A pesticide variable spraying system and method

Abstract

This application discloses a variable-rate pesticide spraying system and method, relating to the field of intelligent plant protection technology. The system includes: an unmanned aerial vehicle (UAV)-borne multispectral imager, a ground control center, and a variable-rate spraying device. The UAV-borne multispectral imager acquires remote sensing image data of the area to be sprayed. The ground control center uses a segmentation model to segment trees in the area to be sprayed based on the remote sensing image data, obtaining segmented images. It determines the ground sampling distance of the UAV-borne multispectral imager. For any tree, its actual geographical location is determined based on the bounding box, the width and height of the segmented image, the ground sampling distance, longitude, latitude, and the latitude radian value of the center pixel of the segmented image. The pesticide application rate is determined based on the mask, the ground sampling distance, red band image data, and near-infrared band image data. The variable-rate spraying device sprays pesticide onto the corresponding trees based on their actual geographical locations and the pesticide application rate. This application achieves precise variable-rate pesticide spraying.

Description

Technical Field

[0001] This application relates to the field of intelligent plant protection technology, and in particular to a variable-rate pesticide spraying system and method. Background Technology

[0002] In modern agriculture, plant protection operations in orchards play a crucial role in the control of pests and diseases. Traditional plant protection methods suffer from problems such as large pesticide usage, uneven application, and low efficiency, leading to resource waste and environmental pollution. Particularly during pesticide spraying, traditional methods cannot precisely target the specific needs of crops, often resulting in uneven spraying or excessive pesticide use, leading to waste of pesticide solutions and environmental pollution. Harmful chemicals in pesticides not only toxicize crops but also enter the air and water bodies through evaporation and runoff, harming the ecological environment and human health.

[0003] Therefore, the efficiency and accuracy of pesticide spraying have become an urgent agricultural technology problem to be solved. Currently, the main solutions to the problems of uneven pesticide spraying, overuse, and low spraying efficiency include GPS navigation-assisted spraying systems and drone spraying systems. However, these methods have drawbacks such as limited coverage and insufficient real-time response capabilities, and cannot achieve efficient and precise spraying in large-scale orchards.

[0004] Chinese patent document CN116142462A discloses a variable-rate pesticide spraying drone, comprising the following steps: Step 1, several first spray pipes are rotatably mounted at the lower ends of several liquid guiding pipes; Step 2, several second spray pipes are fixedly connected to the bottom of the pesticide tank; Step 3, two landing bars are located on the front and rear sides of the pesticide tank, respectively, and the landing bars are connected to the lower end of the drone shell through two drive rod assemblies; Step 4, several connecting ropes slide through positioning rings, and the two ends of the connecting ropes are fixedly connected to the first spray pipes and drive rod assemblies at corresponding positions, and the drive rod assemblies are used to pull the connecting ropes when the landing bars contact the ground, so that the connecting ropes pull the first spray pipes to rotate; Step 5, when the landing bars leave the ground, the first spray pipes rotate to a horizontal position under their own gravity and pull the connecting ropes.

[0005] Among the variable-rate pesticide spraying drone methods described above, variable-rate spraying by a single drone involves high requirements for variables, necessitating precise control of the spraying amount and range. This design requires a trade-off between spraying effectiveness and structural complexity. Furthermore, while the landing stick and connecting rope design allows for the rotation of the spray nozzle, it lacks fine-grained control over variable-rate spraying in actual operation, making it difficult to adjust spraying parameters according to different crops and environmental conditions, thus failing to achieve truly precise variable-rate spraying. These limitations increase system complexity, make it prone to mechanical failures, and increase maintenance difficulty, affecting the overall reliability and operational stability of the drone. Summary of the Invention

[0006] The purpose of this application is to provide a pesticide variable spraying system and method to solve the problem that related technologies cannot achieve precise variable spraying of pesticides.

[0007] To achieve the above objectives, this application provides the following solution:

[0008] In a first aspect, this application provides a pesticide variable spraying system, comprising: an unmanned aerial vehicle (UAV)-borne multispectral imager, a ground control center, and a variable spraying device; the UAV-borne multispectral imager and the variable spraying device are wirelessly connected to the ground control center, respectively.

[0009] The UAV-borne multispectral imager is used to acquire remote sensing image data of the area to be sprayed; the remote sensing image data includes: green band image data, red band image data, red edge band image data, and near-infrared band image data.

[0010] The ground control center is used for:

[0011] Using a segmentation model, the trees in the area to be sprayed are segmented based on the remote sensing image data to obtain a segmented image; the segmented image is an image containing the mask and bounding box of each tree; the segmentation model is obtained by training an improved Yolov8 neural network;

[0012] The imaging parameters and position parameters of the UAV-borne multispectral imager are obtained; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude and relative altitude to the ground;

[0013] Based on the imaging parameters and the relative altitude, the ground sampling distance of the UAV-borne multispectral imager is determined;

[0014] Select any tree as the current tree;

[0015] Based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, determine the pixel position offset of the current tree;

[0016] The geographic location offset of the current tree is determined based on the pixel position offset of the current tree and the ground sampling distance;

[0017] The actual geographical location of the current tree is determined based on the current tree's geographical location offset, the longitude, the latitude, and the latitude radian value of the center pixel of the segmented image;

[0018] The canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance;

[0019] Based on the current canopy area of ​​the trees, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the trees.

[0020] The variable spraying device is used to spray pesticides onto the corresponding trees based on their actual geographical location and the amount of pesticide to be sprayed.

[0021] Optionally, the wavelength range of the green band is 560nm±16nm, the wavelength range of the red band is 650nm±16nm, the wavelength range of the red edge band is 730nm±16nm, and the wavelength range of the near-infrared band is 860nm±26nm.

[0022] Optionally, the process of determining the segmentation model includes:

[0023] Obtain a training set; the training set includes: remote sensing image data of multiple training areas and corresponding segmented images;

[0024] An improved Yolov8 neural network is constructed. The improved Yolov8 neural network includes a backbone network, a feature fusion network, and a detection network. The backbone network includes 5 GhostConv modules, 2 RepVGG modules, 4 C2f modules, and 1 SPPF module. The feature fusion network includes 6 feature concatenation modules, 3 upsampling modules, 6 C2f modules, and 3 GhostConv modules. The detection network includes 4 detection heads.

[0025] Using remote sensing image data of each training area as input and the corresponding segmented image as output, the improved Yolov8 neural network is trained to obtain the segmentation model.

[0026] Optionally, determining the ground sampling distance of the UAV-borne multispectral imager based on the imaging parameters and the relative altitude includes:

[0027] The ground sampling distance is calculated using the ground sampling distance calculation formula, based on the imaging parameters and the relative height; the ground sampling distance calculation formula includes:

[0028]

[0029] Where H is the ground sampling distance; h1 is the relative height; s is the pixel size; and f is the true focal length.

[0030] Optionally, the pixel position offset of the current tree is determined based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, including:

[0031] Using the pixel position offset calculation formula, the pixel position offset of the current tree is calculated based on the bounding box of the current tree in the segmented image and the width and height of the segmented image; the pixel position offset calculation formula includes:

[0032]

[0033] Among them, P x N represents the x-axis component of the pixel position offset. x is the x-axis component of the center pixel position of the bounding box; w is the width of the segmented image; P y The y-axis component represents the pixel position offset; h represents the height of the segmented image; N y This is the y-axis component of the position of the center pixel of the bounding box.

[0034] Optionally, the geographic location offset of the current tree is determined based on the pixel location offset of the current tree and the ground sampling distance, including:

[0035] Using the geographic location offset calculation formula, the geographic location offset of the current tree is calculated based on the pixel position offset of the current tree and the ground sampling distance; the geographic location offset calculation formula includes:

[0036] G x =P x ·H;

[0037] G y =P y ·H;

[0038] Among them, G x The x-axis component of the geographic location offset; G y The y-axis component represents the geographic location offset.

[0039] Optionally, the actual geographical location of the current tree is determined based on the current tree's geographical location offset, the longitude, the latitude, and the latitude radian value of the center pixel of the segmented image, including:

[0040] Using the actual geographic location calculation formula, the actual geographic location of the current tree is calculated based on its geographic location offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image; the actual geographic location calculation formula includes:

[0041]

[0042] Where α2 is the longitude in the actual geographical location; α1 is the longitude in the location parameters; γ is the latitude radian value of the center pixel of the segmented image; β2 is the latitude in the actual geographical location; and β1 is the latitude in the location parameters.

[0043] Optionally, the canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance, including:

[0044] Using the canopy area calculation formula, the canopy area of ​​the current tree is calculated based on the mask of the current tree in the segmented image and the ground sampling distance; the canopy area calculation formula includes:

[0045] A = M·H 2 ;

[0046] Where A is the canopy area; M is the total number of pixels in the tree's mask.

[0047] Optionally, based on the current canopy area of ​​the tree, red band image data, and near-infrared band image data, the current amount of pesticide to be sprayed on the tree is determined, including:

[0048] The amount of pesticide to be sprayed is calculated using a formula based on the tree's canopy area, red band image data, and near-infrared band image data. The formula includes:

[0049]

[0050] Where Q is the amount of pesticide sprayed; k is the correction coefficient; NDVI is the normalized vegetation index; NIR is the image data in the near-infrared band; and R is the image data in the red band.

[0051] Secondly, this application provides a method for variable-rate pesticide application, which is implemented using a variable-rate pesticide application system, and includes:

[0052] Remote sensing image data of the area to be sprayed is acquired by an unmanned aerial vehicle (UAV) equipped with a multispectral imager. The remote sensing image data includes: green band image data, red band image data, red edge band image data, and near-infrared band image data.

[0053] Using a segmentation model, the trees in the area to be sprayed are segmented based on the remote sensing image data to obtain a segmented image; the segmented image is an image containing the mask and bounding box of each tree; the segmentation model is obtained by training an improved Yolov8 neural network;

[0054] The imaging parameters and position parameters of the UAV-borne multispectral imager are obtained; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude and relative altitude to the ground;

[0055] Based on the imaging parameters and the relative altitude, the ground sampling distance of the UAV-borne multispectral imager is determined;

[0056] Select any tree as the current tree;

[0057] Based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, determine the pixel position offset of the bounding box of the current tree;

[0058] The geographic location offset of the current tree is determined based on the pixel position offset of the current tree's bounding box and the ground sampling distance;

[0059] The actual geographical location of the current tree is determined based on the current tree's geographical location offset, the longitude, the latitude, and the latitude radian value of the center pixel of the segmented image;

[0060] The canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance;

[0061] Based on the current canopy area of ​​the trees, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the trees.

[0062] Pesticides are sprayed onto the corresponding trees based on their actual geographical location and the amount of pesticide to be applied.

[0063] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0064] This application discloses a variable-rate pesticide spraying system and method. The system includes: an unmanned aerial vehicle (UAV)-borne multispectral imager, a ground control center, and a variable-rate spraying device. The UAV-borne multispectral imager is used to acquire remote sensing image data of the area to be sprayed. The ground control center is used to: segment trees in the area to be sprayed using a segmentation model based on the remote sensing image data to obtain segmented images; acquire imaging parameters and position parameters of the UAV-borne multispectral imager; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude, and relative altitude to the ground; determine the ground sampling distance of the UAV-borne multispectral imager based on the imaging parameters and relative altitude; identify any tree as the current tree; and based on the segmentation... The method involves determining the pixel position offset of the current tree based on its bounding box in the image and the width and height of the segmented image; determining the geographic location offset of the current tree based on its pixel position offset and ground sampling distance; determining the actual geographic location of the current tree based on its geographic location offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image; determining the canopy area of ​​the current tree based on the mask of the current tree in the segmented image and ground sampling distance; and determining the pesticide application rate for the current tree based on its canopy area, red band image data, and near-infrared band image data. A variable-rate spraying device is used to spray pesticides onto the corresponding trees based on their actual geographic location and the application rate. This application, based on the collaborative operation of an UAV-borne multispectral imager, a ground control center, and a variable-rate spraying device, enables multiple UAV-borne variable-rate spraying devices to work together, accurately locating trees and calculating pesticide application rates, improving pesticide utilization, reducing environmental pollution, and increasing operational efficiency and coverage. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 This is a schematic diagram of a pesticide variable spraying system provided in an embodiment of this application;

[0067] Figure 2 A schematic diagram of a variable-rate pesticide spraying system architecture;

[0068] Figure 3 To segment the image;

[0069] Figure 4 A schematic diagram of the improved Yolov8 neural network structure;

[0070] Figure 5 A schematic diagram of the spraying process of a variable spraying device. Detailed Implementation

[0071] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] The purpose of this application is to provide a pesticide variable spraying system and method, which aims to achieve precise variable spraying of pesticides.

[0073] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0074] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, the pesticide variable spraying system in this embodiment includes: an unmanned aerial vehicle (UAV)-borne multispectral imager, a ground control center, and a variable spraying device; the UAV-borne multispectral imager and the variable spraying device are wirelessly connected to the ground control center.

[0075] Specifically, the drone-borne multispectral imager uses the DJI M3M multispectral camera, with a maximum flight time of 43 minutes and a maximum range of 32km. The ground control center uses a high-performance computer for real-time data processing and analysis, requiring an i7-7700HQ processor or higher, 4GB or more of RAM, and 50GB or more of hard drive space; the software requirements are a Windows 10 64-bit operating system running Python 3.8 or later.

[0076] The UAV-borne multispectral imager is used to acquire remote sensing image data of the area to be sprayed; the remote sensing image data includes: green band image data, red band image data, red edge band image data, and near-infrared band image data.

[0077] As an optional implementation, the wavelength range of the green band is 560nm±16nm, the wavelength range of the red band is 650nm±16nm, the wavelength range of the red edge band is 730nm±16nm, and the wavelength range of the near-infrared band is 860nm±26nm.

[0078] Ground control center, used for:

[0079] Step 01: Using a segmentation model, segment each tree in the area to be sprayed based on remote sensing image data to obtain a segmented image; the segmented image is an image containing the mask and bounding box of each tree; the segmentation model is obtained by training an improved Yolov8 neural network.

[0080] like Figure 3 As shown, each red rectangle in the segmented image is the bounding box of each tree, and the red covered area within each bounding box is the mask of each tree.

[0081] As an optional implementation method, the process of determining the segmentation model includes:

[0082] Step 011: Obtain the training set; the training set includes: remote sensing image data of multiple training areas and corresponding segmented images.

[0083] Step 012: Construct the improved Yolov8 neural network. (For example...) Figure 4 As shown, the improved Yolov8 neural network includes: a backbone network, a feature fusion network, and a detection network; the backbone network includes: 5 GhostConv modules, 2 RepVGG modules, 4 C2f modules, and 1 SPPF module; the feature fusion network includes: 6 feature splicing modules, 3 upsampling modules, 6 C2f modules, and 3 GhostConv modules; and the detection network includes: 4 detection heads.

[0084] The GhostConv module is designed to reduce the computational cost of convolutional neural networks. It generates multiple virtual feature maps by producing a small number of intrinsic feature maps and applying a simple linear transformation, thus reducing the need for convolutional operations. Specifically, GhostConv first generates a small number of intrinsic feature maps through primary convolutions, then applies linear operations to each intrinsic feature map to generate virtual feature maps, thereby reducing filters and computational cost while maintaining the network's expressive power. The RepVGG module is a VGG-style convolutional network structure with five stages of 3×3 convolutional layers. The first layer of each stage performs downsampling through a convolution with a stride of 2. To reduce computation and parameter count, RepVGG inserts grouped convolutions in some convolutional layers and adjusts the channel width of each stage using scaling factors. The final stage has a larger number of channels to enhance feature representation. Furthermore, this module employs global average pooling and fully connected layers as the head structure for image classification tasks.

[0085] Step 013: Using remote sensing image data of each training area as input and the corresponding segmented image as output, train the improved Yolov8 neural network to obtain the segmentation model.

[0086] Step 02: Obtain the imaging parameters and position parameters of the UAV-borne multispectral imager; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude and relative altitude to the ground.

[0087] Specifically, the imaging parameters and position parameters of the UAV-borne multispectral imager are obtained using EXIF ​​data files that store additional information from remote sensing image data.

[0088] Step 03: Determine the ground sampling distance of the UAV-borne multispectral imager based on imaging parameters and relative altitude.

[0089] As an optional implementation, the ground sampling distance of the UAV-borne multispectral imager is determined based on imaging parameters and relative altitude, including:

[0090] The ground sampling distance is calculated using the ground sampling distance calculation formula, based on imaging parameters and relative altitude. The ground sampling distance calculation formula includes:

[0091]

[0092] Where H is the ground sampling distance; h1 is the relative height; s is the pixel size; and f is the true focal length.

[0093] Step 04: Select any tree as the current tree.

[0094] Step 05: Determine the pixel position offset of the current tree based on the bounding box of the current tree in the segmented image and the width and height of the segmented image.

[0095] As an optional implementation, the pixel position offset of the current tree is determined based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, including:

[0096] Using the pixel position offset calculation formula, the pixel position offset of the current tree is calculated based on the bounding box of the current tree in the segmented image and the width and height of the segmented image. The pixel position offset calculation formula includes:

[0097]

[0098] Among them, P x N represents the x-axis component of the pixel position offset. x is the x-axis component of the center pixel position of the bounding box; w is the width of the segmented image; P y The y-axis component represents the pixel position offset; h represents the height of the segmented image; N y This is the y-axis component of the position of the center pixel of the bounding box.

[0099] Step 06: Determine the geographic location offset of the current tree based on the pixel position offset of the current tree and the ground sampling distance.

[0100] As an optional implementation, the geographic location offset of the current tree is determined based on the pixel position offset of the current tree and the ground sampling distance, including:

[0101] Using the geographic location offset calculation formula, the geographic location offset of the current tree is calculated based on its pixel position offset and the ground sampling distance. The geographic location offset calculation formula includes:

[0102] G x =P x ·H.

[0103] G y =P y ·H.

[0104] Among them, G x The x-axis component of the geographic location offset; G y The y-axis component represents the geographic location offset.

[0105] Step 07: Determine the actual geographical location of the current tree based on its geographical location offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image.

[0106] As an optional implementation, the actual geographical location of the current tree is determined based on its geographical location offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image, including:

[0107] Using the actual geographic location calculation formula, the actual geographic location of the current tree is calculated based on its geographic location offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image. The actual geographic location calculation formula includes:

[0108]

[0109] Where α2 is the longitude in the actual geographical location; α1 is the longitude in the location parameters; γ is the latitude radian value of the center pixel of the segmented image; β2 is the latitude in the actual geographical location; and β1 is the latitude in the location parameters.

[0110] Step 08: Determine the canopy area of ​​the current tree based on the mask of the current tree in the segmented image and the ground sampling distance.

[0111] As an optional implementation, the canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance, including:

[0112] Using the canopy area calculation formula, the canopy area of ​​the current tree is calculated based on the mask of the current tree in the segmented image and the ground sampling distance. The canopy area calculation formula includes:

[0113] A = M·H 2 .

[0114] Where A is the canopy area; M is the total number of pixels in the tree's mask.

[0115] Step 09: Based on the current tree canopy area, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the tree.

[0116] As an optional implementation method, the amount of pesticide to be sprayed on the current tree is determined based on the current tree canopy area, red band image data, and near-infrared band image data, including:

[0117] The amount of pesticide to be sprayed is calculated using a formula based on the tree's canopy area, red band image data, and near-infrared band image data. The formula includes:

[0118]

[0119] Where Q is the amount of pesticide sprayed; k is the correction coefficient; NDVI is the normalized vegetation index; NIR is the image data in the near-infrared band; and R is the image data in the red band.

[0120] Specifically, the correction coefficient is related to the species of trees and the types of pests and diseases in the area to be sprayed. Once the area to be sprayed is determined, the correction coefficient is also determined.

[0121] Step 10: The variable spraying device is used to spray pesticides onto the corresponding trees based on their actual geographical location and the amount of pesticide to be sprayed.

[0122] Specifically, the hardware structure of the variable spraying device includes: a spray boom, a water pump, an electrical control unit, a water tank, and its piping. The quadcopter equipped with the variable spraying device uses a quick-release horizontal spray boom. The water pump is powered by an electronic speed controller and its pressure can be controlled according to the PWM signal provided by the flight control system on the quadcopter. The electrical control part mainly consists of a power supply and an electronic speed controller. The power supply uses a lithium battery. The input of the electronic speed controller is connected to the battery, the signal line is connected to the flight control unit, and the output is connected to the water pump. The water tank and piping use a special agricultural plant protection water tank and a high-pressure water pipe system.

[0123] like Figure 5As shown, when there is a need for spraying operations, the PWM signal pulse width output by the flight control board is adjusted based on the actual geographical location of each tree and the amount of pesticide to be sprayed, thereby controlling the water pump pressure and achieving variable-rate spraying. For quadcopter drones, the spraying system uses a vertical spray boom and a long-tube fan-shaped high-pressure nozzle, which is connected to a brushless water pump and water tank through a high-pressure water pipe. The spray boom is fixed on the four arms, located directly below the electronic speed controller. The water tank is fixed on the base plate of the frame, and the water pump is fixed on the water tank.

[0124] In one exemplary embodiment, a pesticide variable spraying method is provided. The pesticide variable spraying method is implemented using a pesticide variable spraying system and includes:

[0125] Remote sensing image data of the area to be sprayed is acquired using a multispectral imager mounted on an unmanned aerial vehicle (UAV). The remote sensing image data includes: green band image data, red band image data, red edge band image data, and near-infrared band image data.

[0126] Using a segmentation model, trees in the area to be sprayed are segmented based on remote sensing image data to obtain segmented images; the segmented images are images containing the masks and bounding boxes of each tree; the segmentation model is obtained by training an improved Yolov8 neural network.

[0127] Acquire the imaging and position parameters of the UAV-borne multispectral imager; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude and relative altitude to the ground.

[0128] Based on imaging parameters and relative altitude, the ground sampling distance of the UAV-borne multispectral imager is determined.

[0129] Select any tree as the current tree.

[0130] The pixel position offset of the current tree's bounding box is determined based on the bounding box of the current tree in the segmented image, as well as the width and height of the segmented image.

[0131] The geographic location offset of the current tree is determined based on the pixel position offset of the current tree's bounding box and the ground sampling distance.

[0132] The actual geographical location of the current tree is determined based on its geographical offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image.

[0133] The canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance.

[0134] Based on the current canopy area of ​​the trees, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the trees.

[0135] Pesticides are sprayed onto the corresponding trees based on their actual geographical location and the amount of pesticide to be applied.

[0136] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the system, method, and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A variable-rate pesticide spraying system, characterized in that, The pesticide variable spraying system includes: an unmanned aerial vehicle (UAV)-borne multispectral imager, a ground control center, and a variable spraying device; the UAV-borne multispectral imager and the variable spraying device are wirelessly connected to the ground control center. The UAV-borne multispectral imager is used to acquire remote sensing image data of the area to be sprayed; the remote sensing image data includes: green band image data, red band image data, red edge band image data, and near-infrared band image data. The ground control center is used for: Using a segmentation model, the trees in the area to be sprayed are segmented based on the remote sensing image data to obtain segmented images; the segmented images are images containing the masks and bounding boxes of each tree; the segmentation model is obtained by training an improved Yolov8 neural network; the improved Yolov8 neural network includes: a backbone network, a feature fusion network, and a detection network; the backbone network includes: 5 GhostConv modules, 2 RepVGG modules, 4 C2f modules, and 1 SPPF module; the feature fusion network includes: 6 feature stitching modules, 3 upsampling modules, 6 C2f modules, and 3 GhostConv modules; the detection network includes: 4 detection heads; The imaging parameters and position parameters of the UAV-borne multispectral imager are obtained; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude and relative altitude to the ground; Based on the imaging parameters and the relative altitude, the ground sampling distance of the UAV-borne multispectral imager is determined; Select any tree as the current tree; Based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, determine the pixel position offset of the current tree; The geographic location offset of the current tree is determined based on the pixel position offset of the current tree and the ground sampling distance; The actual geographical location of the current tree is determined based on the current tree's geographical location offset, the longitude, the latitude, and the latitude radian value of the center pixel of the segmented image; The canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance; Based on the current canopy area of ​​the trees, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the trees. The variable spraying device is used to spray pesticides onto the corresponding trees based on their actual geographical location and the amount of pesticide to be sprayed.

2. The pesticide variable spraying system according to claim 1, characterized in that, The wavelength range of the green band is 560nm±16nm, the wavelength range of the red band is 650nm±16nm, the wavelength range of the red edge band is 730nm±16nm, and the wavelength range of the near-infrared band is 860nm±26nm.

3. The pesticide variable spraying system according to claim 2, characterized in that, The process of determining the segmentation model includes: Obtain a training set; the training set includes: remote sensing image data of multiple training areas and corresponding segmented images; Construct an improved Yolov8 neural network; Using remote sensing image data of each training area as input and the corresponding segmented image as output, the improved Yolov8 neural network is trained to obtain the segmentation model.

4. The pesticide variable spraying system according to claim 1, characterized in that, Determining the ground sampling distance of the UAV-borne multispectral imager based on the imaging parameters and the relative altitude includes: The ground sampling distance is calculated using the ground sampling distance calculation formula, based on the imaging parameters and the relative height; the ground sampling distance calculation formula includes: Where H is the ground sampling distance; h1 is the relative height; s is the pixel size; and f is the true focal length.

5. The pesticide variable spraying system according to claim 4, characterized in that, Based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, the pixel position offset of the current tree is determined, including: Using the pixel position offset calculation formula, the pixel position offset of the current tree is calculated based on the bounding box of the current tree in the segmented image and the width and height of the segmented image; the pixel position offset calculation formula includes: Among them, P x N represents the x-axis component of the pixel position offset. x is the x-axis component of the center pixel position of the bounding box; w is the width of the segmented image; P y The y-axis component represents the pixel position offset; h represents the height of the segmented image; N y This is the y-axis component of the position of the center pixel of the bounding box.

6. The pesticide variable spraying system according to claim 5, characterized in that, Based on the pixel position offset of the current tree and the ground sampling distance, the geographical location offset of the current tree is determined, including: Using the geographic location offset calculation formula, the geographic location offset of the current tree is calculated based on the pixel position offset of the current tree and the ground sampling distance; the geographic location offset calculation formula includes: G x =P x ·H; G y =P y ·H; Among them, G x The x-axis component of the geographic location offset; G y The y-axis component represents the geographic location offset.

7. The pesticide variable spraying system according to claim 6, characterized in that, Based on the current tree's geographical location offset, the longitude, the latitude, and the latitude radian value of the center pixel of the segmented image, the actual geographical location of the current tree is determined, including: Using the actual geographic location calculation formula, the actual geographic location of the current tree is calculated based on its geographic location offset, longitude, latitude, and the latitude radian value of the center pixel of the segmented image; the actual geographic location calculation formula includes: Where α2 is the longitude in the actual geographical location; α1 is the longitude in the location parameters; γ is the latitude radian value of the center pixel of the segmented image; β2 is the latitude in the actual geographical location; and β1 is the latitude in the location parameters.

8. The pesticide variable spraying system according to claim 7, characterized in that, Based on the mask of the current tree in the segmented image and the ground sampling distance, the canopy area of ​​the current tree is determined, including: Using the canopy area calculation formula, the canopy area of ​​the current tree is calculated based on the mask of the current tree in the segmented image and the ground sampling distance; the canopy area calculation formula includes: A=M·H 2 ; Where A is the canopy area; M is the total number of pixels in the tree's mask.

9. The pesticide variable spraying system according to claim 8, characterized in that, Based on the current canopy area of ​​the trees, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the trees, including: The amount of pesticide to be sprayed is calculated using a formula based on the tree's canopy area, red band image data, and near-infrared band image data. The formula includes: Where Q is the amount of pesticide sprayed; k is the correction coefficient; NDVI is the normalized vegetation index; NIR is the image data in the near-infrared band; and R is the image data in the red band.

10. A method for variable-rate pesticide application, wherein the method is implemented using the variable-rate pesticide application system as described in any one of claims 1-9, characterized in that, The pesticide variable spraying method includes: Remote sensing image data of the area to be sprayed is acquired by an unmanned aerial vehicle (UAV) equipped with a multispectral imager. The remote sensing image data includes: green band image data, red band image data, red edge band image data, and near-infrared band image data. Using a segmentation model, the trees in the area to be sprayed are segmented based on the remote sensing image data to obtain segmented images; the segmented images are images containing the masks and bounding boxes of each tree; the segmentation model is obtained by training an improved Yolov8 neural network; the improved Yolov8 neural network includes: a backbone network, a feature fusion network, and a detection network; the backbone network includes: 5 GhostConv modules, 2 RepVGG modules, 4 C2f modules, and 1 SPPF module; the feature fusion network includes: 6 feature stitching modules, 3 upsampling modules, 6 C2f modules, and 3 GhostConv modules; the detection network includes: 4 detection heads; The imaging parameters and position parameters of the UAV-borne multispectral imager are obtained; the imaging parameters include: pixel size and true focal length; the position parameters include: longitude, latitude and relative altitude to the ground; Based on the imaging parameters and the relative altitude, the ground sampling distance of the UAV-borne multispectral imager is determined; Select any tree as the current tree; Based on the bounding box of the current tree in the segmented image and the width and height of the segmented image, determine the pixel position offset of the bounding box of the current tree; The geographic location offset of the current tree is determined based on the pixel position offset of the current tree's bounding box and the ground sampling distance; The actual geographical location of the current tree is determined based on the current tree's geographical location offset, the longitude, the latitude, and the latitude radian value of the center pixel of the segmented image; The canopy area of ​​the current tree is determined based on the mask of the current tree in the segmented image and the ground sampling distance; Based on the current canopy area of ​​the trees, red band image data, and near-infrared band image data, determine the current amount of pesticide to be sprayed on the trees. Pesticides are sprayed onto the corresponding trees based on their actual geographical location and the amount of pesticide to be applied.

Images