Perimeter spraying system for agricultural drone

Abstract

The present invention discloses an outer pest control system that simultaneously detects abnormal conditions on the outer edge of farmland and performs precise pest control by detecting the boundary of a paddy field embankment within a set local analysis area in a real-time image acquired by a drone's shooting means, calculating an error vector with a path based on a cadastral map, generating a corrected path by reflecting the predicted movement amount of the drone due to computational delay, and controlling the flight trajectory of the drone and the pest control timing in real-time by synchronizing them according to the corrected path.

Description

Technology Field

[0001] The present invention relates to an perimeter pest control system for agricultural drones, and more specifically, to a technology that precisely tracks and controls the outer boundaries of farmland by organically combining a flight path based on a cadastral map with real-time image information acquired from a capturing means. Background Technology

[0002] Figure 1 illustrates the problems of conventional GPS offset correction and pest control techniques after perimeter scanning. v1 (GPS offset correction method) involves the user manually calculating camera offset values ​​to overcome the physical location difference between a KML file generated based on a cadastral map (Vworld) and the actual terrain, and then re-importing the modified KML file to perform pest control flight. This initial method has the problem that not only does it take excessive time during the data conversion and re-entry process, but it also cannot flexibly respond to real-time location errors during flight, resulting in very high overall inefficiency.

[0003] To address this, the proposed v2 (pest control method after perimeter scanning) mandatorily includes a separate 'perimeter scanning flight' step to verify the actual boundaries of the farmland prior to the pest control flight. Since the main pest control flight is only possible after learning the boundaries based on terrain information acquired through the scanning flight and regenerating the KML file, it has a structural limitation requiring the drone to fly at least twice for pest control of a single plot. This wastes the drone's limited battery resources and delays the overall operation time, leading to an urgent demand for technical improvements in the field.

[0004] Patent document 1, described as prior art, discloses a technology that updates a second pest control path by detecting a vanishing point and the direction of planting of crops in an acquired image while flying along a first pest control path generated based on a cadastral map or a satellite map.

[0005] However, since this requires a process of analyzing images and regenerating paths to acquire crop planting distribution information, it does not effectively overcome the limitations of the 'scan-fly (2-pass)' method, and there is inefficiency in terms of real-time synchronization between perceived image information and control commands, as well as immediate path correction. In addition, as it focuses on recognizing crop planting rows, it is difficult to identify abnormal conditions of the outer terrain, such as the loss or damage of the paddy field embankment itself.

[0006] Patent Document 2 proposes a method of automatically collecting GPS coordinate information while flying over the perimeter of farmland and designing a flight pattern by mapping it to map information. However, this method involves a manual procedure of collecting and registering coordinates in advance before the pest control flight, and does not provide feedback control through real-time image processing during flight.

[0007] As a result, it is impossible to flexibly modify the route in response to minute discrepancies between the cadastral map and the actual terrain that may occur in the field, or sudden collapses of rice paddy dikes.

[0008] Furthermore, the aforementioned existing technologies focus solely on ensuring the execution capability of pest control operations, and there is a complete lack of a system to verify operational precision retrospectively by storing the history of deviations between the initial cadastral map-based path and the actual corrected flight path as standardized data by waypoint after the operation is completed. Prior art literature

[0009] Korean Published Patent No. 10-2025-0178817 Korean Published Patent No. 10-2019-0076181 The problem to be solved

[0010] To solve the above problems, the present invention aims to solve the following technical challenges in order to overcome physical errors occurring between theoretical routes based on cadastral maps and actual terrain, and to maximize pest control efficiency.

[0011] First, we aim to resolve the problem of position drift of 1 to 3 meters occurring in the existing manual GPS correction method and the resulting high pest control omission rate of 15 to 20 percent, thereby establishing a precise pest control environment with an omission rate of 5 percent or less.

[0012] Second, we aim to present a technology that improves the inefficiency of taking nearly twice as long to work by eliminating the pre-scan flight procedure for checking the terrain before pest control, and enables precise correction and pest control to be completed simultaneously with only a single flight process.

[0013] Third, we aim to solve the problem of discrepancies in flight trajectory and spraying timing that occur because the distance traveled by the drone cannot be accurately corrected during the computational delay time that inevitably occurs during the video data processing process.

[0014] Fourth, we aim to overcome the limitations of drones deviating from their routes or interrupting missions due to the loss of visual indicators in abnormal terrain sections where rice paddy dikes are lost or collapsed due to natural disasters, and to ensure uninterrupted pest control continuity.

[0015] Fifth, we intend to establish a standardized data logging and visualization report generation system capable of objectively verifying the accuracy of the actual flight path and the condition of terrain damage after the work is completed. means of solving the problem

[0016] An perimeter pest control system for an agricultural drone according to an embodiment of the present invention for the above-mentioned problem to be solved comprises: a shooting means (120) that acquires image information through a flight path including boundary information of agricultural land based on a cadastral map; a processing means (130) that generates paddy field dike information using the image information and generates a correction path by calculating an error between the flight path and the paddy field dike information; and a control means (190) that controls the operation of a flight means (110) and a pest control means (140) based on the correction path. The shooting means is characterized by removing noise by performing spatial filtering on pixel data of an input image through a hardware acceleration algorithm, and generating image information by reducing it to a resolution set in correspondence with the computational specifications of a main processor provided in the processing means.

[0017] The above processing means generates a correction path through a local correction algorithm, a boundary detection algorithm, and an error correction algorithm; the local correction algorithm projects coordinate information of a cadastral map-based flight path from image information of a set resolution onto pixel coordinates within an image frame, and sets a local analysis area only for a certain range of pixel sections where a paddy field embankment is predicted to exist, using the projected coordinates as a center point; the boundary detection algorithm performs threshold operations of a predefined color only on pixels within the local analysis area to extract paddy field embankment information corresponding to the actual paddy field embankment boundary line in real time; the error correction algorithm calculates an error vector between the flight path and the paddy field embankment information, calculates the delay time from the time the image information is received to the time the error vector is calculated, and generates a correction path by applying the error vector to the predicted movement amount moved by the flight means during the delay time; and the control means operates with a flight pest control synchronization algorithm that synchronizes the spraying timing by a time point corresponding to the predicted movement amount along with the modification of the flight trajectory according to the correction path, and together with position correction through the time-series organic operation of each algorithm It can be characterized by the simultaneous performance of pest control operations.

[0018] The boundary detection algorithm described above may operate by: a step of extracting an initial boundary candidate group by applying a color threshold to pixels within the local analysis area; a step of determining a final measured boundary in which shadows and exposure interference caused by external light sources are removed by applying a filtering method that assigns weights by analyzing the distribution of saturation (S) and brightness (V) components to the extracted initial boundary candidate group; and a step of determining an abnormal state, including the loss or damage of a paddy field embankment, based on whether there is a discrepancy in physical form by comparing the pixel continuity and pixel density per unit area of ​​the final measured boundary in real time with boundary information of a flight path based on a cadastral map. The flight control synchronization algorithm described above may be characterized by: a step of maintaining continuity between flight and control by referring to the flight path based on a cadastral map when an abnormal state is determined; and a step of returning to control based on a correction path at the point where a normal paddy field embankment boundary is re-detected by the boundary detection algorithm. Effects of the invention

[0019] The present invention can improve pest control quality by drastically shortening the average paddy field embankment deviation distance to 0.3m or less and lowering the pest control omission rate to less than 5% through error correction based on real-time vision recognition.

[0020] The present invention reduces the total work time by approximately 40% compared to conventional methods by performing pest control and route correction in parallel in real time without a separate pre-scan flight, and can complete the mission with only one set of batteries.

[0021] The present invention precisely matches the spraying location with the actual terrain even in a high-speed flight environment of a drone by reflecting the predicted movement amount at the time of computational delay in the control value through an error correction algorithm and a flight pest control synchronization algorithm.

[0022] The present invention enables continuous pest control without mission interruption and allows for the real-time collection of terrain damage information by immediately performing fallback control via a cadastral map-based path when abnormal conditions, such as the collapse of a rice paddy embankment, are detected.

[0023] The present invention can verify pest control precision afterward by utilizing the waypoint-specific location deviation history and terrain diagnosis report generated through the control device (200), and can make precise data for agricultural land maintenance into assets. Brief explanation of the drawing

[0024] Figure 1 illustrates the problems of conventional GPS offset correction and pest control technology after perimeter scanning. FIG. 2 is a block diagram illustrating an perimeter pest control system according to an embodiment of the present invention. Figure 3 is a block diagram illustrating the pest control device of Figure 2 in detail. FIG. 4 is a flowchart illustrating the operation method of a pest control device according to an embodiment of the present invention. Figure 5 is a flowchart illustrating in detail the operation method of the boundary detection algorithm of Figure 4. Figure 6 is an example of extracting the boundary line of a rice paddy dike within a local analysis area. Figure 7 is an example of generating a correction path based on computational delay. Figure 8 is an example of controlling the path when a rice paddy embankment collapses. Figure 9 is an example illustrating a diagnosis report on the condition of a rice paddy embankment. Figure 10 is an example of waypoint-by-waypoint data logging and post-hoc verification. FIG. 11 illustrates the hierarchical hardware and software architecture of an perimeter fire prevention system according to an embodiment of the present invention. Figure 12 compares the work efficiency and performance indicators of the perimeter pest control system according to the present invention with conventional technologies. Figure 13 quantitatively compares the performance and efficiency of conventional perimeter control methods and the real-time fusion correction method according to the present invention. FIG. 14 shows a multi-faceted comparison of the technical elements of the prior art and the present invention. Specific details for implementing the invention

[0025] Embodiments of the present invention will be described in detail below with reference to the attached drawings and the contents described therein, but the present invention is not limited or restricted by the embodiments.

[0026] FIG. 2 is a block diagram illustrating an outer perimeter pest control system according to an embodiment of the present invention, wherein the outer perimeter pest control system (10) includes a drone (50), a remote controller (80), a pest control device (100), and a control device (200).

[0027] The outer perimeter pest control system (10) refers to an entire system that fuses real-time vision information with an initial flight path set based on a cadastral map to correct path errors and diagnoses the terrain conditions outside the farmland to generate a report.

[0028] The drone (50) performs flight by means of the pest control device (100), and the controller (80) is a device that is wirelessly connected to the drone (50) to monitor the flight status and transmits control commands to the user to support switching flight modes and manual operation.

[0029] The pest control device (100) is a hardware configuration mounted on a drone (50) that sprays pesticides in synchronization with a corrected path and real-time flight speed, and performs pest control missions without interruption even in abnormal conditions such as the collapse of a rice paddy embankment. The pest control device (100) may be manufactured as an integrated or separate unit together with the drone (50), but is not limited thereto.

[0030] The control device (200) is a remote device that collects waypoint-by-waypoint location deviation history and terrain diagnosis information from the drone (50), stores this in a standardized data format, and generates post-verification information and a final diagnosis report.

[0031] FIG. 3 is a block diagram illustrating the pest control device of FIG. 2 in detail, wherein the pest control device (100) includes a flight means (110), a shooting means (120), a processing means (130), a pest control means (140), a communication means (180), and a control means (190).

[0032] The flight means (110) is a drone (50) and is controlled by the control means (190) to form a movement and flight trajectory based on a correction path. The shooting means (120) is typically a camera and generates image information.

[0033] The shooting means (120) acquires image information through a flight path that includes boundary information of farmland based on a cadastral map. A cadastral map is a thematic map created to clarify the boundaries of land registered in a land register, and is a type of cadastral record that includes land parcel numbers, land categories, and boundary information.

[0034] Boundary information is data included within the cadastral map that indicates the physical divisions and boundaries of the farmland, and serves as a criterion for identifying the area where the drone (50) must perform actual pest control work.

[0035] The flight path refers to a movement path created based on boundary information on a cadastral map, which enables the drone (50) to fly back and forth in a straight line over agricultural land. The image information is terrain image data acquired in real time through a shooting means (120) while the drone moves along the flight path, and is information transmitted to a processing means (130) after undergoing noise removal and resolution reduction processes through hardware acceleration.

[0036] The processing means (130) generates paddy field embankment information using image information and generates a correction path by calculating the error between the flight path and the paddy field embankment information. The paddy field embankment information refers to the physical boundary line data of the actual paddy field embankment extracted by analyzing real-time image information acquired by the shooting means (120). This is an actual measurement indicator for comparison with the theoretical boundary on the cadastral map, and is information that is quantified and generated in real-time through the processing means (130).

[0037] The correction path is an optimized flight path generated by calculating the positional error between the initial flight path based on cadastral maps and the actual paddy field embankment information extracted from images. It serves to precisely match the location of the actual terrain with the drone's flight trajectory.

[0038] FIG. 4 is a flowchart illustrating the operation method of a pest control device according to an embodiment of the present invention, wherein the pest control device (100) is implemented through a hardware acceleration algorithm, a local correction algorithm, a boundary detection algorithm, an error correction algorithm, and a flight pest control synchronization algorithm.

[0039] The hardware acceleration algorithm is performed in the shooting means (120). The local correction algorithm, the boundary detection algorithm, and the error correction algorithm are performed in the processing means (130). And the flight control synchronization algorithm is performed in the control means (190).

[0040] The shooting means (120) performs spatial filtering on the pixel data of the input image through a hardware acceleration algorithm to remove noise, and generates image information by reducing it to a resolution set in correspondence with the operation specifications of the main processor provided in the processing means (130).

[0041] Hardware acceleration algorithms are a technology that increases the overall response speed of a system by performing complex image processing operations, which were previously handled in software, at high speed using dedicated hardware components or parallel processing units.

[0042] Spatial filtering is an image processing technique that recalculates the value of each pixel in an image based on its relationship with adjacent pixels to suppress noise that impedes terrain recognition accuracy or to clearly correct the boundaries of rice paddy embankments.

[0043] The main processor is a processing unit equipped in the processing means (130) that oversees high-dimensional operations such as local correction, boundary detection, and error correction algorithms, and is responsible for real-time judgment and control command generation of the system.

[0044] The original image acquired by the shooting means (120) over the farmland contains a large amount of unnecessary pixel noise due to vibrations of the drone or external light source conditions, and if transmitted as is, it may place an excessive load on the subsequent computing device. Therefore, the present invention can secure an image of clear quality and prevent the possibility of recognition errors in advance by immediately performing spatial filtering on pixel data by operating a hardware acceleration algorithm at the shooting means (120) stage.

[0045] The shooting means (120) can generate image information by reducing the image to a resolution optimized for the computational capacity and specifications of the processor so that the main processor equipped in the processing means (130) can execute the algorithm without delay. This structure significantly reduces the computational load of the main processor and allows precise position correction and pest control operations to be performed simultaneously in real time with only a single flight process (1-pass) without a separate pre-scan flight.

[0046] The local correction algorithm projects the coordinate information of a cadastral map-based flight path from image information of a set resolution onto pixel coordinates within the image frame, and uses the projected coordinates as a center point to set a local analysis area that includes only a certain range of pixel sections where a rice paddy embankment is predicted to exist.

[0047] The local correction algorithm can perform a key alignment process that converts cadastral map information, which is a global coordinate system, into a local coordinate system, which is the drone's camera viewpoint. Instead of analyzing the entire video frame, the local analysis area can be limited to only the narrow area where the rice paddy embankment is most likely to appear based on the projected coordinates.

[0048] This method prevents data processing waste that occurs when processing the entire high-resolution image and enables efficient management of the main processor's computational load. As a result, it provides a technical foundation that can dramatically increase image processing speed even in environments where drones fly at high speeds, thereby reducing recognition errors and ensuring the real-time capability of the system. In addition, by setting a local analysis area, interference from other terrain features or noise that may be included in the background image can be minimized, thereby improving the precision of the subsequent boundary detection step.

[0049] FIG. 5 is a flowchart illustrating in detail the operation method of the boundary detection algorithm of FIG. 4, wherein the boundary detection algorithm can extract paddy field information corresponding to the actual paddy field boundary line in real time by performing a threshold operation of a predefined color only on pixels within a local analysis area.

[0050] More specifically, the boundary detection algorithm extracts an initial set of boundary candidates by applying color thresholds to pixels within a local analysis area, and determines the final measured boundary from which shadows and exposure interference caused by external light sources are removed by applying a filtering process that assigns weights by analyzing the distribution of saturation (S) and brightness (V) components to the extracted initial boundary candidates. Furthermore, it can identify abnormal conditions, including the loss or damage of paddy field embankments, based on whether there is a discrepancy in physical form by comparing the pixel continuity and pixel density per unit area of ​​the final measured boundary with boundary information of a flight path based on a cadastral map in real time.

[0051] The primary reason for the loss or damage of paddy field embankments in agricultural land is the collapse of the terrain due to sudden natural disasters such as the rainy season. In actual agricultural fields, not only are there minute discrepancies between data on cadastral maps and the actual terrain, but unexpected and sudden embankment collapses also occur frequently. Such loss or damage to embankments can result in the disappearance of visual indicators that drones need to refer to in order to precisely track their flight paths.

[0052] Ultimately, if the indicator of the paddy field embankment is lost, structural limitations may occur, such as the drone deviating from its existing path or the pest control mission in progress being interrupted. In this invention, damage to the paddy field embankment can be determined by analyzing phenomena in which the continuity of boundary lines is broken in the captured image or the pixel density per unit area is measured to be lower than a set threshold.

[0053] This technology enables drones to maintain the continuity of pest control operations by referencing cadastral map-based routes, even in abnormal terrain sections where paddy field embankments have been damaged by natural disasters. Therefore, by automatically detecting and recording the state of embankment erosion, it is possible to provide farmer users with foundational data for a precise diagnostic report that objectively verifies the integrity of the terrain after the work is completed.

[0054] Figure 6 is an example of extracting a rice paddy boundary line within a local analysis area. The boundary detection algorithm performs operations on pixels within the local analysis area rather than the entire image to extract information corresponding to the actual rice paddy boundary line in real time. This local processing method blocks interference from unnecessary background information and reduces computational load, thereby enabling rapid data extraction even in high-speed flight environments of drones.

[0055] The boundary detection algorithm can first extract an initial set of boundary candidates containing the basic shape of the paddy field embankment by applying a predefined color threshold to pixels within a local analysis area. For the extracted candidates, a filtering process can be applied in conjunction with a precise analysis of the distribution of saturation (S) and brightness (V) components—rather than relying solely on color information—to assign weights to each pixel. This multi-stage filtering effectively eliminates shadow areas caused by sunlight or interference phenomena due to overexposure, thereby enabling the determination of a reliable final measured boundary regardless of external light source conditions.

[0056] The final confirmed measured boundary undergoes an analysis of continuity, which indicates the connectivity between pixels, and pixel density per unit area. By comparing the analyzed physical values ​​in real-time with boundary information of the flight path based on the cadastral map, it is possible to determine whether there is a morphological discrepancy between the two pieces of information. If the continuity of the boundary line recognized in real-time is broken or the pixel density is measured to be lower than a threshold, this can be identified as an abnormal state defined as the loss or damage of the paddy field embankment. Consequently, the present invention can automatically detect the condition of agricultural land damage in addition to the function of correcting the path through vision recognition.

[0057] FIG. 7 is an example of generating a correction path based on computational delay. The error correction algorithm calculates an error vector between the flight path and the paddy field information, calculates the delay time from the time when image information is received until the time when the error vector is calculated, and can generate a correction path by applying the error vector to the predicted amount of movement moved by the flight means (110) during the delay time.

[0058] The error vector is a numerical value representing the difference in physical distance and direction between the initial flight path set based on the cadastral map and the paddy field information extracted in real time from the captured video. This serves as an indicator that quantitatively shows how far the drone (50) deviates from the planned path.

[0059] The delay time refers to the physical time elapsed from the point in time when the drone’s shooting means (120) acquires image information and receives it in the system until the point in time when the final error vector is calculated through algorithmic calculations within the processing means (130). In a high-speed flight environment, the position of the drone changes even during this short calculation time, so it is important to measure it accurately.

[0060] The predicted movement amount refers to distance information estimated to have actually moved by the drone at its current flight speed by the flight means (110) during the delay time when the calculation is being performed. This serves to bridge the gap between the location at a past point in time when the video was captured and the actual location of the drone at the present time.

[0061] The correction path is the final flight path generated by applying the calculated error vector to the predicted displacement during the delay time. Consequently, by offsetting position errors caused by image processing parallax, it serves to match the actual terrain boundaries with the drone's flight and spraying trajectory in real time.

[0062] The control means (190) can be driven by a flight control synchronization algorithm that synchronizes the spraying timing by a time point corresponding to the predicted amount of movement along with the correction of the flight trajectory according to the correction path.

[0063] The flight control synchronization algorithm mounted on the control means (190) performs the role of precisely matching spatial flight position correction and temporal spray control within a single integrated loop.

[0064] First, in the flight path correction stage, the flight means (110) of the drone is controlled based on the correction path information transmitted from the processing means (130) to correct the distance from the actual rice paddy boundary in real time. At this time, not only is the path simply corrected, but the predicted amount of movement, which is the distance the drone has already flown forward during the delay time in the image processing process, is immediately reflected in the flight command, thereby preventing physical path deviation caused by control time lag in advance.

[0065] At the same time, during the synchronization phase of the spraying timing, the operating time of the pest control means (140) can be adjusted by a time difference corresponding to the predicted amount of movement. By correcting the physical interval between the past time point when image information was captured by the shooting means (120) and the current time point when actual spraying occurs by the predicted amount of movement, the timing of the generation of the spraying signal can be synchronized with the flight speed.

[0066] As a result, through the operation of this organic algorithm, position correction and pest control operations can be integrated into a single process without being out of sync in time series. This enables precise perimeter pest control by perfectly offsetting spraying errors caused by image processing delays, even in environments where the drone (50) flies at high speed.

[0067] Figure 8 is an example of controlling the path when a rice paddy dike collapses. When the flight control synchronization algorithm is determined to be in an abnormal state, it refers to a flight path based on a cadastral map to maintain the continuity of flight and control, and can return to control based on a corrected path at the point where a normal rice paddy dike boundary is re-detected by the boundary detection algorithm.

[0068] The control means (190) can immediately switch the flight standard from a real-time correction path to a pre-set cadastral map-based flight path when the boundary detection algorithm analyzes the pixel continuity and pixel density per unit area of ​​the actual boundary and confirms an abnormal condition, such as loss or damage to the paddy field embankment. This is to prevent the drone from losing control indicators and deviating from the path or stopping flight in a section where the paddy field embankment, which is a visual indicator for vision recognition, is lost.

[0069] The control means (190) follows a planned flight trajectory by referring to cadastral map coordinates even while passing through an abnormal terrain section, and at the same time maintains the operation of the pest control means (140) without interruption, thereby ensuring the continuity of pest control missions for the entire agricultural land. Through this, pest control work can be completed within a single flight process regardless of whether the actual terrain is damaged, and terrain analysis results and spraying data can be recorded in a time-series synchronized manner.

[0070] The control means (190) can activate a local correction process to realign vision sensor information and cadastral map coordinates when the drone moves out of the collapsed section and reaches a point where the normal rice paddy embankment boundary is re-detected within the frame by the boundary detection algorithm. As a result, the drone automatically returns to a control system based on a correction path with secured precision and continues optimized outer-following pest control and terrain diagnosis. The processing means (130) can receive information regarding the cadastral map from the control device (200) through the communication means (180).

[0071] FIG. 9 illustrates an example of a paddy field condition diagnosis report, and FIG. 10 illustrates an example of waypoint-by-waypoint data logging and post-verification. The control device (200) calculates the history of position deviation between the initial flight path based on the cadastral map and the actual corrected flight path for each waypoint and stores it in a standardized data format. It also generates post-verification information regarding the precision of pest control and terrain soundness by matching abnormal condition information of the paddy field determined by the boundary detection algorithm with the location data of the corresponding waypoint, and can visualize the location of the paddy field collapse area and the pest control error based on the post-verification information.

[0072] The control device (200) can precisely calculate the history of position deviation between the initial flight path based on the cadastral map and the corrected path actually flown through real-time recognition for each waypoint, based on flight data received from the drone through the communication means (180).

[0073] The calculated location deviation data can be converted into a standardized data format including a waypoint identifier (ID), an error value, and topographic condition information of the corresponding point, and stored in memory. At this time, a process is performed to immediately match abnormal condition information, such as the loss or damage of a paddy field embankment determined by the boundary detection algorithm of the processing means (130), with the location data of the corresponding waypoint. Through this matching process, integrated post-verification information is generated regarding pest control precision, which indicates the accuracy of the point where actual spraying took place, and topographic soundness, which indicates the physical preservation state of the outer perimeter of the farmland.

[0074] The control device (200) visualizes the generated post-verification information and provides it to the user. As shown in FIG. 9, the actual flight trajectory of the drone is projected onto the map data, and a specific marker (e.g., '!') is displayed at the point where an abnormal condition is detected, thereby intuitively indicating the exact location of the rice paddy embankment collapse area.

[0075] In addition, as illustrated in Fig. 10, the report quantifies and visualizes pest control errors occurring for each section and waypoint in a chart or list format, thereby enabling users to objectively review the possibility of minute path deviations or pest control omissions that are difficult to identify with the naked eye. Consequently, this system goes beyond simple pest control outsourcing to generate digital diagnostic data for the precise management of farmland and terrain maintenance, thereby improving the efficiency of agricultural management.

[0076] FIG. 11 illustrates the hierarchical hardware and software architecture of an perimeter pest control system according to an embodiment of the present invention, wherein the system (10) can perform real-time path correction and pest control through the organic combination of an input layer, a processing layer, a fusion layer, an output layer, and a driving layer.

[0077] The input layer is a stage for collecting basic data of the system, receiving control commands from the pilot through the VK V9 controller, acquiring KML boundary data of the cadastral map of the farmland from the Vworld API, and obtaining real-time RGB image information through the OAK-1 camera.

[0078] The processing layer is a step for primarily processing input data. The processing means (130) can convert KML data into waypoints and generate a reference path by calculating the coordinates of the outline flight path that the drone must follow from the KML polygons. At the same time, the image processing module can activate an HSV filter to apply a predefined threshold value of the paddy field color and determine the visual location of the paddy field through edge detection based on the color range.

[0079] As a processing means (130) for the fusion layer, the layer where the core operations of the present invention are performed may include a data fusion engine based on Raspberry Pi Zero 2W. The data fusion engine can align the actual paddy field location detected by the KML reference coordinates and the HSV filter by auxiliaryly referencing the current location information received from the GPS module (170). Through this, it is possible to precisely calculate the error vector between the two pieces of information and generate a real-time correction command to modify the flight trajectory of the drone.

[0080] The output layer also serves as a processing means (130) to convert the correction command calculated in the fusion layer into a format that the flight controller can understand. Specifically, it can generate and transmit a real-time path correction packet containing a target position setting command (SET_POSITION_TARGET_LOCAL_NED) based on the local coordinate system using the MAVLink protocol.

[0081] The driving layer performs the final physical operation as a control means (190). The driving layer can correct the actual flight path by adjusting the output of the drone motor through the VK V9 flight controller (FC). At the same time, spraying on the outer edge of the paddy field synchronized with the corrected path can be performed through the control of the pest control pump. The current flight status information generated by the driving layer is fed back to the GPS module (170) to form a real-time control loop, thereby enabling continuous path following.

[0082] The system of the present invention secures precise boundary information of agricultural land by linking in real-time with the Vworld API, a public data platform, without a separate mapmaking process or manual coordinate input. Referring to the hierarchical architecture of FIG. 11, the data acquisition module provided in the input layer can automatically receive KML (Keyhole Markup Language) boundary data of the corresponding plot by connecting to the Vworld Open API server based on the address or current location information of the agricultural land to be treated entered by the user.

[0083] The received KML data is passed to the path planning module in the processing layer, where it is immediately calculated into reference path (flight path) coordinates for flying along the outlines of the farmland. This automated data linkage structure completely eliminates the need for pilots to manually register waypoints in the field or generate complex terrain files in advance.

[0084] The real-time cadastral map linkage technology of the present invention has the following key differences compared to conventional prior art. First, the method proposed in Patent Document 2 involves a cumbersome preliminary procedure in which a pilot manually flies a drone and collects and records the outer GPS coordinates of the farmland one by one. On the other hand, the present invention immediately calls boundary data on the cadastral map via software through API linkage, thereby realizing 'responsiveness without preliminary work' that allows pest control missions to begin immediately upon arrival at the site without a separate measurement flight.

[0085] Second, Patent Document 1 discloses a technology that generates a first pest control path by collecting cadastral maps or satellite maps from an external server, but it has the limitation of a 'scan-flight (2-pass)' method that analyzes images and updates the path to verify the actual planting distribution of crops. In contrast, the present invention implements a single flight process (1-pass) that uses highly reliable cadastral map boundaries obtained through the Vworld API as initial flight indicators and simultaneously performs real-time vision correction through the OAK-1 camera.

[0086] As a result, the present invention ensures the timeliness of data through organic integration with public intellectual data APIs, while simultaneously maximizing operational efficiency for the realization of precision agriculture by drastically reducing pilot intervention and work time.

[0087] The system according to the present invention is linked in real-time with the Vworld API, a public data platform, to automatically acquire precise boundary information for agricultural land. A data acquisition module equipped in the input layer of the system connects to the Vworld Open API server based on an address or current location entered by the user to receive KML boundary data for the corresponding parcel. This automatic linkage structure can completely eliminate the cumbersome procedure of requiring an operator to directly register waypoints or manually generate terrain files in the field.

[0088] This serves as the basis for realizing superior flight convenience without prior work compared to the prior art, Patent Document 2, which relies on manual GPS coordinate collection by the pilot. In addition, unlike Patent Document 1, which requires a separate scan flight for image analysis, the present invention enables a single flight process that performs correction simultaneously with pest control by using highly reliable data obtained through an API as an initial indicator.

[0089] Boundary detection algorithms can use two quantitative indicators, pixel continuity and pixel density per unit area, to identify abnormal conditions such as the collapse or loss of paddy field embankments. Pixel continuity is a numerical representation of the degree to which detected edge pixels are connected in a single linear trajectory, while pixel density per unit area refers to the proportion of effective pixels constituting the shape of the paddy field embankment within a specific area.

[0090] The algorithm compares these figures in real-time with a standard model based on cadastral maps, confirming paddy field embankment damage if pixel connections are severed or density drops below a threshold. These quantitative criteria enable error-free diagnosis of the actual terrain condition, even in environments with interference such as shadows from external light sources or weeds. Information identified as abnormal is matched with the location data of the corresponding waypoints to generate post-verification data; ultimately, this data can be utilized to create a precise diagnostic report that visualizes the location of embankment collapse areas and control errors.

[0091] Figure 12 compares the work efficiency and performance indicators of the perimeter pest control system according to the present invention with conventional technologies, and the comparative analysis to prove the technical superiority of the present invention is as follows.

[0092] The first method, v1, is an early-stage technology in which the user manually inputs GPS offset correction values ​​into a cadastral map-based route before flying. This method has technical flaws, such as the constant occurrence of GPS drift of 1 to 3 meters and the requirement for manual offset calculation every time due to varying correction values ​​depending on field weather conditions. As a result, deviations from the paddy field embankment by more than 2 meters occur frequently, and the pest control omission rate reaches approximately 15 to 20%, making rework unavoidable; consequently, it was determined that field application is impossible.

[0093] The second method, v2, is a technology that first generates a path through a primary scan flight and then performs a secondary pest control flight. This method has structural problems, such as requiring approximately twice the amount of time for operation and experiencing a sharp increase in battery consumption; in particular, it is difficult to respond flexibly to changes in paddy field conditions between the time of scanning and the time of actual pest control. Numerically, it requires about 15 to 20 minutes of additional time per hectare, which increases operating costs and exacerbates fatigue in the field; therefore, it was rejected due to limitations in terms of efficiency.

[0094] In contrast, the present invention v3 adopts a single-pass flight process method that simultaneously performs real-time HSV recognition and pest control using an OAK-1 camera. It achieves technical improvements by fusing Vworld cadastral map data and camera information through a lightweight computing engine based on Raspberry Pi Zero 2W (RPi Zero 2W) and correcting the flight controller (FC) in real time using the MAVLink protocol.

[0095] Figure 13 quantitatively compares the performance and efficiency of conventional perimeter pest control methods with the real-time fusion correction method according to the present invention. v1 (GPS offset method) has an average paddy field embankment deviation distance of approximately 1.8m and records a high pest control omission rate of 15–20%. This method has low operational precision, to the extent that re-pest control occurs in about 3 out of 5 operations. Although it requires only one set of batteries, it exhibits performance at a level that makes actual field application difficult due to frequent path deviations.

[0096] v2 (pre-scan method) is a variant that checks the terrain in advance of flight to reduce the distance of deviation from the paddy ridge by approximately 0.6m and improve the pest control miss rate to 5–8%. However, because it involves a separate scan flight prior to the pest control flight, it incurs an additional 10–15 minutes of work time per hectare and increases battery consumption to two sets, resulting in economic and time inefficiency. Although the frequency of re-treatment has decreased to one in five times, the increased work time hinders overall pest control efficiency.

[0097] In contrast, v3 (real-time 1-pass method) according to the present invention drastically reduces the average paddy field embankment deviation distance to 0.3m or less through real-time vision recognition and correction algorithms. It also exhibits the highest precision with a pest control omission rate of 5% or less, and in particular, it achieves high efficiency by reducing working time by approximately 40% compared to v1 while enabling the completion of all tasks with only one set of batteries. Analysis shows that the frequency of re-pest control is almost non-existent, proving that it achieves technical effects that outperform all conventional technologies in terms of operational reliability and economic efficiency.

[0098] When applying v3, a real-time fusion correction method according to the present invention, the average distance of deviation from the paddy field embankment, which was approximately 1.8m in v1, a conventional manual GPS offset correction method, is drastically reduced to 0.3m or less. This figure is achieved through the organic combination of real-time vision recognition and an error correction algorithm, and is a quantitative indicator proving that the drone follows the boundary of the actual terrain with great precision without deviating from the planned flight trajectory.

[0099] The pest control omission rate, a key factor determining the quality of pest control, can also be significantly improved to 5% or less compared to the 15% to 20% level of conventional methods. This high precision enables the minimization of errors in spraying positions and the securing of uniform pest control performance by reflecting the predicted movement amount resulting from computational delays into the control value in real time, even during high-speed flight.

[0100] In terms of operational efficiency, it demonstrates excellent performance improvement by reducing the total operation time by approximately 40% compared to conventional technology through a single flight process (1-pass) that does not require a separate pre-scan flight. This not only maximizes the operational efficiency of the drone but also serves as a technical foundation for completing pest control missions over a wider area within the same amount of time.

[0101] By optimizing resource utilization, high resource efficiency is realized, enabling a given pest control mission to be completely completed with only one set of drone batteries. In contrast to conventional pre-scan methods that required the consumption of two or more sets of batteries due to an increase in the number of flights, this invention enables high-precision pest control while minimizing power consumption, thereby contributing to the reduction of operating costs and improvement of convenience for farms.

[0102] FIG. 14 shows a multi-faceted comparison of the technical elements of the prior art and the present invention, and the distinctive features and technical advantages of the present invention compared to the prior art are as follows.

[0103] Unlike Patent Document 1, which recognizes crop planting rows or general flight paths, and Patent Document 2, which uses only pre-collected GPS coordinate points as recognition targets, the present invention possesses a recognition algorithm specialized for the ridge boundary itself. Through this, the most precise perimeter tracking based on the actual terrain boundary is possible.

[0104] In terms of processing methods, Patent Document 1 adopts a 2-pass method in which a pest control flight is performed after a scan flight, resulting in a long working time, and Patent Document 2 has the inconvenience of requiring manual registration of coordinates before flight. On the other hand, the present invention can reduce working time by implementing a 1-pass method that corrects the path in real time simultaneously with pest control.

[0105] Regarding image processing techniques, Patent Document 1 utilizes deep learning or general-purpose computer vision (CV), which can result in a relatively high computational load, while Patent Document 2 does not utilize image processing technology at all. In contrast, the present invention applies a lightweight real-time HSV computation algorithm to enable delay-free real-time correction computation even on low-spec hardware.

[0106] In terms of map data utilization, Patent Document 1 requires the prior creation of a separate satellite map or drone map, and Patent Document 2 has limitations in that it relies on the user's manual GPS marking. By automatically linking the Vworld public cadastral map API with the system, the present invention can provide the convenience of immediately reflecting the latest cadastral information into the pest control route without any separate prior work.

[0107] Finally, in contrast to conventional flight controller (FC) command methods that remain at the level of static control simply executing pre-planned paths or waypoints, the present invention can perform real-time dynamic correction commands using the MAVLink protocol. This allows for the immediate correction of path deviations caused by minute terrain errors or external forces occurring during flight, thereby maintaining uniform pest control quality.

[0108] The following is a detailed description of the MAVLink protocol and the SET_POSITION_TARGET_LOCAL_NED message, which are the core mechanisms for transmitting correction commands from the output layer of the present invention to the flight controller (FC).

[0109] [MAVLink-based Real-time Dynamic Correction Command System]

[0110] The system of the present invention has an output structure using the MAVLink communication protocol so that the flight controller (FC) of the drone can understand and execute the correction path data calculated in the fusion layer in real time. MAVLink is a lightweight message marshalling protocol for data exchange between a drone and a ground station or an onboard computing unit, and in the present invention, it can be utilized as a high-speed communication channel between a data fusion engine (Raspberry Pi Zero 2W) and a flight controller (VK V9 FC).

[0111] The output layer can receive the correction vector and target coordinates generated by the data fusion engine of the fusion layer, convert them into a specific command packet of MAVLINK, the SET_POSITION_TARGET_LOCAL_NED message, and transmit them.

[0112] [Technical role of the SET_POSITION_TARGET_LOCAL_NED message]

[0113] SET_POSITION_TARGET_LOCAL_NED is a message for precisely controlling the target position, velocity, and acceleration based on the drone's local coordinate system (Local NED: North-East-Down). The implementation importance of this message in the present invention is as follows.

[0114] Coordinate System Alignment and Position Control: By calculating the relative error based on vision sensors and global coordinates (GPS) based on cadastral maps, the derived final correction point is converted into a local coordinate system based on the drone's current location and transmitted, enabling precise positional movement without error.

[0115] Dynamic trajectory modification: Unlike the existing method of simply executing predefined waypoints sequentially, this becomes a key means of realizing 'in-flight path modification' by dynamically changing the target position in real-time during flight and transmitting packets.

[0116] Delay time compensation reflection: As the final coordinate value reflecting the predicted movement amount calculated by the error correction algorithm is included in this packet and transmitted, the flight controller can immediately follow the corrected trajectory without physical time lag caused by computational delay.

[0117] [Organic Interaction with the Driving Layer]

[0118] The converted real-time path correction packet is transmitted to the VK V9 flight controller (FC) of the driving layer, and the flight controller corrects the actual flight trajectory by fine-tuning the output of each motor according to the received position setting command. At the same time, the control means controls the spraying timing of the pest control pump in synchronization with the corrected position data, thereby enabling precise perimeter pest control with terrain errors corrected in real time.

[0119] Consequently, the use of the SET_POSITION_TARGET_LOCAL_NED message serves as a practical data interface connecting the intelligent computation unit (fusion layer) and the physical driving unit (driving layer), thereby enabling real-time dynamic control performance that differentiates the present invention from conventional technology.

[0120] The image analysis and flight control algorithms applied in the present invention are closely related to modern Lightweight AI technology. Specifically, the boundary detection and terrain diagnosis process performed in the processing means (130) is based on an AI-based vision recognition mechanism that classifies pixel-unit data according to learned thresholds and identifies patterns.

[0121] Unlike general high-performance deep learning models, the artificial intelligence technology adopted by the present invention has a structure optimized to achieve low latency and minimize computational complexity in the limited resource environment of a drone. This is an essential technical choice for performing immediate path correction and pest control within a single flight process (1-pass) even while the drone is flying at high speed.

[0122] The artificial intelligence algorithm of the present invention significantly reduces the dimensionality of input data by performing calculations by limiting the operation to only the local area of ​​interest (ROI) estimated based on cadastral map coordinates, instead of analyzing the entire image frame. This local correction technique eliminates unnecessary background calculations, thereby accelerating the inference speed of the artificial intelligence model and lowering the computational complexity.

[0123] The boundary detection algorithm adopts a lightweight structure that combines HSV color space analysis and a pixel density-based morphological diagnostic model instead of a complex multilayer neural network. This enables latency-free real-time computation even on ultra-small, low-power embedded hardware such as the Raspberry Pi Zero 2W, and ensures high reliability through weighted filtering that actively responds to changes in external light sources.

[0124] The error correction and flight control synchronization algorithm utilizes AI predictive modeling technology to offset time lags occurring during image processing. Rather than simply relying on past image information, it estimates the drone's flight state during computational delay based on data to calculate predicted movement, and by summing this to the control value, it can preemptively correct physical errors.

[0125] Accordingly, the present invention can provide an intelligent perimeter pest control solution capable of precise terrain tracking and pest control while ensuring real-time performance by lightweighting and optimizing high-performance server-class artificial intelligence technology to fit the specifications of drone-mounted hardware.

[0126] The artificial intelligence-based image analysis algorithm applied in the present invention has an optimized structure to achieve low latency and minimize computational complexity in the limited resource environment of a drone. Instead of a deep learning model based on a multilayer neural network that generally requires high computational resources, a lightweight algorithm combining HSV color space analysis and a morphological diagnostic model with relatively low computational load can be adopted.

[0127] This technical configuration serves as the key basis for enabling real-time computation without physical latency, even on ultra-small and low-spec embedded hardware such as the Raspberry Pi Zero 2W. In particular, by limiting processing to only the local area of ​​interest (ROI) set in conjunction with cadastral map coordinates instead of analyzing the entire video frame, the amount of data processed can be drastically reduced and the inference speed of artificial intelligence models can be accelerated.

[0128] Accordingly, the present invention can provide technical effects that perfectly guarantee real-time responsiveness of the system even in low-spec hardware environments, and enable the simultaneous execution of immediate path correction and precision pest control missions without error within a single flight process, even while the drone is flying at high speed. Explanation of the symbols

[0129] 10: Perimeter pest control system 50: Drone 80: Controller 100: Pest control device 110: Means of flight 120: Means of filming 130: Treatment means 140: Control means 170: GPS module 180: Communication means 190: Control means 200: Control device

Claims

Claim 1 The apparatus includes a shooting means (120) for acquiring image information through a flight path containing boundary information of agricultural land based on a cadastral map; a processing means (130) for generating paddy field dike information using the image information and generating a correction path by calculating an error between the flight path and the paddy field dike information; and a control means (190) for controlling the operation of a flight means (110) and a pest control means (140) based on the correction path. The shooting means performs spatial filtering on pixel data of an input image through a hardware acceleration algorithm to remove noise, and generates image information by reducing it to a resolution set in correspondence with the computational specifications of a main processor provided in the processing means. The processing means generates a correction path through a local correction algorithm, a boundary detection algorithm, and an error correction algorithm. The local correction algorithm projects coordinate information of a cadastral map-based flight path onto pixel coordinates within an image frame from image information of a set resolution, and sets a local analysis area only for a certain range of pixel sections where a paddy field dike is predicted to exist using the projected coordinates as a center point. The boundary detection algorithm [applies to] pixels within the local analysis area An perimeter pest control system for an agricultural drone, characterized by performing a threshold operation of a predefined color only for the purpose of extracting paddy field information corresponding to the actual paddy field boundary line in real time, the error correction algorithm calculates an error vector between the flight path and the paddy field information, calculates a delay time from the time when image information is received to the time when the error vector is calculated, and generates a correction path by applying the error vector to the predicted amount of movement moved by the flight means during the delay time, and the control means operates with a flight pest control synchronization algorithm that synchronizes the spraying timing by a time point corresponding to the predicted amount of movement along with the modification of the flight trajectory according to the correction path, and the pest control operation is performed together with position correction through the time-series organic operation of each algorithm. Claim 2 delete Claim 3 An perimeter pest control system for an agricultural drone according to claim 1, wherein the boundary detection algorithm operates by: a step of extracting an initial boundary candidate group by applying a color threshold to pixels within the local analysis area; a step of determining a final measured boundary in which shadows and exposure interference caused by an external light source are removed by applying a filtering method that applies weights by analyzing the distribution of saturation (S) and brightness (V) components to the extracted initial boundary candidate group; and a step of determining an abnormal state, including loss or damage of a paddy field embankment, based on whether there is a discrepancy in physical form by comparing the pixel continuity and pixel density per unit area of ​​the final measured boundary in real time with boundary information of a flight path based on a cadastral map; and wherein the flight pest control synchronization algorithm operates by: a step of maintaining continuity of flight and pest control by referring to a flight path based on a cadastral map when an abnormal state is determined; and a step of returning to control based on a correction path at the point where a normal paddy field embankment boundary is re-detected by the boundary detection algorithm.

Images