A Beidou-based multi-level UAV adaptive patrol method and system

Abstract

The present invention discloses a Beidou-based multi-level UAV adaptive patrol method and system, including: the UAV obtains the UAV position and forest farm image; the command center sets the starting patrol point, and the position consistency is judged by the starting patrol point position and the UAV position; an adaptive screening algorithm is established by the position of the first-level forest farm image to obtain a first-level effective forest farm image, and a depth-grayscale dual threshold monitoring model is established by the first-level effective forest farm image to determine whether the forest farm has a risk of pine wood nematode infection; it is determined whether the UAV performs a second-level patrol, and manual intervention is performed on key risk areas. The present invention removes redundant image information by establishing an adaptive screening algorithm, establishes a depth-grayscale dual threshold monitoring model to determine whether the forest farm has a risk of pine wood nematode infection, and performs manual intervention on key risk areas, thereby achieving the task of accurately protecting against pine wood nematode disease.

Description

Technical Field

[0001] The present invention relates to the technical field of cross-application of Beidou navigation system and unmanned aerial vehicle technology, and in particular to a Beidou-based multi-level unmanned aerial vehicle adaptive patrol method and system. Background Art

[0002] Pine wood nematode is one of the most harmful organisms to forestry production. Due to its rapid spread and high mortality rate, pine wood nematode disease can kill infected healthy trees in as little as four weeks. For a pine forest, once a pine tree is infected with pine wood nematode disease, the entire pine forest will be destroyed within five years, causing huge economic losses.

[0003] At present, the prevention and control of pine wood nematode disease mainly uses drones, and pine wood nematode disease is diagnosed through remote sensing images collected by drones. Although the above method can achieve the purpose of pine wood nematode disease protection, the current classic method often uses deep learning to diagnose pine wood nematode disease, and the premise of deep learning judgment is to find a suitable data set, otherwise the diagnostic accuracy will be greatly discounted. In addition, in the process of using drones to patrol pine wood nematode disease, the cost is high due to the large area of ​​the forest farm. It is difficult to accurately locate the forest farm using only the sensor information of the drone, and there are many images collected by the drone. For the direct processing of these images, not only will the computing power be consumed, but also the real-time monitoring of pine wood nematode disease will be affected. For those key areas, a single patrol will inevitably have omissions. It is best to realize multiple patrols of key areas by drones to further improve the protection accuracy. To solve the above problems, the present invention proposes a multi-level drone adaptive patrol method and system based on Beidou. Summary of the invention

[0004] The present invention provides a Beidou-based multi-level UAV adaptive patrol method and system to solve the technical problems of image information redundancy, low accuracy and high cost of UAVs in the process of using UAVs for pine wood nematode protection.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] On the one hand, the present invention provides a Beidou-based multi-level UAV adaptive patrol method, the method comprising:

[0007] S1: The drone is equipped with a Beidou receiver and a depth camera. The Beidou receiver obtains the drone's position, and the depth camera obtains forest images.

[0008] The drone is equipped with a BeiDou receiver, which obtains the drone position P using the following calculation method:

[0009] ;

[0010] in, represents the jth position of the i-th UAV, i represents the serial number, j represents the serial number of the UAV position, i and j are both integers, represents the x-axis coordinate of the i-th UAV at the j-th position, represents the y-axis coordinate of the i-th UAV at the j-th position, represents the z-axis coordinate of the i-th UAV at the j-th position;

[0011] The drone is equipped with a depth camera, which acquires forest images;

[0012] Each pixel in the forest image F contains RGB color and depth H, which are calculated as follows:

[0013] ;

[0014] in, represents the forest image at the jth position of the ith UAV, represents the red component of the rth pixel at the jth position of the i-th UAV, represents the green component of the rth pixel at the jth position of the i-th UAV, represents the blue component of the rth pixel at the jth position of the i-th UAV, represents the depth of the rth pixel at the jth position of the i-th UAV;

[0015] After the UAV acquires the forest farm image, the forest farm image is grayed out. The specific formula for graying out is as follows:

[0016] ;

[0017] in, represents the grayscale value of the rth pixel in the forest image after grayscale processing when the i-th UAV is at the j-th position;

[0018] S2: The command center divides the forest farm into regions and sets the starting patrol points for each region. When the drone arrives near the forest farm, the starting patrol point and the drone position are used to determine the consistency of the positions. If the positions are consistent, the drone performs a first-level patrol and obtains a first-level forest farm image. If the positions are inconsistent, the drone continues to move toward the starting patrol point.

[0019] The command center divides the forest farm into N areas according to the scope and distribution of the forest farm, and sets a starting patrol point in each area;

[0020] The command center receives the forest farm location signal of the Beidou system and obtains the location U of each starting patrol point. The calculation method is:

[0021] ;

[0022] in, represents the position of the i-th starting patrol point, Indicates the position of the i-th starting patrol point The x-axis coordinate of Indicates the position of the i-th starting patrol point The y-axis coordinate of Indicates the position of the i-th starting patrol point The z-axis coordinate of

[0023] When the UAV arrives near the forest farm, the position consistency is judged by the starting patrol point position and the UAV position, and the position consistency coefficient W is calculated. The position consistency coefficient of the i-th UAV at the jth moment is The specific expression is as follows:

[0024] ;

[0025] Set the initialization position consistency threshold ;

[0026] Set the initialization flag FLAG;

[0027] The position consistency is determined by the position consistency coefficient and the position consistency threshold, and the calculation method is:

[0028] ;

[0029] if Less than or equal to , Equal to 1, indicating that the position of the i-th UAV and the i-th starting patrol point are consistent at this time, the UAV performs a first-level patrol and obtains a first-level forest image ;

[0030] if Greater than , Equal to 0, indicating that the positions of the i-th UAV and the i-th starting patrol point are inconsistent at this time, and the i-th UAV continues to move towards the i-th starting patrol point;

[0031] S3: Obtain the position of the first-level forest farm image from the first-level forest farm image, and establish an adaptive screening algorithm based on the position of the first-level forest farm image to obtain a first-level effective forest farm image, and establish a depth-grayscale dual threshold monitoring model based on the first-level effective forest farm image. Use the depth-grayscale dual threshold monitoring model to determine whether the forest farm has a risk of pine wood nematode infection, and complete the risk judgment of the first-level effective forest farm image;

[0032] S4: If the first-level valid forest image is of low risk, a secondary health check is performed; if the first-level valid forest image is of medium or high risk, the positions of the first-level valid forest images of medium and high risk are clustered to obtain the risk clustering area, and the drone performs secondary patrol of the risk clustering area;

[0033] S5: The command center conducts real-time monitoring of the secondary forest images collected by the secondary patrol, sets key risk areas based on the number of suspicious pixels in the secondary forest images, and conducts manual intervention in the key risk areas.

[0034] It should be further explained that the UAV determines its own position through the Beidou signal obtained by the Beidou receiver. Since the forest images are collected by the UAV, the position of each forest image can be further determined. In this scheme, by graying the forest images, the computing power of the forest images in the subsequent processing process can be reduced, thereby improving the computing efficiency of this scheme.

[0035] It should be further explained that forest farms are generally large in scope. By dividing the forest farm into multiple small areas and determining the starting patrol point of each area through the Beidou system, in order to avoid the loss of electricity, memory and computing power caused by the depth camera collecting images during non-patrol time periods, this solution uses the starting patrol point location and the drone location to perform position consistency judgment, thereby improving patrol efficiency.

[0036] Further, the S3 includes:

[0037] S310: When the UAV performs a first-level patrol, it will collect multiple first-level forest images, obtain the location of the first-level forest image from the first-level forest image, and establish an adaptive screening algorithm based on the location of the first-level forest image to screen the first-level forest image to obtain a first-level effective forest image. The specific steps are as follows:

[0038] S311: Calculate the location of the first-level forest image :

[0039] The first-level forest image collected by the i-th UAV is ,in , M represents the number of first-level forest images collected by the i-th UAV, and the location of each first-level forest image is obtained , calculated as:

[0040] ;

[0041] S312: Establishing an adaptive screening algorithm based on the position of the first-level forest image:

[0042] Set the initialization filter coefficient , calculated as:

[0043] ;

[0044] in, Indicates the maximum value of the x-axis coordinate in the position of the first-level forest image. Indicates the minimum value of the x-axis coordinate in the position of the first-level forest image. Represents the absolute value operator;

[0045] Set the initialization adaptive threshold , calculate the adaptive screening interval E, where E is in meters, and establish an adaptive screening algorithm:

[0046] if Greater than or equal to ,but ;

[0047] if Less than ,but ;

[0048] S313: Screening the first-level forest farm image to obtain a first-level effective forest farm image

[0049] right In Sort from small to large, Interval with adaptive screening interval E, in addition to selecting The corresponding first-level forest image and In addition to the corresponding first-level forest farm image, in each adaptive screening interval E, the residual sum of the position of each first-level forest farm image and the positions of other first-level forest farm images in the interval is calculated, and the first-level forest farm image corresponding to the minimum value of the residual sum is taken as the first-level effective forest farm image of the interval, and then all the first-level effective forest farm images are obtained;

[0050] It should be further explained that, because the depth camera of the drone collects forest images, multiple forest images have overlapping areas. Direct calculation will not only reduce the calculation accuracy, but also seriously consume the computing power of the drone. Therefore, the adaptive screening algorithm can remove redundant information and increase the patrol range of low-cost drones, which has higher economic value and practicality.

[0051] Furthermore, the depth-grayscale dual threshold monitoring model is established based on the primary effective forest farm image to complete the risk judgment of the primary effective forest farm image, including the following steps:

[0052] S320: After obtaining a first-level effective forest image Finally, a depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image. The depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has the risk of pine wood nematode infection. The specific steps are as follows:

[0053] S321: Calculate the first-level effective forest image collected by the i-th UAV: , calculated as:

[0054] ;

[0055] Among them, T represents the number of first-level effective forest images collected by the i-th UAV;

[0056] ;

[0057] in, represents the jth first-level effective forest image collected by the i-th UAV;

[0058] S322: Establish a depth-grayscale dual threshold monitoring model based on the first-level effective forest image:

[0059] First, set the depth score Grayscale score , initialize the depth threshold Grayscale threshold ;

[0060] if Less than or equal to , then the depth score of the pixel is 1 if Less than or equal to , then the depth score of the pixel is 0;

[0061] Secondly, calculate the grayscale residual of the grayscale of the rth pixel in the jth first-level effective forest image collected by the i-th UAV and the grayscale of the adjacent pixel , the calculation formula is as follows:

[0062] ;

[0063] in, represents the first effective forest image in the jth primary forest image collected by the i-th UAV The gray level of the g-th pixel adjacent to the g-th pixel, m represents The number of adjacent pixels;

[0064] if Less than or equal to , then the grayscale score of the pixel is 0, if Less than or equal to , then the grayscale score of the pixel is 1;

[0065] Finally, a depth-grayscale dual threshold monitoring model is established :

[0066] ;

[0067] in, represents the first effective forest image in the jth primary forest image collected by the i-th UAV The comprehensive score of pixels;

[0068] if If it is 2, it means that the pixel is a suspicious pixel, otherwise it is a healthy pixel;

[0069] S323: Determine whether there is a risk of pine wood nematode infection in a forest farm through the depth-grayscale dual threshold monitoring model:

[0070] Calculate the number of suspicious pixels J in each first-level valid forest image;

[0071] if , then the first-level effective forest image may have the risk of pine wood nematode infection, which is low risk;

[0072] if , then the first-level effective forest image has the risk of pine wood nematode infection, which is medium risk;

[0073] if , then the first-level effective forest image has a serious risk of pine wood nematode infection, which is a high risk;

[0074] Further, the S4 includes:

[0075] S410: If the first-level valid forest image is low risk, a secondary health check is performed; specifically:

[0076] Initialize the health threshold, manually obtain the healthy forest image, and calculate the grayscale average of each pixel in the healthy forest image , the grayscale and Difference is calculated, and the maximum value of the above difference is calculated. If the maximum value of the above difference is less than the health threshold, the low-risk first-level effective forest farm image is healthy; if the maximum value of the above difference is greater than or equal to the health threshold, the low-risk first-level effective forest farm image is medium risk;

[0077] S420: If the first-level effective forest farm image is of medium or high risk, the positions of the first-level effective forest farm images of medium or high risk are clustered to obtain the risk clustering area, and the UAV performs a second-level patrol of the risk clustering area.

[0078] It should be further explained that the grayscale average value of each pixel in the healthy forest image is calculated , the grayscale and Make a difference and calculate the maximum value of the above difference. The specific expression is as follows:

[0079] ;

[0080] in, represents the jth low-risk first-level effective forest image collected by the i-th UAV The gray value of a pixel, express and The difference between

[0081] calculate The maximum value of ;

[0082] To further explain, the clustering method used in this scheme is K-means clustering. Since K-means clustering is a conventional method in statistics, this scheme will not be described in detail. Different from the use of traditional K-means clustering method, the purpose of this scheme is not to distinguish the types of clustering, but to use the K-means clustering method to determine the number of risk clustering areas. The drone performs secondary patrols on the risk clustering areas. It should be noted that in order to avoid missed secondary patrols, the actual area of ​​the secondary patrol will generally be 1.1 to 1.2 times the risk clustering area. The actual area is the area that is proportionally enlarged from the risk clustering area.

[0083] On the other hand, the present invention also provides a Beidou-based multi-level UAV adaptive patrol system, which includes the following modules:

[0084] Data acquisition module: The drone is equipped with a Beidou receiver and a depth camera. The Beidou receiver obtains the drone's location, and the depth camera obtains forest images.

[0085] Consistency judgment module: The command center divides the forest farm into areas and sets the starting patrol points for each area. When the drone arrives near the forest farm, the starting patrol point and the drone position are used to make a position consistency judgment. If the positions are consistent, the drone performs a first-level patrol and obtains a first-level forest farm image. If the positions are inconsistent, the drone continues to move to the starting patrol point.

[0086] Risk judgment module: The position of the first-level forest farm image is obtained from the first-level forest farm image, and an adaptive screening algorithm is established based on the position of the first-level forest farm image to obtain a first-level effective forest farm image. A depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image. The depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has a risk of pine wood nematode infection, and complete the risk judgment of the first-level effective forest farm image;

[0087] Secondary patrol module: If the first-level effective forest farm image is of low risk, a secondary health check is performed; if the first-level effective forest farm image is of medium or high risk, the positions of the first-level effective forest farm images with medium or high risk are clustered to obtain the risk clustering area. The drone performs secondary patrols on the risk clustering area, and the command center conducts real-time monitoring of the secondary forest farm images collected by the secondary patrols. The key risk areas are set according to the number of suspicious pixels in the secondary forest farm images, and manual intervention is performed on the key risk areas.

[0088] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0089] 1. The present invention provides a multi-level unmanned aerial vehicle adaptive patrol method based on Beidou. The unmanned aerial vehicle carries a Beidou receiver and a depth camera to obtain the unmanned aerial vehicle's own position and forest farm image, and grayscales the forest farm image, which can further improve the calculation efficiency of subsequent forest farm image processing; the command center divides the forest farm into regions according to the Beidou position, reasonably divides the forest farm into multiple regions, and sets a starting patrol point in each region. The use of multiple unmanned aerial vehicles to perform patrol tasks at the same time can improve the patrol efficiency, and then timely discover the area infected with pine wood nematode disease, and the position consistency is judged by the starting patrol point position and the unmanned aerial vehicle position. Only when the unmanned aerial vehicle starts patrolling, the image acquisition task is performed, which reduces the loss of unmanned aerial vehicle power, computing power and memory caused by non-patrol. Compared with the traditional patrol method, the patrol efficiency of this scheme is higher, and multiple low-cost unmanned aerial vehicles can be used to achieve rapid patrol, which has higher economic value.

[0090] 2. The present invention establishes an adaptive screening algorithm based on the position of the first-level forest images collected by the UAV, and adaptively calculates the screening interval based on the number of first-level forest images collected by the UAV and the screening coefficient. Compared with the estimated fixed screening interval, the present solution can be adaptively adjusted according to the collection frequency of the UAV and the patrol range of the UAV, and has higher intelligence. Through the adaptive screening algorithm, redundant forest image information can be removed. Under the premise of ensuring the accuracy of the method, the calculation cost of the UAV is reduced to obtain a first-level effective forest image. A depth-grayscale dual threshold monitoring model is established from the first-level effective forest image. The depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has the risk of pine wood nematode infection, and complete the risk judgment of the first-level effective forest image.

[0091] 3. In the present invention, if the first-level effective forest image is low-risk, a secondary health check is performed. In the secondary health check, a healthy forest image is manually obtained, and the grayscale of pixels in the healthy forest image and the low-risk forest image is compared to further determine whether the low-risk area is truly healthy, and then determine whether to perform a secondary patrol. Compared with the traditional method, this scheme combines sensor information and manual intervention information to further improve the accuracy of the scheme with less human interference; cluster the positions of medium and high-risk first-level effective forest images to obtain risk clustering areas, and the drone performs secondary patrols on the risk clustering areas. The command center monitors the secondary forest images collected by the secondary patrols in real time, sets key risk areas according to the number of suspicious pixels in the secondary forest images, and performs manual intervention on the key risk areas to achieve timely protection against pine wood nematodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 A flowchart of a Beidou-based multi-level UAV adaptive patrol method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0093] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0094] Example 1

[0095] This embodiment will describe in more detail a Beidou-based multi-level UAV adaptive patrol method of the present invention in conjunction with the corresponding drawings. Figure 1 Specifically, the method of this embodiment includes the following steps:

[0096] S1: The drone is equipped with a BeiDou receiver and a depth camera. The BeiDou receiver obtains the drone's position, and the depth camera obtains forest images, including:

[0097] The drone is equipped with a BeiDou receiver, which acquires the drone position P. represents the jth position of the ith UAV, , where i represents the serial number, j represents the serial number of the drone position, and both i and j are integers. represents the x-axis coordinate of the i-th UAV at the j-th position, represents the y-axis coordinate of the i-th UAV at the j-th position, represents the z-axis coordinate of the i-th UAV at the j-th position;

[0098] The drone is equipped with a depth camera, which acquires forest images;

[0099] Each pixel in the forest image F contains RGB color and depth H. , represents the forest image at the jth position of the ith UAV, , represents the red component of the rth pixel at the jth position of the i-th UAV, represents the green component of the rth pixel at the jth position of the i-th UAV, represents the blue component of the rth pixel at the jth position of the i-th UAV, represents the depth of the rth pixel at the jth position of the i-th UAV;

[0100] After the UAV acquires the forest farm image, the forest farm image is grayed out. The specific formula for graying out is as follows:

[0101] ;

[0102] in, represents the grayscale value of the rth pixel in the forest image after grayscale processing when the i-th UAV is at the j-th position;

[0103] It should be further explained that in this solution, when the drone collects depth images, the default height is consistent, which can avoid the difference in viewing angle caused by height difference, and thus avoid the difference in the screening of forest images.

[0104] S2: The command center divides the forest farm into regions and sets the starting patrol points for each region. When the drone arrives near the forest farm, the starting patrol point and the drone position are used to determine the consistency of the positions. If the positions are consistent, the drone performs a first-level patrol and obtains a first-level forest farm image. If the positions are inconsistent, the drone continues to move toward the starting patrol point, including:

[0105] The command center divides the forest farm into N areas according to the scope and distribution of the forest farm, and sets a starting patrol point in each area;

[0106] According to the forest location signal received by the command center from the Beidou system, the location U of each starting patrol point can be obtained. represents the position of the i-th starting patrol point, , Indicates the position of the i-th starting patrol point The x-axis coordinate of Indicates the position of the i-th starting patrol point The y-axis coordinate of Indicates the position of the i-th starting patrol point The z-axis coordinate of

[0107] When the UAV arrives near the forest farm, the position consistency is judged by the starting patrol point position and the UAV position, and the position consistency coefficient W is calculated. The position consistency coefficient of the i-th UAV at the jth moment is The specific expression is as follows:

[0108] ;

[0109] Set the initialization position consistency threshold ;

[0110] Set the initialization flag FLAG;

[0111] The position consistency coefficient and the position consistency threshold are used to determine whether the UAV can perform a first-level patrol. The specific expression for the position consistency judgment of the i-th UAV at the j-th moment is as follows:

[0112] ;

[0113] if Less than or equal to , Equal to 1, indicating that the position of the i-th UAV and the i-th starting patrol point are consistent at this time, the UAV performs a first-level patrol and obtains a first-level forest image ;

[0114] if Greater than , is equal to 0, indicating that the positions of the i-th UAV and the i-th starting patrol point are inconsistent at this time, and the i-th UAV continues to move towards the i-th starting patrol point;

[0115] It should be further explained that the starting patrol point is generally set at the edge of the area, that is, as close as possible to the direction of the drone's take-off. This can improve the patrol efficiency of the drone and achieve rapid patrol of the forest farm.

[0116] S3: Obtain the location of the first-level forest farm image from the first-level forest farm image, and establish an adaptive screening algorithm based on the location of the first-level forest farm image to obtain a first-level effective forest farm image. Establish a depth-grayscale dual threshold monitoring model based on the first-level effective forest farm image, and determine whether the forest farm has a risk of pine wood nematode infection through the depth-grayscale dual threshold monitoring model, and complete the risk judgment of the first-level effective forest farm image, including:

[0117] S310: When the UAV performs a first-level patrol, it will collect multiple first-level forest images, obtain the location of the first-level forest image from the first-level forest image, and establish an adaptive screening algorithm based on the location of the first-level forest image to screen the first-level forest image to obtain a first-level effective forest image. The specific steps of the adaptive screening algorithm are as follows:

[0118] S311: Calculate the location of the first-level forest image :

[0119] The first-level forest image collected by the i-th UAV is ,in , M represents the number of first-level forest images collected by the i-th UAV, and the location of each first-level forest image is obtained ,in ;

[0120] S312: Establishing an adaptive screening algorithm based on the position of the first-level forest image:

[0121] Set the initialization filter coefficient , The specific expression is as follows:

[0122] ;

[0123] in, Indicates the maximum value of the x-axis coordinate in the position of the first-level forest image. Indicates the minimum value of the x-axis coordinate in the position of the first-level forest image. Represents the absolute value operator;

[0124] Set the initialization adaptive threshold , calculate the adaptive screening interval E, the unit of E is meter, Greater than or equal to ,but if Less than ,but ;

[0125] S313: Establishing an adaptive screening algorithm based on the position of the first-level forest image:

[0126] right In Sort from small to large, Interval with adaptive screening interval E, in addition to selecting The corresponding first-level forest image and In addition to the corresponding first-level forest farm image, in each adaptive screening interval E, the residual sum of the position of each first-level forest farm image and the positions of other first-level forest farm images in the interval is calculated, and the first-level forest farm image corresponding to the minimum value of the residual sum is taken as the first-level effective forest farm image of the interval, and then all the first-level effective forest farm images are obtained;

[0127] S320: After obtaining a first-level effective forest image Finally, a depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image to complete the risk judgment of the first-level effective forest farm image;

[0128] After obtaining the first-level effective forest farm image, a depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image. The depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has the risk of pine wood nematode infection. The specific steps are as follows:

[0129] S321: Calculate the first-level effective forest image collected by the i-th UAV: ,in , T represents the number of first-level effective forest images collected by the i-th UAV, ,in represents the jth first-level effective forest image collected by the i-th UAV;

[0130] S322: Establish a depth-grayscale dual threshold monitoring model based on the first-level effective forest image:

[0131] First, set the depth score Grayscale score , initialize the depth threshold Grayscale threshold ;

[0132] if Less than or equal to , then the depth score of the pixel is 1 if Less than or equal to , then the depth score of the pixel is 0;

[0133] Secondly, calculate the grayscale residual of the grayscale of the rth pixel in the jth first-level effective forest image collected by the i-th UAV and the grayscale of the adjacent pixel , the calculation formula is as follows:

[0134] ;

[0135] in, represents the first effective forest image in the jth primary forest image collected by the i-th UAV The gray level of the g-th pixel adjacent to the g-th pixel, m represents The number of adjacent pixels;

[0136] if Less than or equal to , then the grayscale score of the pixel is 0, if Less than or equal to , then the grayscale score of the pixel is 1;

[0137] Finally, a depth-grayscale dual threshold monitoring model is established :

[0138] ;

[0139] in, represents the first effective forest image in the jth primary forest image collected by the i-th UAV The comprehensive score of pixels;

[0140] if If it is 2, it means that the pixel is a suspicious pixel, otherwise it is a healthy pixel;

[0141] S323: Determine whether there is a risk of pine wood nematode infection in a forest farm through the depth-grayscale dual threshold monitoring model:

[0142] Calculate the number of suspicious pixels J in each first-level valid forest image;

[0143] if , then the first-level effective forest image may have the risk of pine wood nematode infection, which is low risk;

[0144] if , then the first-level effective forest image has the risk of pine wood nematode infection, which is medium risk;

[0145] if , then the first-level effective forest image has a serious risk of pine wood nematode infection and is high risk.

[0146] It should be further explained that the traditional identification of pine wood nematode disease is often judged by deep learning or hyperspectral, and the above-mentioned judgment method is often only effective for the identification of symptoms in the middle and late stages, and is relatively complicated to implement. This solution only uses the data from the depth camera to distinguish between pine forests and land, and establishes a depth-grayscale dual threshold monitoring model to judge pine wood nematode disease. It is easy and fast to operate, and has a high accuracy.

[0147] S4: If the first-level valid forest image is of low risk, a secondary health check is performed. If the first-level valid forest image is of medium or high risk, the positions of the first-level valid forest images with medium or high risk are clustered to obtain the risk clustering area. The drone performs secondary patrols on the risk clustering area, including:

[0148] If the first-level effective forest image is low risk, a secondary health check is performed; specifically:

[0149] S410: Initialize the health threshold, manually obtain the healthy forest image, and calculate the grayscale average of each pixel in the healthy forest image , the grayscale and Make a difference and calculate the maximum value of the above difference. The specific expression is as follows:

[0150] ;

[0151] in, represents the jth low-risk first-level effective forest image collected by the i-th UAV The gray value of a pixel, express and The difference between

[0152] calculate The maximum value of ;

[0153] If the above maximum value is less than the health threshold, the low-risk first-level valid forest farm image is healthy; if the above maximum value is greater than or equal to the health threshold, the low-risk first-level valid forest farm image is medium risk;

[0154] S420: If the first-level effective forest farm image is of medium or high risk, the positions of the first-level effective forest farm images of medium or high risk are clustered to obtain a risk clustering area, and the UAV performs a second-level patrol of the risk clustering area;

[0155] It should be further explained that by conducting secondary health checks on low-risk first-level effective forest farm images, the first-level effective forest farm images in medium-risk areas can be further identified and screened, thereby obtaining accurate risk clustering areas and achieving precise patrols.

[0156] S5: The command center conducts real-time monitoring of the secondary forest images collected by the secondary patrol, sets key risk areas based on the number of suspicious pixels in the secondary forest images, and conducts manual intervention in the key risk areas.

[0157] It should be further explained that the pixel size of the forest image is 1280*1040. When the number of suspicious pixels in the secondary forest image exceeds 100, the risk clustering area where the secondary forest image is located is a key risk area. In the actual testing process, the so-called suspicious pixels can be more accurately said to be the pixel intersection points of diseased trees and healthy trees. Due to the rapid spread of pine wilt disease, other trees in the area may be at great risk of infection. Therefore, artificial intervention is needed in key risk areas to avoid further spread and spread of pine wilt disease.

[0158] Example 2

[0159] This embodiment provides a Beidou-based multi-level UAV adaptive patrol system:

[0160] Data acquisition module: The drone is equipped with a Beidou receiver and a depth camera. The Beidou receiver obtains the drone's location, and the depth camera obtains forest images.

[0161] Consistency judgment module: The command center divides the forest farm into areas and sets the starting patrol points for each area. When the drone arrives near the forest farm, the starting patrol point and the drone position are used to make a position consistency judgment. If the positions are consistent, the drone performs a first-level patrol and obtains a first-level forest farm image. If the positions are inconsistent, the drone continues to move to the starting patrol point.

[0162] Risk judgment module: The position of the first-level forest farm image is obtained from the first-level forest farm image, and an adaptive screening algorithm is established based on the position of the first-level forest farm image to obtain a first-level effective forest farm image. A depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image. The depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has a risk of pine wood nematode infection, and complete the risk judgment of the first-level effective forest farm image;

[0163] Secondary patrol module: If the first-level effective forest farm image is of low risk, a secondary health check is performed; if the first-level effective forest farm image is of medium or high risk, the positions of the first-level effective forest farm images with medium or high risk are clustered to obtain the risk clustering area. The drone performs secondary patrols on the risk clustering area, and the command center conducts real-time monitoring of the secondary forest farm images collected by the secondary patrols. The key risk areas are set according to the number of suspicious pixels in the secondary forest farm images, and manual intervention is performed on the key risk areas.

[0164] The Beidou-based multi-level UAV adaptive patrol system of the present embodiment corresponds to the Beidou-based multi-level UAV adaptive patrol method of the above-mentioned embodiment; wherein, the functions implemented by each functional module in the Beidou-based multi-level UAV adaptive patrol system of the present embodiment correspond one-to-one to each process step in the Beidou-based multi-level UAV adaptive patrol method of the above-mentioned embodiment; therefore, they will not be repeated here.

[0165] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0166] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0167] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A Beidou-based multi-level UAV adaptive patrol method, characterized in that: The following steps are involved: S1: The drone is equipped with a Beidou receiver and a depth camera. The Beidou receiver obtains the drone's position, and the depth camera obtains forest images. The drone is equipped with a BeiDou receiver, which obtains the drone position P using the following calculation method: ； in, represents the jth position of the i-th UAV, i represents the serial number, j represents the serial number of the UAV position, i and j are both integers, represents the x-axis coordinate of the i-th UAV at the j-th position, represents the y-axis coordinate of the i-th UAV at the j-th position, represents the z-axis coordinate of the i-th UAV at the j-th position; The drone is equipped with a depth camera, which acquires forest images; Each pixel in the forest image F contains RGB color and depth H, which are calculated as follows: ； in, represents the forest image at the jth position of the ith UAV, represents the red component of the rth pixel at the jth position of the i-th UAV, represents the green component of the rth pixel at the jth position of the i-th UAV, represents the blue component of the rth pixel at the jth position of the i-th UAV, represents the depth of the rth pixel at the jth position of the i-th UAV; The forest farm image is grayed out, and the graying formula is as follows: ； in, It represents the gray value of the rth pixel in the forest image after grayscale processing when the i-th UAV is at the j-th position; S2: The command center divides the forest farm into regions and sets the starting patrol points for each region. When the drone arrives near the forest farm, the starting patrol point and the drone position are used to determine the consistency of the positions. If the positions are consistent, the drone performs a first-level patrol and obtains a first-level forest farm image. If the positions are inconsistent, the drone continues to move toward the starting patrol point. The command center divides the forest farm into N areas according to the scope and distribution of the forest farm, and sets a starting patrol point in each area; The command center receives the forest farm location signal of the Beidou system and obtains the location U of each starting patrol point. The calculation method is: ； in, represents the position of the i-th starting patrol point, Indicates the position of the i-th starting patrol point The x-axis coordinate of Indicates the position of the i-th starting patrol point The y-axis coordinate of Indicates the position of the i-th starting patrol point The z-axis coordinate of When the UAV arrives near the forest farm, the position consistency is judged by the starting patrol point position and the UAV position, and the position consistency coefficient W is calculated. The position consistency coefficient of the i-th UAV at the jth moment is The specific expression is as follows: ； Set the initialization position consistency threshold; Set the initialization flag FLAG; The position consistency is determined by the position consistency coefficient and the position consistency threshold, and the calculation method is: ； if Less than or equal to , Equal to 1, indicating that the position of the i-th UAV and the i-th starting patrol point are consistent at this time, the UAV performs a first-level patrol and obtains a first-level forest image ; if Greater than , Equal to 0, indicating that the positions of the i-th UAV and the i-th starting patrol point are inconsistent at this time, and the i-th UAV continues to move towards the i-th starting patrol point; S3: Obtain the position of the first-level forest farm image from the first-level forest farm image, and establish an adaptive screening algorithm based on the position of the first-level forest farm image to obtain a first-level effective forest farm image, and establish a depth-grayscale dual threshold monitoring model based on the first-level effective forest farm image. Use the depth-grayscale dual threshold monitoring model to determine whether the forest farm has a risk of pine wood nematode infection, and complete the risk judgment of the first-level effective forest farm image; S4: If the first-level valid forest image is of low risk, a secondary health check is performed; if the first-level valid forest image is of medium or high risk, the positions of the first-level valid forest images of medium and high risk are clustered to obtain the risk clustering area, and the drone performs secondary patrol of the risk clustering area; S5: The command center conducts real-time monitoring of the secondary forest images collected by the secondary patrol, sets key risk areas based on the number of suspicious pixels in the secondary forest images, and conducts manual intervention in the key risk areas.

2. The Beidou-based multi-level UAV adaptive patrol method according to claim 1 is characterized in that: The S3 includes: S310: When the UAV performs a first-level patrol, it will collect multiple first-level forest images, obtain the location of the first-level forest image from the first-level forest image, and establish an adaptive screening algorithm based on the location of the first-level forest image to screen the first-level forest image to obtain a first-level effective forest image. The specific steps are as follows: S311: Calculate the location of the first-level forest image : The first-level forest image collected by the i-th UAV is ,in , M represents the number of first-level forest images collected by the i-th UAV, and the location of each first-level forest image is obtained , calculated as: ； S312: Establishing an adaptive screening algorithm based on the position of the first-level forest image: Set the initialization filter coefficient , calculated as: ； in, Indicates the maximum value of the x-axis coordinate in the position of the first-level forest image. Indicates the minimum value of the x-axis coordinate in the position of the first-level forest image. Represents the absolute value operator; Set the initialization adaptive threshold , calculate the adaptive screening interval E, where E is in meters, and establish an adaptive screening algorithm: if Greater than or equal to ,but ; if Less than ,but ; S313: Screening the first-level forest farm image to obtain a first-level valid forest farm image: right In Sort from small to large, Interval with adaptive screening interval E, in addition to selecting The corresponding first-level forest image and In addition to the corresponding first-level forest farm image, in each adaptive screening interval E, the residual sum of the position of each first-level forest farm image and the positions of other first-level forest farm images in the interval is calculated, and the first-level forest farm image corresponding to the minimum value of the residual sum is taken as the first-level valid forest farm image of the interval, and then all the first-level valid forest farm images are obtained.

3. The Beidou-based multi-level UAV adaptive patrol method according to claim 2 is characterized in that: The depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image to complete the risk judgment of the first-level effective forest farm image, including the following steps: S320: After obtaining the first-level effective forest farm image, a depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image, and the depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has a risk of pine wood nematode infection. The specific steps are as follows: S321: Calculate the first-level effective forest image collected by the i-th UAV: , calculated as: ； Among them, T represents the number of first-level effective forest images collected by the i-th UAV; ； in, represents the jth first-level effective forest image collected by the i-th UAV; S322: Establish a depth-grayscale dual threshold monitoring model based on the first-level effective forest image: First, set the depth score Grayscale score , initialize the depth threshold Grayscale threshold ; if Less than or equal to , then the depth score of the pixel is 1 if Less than or equal to , then the depth score of the pixel is 0; Secondly, calculate the grayscale residual of the grayscale of the rth pixel in the jth first-level effective forest image collected by the i-th UAV and the grayscale of the adjacent pixel , the calculation formula is as follows: ； in, represents the first effective forest image in the jth primary forest image collected by the i-th UAV The gray level of the g-th pixel adjacent to the g-th pixel, m represents The number of adjacent pixels; if Less than or equal to , then the grayscale score of the pixel is 0, if Less than or equal to , then the grayscale score of the pixel is 1; Finally, a depth-grayscale dual threshold monitoring model is established : ； in, represents the first effective forest image in the jth primary forest image collected by the i-th UAV The comprehensive score of pixels, if If it is 2, it means that the pixel is a suspicious pixel, otherwise it is a healthy pixel; S323: Determine whether there is a risk of pine wood nematode infection in a forest farm through the depth-grayscale dual threshold monitoring model: Calculate the number of suspicious pixels J in each first-level valid forest image; if , then the first-level effective forest image may have the risk of pine wood nematode infection, which is low risk; if , then the first-level effective forest image has the risk of pine wood nematode infection, which is medium risk; if , then the first-level effective forest image has a serious risk of pine wood nematode infection and is high risk.

4. The Beidou-based multi-level UAV adaptive patrol method according to claim 3 is characterized in that: The S4 includes: S410: If the first-level valid forest image is low risk, a secondary health check is performed; specifically: Initialize the health threshold, manually obtain the healthy forest image, and calculate the grayscale average of each pixel in the healthy forest image , the grayscale and Difference is calculated, and the maximum value of the above difference is calculated. If the maximum value of the above difference is less than the health threshold, the low-risk first-level effective forest farm image is healthy; if the maximum value of the above difference is greater than or equal to the health threshold, the low-risk first-level effective forest farm image is medium risk; S420: If the first-level effective forest farm image is of medium or high risk, the positions of the first-level effective forest farm images of medium or high risk are clustered to obtain the risk clustering area, and the drone performs secondary patrol of the risk clustering area.

5. A Beidou-based multi-level UAV adaptive patrol system, characterized in that: Includes the following modules: Data acquisition module: The drone is equipped with a Beidou receiver and a depth camera. The Beidou receiver obtains the drone's location, and the depth camera obtains forest images. Consistency judgment module: The command center divides the forest farm into areas and sets the starting patrol points for each area. When the drone arrives near the forest farm, the starting patrol point and the drone position are used to make a position consistency judgment. If the positions are consistent, the drone performs a first-level patrol and obtains a first-level forest farm image. If the positions are inconsistent, the drone continues to move to the starting patrol point. Risk judgment module: The position of the first-level forest farm image is obtained from the first-level forest farm image, and an adaptive screening algorithm is established based on the position of the first-level forest farm image to obtain a first-level effective forest farm image. A depth-grayscale dual threshold monitoring model is established based on the first-level effective forest farm image. The depth-grayscale dual threshold monitoring model is used to determine whether the forest farm has a risk of pine wood nematode infection, and complete the risk judgment of the first-level effective forest farm image; Secondary patrol module: If the first-level effective forest farm image is of low risk, a secondary health check is performed; if the first-level effective forest farm image is of medium or high risk, the positions of the first-level effective forest farm images of medium or high risk are clustered to obtain the risk clustering area. The drone performs secondary patrols on the risk clustering area. The command center monitors the secondary forest farm images collected by the secondary patrols in real time, sets key risk areas based on the number of suspicious pixels in the secondary forest farm images, and conducts manual intervention on the key risk areas. To realize a Beidou-based multi-level UAV adaptive patrol method as described in any one of claims 1-4.

Images