A precise target spraying method and device based on K-Means clustering of surface array point cloud

Abstract

The present invention discloses a method and device for precise target spraying based on K-Means clustering of array point cloud, the method comprising the following steps: filtering out ground point cloud data through a pitch angle threshold, and the array point cloud data after removing the ground point cloud data is fruit tree point cloud data; processing the fruit tree point cloud data through a projection method; taking the minimum sum of squares of clustering errors of a target set as an optimization target, combining the minimum-maximum distance method, dividing the processed fruit tree point cloud data into k clusters through K-Means clustering, each cluster represents a different fruit tree; positioning each fruit tree through samples of each cluster, and calculating the canopy spray distance of the fruit tree; the sprayer sprays the fruit tree precisely according to the canopy spray distance of the fruit tree. The method can locate and detect fruit trees, calculate the canopy spray distance, so as to achieve precise target spraying operation, and has a short response time, little environmental impact, and high detection accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection and positioning, and in particular to a precise target spraying method and device based on surface array point cloud K-Means clustering. Background Art

[0002] At present, the main methods for controlling orchard pests and diseases are biological control, agricultural ecological control and chemical control. Among them, chemical control is the most widely used and fastest-acting method. Most of the mainstream orchard pesticide application machinery in China adopts quantitative application, which is often a continuous spraying method, resulting in a large amount of pesticide deposition on the ground, even more than the amount deposited on the target. This method causes a large amount of pesticide waste, which pollutes the environment on the one hand, and causes excessive pesticide residues on the other hand, posing a very serious safety hazard.

[0003] In order to solve the above problems and address the situation of discontinuous planting of orchard plants, precision target spraying technology can achieve better pesticide application effects. In chemical control, precision application technology is the development direction of chemical control in orchard production. Its basic principle is to change the spraying amount in real time according to the different characteristic information of crops or fruit trees to achieve on-demand application. Compared with traditional large-scale spraying technology, precision application technology can effectively reduce the excessive use of pesticides through variable spraying, reduce the drift of droplets during spraying, improve the control effect of pesticides on fruit tree diseases and pests, save operating costs and reduce the pollution of pesticides to the environment.

[0004] The most critical work in precision target spraying technology is fruit tree detection technology. At present, the main method used by agricultural machinery is fruit tree detection technology based on machine vision. However, due to practical problems such as long response time, great influence of light and great influence of water mist refraction, the application of visual detection in actual orchard scenes is greatly restricted by conditions; and the detection accuracy is low, and the canopy position of the tree crown cannot be accurately detected. Summary of the invention

[0005] The purpose of the present invention is to overcome the above-mentioned problems and provide a precise target spraying method based on K-Means clustering of array point cloud. The method can locate and detect fruit trees and calculate the canopy spray distance to achieve precise target spraying operation, and has short response time, little environmental impact and high detection accuracy.

[0006] Another object of the present invention is to provide a precise target spray device based on surface array point cloud K-Means clustering.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A precise target spraying method based on K-Means clustering of surface array point cloud, characterized by comprising the following steps:

[0009] (1) Detect fruit trees through radar and obtain surface array point cloud data;

[0010] (2) After the array point cloud data is read into the radar, the ground point cloud data is filtered out by the pitch angle threshold, and the array point cloud data after the ground point cloud data is removed is the fruit tree point cloud data;

[0011] (3) Processing fruit tree point cloud data by projection method;

[0012] (4) Taking the minimum sum of squares of clustering errors of the target set as the optimization goal, combined with the minimum maximum distance method, the processed fruit tree point cloud data is divided into k clusters through K-Means clustering, and each cluster represents a different fruit tree;

[0013] (5) Using samples from each cluster, locate each fruit tree and calculate the relative distance between the edge of the fruit tree canopy and the nozzle plane and the spray distance of the fruit tree canopy;

[0014] (6) The sprayer precisely targets the fruit trees based on the relative distance between the edge of the fruit tree canopy and the nozzle plane and the spray distance of the fruit tree canopy.

[0015] The working principle of the above-mentioned precise target spraying method based on K-Means clustering of surface array point cloud is:

[0016] First, the ground point cloud data is filtered out through the pitch angle threshold, and then the fruit tree point cloud data is processed through the projection method. The minimum sum of squares of clustering errors of the target set is taken as the optimization goal. Combined with the minimum-maximum distance method, the processed fruit tree point cloud data is divided into k clusters through K-Means clustering. Each cluster represents a different fruit tree. Through the samples of each cluster, each fruit tree can be located and the canopy spray distance of the fruit tree can be calculated. The projection method reduces the computational complexity of the traditional K-Means clustering algorithm, reduces the dimensionality of the three-dimensional point cloud data, saves processor computing resources, and is conducive to the application of the target algorithm. The pitch angle threshold is set by radar internal and external parameters, and the ground point cloud data (ground point cloud noise) is filtered out with minimal time consumption to improve the algorithm execution efficiency. Before performing the clustering algorithm, the three-dimensional point cloud data is reduced in dimension by utilizing the characteristics of fruit tree planting and projected onto the target bird's-eye view. A simple relative distance is used as the clustering standard to simplify the amount of calculation, which is helpful for the practical application of the algorithm. When performing K-Means clustering, the initial clustering center is set by human intervention to avoid the clustering result falling into local optimization. The minimum sum of squares of clustering errors is used as the optimization goal to iterate clustering towards a better effect, which has the effect of adaptive iteration.

[0017] In a preferred embodiment of the present invention, the specific operation of step (2) is as follows:

[0018] (2.1) Assume that the detection range deflection angle of the array point cloud data is θ 0 , the pitch angle is ω 0 ;

[0019] (2.2) The information of each point in the array point cloud data includes the distance ρ from the radar to the scanning point, the deflection angle θ of the point, and the pitch angle ω of the point, and is represented by p = (ρ, θ, ω);

[0020] (2.3) Measure the distance D between the fruit tree trunks, the radar installation height H, and the sprayer width W, and calculate the pitch angle threshold that needs to be filtered out of the ground point cloud data: If the pitch angle in the surface array point cloud data is less than It is judged to be ground point cloud data, and the ground point cloud data is cleared, and the surface array point cloud data after clearing the ground point cloud data is defined as fruit tree point cloud data. The fruit tree point cloud data set is T = (p 0 , p 1 , p 2 , ..., p c ).

[0021] Preferably, the specific operation of step (3) is: projecting the fruit tree point cloud data set T onto the target top view by projection method, establishing a rectangular coordinate system with the center point of the radar as the origin, calculating the projection component l = ρcosωsinθ of each point in the fruit tree point cloud data in the forward direction of the sprayer, and defining the projection data set L = (l 1 , ..., l c ).

[0022] Preferably, in step (4), the target set C for storing clusters is defined as an empty set, the projected data set L is used as the standard for K-Means clustering, the minimum sum of squared clustering errors of the target set C is used as the optimization goal, and the minimum-maximum method is used to find k clusters in the projected data set L.

[0023] Preferably, in step (4), the target set C for storing clusters is defined as an empty set, the projected data set L is used as the standard for K-Means clustering, the minimum sum of squares of clustering errors of the target set C is used as the optimization goal, and the specific steps of using the minimum-maximum method to find k clusters in the projected data set L are:

[0024] (4.1) Calculate the distance between two points in the projected data set L |l a -l b |, (a, b∈{1, 2, ..., c}), find the two points with the largest distance l m and l n , as the initial cluster center;

[0025] (4.2) Calculate the projection data set L points to l respectively m , l n The distance D(l,l m ), D(l,l n ), and classify each point into the category of the point with the smaller distance value, and recalculate the cluster center μ 1 , μ 2 ;

[0026] (4.3) Calculate the distance from each point in the projected data set L to the cluster center (μ 1 , μ 2 , ..., μ k ) 1 ), D(l, μ 2 ), ..., D(l, μ k ), traverse each point and classify it into the category of the point with a small distance value, record the points with a small distance value and form a minimum distance set M;

[0027] (4.4) Update the target set C = {C 1 , C 2 , ..., C k}, calculate the sum of squares of clustering errors based on the current target set C (μ i Category C i 's center of mass);

[0028] (4.5) Select the point μ with the largest distance value in the minimum distance set M x , as the new cluster center, execute steps (4.3) and (4.4), and then go to step (4.6);

[0029] (4.6) Compare the SSE obtained from the two iterations. If SSE old >SSE new , it proves that the selected new cluster center has a good clustering effect, and go to step (4.7), otherwise the clustering ends and the output target set C = {C 1 , C 2 , ..., C k} as the final clustering result;

[0030] (4.7) According to the partition of the current target set C, recalculate C respectively i The cluster center of (i=1, 2, ..., k) (n i Indicates category C i ), go to step (4.3).

[0031] Preferably, in step (5), the target data set C obtained in step (4) = {C 1 , C 2 , ..., C k The k clusters in} are the point cloud clustering results of the relative distances of the k fruit trees in the direction of the sprayer. The relative distances between the centroids of the k fruit trees and the nozzle plane of the sprayer are equivalent to the values ​​of the cluster centers.

[0032] Calculate C i The minimum value of the sample points in each cluster (i=1, 2, ..., k) is min{l i}, (l∈C i ), and obtain the relative distance between the canopy edge of k fruit trees and the sprinkler plane;

[0033] Calculate C i The maximum distance between the sample points in each cluster (i=1, 2, ..., k) is max{|l m -l n |},(l m , l n ∈{C i}), and obtain the canopy spray distance of k fruit trees.

[0034] Preferably, in step (6), the sprayer reaches the relative distance min{l}, (l∈C i ), the sprayer control system opens the nozzle solenoid valve, and the sprayer starts spraying. The canopy spray distance is max{|l m -l n |},(l m , l n ∈{C i}), when the sprayer travel exceeds the canopy spray distance, the sprayer control system closes the nozzle solenoid valve and the spraying ends until it reaches the canopy edge of the next fruit tree.

[0035] A precise target spraying device based on surface array point cloud K-Means clustering comprises a sprayer, a radar arranged on the sprayer and a sprayer control system, wherein the sprayer and the radar are both connected to the sprayer control system.

[0036] Preferably, the radar is an array radar.

[0037] Preferably, the sprayer includes a spray head and a spray head solenoid valve for controlling the spray head switch, and a moving mechanism is provided at the bottom of the sprayer; and the sprayer control system is a processor.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The precise target spraying method based on K-Means clustering of array point cloud in the present invention first filters out the ground point cloud data through the pitch angle threshold, and then processes the fruit tree point cloud data through the projection method. The minimum sum of squares of clustering errors of the target set is taken as the optimization goal. Combined with the minimum and maximum distance method, the processed fruit tree point cloud data is divided into k clusters through K-Means clustering. Each cluster represents a different fruit tree. Through the samples of each cluster, each fruit tree can be located, and the canopy spray distance of the fruit tree can be calculated, thereby realizing the precise positioning of each fruit tree, with high detection accuracy, and finally realizing precise target spraying operation.

[0040] 2. The precise target spraying method based on K-Means clustering of the surface array point cloud in the present invention detects fruit trees through radar, which is less affected by light and water mist refraction, further improving the detection accuracy.

[0041] 3. The precise target spraying method based on K-Means clustering of array point cloud in the present invention uses projection processing to reduce the computational complexity of the traditional K-Means clustering algorithm, reduces the dimensionality of the three-dimensional point cloud data, saves processor computing resources, and is conducive to the application of the target algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the radar detection range and coordinate system establishment in a specific implementation of a precise target spraying method based on K-Means clustering of array point clouds in the present invention.

[0043] Figure 2 It is a cross-sectional schematic diagram of the target operation in the precise target spraying method based on K-Means clustering of the surface array point cloud in the present invention.

[0044] Figure 3 It is a top view of the target operation in the precise target spraying method based on K-Means clustering of the surface array point cloud in the present invention. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0046] See also Figure 1-Figure 3 This embodiment discloses a precise target spraying method based on surface array point cloud K-Means clustering, comprising the following steps:

[0047] (1) The fruit tree 4 is detected by radar 1 to obtain the surface array point cloud data. Without loss of generality, it is assumed that the detection range deflection angle of the surface array point cloud data is θ 0 , the pitch angle is ω0 ; The information of each point in the surface array point cloud data includes the distance ρ from the radar 1 to the scanning point, the deflection angle θ of the point and the pitch angle ω of the point, and is represented by p = (ρ, θ, ω).

[0048] (2) After the array point cloud data is read into radar 1, the ground point cloud data is filtered out by the pitch angle threshold. The array point cloud data after the ground point cloud data is removed is the fruit tree point cloud data:

[0049] When the distance D between the trunks of fruit trees, the installation height H of the radar 1 and the width W of the sprayer 2 are measured, the pitch angle threshold for filtering out the ground point cloud data is calculated as follows: If the pitch angle in the surface array point cloud data is less than It is judged to be ground point cloud data (point cloud data below the canopy, i.e. Figure 2 The point cloud data below the radar level) is cleared, and the surface array point cloud data after clearing the ground point cloud data is defined as the fruit tree point cloud data. The fruit tree point cloud data set is T = (p 0 , p 1 , p 2 , ..., p c ). The pitch angle threshold is set by using the internal and external parameters of radar 1 to filter out the ground point cloud data (ground point cloud noise) with minimal time consumption, thereby improving the algorithm execution efficiency.

[0050] (3) The fruit tree point cloud data set T is projected onto the top view of the target using the projection method. A rectangular coordinate system is established with the center point of radar 1 as the origin. The projection component l = ρcosωsinθ of each point in the fruit tree point cloud data in the forward direction of the sprayer 2 is calculated, and the projection data set L = (l 1 , ..., l c ). Before clustering algorithm is implemented, the dimensionality of 3D point cloud data is reduced by using the four planting characteristics of fruit trees and projected onto the top view of the target. Simple relative distance is used as the clustering standard to simplify the amount of calculation, which is helpful for the practical application of the algorithm.

[0051] (4) Define the target set C for storing clusters as an empty set, use the projected data set L as the standard for K-Means clustering, take the minimum sum of squares of clustering errors of the target set C as the optimization goal, and use the minimum-maximum method to find k clusters in the projected data set L:

[0052] (4.1) In order to reduce the impact of falling into the local optimal solution due to the random selection of the initial cluster center, the distance between two points in the projected data set L is calculated. a -l b |, (a, b∈{1, 2, ..., c}), find the two points with the largest distance l m and l n , as the initial cluster center;

[0053] (4.2) Calculate the projection data set L points to l respectively m , l n The distance D(l,l m ), D(l,l n ), and classify each point into the category of the point with the smaller distance value, and recalculate the cluster center μ 1 , μ 2 ;

[0054] (4.3) Calculate the distance from each point in the projected data set L to the cluster center (μ 1 , μ 2 , ..., μ k ) 1 ), D(l, μ 2 ), ..., D(l, μ k ), traverse each point and classify it into the category of the point with a small distance value, record the points with a small distance value and form a minimum distance set M;

[0055] (4.4) Update the target set C = {C 1 , C 2 , ..., C k}, calculate the sum of squares of clustering errors based on the current target set C (μ i Category C i 's center of mass);

[0056] (4.5) Select the point μ with the largest distance value in the minimum distance set M x , as the new cluster center, execute steps (4.3) and (4.4), and then go to step (4.6);

[0057] (4.6) Compare the SSE obtained from the two iterations. If SSE old >SSE new , it proves that the selected new cluster center has a good clustering effect, and go to step (4.7), otherwise the clustering ends and the output target set C = {C 1 , C 2 , ..., C k} as the final clustering result; among them, SSE old Represents the SSE of the previous iteration; SSE new Indicates the SSE of subsequent iterations;

[0058] (4.7) According to the partition of the current target set C, recalculate C respectively i The cluster center of (i=1, 2, ..., k) (n i Indicates category Ci ), go to step (4.3).

[0059] When performing K-Means clustering, the initial cluster center is set by human intervention to prevent the clustering results from falling into local optimization; the minimum sum of squared clustering errors is used as the optimization goal, so that clustering can be iterated towards a better result, which has the effect of adaptive iteration.

[0060] (5) The target data set C obtained in step (4) = {C 1 , C 2 , ..., C k The k clusters in} are the point cloud clustering results of the relative distances of the k fruit trees 4 in the direction of the sprayer 2. The relative distances between the centroids of the k fruit trees 4 and the nozzle plane of the sprayer 2 are equivalent to the values ​​of the cluster centers.

[0061] Calculate C i The minimum value of the sample points in each cluster (i=1, 2, ..., k) is min{l i}, (l∈C i ), and obtain the relative distance between the canopy edge of k fruit trees and the sprinkler plane;

[0062] Calculate C i The maximum distance between the sample points in each cluster (i=1, 2, ..., k) is max{|l m -l n |},(l m , l n ∈{C i}), and obtain the canopy spray distance of k fruit trees 4.

[0063] (6) Sprayer precision target operation process:

[0064] Sprayer 2 reaches the relative distance min{l}, (l∈C i ), the sprayer control system opens the nozzle solenoid valve, and sprayer 2 starts spraying. The canopy spray distance is max{|l m -l n |},(l m , l n ∈{C i}), when the sprayer 2 travels beyond the canopy spray distance, the sprayer control system closes the nozzle solenoid valve and the spraying ends until it reaches the canopy edge of the next fruit tree 4.

[0065] See also Figure 3 , the direction indicated by arrow 4 is the forward direction of the sprayer.

[0066] See also Figure 1-Figure 3The working principle of the above-mentioned precise target spraying method based on K-Means clustering of surface array point cloud is:

[0067] First, the ground point cloud data is filtered out through the pitch angle threshold, and then the fruit tree point cloud data is processed through the projection method. The minimum sum of squares of clustering errors of the target set is taken as the optimization goal. Combined with the minimum and maximum distance method, the processed fruit tree point cloud data is divided into k clusters through K-Means clustering. Each cluster represents a different fruit tree 4. Through the samples of each cluster, each fruit tree 4 can be located and the canopy spray distance of the fruit tree can be calculated. The projection method reduces the computational complexity of the traditional K-Means clustering algorithm, reduces the dimensionality of the three-dimensional point cloud data, saves processor computing resources, and is conducive to the application of the target algorithm.

[0068] See also Figure 2 This embodiment also discloses a precise target spraying device based on K-Means clustering of surface array point cloud, including a sprayer 2, a radar 1 arranged on the sprayer 2, and a sprayer control system, wherein the sprayer 2 and the radar 1 are both connected to the sprayer control system.

[0069] See also Figure 2 The radar 1 is a planar array radar, specifically, a millimeter wave radar. During measurement, the radar 1 stays at one position and can measure a point cloud within a range. All fruit trees 4 within the range of pitch angle and heading angle are measured.

[0070] See also Figure 2 The sprayer 2 includes a spray head and a spray head solenoid valve for controlling the spray head switch, and a moving mechanism is provided at the bottom of the sprayer 2; the sprayer control system is a processor, and the processor is an STM32 microcontroller.

[0071] The above is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above content. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A precise target spraying method based on K-Means clustering of surface array point cloud, characterized in that: The following steps are involved: (1) Detect fruit trees through radar and obtain surface array point cloud data; (2) After the array point cloud data is read into the radar, the ground point cloud data is filtered out by the pitch angle threshold, and the array point cloud data after the ground point cloud data is removed is the fruit tree point cloud data; (3) Processing fruit tree point cloud data by projection method; (4) Taking the minimum sum of squares of clustering errors of the target set as the optimization goal, combined with the minimum maximum distance method, the processed fruit tree point cloud data is divided into k clusters through K-Means clustering, and each cluster represents a different fruit tree; (5) Using samples from each cluster, locate each fruit tree and calculate the relative distance between the edge of the fruit tree canopy and the nozzle plane and the spray distance of the fruit tree canopy; (6) The sprayer sprays the fruit trees precisely based on the relative distance between the edge of the fruit tree canopy and the nozzle plane and the spray distance of the fruit tree canopy; The specific operation of step (2) is: (2.1) Assume that the detection range angle of the array point cloud data is θ0 and the pitch angle is ω0; (2.2) The information of each point in the array point cloud data includes the distance ρ from the radar to the scanning point, the deflection angle θ of the point, and the pitch angle ω of the point, and is represented by p = (ρ, θ, ω); (2.3) Measure the distance D between the fruit tree trunks, the radar installation height H, and the sprayer width W, and calculate the pitch angle threshold that needs to be filtered out of the ground point cloud data: If the pitch angle in the surface array point cloud data is less than It is judged to be ground point cloud data, and the ground point cloud data is cleared, and the surface array point cloud data after clearing the ground point cloud data is defined as fruit tree point cloud data. The fruit tree point cloud data set is T = (p0, p1, p2, ..., p c ); The specific operation of step (3) is as follows: project the fruit tree point cloud data set T onto the top view of the target using the projection method, establish a rectangular coordinate system with the center point of the radar as the origin, calculate the projection component l = ρcosωsinθ of each point in the fruit tree point cloud data in the forward direction of the sprayer, and define the projection data set L = (l1, ..., l c ).

2. According to claim 1, a precise target spraying method based on surface array point cloud K-Means clustering is characterized in that: In step (4), the target set C for storing clusters is defined as an empty set, the projected data set L is used as the standard for K-Means clustering, the minimum sum of squared clustering errors of the target set C is used as the optimization goal, and the minimum-maximum method is used to find k clusters in the projected data set L.

3. The precise target spraying method based on surface array point cloud K-Means clustering according to claim 2 is characterized in that: In step (4), the target set C for storing clusters is defined as an empty set, the projected data set L is used as the standard for K-Means clustering, the minimum sum of squares of clustering errors of the target set C is used as the optimization goal, and the specific steps of using the minimum-maximum method to find k clusters in the projected data set L are as follows: (4.1) Calculate the distance between two points in the projected data set L |l a -l b |, a, b∈{1,2,...,c}, find the two points with the largest distance l m and l n , as the initial cluster center; (4.2) Calculate the projection data set L points to l respectively m , l n The distance D(l,l m ), D(l,l n ), and classify each point into the category of the point with the smaller distance value, and recalculate the cluster centers μ1, μ2; (4.3) Calculate the distance from each point in the projected data set L to the cluster center (μ1, μ2, ..., μ k ) of the distance D(l, μ1), D(l, μ2), ..., D(l, μ k ), traverse each point and classify it into the category of the point with a small distance value, record the points with a small distance value and form a minimum distance set M; (4.4) Update the target set C = {C1, C2, ..., C k }, calculate the sum of squares of clustering errors based on the current target set C μ i Category C i The centroid of (4.5) Select the point μ with the largest distance value in the minimum distance set M x , as the new cluster center, execute steps (4.3) and (4.4), and then go to step (4.6); (4.6) Compare the sum of squares of clustering errors obtained from the two iterations before and after. If SSE old >SSE new , it proves that the selected new cluster center has a good clustering effect, and go to step (4.7), otherwise the clustering ends and the output target set C = {C1, C2, ..., C k } as the final clustering result; among them, SSE old Represents the sum of squares of clustering errors in the previous iteration; SSE new Represents the sum of squares of clustering errors in the next iteration; (4.7) According to the partition of the current target set C, recalculate C respectively i The cluster center n i Indicates category C i The number of samples, i = 1, 2, ..., k, go to step (4.3).

4. The method for precise target spraying based on K-Means clustering of surface array point cloud according to claim 3, characterized in that: In step (5), the target data set C obtained in step (4) = {C1, C2, ..., C k The k clusters in} are the point cloud clustering results of the relative distances of the k fruit trees in the direction of the sprayer. The relative distances between the centroids of the k fruit trees and the nozzle plane of the sprayer are equivalent to the values ​​of the cluster centers. Calculate C i The minimum value of each sample point in each cluster min{l i }, l∈C i , get the relative distance between the canopy edge of k fruit trees and the sprinkler plane; Calculate C i The maximum distance between sample points in each cluster is max{|l m -l n |}, l m , l n ∈{C i }, and obtain the canopy spray distance of k fruit trees.

5. The method for precise target spraying based on K-Means clustering of surface array point cloud according to claim 4, characterized in that: In step (6), the sprayer reaches the relative distance min{l}, l∈C i After that, the sprayer control system opens the nozzle solenoid valve and the sprayer starts spraying. The canopy spray distance is max|l m -l n |}, l m , l n ∈{C i }, when the sprayer stroke exceeds the canopy spray distance, the sprayer control system closes the nozzle solenoid valve and the spraying ends until it reaches the canopy edge of the next fruit tree.

6. A precise target spray device based on K-Means clustering of surface array point cloud, characterized in that: The precise target spraying device is used to implement the precise target spraying method based on surface array point cloud K-Means clustering as described in any one of claims 1 to 5, comprising a sprayer, a radar arranged on the sprayer, and a sprayer control system, wherein the sprayer and the radar are both connected to the sprayer control system.

7. The precise target spraying device based on surface array point cloud K-Means clustering according to claim 6, characterized in that: The radar is an array radar.

8. The precise target spraying device based on surface array point cloud K-Means clustering according to claim 7, characterized in that: The sprayer comprises a spray head and a spray head electromagnetic valve for controlling the spray head switch; a moving mechanism is arranged at the bottom of the sprayer; and the sprayer control system is a processor.

Images