can extract a PDOP (Position Dilution of Precision) value or a Horizontal Accuracy value from the GNSS (Global Navigation Satellite System) log data included in the metadata and set it as the position error value (P_meas), call the legal parcel boundary (Cadastral Boundary) corresponding to the location from the cadastral map database stored based on the coordinate information of the farmland image, and overlay the object area inferred by the artificial intelligence module with the legal parcel boundary, and calculate the ratio of the intersection area to the union area of ​​the two areas (Intersection over Union) to derive the intersection rate (I_IoU).

[0127] At this time, the above mathematical formula 1 was designed using a Euclidean Norm method in which the normalized value of each element is a vector, in order to convert heterogeneous evaluation elements having different physical units and scales into the same scale to calculate overall quality.

[0128] Unlike the method of simply calculating the arithmetic mean of each item's score, this is a geometric approach that sets each evaluation element as an independent axis in a multidimensional quality space and calculates the vector magnitude from the origin. This design structure has the advantage of effectively identifying data where the overall quality balance is disrupted, by reflecting the characteristic that the overall vector magnitude decreases when the quality of other items is significantly low, even if the quality of a specific item is very high.

[0129] In addition, each term includes a normalization process that divides the measured value by a reference value to convert it into a ratio between 0 and 1, so that clarity, which has integer values ​​in the thousands, and cross-rate, which has ratio values ​​in the decimal units, can be summed fairly.

[0130] In addition, looking at the specific meaning and examples of each parameter, the term related to sharpness is designed to have a positive effect on the quality score as the image clarity increases. For example, the closer the image edge intensity derived through the Laplacian variance algorithm is to the reference value, the closer the value will be to 1.

[0131] On the other hand, for items where lower values ​​indicate superiority, such as positional error values ​​or GSD differences, an inverse correction method is applied by subtracting the ratio obtained by dividing the measured error by the allowable threshold from 1. For example, in the case of RTK reliability values, the smaller the measured PDOP value is compared to the allowable limit, the closer the error rate approaches 0; by subtracting this from 1, the system is configured so that high positional accuracy is consequently converted into a high quality score.

[0132] In addition, regarding the intersection rate, the closer the object recognized by artificial intelligence is to the actual cadastral map, the closer the value is to 1, which serves as an indicator guaranteeing that the data captured the geographically correct location.

[0133] Finally, the GSD Stability Index determines whether the drone flew in compliance with the planned altitude, and it has a structure in which points are deducted as the discrepancy between the target resolution and the actual captured resolution increases.

[0134] Consequently, the above processor derives the highest quality index when all four of these elements are close to perfect, and operates to lower the overall index and evaluate the suitability for artificial intelligence learning as low if any of the defects exceeding the acceptable range occur.

[0135] In this context, the aforementioned RTK (Real Time Kinematic) refers to a real-time mobile positioning technology that acquires ultra-precise position information in centimeters by calculating the carrier phase difference between a fixed base station and a moving drone (Rover) in real time to correct errors occurring in the Global Navigation Satellite System (GNSS).

[0136] Furthermore, PDOP (Position Dilution of Precision) is an indicator representing the rate of positional accuracy degradation, indicating how evenly the satellites used for positioning are distributed in the sky. Generally, the more widely the satellites are spread out, the lower the PDOP value, signifying higher positioning precision; conversely, the more the satellites are concentrated in a specific area, the higher the PDOP value, suggesting a higher probability of increased positional error. Therefore, in this invention, the reliability of positional information is quantitatively determined through this PDOP value.

[0137] In addition, GSD (Ground Sample Distance) refers to the ground sample distance and represents the physical horizontal and vertical distance occupied by a single pixel of a digital image captured by a drone on the actual ground surface. For example, if the GSD is 5cm, it means that one pixel represents a 5cm area of ​​the actual ground, and a lower value indicates higher resolution. In this invention, since the GSD value fluctuates when the flight altitude of the drone is not constant and can degrade AI analysis performance, the difference from the target GSD is utilized as a key indicator for quality evaluation.

[0139] In addition, the processor can calculate the uncertainty index based on the ensemble variance calculated through the variance between the inferred results obtained by applying a plurality of different artificial intelligence models to the dataset, the error propagation value calculated according to the error propagation law regarding the degree of influence of input noise on the result value according to the sensor specifications of the unmanned aerial vehicle, and the confidence interval width calculated through the difference between the upper limit and the lower limit of the estimate distribution derived by repeatedly performing bootstrap sampling on the dataset.

[0140] Looking at it in more detail, the above uncertainty index can be derived by the following mathematical formula 2.

[0141] [Mathematical Formula 2]

[0142]

[0143] At this time, U represents the uncertainty index for the dataset, V_ens represents the ensemble variance value, V_ref represents the preset maximum allowable variance threshold, E_prop represents the error propagation value, E_ref represents the preset allowable error propagation threshold, W_CI represents the width of the confidence interval, and W_ref represents the preset maximum allowable width of the confidence interval.

[0144] In this case, the above mathematical formula 2 was designed using the Euclidean Norm method to geometrically calculate the total amount of uncertainty (Magnitude) by considering the statistical, physical, and probabilistic factors constituting uncertainty as independent vector components and integrating them. Similar to the quality index mentioned earlier, this has a structure that normalizes variables with different units and ranges to the same scale and sums them.

[0145] However, unlike the quality index, which aims for a 'higher is better' value, the uncertainty index indicates that 'the closer the value approaches zero, the more stable and reliable the data is (lower is better),' and signifies that data ambiguity increases as the value grows.

[0146] In particular, each term is structured as a ratio that divides the measured value by the reference limit, so that if the uncertainty of a specific factor exceeds the allowable range, the ratio becomes greater than 1, and by squaring this, a weighted penalty is applied to the entire index to sensitively detect risk data.

[0147] Looking at the specific meaning of each parameter, first, the ensemble variance (V_ens) represents "model uncertainty" and measures how consistent the views of multiple AI models with different structures or algorithms are regarding the same input data. If the variance is large due to differing inference results between models, it suggests that the data is an ambiguous object that is difficult to interpret mechanically.

[0148] Furthermore, the error propagation value (E_prop) is associated with "data uncertainty" and indicates how much sensor noise or initial measurement errors are amplified when reaching the final result (e.g., area calculation). A high value signifies low system stability, meaning that even minute fluctuations in the input data can drastically alter the outcome.

[0149] Furthermore, the width of the confidence interval (W_CI) represents "statistical uncertainty." If the estimates derived from bootstrap sampling are clustered within a narrow range, it indicates a high level of confidence, whereas if they are widely spread, it indicates that the estimation results are difficult to trust.

[0150] Finally, the threshold values ​​(V_ref, E_ref, W_ref) located in the denominator of each term represent technical limits acceptable to the system or tolerance standards set by the administrator. For example, V_ref is a value that specifies the extent to which differences in opinion between AI models are allowed, and E_ref represents the allowable limit of the impact of sensor errors on the results. If the measured values ​​remain smaller than these threshold values, each term will have a decimal point less than 1, keeping the total uncertainty index (U) low; however, if any of the threshold values ​​are exceeded, the total index rises sharply, serving as grounds for classifying the dataset as 'inspection required' or 'non-conforming'. Specifically, these values ​​may be arbitrarily set by the administrator of the present invention.

[0152] FIG. 7 is a flowchart of a method for AI adaptation of farmland automatic reporting data and farm map linkage according to one embodiment of the present invention.

[0153] Referring to FIG. 7, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention can receive metadata including a farmland image captured by an unmanned aerial vehicle and environmental information at the time of capture (S101).

[0154] In addition, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention can generate a dataset by normalizing and mutually mapping the farmland image and the metadata according to a preset standard schema (S103).

[0155] In addition, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention can derive a quality index and an uncertainty index for the dataset (S105).

[0156] In addition, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention can determine the AI ​​learning suitability of the dataset based on the quality index and the uncertainty index (S107).

[0157] In addition, the method for AI adaptation of farmland automatic reporting data and farm map linkage according to one embodiment of the present invention can derive a preset object through an artificial intelligence module from a farmland image included in the dataset when the AI ​​learning suitability is determined to be suitable (S109).

[0158] In addition, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention can perform labeling based on a pre-set domain ontology for the object (S111).

[0159] In addition, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention can transmit a labeled dataset to a farmland information management system (S113).

[0160] In addition, the AI ​​adaptation and farm map linkage method for farmland automatic reporting data according to one embodiment of the present invention may be configured in the same way as the AI ​​adaptation and farm map linkage device for farmland automatic reporting data disclosed in FIGS. 1 to 6.

[0162] The embodiments described above may be implemented as hardware components, software components, and / or combinations of hardware and software components. For example, the devices, methods, and components described in the embodiments may be implemented using one or more general-purpose or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are also possible.

[0163] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiment, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

[0164] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or command the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

[0165] Although the embodiments have been described above with reference to the limited drawings, those skilled in the art can apply various technical modifications and variations based on the above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or if the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0166] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

Claim 1 In an electronic device, memory; and a processor connected to said memory; The processor comprises: receiving metadata including farmland images captured from an unmanned aerial vehicle and environmental information at the time of capture; generating a dataset by normalizing and mutually mapping the farmland images and the metadata according to a preset standard schema; deriving a quality index and an uncertainty index for the dataset; determining the suitability of the dataset for artificial intelligence learning based on the quality index and the uncertainty index; if the determination result indicates that the suitability for artificial intelligence learning is suitable, deriving a preset object from the farmland images included in the dataset through an artificial intelligence module; performing labeling on the object based on a preset domain ontology; transmitting the labeled dataset to a farmland information management system; and the metadata includes latitude and longitude coordinate information indicating the capture location of the farmland images; flight information including the capture altitude of the unmanned aerial vehicle; GNSS log information including PDOP and horizontal position accuracy indicating the positioning precision of the unmanned aerial vehicle; and optical information including the focal length of the camera and the camera sensor size for calculating the ground sample distance of the farmland images. Including, the processor assigns a spatial tile index to the dataset by applying at least one indexing method among H3 or QuadKey, generates one or more packets by dividing into tile units unified by a preset standard coordinate system, and stores the generated one or more packets in the memory as input data for the artificial intelligence module, and the processor performs the labeling, wherein the labeling for the object is performed by referring to a label dictionary constructed by combining preset rules and inference based on a Large Language Model (LLM), and the processor, regarding the dataset, data license information, scope of consent,A machine-readable Rights Token (RT) containing retention period and access rights information is generated and granted, and the processor forms an immutable lineage chain by recording a hash value, parent hash value, processor information, and applied model version for the output of each processing step to track the creation and modification history of the dataset, and the processor compares the calculated quality index and uncertainty index with a pre-set quality threshold and uncertainty threshold, and in the first case where the quality index is greater than or equal to the quality threshold and the uncertainty index is less than or equal to the uncertainty threshold, the AI ​​learning suitability of the dataset is determined to be suitable, and in the second case where the quality index is greater than or equal to the quality threshold and the uncertainty index exceeds the uncertainty threshold, the AI ​​learning suitability of the dataset is determined to be insufficient, the dataset is determined to be data requiring inspection and moved to an Active Learning queue, and a labeling inspection request signal is transmitted to an administrator terminal, and the quality index is the quality threshold In the third case where it is less than, the AI ​​learning suitability of the above dataset is determined to be substandard, the said dataset is discarded or isolated separately, and a signal requesting re-shooting of the area corresponding to the said dataset is transmitted to the said administrator terminal; the processor calculates the quality index based on a sharpness measurement value representing the clarity of the farmland image, a Real Time Kinematic (RTK) reliability value representing the precision of location information included in the metadata, an intersection rate between the object area derived by the AI ​​module and the stored cadastral map area, and a GSD stability index calculated based on the difference between the preset target ground sample distance (GSD) and the actual captured GSD; the quality index is derived by the following mathematical formula, Q represents the quality index for the corresponding dataset, S_meas represents the sharpness measurement value for the corresponding dataset, S_ref represents the preset maximum sharpness reference value, P_meas represents the position error value measured as the RTK reliability value for the corresponding dataset, P_ref represents the preset position error allowable threshold, I_IoU represents the intersection rate, G_tgt represents the target ground sample distance, and G_act represents the actual captured GSD; the processor extracts the PDOP (Position Dilution of Precision) value or the Horizontal Accuracy value from the GNSS (Global Navigation Satellite System) log data included in the metadata and sets it as the position error value (P_meas); based on the coordinate information of the farmland image, it retrieves the legal parcel boundary corresponding to the location from the stored cadastral map database, and mutually [interacts] the object region inferred by the artificial intelligence module with the legal parcel boundary. The above intersection rate (I_IoU) is derived by overlaying and calculating the ratio of the intersection area to the union area of ​​the two regions, and the processor calculates the above uncertainty index based on the ensemble variance calculated through the variance between the inferred results obtained by applying a plurality of different artificial intelligence models to the above dataset, the error propagation value calculated according to the error propagation law regarding the degree of influence of input noise according to the sensor specifications of the above unmanned aerial vehicle on the result value, and the confidence interval width calculated through the difference between the upper limit and the lower limit of the estimate distribution derived by repeatedly performing bootstrap sampling on the above dataset, and the above uncertainty index isDerived by the mathematical formula below, An electronic device characterized in that U represents the uncertainty index for the corresponding dataset, V_ens represents the ensemble variance value, V_ref represents a preset maximum allowable variance threshold, E_prop represents the error propagation value, E_ref represents a preset allowable error propagation threshold, W_CI represents the width of the confidence interval, and W_ref represents a preset maximum allowable width of the confidence interval. Claim 2 delete Claim 3 delete

Images