Artificial intelligence rice farm pest bird monitoring and laser repellent integrated system

Abstract

An integrated AI system for monitoring harmful birds in rice farming and laser repellent is provided, comprising a camera for photographing a rice farming field, a laser repellent that irradiates a laser toward harmful birds within the rice farming field, a receiver that receives images of the rice farming field from the camera, a detection unit that detects harmful birds within the rice farming field using a pre-established object detection model, and an integrated control server that controls the laser repellent to irradiate a laser toward the harmful birds.

Description

Technology Field

[0001] The present invention relates to an integrated AI rice farming harmful bird monitoring and laser eradication system, and provides a system that increases the laser strike rate by identifying not only the appearance but also the behavioral patterns of harmful birds, and reduces resistance to eradication by randomly using ultrasound, LED, and laser multi-beams. Background Technology

[0002] Rice farming forms the backbone of Korean agriculture, but harmful birds such as magpies, sparrows, and pigeons cause economic losses to farmers by eating rice seeds immediately after transplanting and damaging rice ears during the heading and harvesting stages. Recently, due to the effects of climate change, urban expansion, and deforestation, the population of harmful birds has increased by an annual average of 10–15%, and damage is accumulating as birds consume an average of 3–40g of grain per day. Traditional manual, chemical, and mechanical pest control methods show limitations in long-term efficiency due to environmental issues and the high adaptability of birds. Accordingly, sustainable pest control strategies are needed to achieve both environmental conservation and food security simultaneously, and AI-based research is required to address field variability and the learning and tolerance development of birds.

[0003] At this time, a method for detecting harmful birds based on AI and repelling them using a multi-focal laser has been researched and developed. In this regard, prior art Korean Registered Patent No. 10-2805949 (published May 14, 2025) and Korean Published Patent No. 2023-0018062 (published February 7, 2023) disclose a configuration that automatically recognizes harmful birds from CCTV footage using machine learning, precisely targets only birds by irradiating them with a green laser based on extracted eye coordinates, separately recognizes the eye coordinates of humans or cows to exclude them from the target, and ensures safety by issuing commands to prohibit or correct laser irradiation using safety coordinate rules; and a configuration that detects the appearance of birds using radar, enhances repellent effectiveness and coverage by reciprocating the laser pointer with a zigzag guide rail and interlocking link structure to irradiate a wide area with a multi-focal laser, and suppresses the learning effect by applying multi-stage threats through the simultaneous use of acoustic stimulation including a speaker.

[0004] However, in the former case, birds exhibit diverse behavioral patterns that frequently lead to false detections, making precise targeting impossible; conversely, in the latter case, only configurations have been disclosed that cover a wide area but lack a tangible sense of impact. Although damage caused by harmful birds persists every year, solutions remain limited, and it is difficult to devise effective countermeasures due to the high adaptability and resistance to eradication possessed by these birds. Therefore, research and development are required for a system that analyzes not only the appearance but also the behavioral patterns of harmful birds to reduce false detection rates and inhibit their learning effects, thereby lowering their adaptability and resistance to eradication. The problem to be solved

[0005] One embodiment of the present invention provides an integrated AI monitoring and laser eradication system for harmful birds in rice farming, which utilizes different object detection models for day and night when detecting harmful birds, and analyzes not only the appearance but also the behavioral patterns of the harmful birds using Temporal Pattern Analysis, thereby reducing the false positive rate of birds not detected according to their behavioral patterns and increasing accuracy; increases the laser hit rate by predicting the flight direction and speed of birds by operating a real-time object detection model; and reduces adaptability and resistance to eradication by randomly operating not only lasers but also ultrasound, LEDs, and laser multi-beams. However, the technical problem that this embodiment aims to solve is not limited to the technical problem described above, and other technical problems may exist. means of solving the problem

[0006] As a technical means for achieving the technical problem described above, one embodiment of the present invention includes a camera for photographing a rice cultivation field, a laser repellent for irradiating a laser toward harmful birds within the rice cultivation field, a receiving unit for receiving an image of the rice cultivation field from the camera, a detection unit for detecting harmful birds within the rice cultivation field using a pre-established object detection model, and an integrated control server including a control unit for controlling the laser repellent to irradiate a laser toward the harmful birds. Effects of the invention

[0007] According to any one of the means for solving the problem of the present invention described above, when detecting harmful birds, different object detection models are used for day and night, and at the same time, not only the appearance but also the behavioral patterns of the harmful birds are analyzed through temporal pattern analysis. This reduces the false positive rate of birds that are not detected according to their behavioral patterns, thereby increasing accuracy. Additionally, by operating a real-time object detection model to predict the flight direction and speed of the birds, the laser hit rate can be increased. Furthermore, by randomly operating not only the laser but also ultrasound, LEDs, and laser multi-beams, adaptability and resistance to deterrence can be reduced. Brief explanation of the drawing

[0008] FIG. 1 is a drawing illustrating an integrated system for monitoring harmful birds in rice farming and laser control according to one embodiment of the present invention. Figure 2 is a block diagram illustrating an integrated control server included in the system of Figure 1. FIGS. 3 and 4 are drawings for illustrating an embodiment in which an AI rice farming harmful bird monitoring and laser repellent solution according to an embodiment of the present invention is implemented. FIG. 5 is a flowchart illustrating a method for providing an AI rice farming harmful bird monitoring and laser eradication solution according to an embodiment of the present invention. Specific details for implementing the invention

[0009] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0010] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected" but also cases where they are "electrically connected" with other elements interposed between them. Furthermore, when a part is described as "including" a component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components, and it should be understood that this does not preclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0011] Terms such as “about,” “substantially,” etc., used throughout the specification, are used to mean at or near the stated value when inherent manufacturing and material tolerances are presented in the stated meaning, and are used to prevent unscrupulous infringers from unfairly exploiting the disclosure in which precise or absolute values ​​are mentioned to aid in understanding the invention. Terms such as “step” or “step of” used throughout the specification of the invention do not mean “step for”.

[0012] In this specification, the term "part" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Additionally, one unit may be realized using two or more pieces of hardware, and two or more units may be realized by one piece of hardware. Meanwhile, "part" is not limited to software or hardware, and "part" may be configured to reside in an addressable storage medium or configured to run on one or more processors. Accordingly, as an example, "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.

[0013] Some of the operations or functions described herein as being performed by a terminal, device, or device may instead be performed by a server connected to said terminal, device, or device. Likewise, some of the operations or functions described as being performed by a server may also be performed by a terminal, device, or device connected to said server.

[0014] In this specification, some of the operations or functions described as mapping or matching with a terminal may be interpreted as meaning mapping or matching the terminal's unique number or personal identification information, which is the terminal's identifying data.

[0015] The present invention will be described in detail below with reference to the attached drawings.

[0016] FIG. 1 is a drawing illustrating an integrated system for monitoring harmful birds in rice farming and laser repellent according to an embodiment of the present invention. Referring to FIG. 1, the integrated system for monitoring harmful birds in rice farming and laser repellent (1) may include at least one user terminal (100), an integrated control server (300), at least one camera (400), and at least one laser repellent (500). However, since the integrated system for monitoring harmful birds in rice farming and laser repellent (1) of FIG. 1 is merely an embodiment of the present invention, the present invention is not to be interpreted as being limited by FIG. 1.

[0017] At this time, each component of FIG. 1 is generally connected through a network (Network, 200). For example, as shown in FIG. 1, at least one user terminal (100) can be connected to an integrated control server (300) through the network (200). And, the integrated control server (300) can be connected to at least one user terminal (100), at least one camera (400), and at least one laser repellent (500) through the network (200). Also, at least one camera (400) can be connected to the integrated control server (300) through the network (200).

[0018] Here, a network refers to a connection structure capable of exchanging information among individual nodes, such as multiple terminals and servers. Examples of such networks include Local Area Networks (LANs), Wide Area Networks (WANs), the World Wide Web (WWW), wired and wireless data networks, telephone networks, and wired and wireless television networks. Examples of wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), 5GPP (5th Generation Partnership Project), 5G NR (New Radio), 6G (6th Generation of Cellular Networks), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), RF (Radio Frequency), Bluetooth network, NFC (Near-Field Communication) network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, etc.

[0019] In the following, the term "at least one" is defined as a term including both singular and plural forms, and it will be obvious that even if the term "at least one" does not exist, each component may exist in a singular or plural form and may mean singular or plural. Furthermore, whether each component is provided in a singular or plural form may be changed according to the embodiment.

[0020] At least one user terminal (100) may be a terminal of a user who is an administrator and monitors the results of detecting and repelling harmful birds using a web page, app page, program, or application related to AI rice farming harmful bird monitoring and laser repellent solution.

[0021] Here, at least one user terminal (100) may be implemented as a computer capable of connecting to a remote server or terminal via a network. Here, the computer may include, for example, a navigation system, a laptop equipped with a web browser, a desktop, a laptop, etc. At this time, at least one user terminal (100) may be implemented as a terminal capable of connecting to a remote server or terminal via a network. At least one user terminal (100) may include all kinds of handheld-based wireless communication devices, such as navigation, PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartphone, smartpad, tablet PC, etc.

[0022] The integrated control server (300) may be a server that provides a web page, app page, program, or application for AI rice farming harmful bird monitoring and laser repellent solutions. Additionally, the integrated control server (300) may be a server that detects harmful birds after receiving video from a camera (400) and causes a laser to be emitted through a laser repellent (500). Furthermore, the integrated control server (300) may be a server that outputs these results to a dashboard and provides them to a user terminal (100). Here, the integrated control server (300) may be implemented as a computer capable of connecting to a remote server or terminal via a network. Here, the computer may include, for example, a laptop, desktop, or laptop equipped with a navigation system or a web browser.

[0023] At least one camera (400) may be a device that transmits video captured with or without using a web page, app page, program, or application related to AI rice farming harmful bird monitoring and laser repellent solution to an integrated control server (300).

[0024] Here, at least one camera (400) may be implemented as a computer capable of connecting to a remote server or terminal via a network. Here, the computer may include, for example, a navigation system, a laptop equipped with a web browser, a desktop, a laptop, etc. At this time, at least one camera (400) may be implemented as a terminal capable of connecting to a remote server or terminal via a network. At least one camera (400) may include all kinds of handheld-based wireless communication devices, such as navigation, PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartphone, smartpad, tablet PC, etc.

[0025] At least one laser repellent (500) may be a device that irradiates a laser in the direction of harmful birds by receiving control from an integrated control server (300) using a web page, app page, program, or application related to AI rice farming harmful bird monitoring and laser repellent solutions.

[0026] Here, at least one laser repellent (500) may be implemented as a computer capable of connecting to a remote server or terminal via a network. Here, the computer may include, for example, a laptop, desktop, or laptop equipped with a navigation system or a web browser. At this time, at least one laser repellent (500) may be implemented as a terminal capable of connecting to a remote server or terminal via a network. At least one laser repellent (500) may include all kinds of handheld-based wireless communication devices, such as navigation, PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartphone, smartpad, tablet PC, etc.

[0027] FIG. 2 is a block diagram for explaining an integrated control server included in the system of FIG. 1, and FIG. 3 and FIG. 4 are drawings for explaining an embodiment in which an AI rice farming harmful bird monitoring and laser repellent solution according to an embodiment of the present invention is implemented.

[0028] Referring to FIG. 2, the integrated control server (300) may include a receiving unit (310), a detection unit (320), a control unit (330), a bird detection optimization unit (340), and an automatic output adjustment unit (350).

[0029] When an integrated control server (300) or another server (not shown) operating in conjunction with an embodiment of the present invention transmits an AI rice farming harmful bird monitoring and laser repellent solution application, program, app page, web page, etc. to at least one user terminal (100), at least one camera (400), and at least one laser repellent (500), the at least one user terminal (100), at least one camera (400), and at least one laser repellent (500) may install or open the AI ​​rice farming harmful bird monitoring and laser repellent solution application, program, app page, web page, etc. Additionally, a service program may be operated on at least one user terminal (100), at least one camera (400), and at least one laser repellent (500) using a script executed in a web browser. Here, a web browser refers to a program that enables the use of web (WWW: World Wide Web) services and receives and displays hypertext described in HTML (Hyper Text Mark-up Language), and includes, for example, Chrome, Microsoft Edge, Safari, Firefox, Whale, UC Browser, etc. Additionally, an application refers to an application on a terminal, and includes, for example, an app running on a mobile terminal (smartphone).

[0030] Referring to FIG. 2, the receiving unit (310) can receive an image of a rice cultivation site from the camera (400). The camera (400) can photograph the rice cultivation site. In one embodiment of the present invention, a rice cultivation site is used as an example, but it is understood that it may include cultivation sites for other crops or greenhouses.

[0031] The detection unit (320) can detect harmful birds within a rice cultivation area using a pre-established object detection model. In one embodiment of the present invention, harmful birds that cause damage to rice farming are described as examples, but it is obvious that harmful pests, i.e., animals other than birds, may be included. At this time, the object detection model may be a hybrid model that uses ① YOLO in bright daytime environments as shown in FIG. 3 (a), and ② Faster R-CNN (Faster Regional Convolutional Neural Network) in low light, rain, or nighttime environments, as will be described later. In addition, as shown in FIG. 4f, by analyzing movement vectors and shape changes within consecutive frames, ③ Temporal Pattern Analysis is combined with the hybrid model to classify patterns of individual flight, gliding, wing folding, and swarm flight, thereby reducing false positives and determining the threat level early.

[0032] In reality, since harmful birds exhibit various behavioral patterns such as individual flight, gliding, wing folding, and drowsiness, it is difficult to make a judgment based on a single still image, and there is a possibility of false detection. Therefore, time-series pattern analysis is used to identify the type of behavioral pattern in the image. By analyzing changes in shape and direction of movement between consecutive frames, behavioral pattern-based classification can be performed. The process for modeling the above-mentioned model will be described later in the bird detection optimization unit (350).

[0033] The control unit (330) can control the laser repellent (500) to irradiate a laser toward harmful birds. The laser repellent (500) can irradiate a laser toward harmful birds within a rice cultivation field. The laser repellent (500) can be connected to a rotation module based on a servo motor and an encoder motor to follow harmful birds.

[0034] First, the pixel coordinates viewed by the camera (400) are converted into camera coordinates, and the converted camera coordinates are converted into world coordinates using internal and external camera parameters, thereby converting them into coordinates that can be aimed at by the laser repellent (500). Then, the control unit (330) calculates the target angles for pan and tilt to rotate and tilt the servo motor and encoder motor of the laser repellent (500), and aims by rotating and tilting the servo motor and encoder motor accordingly, and can irradiate the laser when it converges to the target angle. At this time, the output is also automatically adjusted, which will be described later in the automatic output adjustment unit (350).

[0035] step explanation tool or model 1. Calibration and Alignment Corrects intrinsic camera parameters (focal length, principal point, distortion) and extrinsic parameters (rotation R, displacement t) between the camera and pan / tilt (laser head)—establishes field reference coordinates (world coordinate system) and defines the origin and orientation using markers (checkerboard / ArUco / AprilTag) on ​​the top of pillars, etc. Utilized OpenCV (Checkerboard / Charuco), AprilTag / ArUco, and camera intrinsic / extrinsic parameter correction tools. 2. Detection and Tracking Find harmful algae in the video using object detection (e.g., YOLO series) and obtain trajectories by assigning IDs using inter-frame tracking (e.g., ByteTrack / Kalman). PyTorch + YOLOv5 / YOLOv8, Detectron2, ByteTrack / Kalman, ONNX / TensorRT edge inference 3. Distance Estimation -Select 1) Estimate depth using stereo (left / right camera) parallax -Select 2) Acquire distance in the ROI direction using LiDAR / ToF -Select 3) Approximate using monocular + scale / ground model (intersection with the corrected ground plane) ▶ Obtain target distance d as a result Stereo (Industrial Global Shutter) + SGBM / ELAS, LiDAR / ToF (or mmWave assist), Ground model fitting (RANSAC) 4. Pixel to Camera Coordinate Conversion Restore detected pixel coordinates (u,v) to camera coordinates Xc ROS 2 + tf2 frame management, OpenCV (distortion correction / projection transformation), Eigen / NumPy linear algebra functions 5. Camera → World (or Gimbal) Coordinate Transformation Converts Xc to laser gimbal (pan / tilt) reference coordinates Xg. 6. Calculation of Pan / Tilt Target Angle - Calculates pan angle (horizontal rotation) and tilt angle (vertical rotation) from Xg using inverse kinematics - Corrects mechanism offset (deviation between laser emission axis and camera optical axis) and reference point error BLDC / Servo + Precision Encoder, STM32 / ESP32 Controller, PID + Slue Limit, Motion Profile (S-curve) 7. Motor Control - Servo / encoder motors are driven by applying PID + slew rate limiting based on the current angle error relative to the target angle. - Feedforward (predicted angular velocity) and friction / backlash compensation are used to reduce transient vibration. 8. Automatic adjustment of laser irradiation and output - When converges to the target angle, it calculates the minimum output that satisfies the required contrast based on distance (d) and ambient illuminance (Ev) (automatic capping in the 0.5–5 mW range) - Maintains safety standards by managing dwell time through scan paths (Lissajous / Luster / Random Walk) Class 2 / 3R diode (≤5 mW), constant current driver, output capacitor (distance·illumination·MPE), scan path generated 9. Safety / Interlock - Immediate shutdown upon detection of people, livestock, or reflectors; masking of prohibited areas; compliance with IEC 60825 standards. - Output down-ramp after verifying reaction effect (dispersion) via log (energy saving and prevention of immunity). Human / livestock detection model, restricted area mask (geofence), hardware interlock relay, E-stop button box

[0036] The mathematical formulas for coordinate transformation and angle calculation are as follows. First, to transform the pixel (u,v) into camera coordinates Xc, the following mathematical formula 1 can be used. Here, u,v are pixel coordinates, fx and fy are focal lengths (in pixels), cx and cy are principal points (the center pixel onto which the optical axis is projected), xn and yn are normalized image coordinates, Zc is the depth (distance) in the camera coordinate system, and Xc is [Xc,Yc,Zc] T However, it refers to a 3D point within the camera coordinate system.

[0037]

[0038] And to convert the camera coordinates to Xg, which is the coordinate of the laser repellent (500), the following mathematical formula 2 can be used. This is a process of converting the camera coordinates to world coordinates, and is a step of converting Xc to the reference coordinate Xg of the laser repellent (500). Here, Rg,c, tg,c are rotations and displacements for converting from camera coordinates to world coordinates, which are the reference of the laser repellent (500). Xg is [Xg,x,Xg,y,Xg,z] T It is a 3D point relative to the gimbal.

[0039]

[0040] At this time, Rgc and tgc can be obtained through field calibration. To calculate the target angles of pan and tilt through this, the following Equation 3 can be used. At this time, the pan angle (horizontal rotation) and tilt angle (vertical rotation) are calculated from Xg using inverse kinematics, and the mechanism offset (deviation between the laser emission axis and the camera optical axis) and reference point error are corrected.

[0041]

[0042] At this time, θpan and θtilt represent the target pan and tilt angles (radians) of the laser repellent (500).

[0043] Motor Control / Laser Irradiation Pseudocode # Input: Frame, internal / external parameters, illuminance Evdets = detect_birds(frame) # YOLO inference track = track_select(dets) # Select highest threat target(u,v) = track.pixel_center() Zc = estimate_depth(u,v) # stereo / LiDAR / ground model Xc = pixel_to_camera(u,v,Zc,K) # (a)Xg = R_gc * Xc + t_gc # (b)pan = atan2(Xg.y, Xg.x) # (c)tilt = atan2(Xg.z, sqrt(Xg.x^2+Xg.y^2)) # (c)motor.pan.setpoint = panmotor.tilt.setpoint = tiltmotor.update_PID_with_slew_limit() P_cmd = auto_power(Ev, Zc) # 0.5~5 mW, safety cap included if safety_clear() and on_target(): laser.emit(P_cmd, scan_path=dwell_safe_random())else: laser.off()

[0044] Additionally, the laser repellent (500) can be connected to an autonomous robot equipped with a 4-wheel arrangement of Mecanum wheels to move within a rice cultivation field. At this time, a Mecanum wheel is a special wheel with rollers arranged at a 45-degree angle around the circumference, which enables omnidirectional (holonomic) driving capable of forward and backward movement, left and right parallel movement (crab movement), and rotation in place. The 4-wheel Mecanum chassis, with four wheels arranged in a rectangular shape, enables fine position correction and diagonal movement even in narrow areas within the cultivation field, making it advantageous for laser aiming alignment.

[0045] The bird detection optimization unit (340) can detect harmful birds using YOLO (You Only Look Once) during the day when modeling the object detection model, and use Faster R-CNN (Faster Regional Convolutional Neural Network) in environments where the light intensity is lower than a preset threshold value, such as at night or rain, and can reduce the false detection rate by analyzing the behavioral patterns of harmful birds, such as shape changes and movement directions between consecutive frames of the image, through temporal pattern analysis.

[0046] <HSV-ROI 필터링 기반 이미지 데이터 수집>

[0047] First, to collect data for modeling, the bird detection optimization unit (340) can establish a distributed bird image collection infrastructure by utilizing a 4K high-resolution camera and an edge computing system based on NVIDIA Jetson Xavier NX as shown in FIG. 4e. Then, the bird detection optimization unit (340) converts the collected image data from RGB to the HSV color space to effectively separate and extract characteristic color patterns of birds. By specifying the HSV range based on the color distribution by species, it automatically detects the Region of Interest (ROI) and selects only the frames containing birds. In addition, by applying morphological operations, it can remove background noise that may occur during shooting and clearly enhance the contours of birds. Furthermore, the bird detection optimization unit (340) can secure diversity of light condition-based data that reflects lighting changes by time of day (sunrise, noon, sunset, night, etc.) through a 24-hour time-lapse video shooting technique. This process is summarized in Table 3.

[0048] step Key processing details tool or model 1. Establishment of Distributed Image Collection Infrastructure 4K high-resolution cameras were installed at multiple locations within the rice cultivation field, and video data was collected in a distributed manner in real time using NVIDIA Jetson Xavier NX-based edge computing devices. 4K IP Camera, RTSP Streaming, NVIDIA Jetson Xavier NX, GStreamer, RTSP Server, FastAPI / Flask Data Ingestion API 2. Color Space Conversion and Automatic ROI Detection Collected video frames are converted from RGB to the HSV color space to separate bird color patterns, and regions of interest (ROI) are automatically detected using species-specific HSV ranges. OpenCV(cv2.cvtColor, inRange), NumPy, CUDA acceleration, TensorRT edge computation 3. Frame Selection Only frames containing detected ROIs are saved to remove unnecessary frames where birds are not present. OpenCV, PyTorch DataLoader, FFmpeg Frame Filtering 4. Morphological Operations and Noise Removal Removes background noise and clearly enhances contours through morphological operations such as erosion / dilation and open / close. OpenCV(cv2.morphologyEx, cv2.erode, cv2.dilate), CUDA parallel processing 5. 24-hour time-lapse data collection Time-lapse shooting techniques were applied to secure data under various lighting conditions, such as sunrise, noon, sunset, and night. Cron Scheduler, GStreamer, FFmpeg Time-lapse Capture, OpenCV Auto Exposure Compensation 6. Data Management and Synchronization Spatiotemporal consistency of the training dataset is ensured through time synchronization between edge terminals (NTP) and image metadata (stored in DB). PostgreSQL, TimescaleDB, NTP synchronization, MQTT / REST data upload

[0049] <Securing Deep Learning Image Data for Harmful Algae by Background Environment>

[0050] Reflecting that bird behavior and cognitive characteristics vary depending on the time of day, the model must be trained using captured images that include changes in lighting conditions by time of day, such as dawn, noon, and night. Additionally, it must be trained using images under various weather conditions, such as clear, cloudy, foggy, rainy, and snowy conditions, to develop a model with high accuracy in actual climate environments. Furthermore, the model must learn patterns that consider bird appearances based on seasonality and topographical characteristics, including environmental changes during the growth stages of agricultural land (planting, growing season, harvest season, and fallow season). Additionally, it must compensate for the limitations of fixed cameras and ensure spatial diversity through multi-angle / multi-distance stereoscopic shooting utilizing mobile robots and mobile pole camera systems. To achieve this, the following processes and tools may be used. Of course, the model is not limited to the following processes and tools.

[0051] step Key processing details tool or model 1. Data Collection by Time Period Image data is collected by automatically adjusting light conditions (solar altitude, color temperature, exposure) according to time of day, such as dawn, noon, and night, and training data is secured that reflects the behavioral differences of birds by time of day. GStreamer / FFmpeg Time-lapse Scheduler, Auto Exposure, Light Sensor Integration, NTP Synchronization, OpenCV Image Meta Tagging 2. Securing Data by Weather Conditions We secured images of various climate environments, including clear, cloudy, foggy, rainy, and snowy conditions, to construct a dataset for training a real-world adaptive model. Weather API (OpenWeatherMap) integration, sensors (temperature, humidity, pressure, visibility), automated image classification tags (PyTorch ImageFolder), augmentation techniques (Random Fog, Rain, Snow) 3. Data Collection by Growth Stage Learning bird appearance patterns based on changes in crop height, color, and reflectance by reflecting the conditions of paddy fields at each stage, such as planting, growth, harvest, and fallow. Satellite / Drone NDVI (Vegetation Index)-based growth classification, seasonal image clustering (K-means), label management (Label Studio, DVC), PostgreSQL meta DB 4. Ensuring spatial diversity (multi-angle / multi-distance) To compensate for the limitations of fixed cameras, stereoscopic data is collected from various angles and distances using a mobile robot and a movable pole camera system. ROS 2-based SLAM (Nav2, RTAB-Map), IMU+LiDAR fusion, RTSP streaming, NVIDIA Jetson Orin NX, 3D Reconstruction (Open3D, COLMAP) 5. Data Integration and Preprocessing Constructed an integrated training set by merging temporal, weather, growth, and spatial information into metadata. Performed removal of redundant and noise frames and normalization. Pandas+NumPy, OpenCV, PyTorch Dataset Class, TimescaleDB (Time Series DB), MLflow / DVC Version Management 6. Construction and Validation of Training Dataset Balanced sampling of collected data and automatic separation of validation sets prior to model training. PyTorch Lightning, TensorFlow Data Pipeline, Weights & Biases (W&B), Data Visualization and Quality Verification

[0052] <Application of object detection model optimized for bird detection>

[0053] Then, the aforementioned hybrid model, namely [YOLO + Faster R-CNN + Time Series Pattern Analysis], is constructed, and an object detection model optimized for real-time detection is selectively used.

[0054] Data Labeling and Data Augmentation

[0055] A multi-level based precision annotation system can be utilized to label various species of harmful birds and include various behavioral patterns (flight, landing, foraging, resting, etc.) and population information (single, small, medium, large colonies). Additionally, considering the specific characteristics of agricultural zones (rice paddy surface reflection, low light, humidity, etc.), augmentation techniques specialized for agricultural environments, such as simulations of water reflection and shadow reflection on the rice paddy surface, can be applied. Furthermore, optical image transformation technologies such as brightness / contrast / saturation adjustment, motion blur, and Gaussian noise can be utilized to respond to changes in light intensity and weather.

[0056] step Key processing details tool or model 1. Establishment of a bird classification system Definition of species labels and establishment of class systems for major bird species (magpie, sparrow, pigeon, crow, brown-eared bulbul, rock pigeon, etc.) Class definition schema (JSON Schema), Pandas / Excel-based meta-definition, Label Studio / Labelbox project settings 2. Establishment of a Multi-Label Annotation System Established a multilabel annotation system that tags behavior (flight, landing, herding, rest) and population size (single, small, medium, large). Label Studio, CVAT, COCO-format JSON, Pascal VOC XML, Python automatic comment script 3. Inspection and Management of Labeled Data annotation quality evaluation and duplicate detection, class imbalance analysis, and inter-reviewer agreement (Cohen's Kappa) check. Pandas, NumPy, Scikit-learn, DVC (Data Version Control), MLflow data version tracking 4. Environmentally Specialized Augmentation (Agricultural Zone Reflection / Shadow) Simulation-type image augmentation was performed to reflect the actual characteristics of agricultural zones by simulating water surface reflections, shadows, and humidity effects. OpenCV, Albumentations(Random Shadow, Reflection Simulation), PyTorch Image Augmentation Pipeline 5. Optical Image Transformation Enhancement Brightness / contrast / saturation adjustments, motion blur, Gaussian noise, etc. applied to respond to changes in illumination and weather. OpenCV, Imgaug, Albumentations (RandomBrightnessContrast, MotionBlur, GaussianNoise) 6. Augmented Data Validation and Dataset Integration Integrates original and augmented images by class and automatically splits into training / validation / test sets (8:1:1). PyTorch Dataset / DataLoader, TensorFlow tf.data, Weights & Biases (W&B) Data Quality Dashboard 7. Construction of final training set and format conversion Convert to formats such as COCO / YOLO / TFRecord for model training FiftyOne, Roboflow, YOLOv5 Dataset Converter, TensorFlow TFRecord Generator

[0057] The automatic output control unit (350) can adjust the laser output of the laser repellent (500) to within a preset output range according to the distance between the laser repellent (500) and harmful birds and the illuminance of the rice cultivation field, while also adjusting the laser output to satisfy the upper limit of the output based on MPE (Maximum Permissible Exposure) according to the laser grade and exposure time. First, the method of automatically adjusting the output (P) in the range of 0.5 to 5 mW according to ① distance (d) and illuminance (Ev) can be designed to ② maintain a safe upper limit (laser grade · MPE) and ③ obtain sufficient contrast with minimum output. This can be implemented with [feedback control + safety cap + adaptive logic].

[0058] Key Idea target Calculate the required spot irradiance (E_req) by considering distance (d) and beam spread (θ) to achieve only the minimum contrast (C*) that birds can easily perceive at ambient brightness (Ev), and command the minimum power (P) that satisfies this. safety limit Apply output upper limit P_cap(d,t) based on MPE (Maximum Permissible Exposure) according to laser class (Class 2 / 3R) and exposure time. Closed-loop correction Fine-tuning is performed using PID + slew rate limiting based on actual brightness feedback (y) received from the camera and photodiode. Adaptive Savings Once the eradication effect is confirmed, a down-ramp policy is applied to gradually reduce output.

[0059] The variables and parameters to be used below are defined first as shown in Table 7 below.

[0060] Variable and Parameter Table sign definition d Distance to target [m] Ev Ambient illuminance [lux] C* Dimensionless contrast (Weber contrast, etc.) ρ_eff Dimensionless effective surface reflectance (cultivated land / rice leaves, etc.) θ Laser half-angle divergence [rad] a0 Exit beam radius (based on 1 / e²) [m] r(d) Spot radius r(d) at distance d = a0 + d · tan(θ / 2) [m] A(d) Spot area A(d)=π·r(d)² [m²] E_req Target Spot Irradiance [W / m²] P Laser Light Output Command [W] (Operates in range of 0.5~5 mW) P_cap(d,t) MPE · Safe upper limit output reflecting hardware constraints [W] y Sensor feedback (camera photometry / photodiode) e Control error e = E_req - y u Controller output (current / duty)

[0061] <Open-loop Calculation (Initial Output Estimation)>

[0062] To calculate the open loop to estimate the initial output, the required illumination level must first be determined. To create the minimum contrast C* at the ambient illumination Ev, a simple approximation Ereq ≈ α·C*·Ev / ρeff can be used. Here, α is the camera-visual correction constant (determined by calibration). Thus, to calculate the initial output, to create the required illumination level at the spot area A(d), P0 = Ereq · A(d) can be used. Since r(d) = a0 + d · tan(θ / 2), as d increases, A(d) increases, and P tends to increase.

[0063] <Safety Limit (Hard Cap) Applied>

[0064] Next, the MPE-based upper limit, the grade limit, and the HW (hardware, laser repellent (500)) limit can be combined to define a safety upper limit as shown in Equation 4 below.

[0065]

[0066] t is the expected exposure time (scan dwell time), and k is a conservative factor (<1) to provide an additional safety margin. Here, dwell time refers to the time the laser stays at a single spot. MPEtoPower is the upper limit of output calculated by converting the standard allowable irradiation limit [MPE] into a spot area A(d), and can be set more conservatively to account for reflection / diffuse reflection scenarios. Accordingly, the final open-loop command becomes Popen=clip(P0, Pmin, Pcap).

[0067] Closed-loop (feedback) fine-tuning

[0068] Regarding sensor feedback, the linear luminous intensity (or HDR restored value) of the ROI (Region of Interest) of the camera (400) can be inversely converted into Lux / Irradiance (pre-calibrated) and used as y. Alternatively, the spot irradiation amount can be directly measured by a photodiode (receiver) to synthesize y (sensor fusion). Here, the feedback control of the laser repellent (500) can be performed as shown in Equation 5 below. That is, the error is fed into a PID controller to operate in a way that minimizes the error.

[0069]

[0070] At this time, slew rate limiting (dP / dt ≤ S_max) and anti-windup can be applied. The output command can be P=clip(Popen+u, Pmin, Pcap). Here, slew rate limiting is a technique that limits the rate of change so that the control output does not change too rapidly over time. It reduces spikes, flickering, and vibration, and stabilizes the transient response of the drive system (motor / laser driver). S_max is the maximum allowable rate of change per unit time (e.g., W / s), and P(t) is the control output at t (e.g., laser output, W). Pmin and Pmax are hard / safe limits (e.g., 0.5–5 mW), and clip(·) is an operation that cuts to the lower and upper limits.

[0071] Anti-windup is a technique that prevents and resolves the windup phenomenon in PID control, where the integral term accumulates excessively due to saturation. It suppresses or reverses the integral in the saturation range to prevent a large overshoot upon return. Among the representative methods, clamping temporarily stops the accumulation of the integral when the actuator becomes saturated, and back-calculation performs the role of returning the difference between the output cut off by saturation and the original output to the integrator as a correction feedback. Accordingly, by limiting the slew rate, sudden changes in output are prevented to reduce laser flicker, current spikes, and motor vibrations, and by using anti-windup, the integral is not accumulated when the laser suppressor (500) is at the upper limit (e.g., 5 mW), thereby preventing excessive overshoot at the moment of release.

[0072] Situational Adaptation Logic Weekly High Light (Ev↑) Even if C* is the same, E_req↑ → P increases, but is safely limited by P_cap. Night / Low Light (Ev↓) Even with the same C*, E_req↓ → P automatically decreases, operating near minimum output (0.5 mW). Down ramp upon confirmation of effect When current dispersion is detected, P is gradually lowered to fix at the minimum holding output. Uncertainty correction If y variability increases due to wind / fog, etc., the integral gain is reduced and stabilized using a moving average. pseudocode loop at 100 Hz: d ← range_sensor() # ToF / LiDAR / stereo / monocular+SfM Ev ← ambient_illuminance() # Multispec / illuminance sensor / camera HDR r ← a0 + d * tan(θ / 2) A ← π * r^2 E_req ← α * C* * Ev / ρ_eff P0 ← E_req * A P_cap ← min(P_HW_max, κ * MPE_to_Power(d, t_dwell)) P_open← clip(P0, P_min, P_cap) y ← fuse( photodiode_irradiance(), camera_photometry_ROI() ) e ← E_req - yu ← PID(e) with anti_windup, slew_limit P_cmd ← clip(P_open + u, P_min, P_cap) driver.set_power(P_cmd) # Apply current / duty safety_interlocks.enforce() # Immediately cut off power upon detection of human / eye / restricted area log(d, Ev, P_cmd, y, alarms)end

[0073] When the output of such a laser is controlled, it does not burn or injure birds such as sparrows. One embodiment of the present invention is designed for non-lethal and non-contact deterrence, in which the laser is operated only within a safe range at Class 2 (≤1 mW) or Class 3R (≤5 mW) levels, and includes automatic shut-off (interlock) in abnormal conditions and restrictions on use in the direction of people / livestock. Accordingly, the irradiation effect of the laser is an avoidance reaction caused by temporary glare and startle, and does not assume thermal damage such as burns or combustion.

[0074] Complex Eradication Mechanism

[0075] Additionally, the AI ​​rice farming harmful bird monitoring and laser repellent integrated system (1) may include a variable ultrasonic transmitter that randomly emits ultrasonic waves of 15 kHz to 25 kHz when harmful birds are detected, a high-brightness LED strobe emitter that induces momentary glare using preset light intensity and pulses, and a laser multi-beam diffusion module that provides visual stimulation to cover the cultivated land by generating multiple beams or light patterns.

[0076] 15~25kHz variable ultrasonic transmitter definition Variable Ultrasonic Transmitter (15–25 kHz): Refers to an active speaker module that emits signals with variable frequency, amplitude, and waveform in the high-frequency band to which birds are sensitive. Corresponds to the sound stimulation component within the "Audiovisual Simultaneous Stimulation Function" of the document. Operation process 1. Contextual Awareness Identifies species, distance, and cluster size using an object detection model (e.g., YOLOv5 / Faster R-CNN). 2. Select Profile Select the frequency sweep range (e.g., 17→23 kHz), duty cycle, and waveform (sine / pulse / pulse train) according to the length, distance, and ambient noise level. 3. Apply Randomization To prevent pattern learning, the frequency, duty cycle, burst length, and interval are randomized non-periodically. 4. Output Control Limits excessive sound pressure with a real-time noise leveler (AGC) and corrects the output with a distance-based attenuation table. 5. Effect Determination Check for dispersion using camera / radar, and if the effect is confirmed, lower the intensity or switch modes. tool or model hardware Ultrasonic transducer (piezo / horn rod), Class D amplifier, microphone array (noise feedback), waterproof housing (IP rating) Firmware / Algorithm PWM / timer-based waveform synthesis, burst scheduler, frequency sweep / hop logic, AGC software Edge Control (ROS 2 / FastAPI gRPC), Rule Engine (Large / Distance → Profile Mapping), Remote Parameter Tuning Dashboard

[0077] High-brightness LED strobe light definition High-intensity LED Strobe: Refers to a light stimulation module that uses high luminous intensity and short pulses to induce momentary glare and visual startle. As the optical stimulation element of "simultaneous audiovisual stimulation," it provides a diffused lighting effect over a wide angle, independent of lasers. Operation process 1. Contextual Awareness Receives detection distance, angle, and cluster density as input. 2. Pulse Design Combining pulse width (e.g., 5–30 ms), emission period (e.g., 5–15 Hz), and multiple channels (white + green / blue), the amount of light is increased or decreased within limits according to ambient light. 3. Apply Randomization Randomizing pulse intervals, channel order, and flash patterns reduces predictability. 4. Output Control Current peak limiting and thermal management (temperature sensor-based derate) are applied to the LED driver. 5. Effect Determination If evasive movement is detected via camera motion analysis, the intensity is lowered, and if the effect is minimal, the pattern is changed. tool or model hardware High-power LED package (optical lens assembly), constant-current LED driver, temperature sensor, diffuse / focus interchangeable optics Firmware / Algorithm High-side current control, pulse sequencer, thermal dirate (SoA curve based) software Pattern preset / random seed management, Event Bus integration, Dashboard (real-time light intensity / temperature monitoring)

[0078] Laser Multibeam Diffusion Module definition Laser Multibeam Diffusion Module: Refers to a laser optical module that provides visual stimulation over a wide area by generating multiple beams or diffused light patterns (lines, gratings, tori, etc.). As the laser element of "simultaneous audiovisual stimulation," it enhances spatial coverage through multibeam and pattern diffusion. Operation process 1. Contextual Awareness Receives target location, density, and direction of movement as input. 2. Select Pattern Select from Single Spot (Precision Aiming) ↔ Multibeam (Area Control) ↔ Pattern Scan (Line / Grid). 3. Output and Scan Control Apply distance and illumination-based automatic output caps (e.g., Class 2 / 3R upper limits) and scan dwell time limits (based on the document's laser safety and output limits and control and aiming system). 4. Apply Randomization Random elements are added to the scan path, pattern size, and rotation speed to make path prediction difficult. 5. Effect Determination Once dispersion is confirmed, output is gradually reduced (energy saving), and combined with ultrasound and strobe if necessary. tool or model hardware Class 2 / 3R diodes (≤5 mW), DOE (Diffractive Optical Element) or galvano / mirror array, pan / tilt drive, safety interlock Firmware / Algorithm Scan path generation (random walk / lustre / Lissajous), dwell management, output cap (grade / distance / illumination) software Target Priority / Pattern Scheduler, Safe Zone Masking, Real-time Log / Replay

[0079] To prevent tolerance, non-periodic and random operation logic is required; by aperiodicly changing the frequency / pulse / pattern / output / scan paths of ultrasound, LED, and laser on the time axis, a control strategy can be implemented to prevent birds from becoming accustomed to or ignoring specific patterns.

[0080] Non-periodic and random operation logic (anti-immunity) 1. Situation Assessment Evaluated species, community, residence time, and environment (illumination, wind speed, rainfall). 2. Tactical Decision Select single stimulus ↔ dual (sound + light) ↔ triple (sound + light + laser) stimulus 3. Injecting Randomization Parameters Frequency, duty cycle, pulse width, flash sequence, scan path, pattern size, and output level are updated frame by frame using a random number seed. 4. Effectiveness Feedback Calculate resistance indicators using dispersion rate, re-access rate, and repeated exposure history, and update the scheduler to avoid repeating the same tactics. 5. Safety and Energy Optimization Automatic output control and temperature / power management are implemented with the goal of providing minimum sufficient stimulation within safety limits. tool or model Policy / Scheduler Priority queue-based multimodal tactical scheduler, random number seed management, reproducible experiment logs Sensor / Feedback Camera (detection / tracking), light sensor, microphone array, IMU / wind speed input service Event-driven architecture (gRPC / MQTT), real-time dashboard (WebRTC / HLS), OTA parameter distribution

[0081] Animal Stress Reduction Logic

[0082] If there are livestock farms nearby, such as those raising cattle, pigs, and chickens, the animals may be unable to sleep properly at night and experience stress. This can lead to reduced birth rates, decreased egg production, and weakened immunity. Accordingly, a usage range limitation mechanism can be added to automatically block the laser in the direction where humans or livestock are detected. Additionally, geofences can be used to prevent the laser from shining in the direction of the livestock barn or to adjust the distance so that it does not reach the barn. Furthermore, a shield can be installed on the side of the barn to prevent the laser from causing glare.

[0083] In addition, in the case of a combined pest control device, the intensity and pattern of each stimulus can be variably controlled according to time of day and environment, thereby automatically attenuating sound and light (Soft Mode) to reduce startling and sleep disturbance in livestock.

[0084] Additional easing measures mitigation factors Implementation method Direction control Designate the corresponding direction angle as a blind zone upon livestock detection. Frequency filtering Avoidance of livestock hearing sensitivity frequencies (e.g., cattle 10–15 kHz) when transmitting ultrasound Night light mode Automatic reduction of LED strobe and laser output upon low-light sensor detection Environmental learning feedback When a reaction (sound or movement) is detected within a certain area, the stimulation intensity is automatically reduced.

[0085] Integrated Management Dashboard

[0086] In one embodiment of the present invention, a dashboard such as FIG. 4h is constructed for real-time monitoring and visualization. It is a multi-view dashboard that provides user-customized screens such as map-based robot pest control locations and statistical summaries. It is a real-time notification feed that displays major events in real time, such as bird appearances, pest control operations, and system status. It is an adaptive UI that provides a responsive interface optimized for various devices, such as PCs, tablets, and mobile phones. It is also a custom widget that allows for customized output of elements provided to each user. Additionally, it is possible to provide real-time video verification of the field camera and highlighting of detected objects through an AI camera live view. Furthermore, remote control is possible. It is a WEB-based responsive application that can be used on both PCs and mobile phones. It is possible to respond immediately to the appearance of harmful birds by using a push notification system that provides differentiated notifications based on importance. It is also possible to directly control the laser pest control device (500) on a mobile phone using a remote control function.

[0087] Hereinafter, the operation process according to the configuration of the integrated control server of FIG. 2 described above will be explained in detail with reference to FIG. 3 and FIG. 4. However, it is obvious that the embodiment is merely one of the various embodiments of the present invention and is not limited thereto.

[0088] Referring to FIG. 3, (a) the integrated control server (300) can build a model of [YOLO + Faster R-CNN + time series pattern analysis] for modeling an object detection model, collect a dataset for this purpose and label it to classify harmful birds by species, size, and behavioral patterns, and augment the data by simulating and applying changes according to the agricultural environment, light intensity, and weather, thereby enabling the identification and extraction of harmful birds in various situations. Then, the integrated control server (300) detects objects in real time and irradiates them with a laser as in (b), and can perform complex eradication using various methods as in (c).

[0089] One embodiment of the present invention implements a system as shown in FIG. 4b to solve the problem as shown in FIG. 4a, and can utilize a Mecanum wheel with a four-wheel array as shown in FIG. 4c as a means of locomotion for an autonomous robot. One embodiment of the present invention can implement an integrated system as shown in FIG. 4d. To this end, data can be collected as shown in FIG. 4e, and modeled by learning, verifying, and testing an object detection model by labeling and augmenting as shown in FIG. 4f. A composite pest control device can also be used as shown in FIG. 4g, and a dashboard including widgets can be provided to a user terminal (100) as shown in FIG. 4h, and a control interface can be provided to control a camera (400) or a laser pest control device (500) from the user terminal (100). A system according to one embodiment of the present invention can be manufactured as shown in FIG. 4i, but is not limited thereto.

[0090] As for the details regarding the AI ​​rice farming harmful bird monitoring and laser eradication solution provision method shown in FIGS. 2 to 4 that are not described, they are identical to or can be easily inferred from the details described above regarding the AI ​​rice farming harmful bird monitoring and laser eradication solution provision method shown in FIG. 1, so further explanation will be omitted.

[0091] FIG. 5 is a diagram illustrating the process of transmitting and receiving data between each component included in the AI ​​rice farming harmful bird monitoring and laser eradication integrated system of FIG. 1 according to an embodiment of the present invention. Hereinafter, an example of the process of transmitting and receiving data between each component will be described through FIG. 5, but the present invention is not to be interpreted as being limited to such an embodiment, and it is obvious to those skilled in the art that the process of transmitting and receiving data illustrated in FIG. 5 may be changed according to various embodiments described above.

[0092] Referring to FIG. 5, the integrated control server receives an image of a rice cultivation field from a camera (S5100) and detects harmful birds within the rice cultivation field using a pre-established object detection model (S5200). Additionally, the integrated control server controls a laser repellent to irradiate a laser toward the harmful birds (S5300).

[0093] The order of the steps described above (S5100~S5300) is merely an example and is not limited thereto. That is, the order of the steps described above (S5100~S5300) may vary, and some of these steps may be executed simultaneously or deleted.

[0094] As for the details regarding the AI ​​rice farming harmful bird monitoring and laser eradication solution provided in Fig. 5 that are not described, they are identical to or can be easily inferred from the details described above regarding the AI ​​rice farming harmful bird monitoring and laser eradication solution provided in Figs. 1 to 4, so further explanation will be omitted.

[0095] The method for providing an AI rice farming harmful bird monitoring and laser repellent solution according to one embodiment described through FIG. 5 may also be implemented in the form of a recording medium containing computer-executable instructions, such as an application or program module executed by a computer. A computer-readable medium may be any available medium accessible by a computer and includes both volatile and non-volatile media, as well as removable and non-removable media. Additionally, a computer-readable medium may include all computer storage media. A computer storage medium includes both volatile and non-volatile, removable and non-removable media implemented by any method or technique for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0096] The method for providing an AI rice farming harmful bird monitoring and laser eradication solution according to one embodiment of the present invention described above may be executed by an application basically installed on a terminal (which may include a program included in a platform or operating system, etc., basically installed on the terminal), or by an application (i.e., a program) directly installed by a user on a master terminal through an application providing server, such as an application store server, an application, or a web server related to the service. In this sense, the method for providing an AI rice farming harmful bird monitoring and laser eradication solution according to one embodiment of the present invention described above may be implemented as an application (i.e., a program) that is basically installed on a terminal or directly installed by a user, and may be recorded on a computer-readable recording medium such as a terminal.

[0097] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0098] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention.

Claims

Claim 1 A camera for photographing a rice cultivation field; a laser repellent for irradiating a laser toward harmful birds within the rice cultivation field; a variable ultrasonic transmitter that randomly emits ultrasonic waves of 15 kHz to 25 kHz when harmful birds are detected; a high-intensity LED strobe emitter that induces momentary glare using preset light intensity and pulses; and a laser multi-beam diffusion module that generates multiple beams or light patterns to provide visual stimulation to cover the cultivation field. and a receiving unit that receives an image of a rice cultivation field from the camera; a detection unit that detects harmful birds within the rice cultivation field using a pre-established object detection model; a control unit that controls the laser repellent to irradiate a laser toward the harmful birds; and a bird detection optimization unit that, when modeling the object detection model, detects the harmful birds using YOLO (You Only Look Once) during the day, and uses Faster R-CNN (Faster Regional Convolutional Neural Network) in environments where the illuminance is lower than a preset threshold value, such as at night or during rainfall, and reduces the false detection rate by analyzing the behavioral patterns of the harmful birds, such as shape changes and movement directions between consecutive frames of the image, using Temporal Pattern Analysis. The integrated control server includes an automatic output control unit that adjusts the laser output of the laser repellent to within a preset output range according to the distance between the laser repellent and the harmful bird and the illuminance of the rice cultivation field, while adjusting the laser output to satisfy an upper limit based on MPE (Maximum Permissible Exposure) according to the laser grade and exposure time; wherein the laser repellent is connected to an autonomous driving robot equipped with a 4-wheel array of Mecanum wheels to move within the rice cultivation field, and the bird detection optimization unit, in order to collect data for modeling,Performing a Step 1 process corresponding to establishing a distributed image acquisition infrastructure, a Step 2 process corresponding to color space conversion and automatic ROI detection, a Step 3 process corresponding to frame selection, a Step 4 process corresponding to morphological operations and noise removal, a Step 5 process corresponding to 24-hour time-lapse data collection, and a Step 6 process corresponding to data management and synchronization, wherein in the above Step 1 process, 4K high-resolution cameras are installed at multiple points within the rice cultivation field, and an infrastructure is established by collecting image data corresponding to bird images in a distributed manner in real time using 4K high-resolution cameras and NVIDIA Jetson Xavier NX-based edge computing devices, wherein in the above Step 2 process, the collected image data is converted from RGB to the HSV color space to separate the color patterns of birds, and a Region of Interest (ROI) is automatically detected through HSV range designation based on the color distribution by species, wherein in the above Step 3 process, only frames containing birds corresponding to the detected ROI are selected and stored, while frames in which birds are not present are removed, and in the above Step 4 process, erosion and dilation, By applying morphological operations including openness and closeness, processing is performed to remove background noise that may occur during shooting and to enhance the contours of birds; in the above 5th step, a 24-hour time-lapse video recording technique is applied to secure data under multiple light conditions including sunrise, noon, sunset, and night, thereby enabling the acquisition of light condition-based data that reflects lighting changes by time period corresponding to the multiple light conditions; in the above 6th step, time synchronization between edge terminals and storage of image metadata ensure spatiotemporal consistency of the training dataset; and to secure image data for harmful birds by environment, the process comprises: Step 1, corresponding to data collection by time period; Step 2, corresponding to data acquisition by weather condition; Step 3, corresponding to data collection by growth stage; and Step 4, corresponding to securing spatial diversity.Utilizing a process comprising Step 5, corresponding to data integration and preprocessing, and Step 6, corresponding to the construction and verification of a training dataset, wherein in Step 1, image data is collected by automatically adjusting light conditions, including solar altitude, color temperature, and exposure, for each time period including dawn, noon, and night, and training data reflecting behavioral differences of birds by time period is secured; in Step 2, images of multiple climatic environments, including clear, cloudy, foggy, rainy, and snowy conditions, are secured to construct a dataset for training a real-world adaptive model; in Step 3, bird appearance patterns based on changes in crop height, color, and reflectance are learned by reflecting the conditions of paddy fields at each stage, including planting, growing season, harvesting season, and fallow season; in Step 4, stereoscopic data is collected from multiple angles and distances using a mobile robot and a movable pole camera system to compensate for the limitations of fixed cameras; in Step 5, time, weather, growth, and spatial information are merged into metadata to construct an integrated training set, followed by the removal of duplicates and noise frames, and normalization is performed; and in Step 6, the collected data Sampling is performed, and a validation set is automatically separated prior to model training. Subsequently, for the labeling and augmentation of the collected data, the following steps are performed: Step 1, corresponding to establishing a bird classification system; Step 2, corresponding to constructing a multi-label annotation system; Step 3, corresponding to verifying and managing labeled data; Step 4, corresponding to environment-specific augmentation; Step 5, corresponding to optical image transformation augmentation; Step 6, corresponding to augmented data verification and dataset integration; and Step 7, corresponding to constructing a final training set and format conversion. In Step 1, species-specific label definitions and class systems are constructed for multiple bird species, including magpies, sparrows, pigeons, crows, brown-eared bulbuls, and rock pigeons. In Step 2, behaviors including flight, landing, foraging, and resting, and solitary, small, medium,Establishing an annotation system for multilabel parallel tagging of object numbers including large numbers, and in the third step, performing annotation quality evaluation, duplicate detection, class imbalance analysis, and inter-verifier agreement checks; in the fourth step, performing simulation-type image augmentation that reflects the specific characteristics of actual agricultural zones by simulating the effects of water surface reflection, shadows, and humidity in rice paddies; in the fifth step, applying optical image transformation techniques including adjustments to brightness, contrast, and saturation, motion blur, and Gaussian noise to respond to changes in illumination and weather; in the sixth step, integrating the original and augmented images by class, and automatically splitting the training, validation, and test sets in an 8:1:1 ratio; in the seventh step, converting to COCO, YOLO, and TFRecord formats suitable for model training; subsequently, constructing the object detection model, which is a hybrid model utilizing YOLO and Faster R-CNN, and selectively using the object detection model for real-time object detection; and the automatic output control unit, which is for birds to perceive in ambient brightness An integrated system for monitoring harmful algae in rice farming and laser eradication, which calculates the required target spot irradiance by considering distance and beam spread to achieve minimum contrast, commands a minimum power satisfying the calculated target spot irradiance in the range of 0.5 mW to 5 mW, regulates the laser output by applying an output upper limit based on MPE (Maximum Permissible Exposure) according to laser class and exposure time, performs closed-loop correction by receiving actual brightness feedback from a camera and photodiode and fine-tuning with PID and slew rate limits, and regulates the laser output by applying a down-ramp policy of adaptive reduction that gradually lowers the output when the eradication effect is confirmed. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 delete

Images