Farmland health monitoring system and method based on cooperation of unmanned aerial vehicle and quadruped robot

Abstract

The invention relates to the field of farmland health monitoring, in particular to a farmland health monitoring system and method based on cooperation of an unmanned aerial vehicle and a quadruped robot. The method comprises the steps of firstly obtaining vegetation health indexes of pixel points according to pixel values in a multispectral image collected by an unmanned aerial vehicle, performing grid division on a farmland to be detected, screening out an abnormal grid, then collecting detection data of different parameters of each sampling point of the abnormal grid by using a quadruped robot, and calculating the vegetation health indexes of the pixel points according to the detection data. According to vegetation health indexes of local pixel points of sampling points of the abnormal grids, local health coefficients of the sampling points are obtained, candidate parameters are preliminarily screened out, then regression analysis is carried out on the local health coefficients of all the sampling points and detection data of all the candidate parameters, and the health influence weight of each candidate parameter is obtained; and further obtaining dominant influence factors of the abnormal grid. According to the invention, a mode of cooperation of the unmanned aerial vehicle and the quadruped robot is adopted, so that the problem that high-efficiency and high-precision health monitoring of the farmland cannot be considered at the same time is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of farmland health monitoring, and in particular to a farmland health monitoring system and method based on cooperation of unmanned aerial vehicles and quadruped robots. BACKGROUND

[0002] Health monitoring of farmland is the core of the transformation from traditional agriculture to modern and intelligent agriculture. A modern agricultural management system that is efficient, green and sustainable is built for farmland health monitoring through modern technologies such as remote sensing and the Internet of Things, to ensure the most stable output at the minimum environmental cost. Through monitoring, threats such as diseases and pests, drought, floods, and frost damage can be discovered and warned in a timely manner, and measures can be taken in advance to minimize losses and ensure the stability and sustainability of grain production. Healthy soil and crops can produce safe and high-quality agricultural products.

[0003] In related technologies, an unmanned aerial vehicle is usually used alone for large-area farmland cruising, which has the advantages of high efficiency and wide field of view, but lacks the ability to capture detailed information such as crop roots, leaf backs, and soil micro-conditions. A ground robot (such as a quadruped, tracked, or wheeled robot) is used alone for local fine monitoring, which has the advantage of collecting high-precision data at close range and multiple angles, but has the disadvantage of extremely low efficiency for large-area monitoring, resulting in the inability of existing technologies to simultaneously achieve efficient and high-precision health monitoring of farmland. SUMMARY

[0004] To solve the technical problem that existing technologies cannot simultaneously achieve efficient and high-precision health monitoring of farmland, the present application aims to provide a farmland health monitoring system and method based on cooperation of unmanned aerial vehicles and quadruped robots, and the technical solution adopted is as follows:

[0005] The present application proposes a farmland health monitoring method based on cooperation of unmanned aerial vehicles and quadruped robots, which comprises:

[0006] Collecting multispectral images of the farmland to be tested using an unmanned aerial vehicle;

[0007] Obtaining a vegetation health index of a pixel point according to the pixel value of the pixel point in the multispectral image; dividing the farmland to be tested into grids, and selecting abnormal grids according to the vegetation health index of the pixel points in the grids in the multispectral image;

[0008] Collect detection data of different parameters of each sampling point of the target abnormal grid by the quadruped robot, obtain a local health coefficient of the sampling point according to a vegetation health index of a pixel point in a preset neighborhood of the sampling point of the multispectral image in the target abnormal grid; analyze the correlation between the local health coefficients of all sampling points and the detection data of the parameters, and select candidate parameters; perform regression analysis on the local health coefficients of all sampling points and the detection data of the candidate parameters, and obtain a health influence weight of each candidate parameter;

[0009] According to the health influence weight, a dominant influence factor of the target abnormal grid is obtained.

[0010] Further, the obtaining of the vegetation health index of each pixel point comprises:

[0011] The pixel value of each pixel point in the multispectral image at least comprises a near-infrared band reflectance value, a red light band reflectance value and a green light band reflectance value;

[0012] The difference between the near-infrared band reflectance value and the red light band reflectance value of each pixel point is taken as the numerator, the sum of the near-infrared band reflectance value and the red light band reflectance value of each pixel point is taken as the denominator, and the ratio is taken as the normalized difference vegetation index of each pixel point;

[0013] The difference between the near-infrared band reflectance value and the green light band reflectance value of each pixel point is taken as the numerator, the sum of the near-infrared band reflectance value and the green light band reflectance value of each pixel point is taken as the denominator, and the ratio is taken as the greenness normalized difference vegetation index of each pixel point;

[0014] The difference between the green light band reflectance value and the near-infrared band reflectance value of each pixel point is taken as the numerator, the sum of the near-infrared band reflectance value and the green light band reflectance value of each pixel point is taken as the denominator, and the ratio is taken as the normalized difference water index of each pixel point;

[0015] The normalized difference vegetation index, the greenness normalized difference vegetation index and the normalized difference water index of each pixel point are weighted and summed to obtain the vegetation health index of each pixel point, wherein the weight of the normalized difference water index is a negative number.

[0016] Further, the screening of the abnormal grid from all grids comprises:

[0017] The average value of the vegetation health index of all pixel points in each grid in the multispectral image is normalized to obtain a vegetation health evaluation value of each grid;

[0018] Based on the vegetation health evaluation value, an abnormal grid is screened from all grids.

[0019] Furthermore, the step of filtering out abnormal grids from all grids based on the vegetation health assessment value includes:

[0020] The vegetation health assessment values ​​of all grids are input into the Otsu threshold segmentation algorithm, which outputs the optimal threshold.

[0021] Grids whose vegetation health assessment values ​​are less than the optimal threshold are considered abnormal grids.

[0022] Furthermore, obtaining the local health coefficient for each sampling point includes:

[0023] The average value of the vegetation health index of all pixels in a preset neighborhood of each sampling point in the target anomaly grid of the multispectral image is used as the local health coefficient of each sampling point.

[0024] Furthermore, the candidate parameters for selecting the target anomaly mesh from all parameters include:

[0025] The sequence of local health coefficients of all sampling points of the target anomaly grid is taken as the local health coefficient sequence of the target anomaly grid;

[0026] Using any parameter as the target parameter, the sequence of detection data of the target parameter of all sampling points of the target anomaly grid is taken as the detection data sequence of the target anomaly grid with respect to the target parameter, wherein the values ​​at the same position in the local health coefficient sequence and the detection data sequence correspond to the same sampling point.

[0027] The absolute value of the Pearson correlation coefficient between the local health coefficient sequence of the target anomaly grid and the detection data sequence with respect to the target parameter is used as the correlation parameter of the target parameter in the target anomaly grid.

[0028] In the target anomaly network, parameters whose correlation parameters are greater than a preset correlation threshold are selected as candidate parameters.

[0029] Furthermore, the health impact weights for each candidate parameter of the target anomaly grid include:

[0030] Construct a multiple linear regression model for the target anomaly grid;

[0031] Using the least squares method, the local health coefficients of all sampling points of the target abnormal grid and the detection data of each candidate parameter are used as a sample set, and the multiple linear regression model is fitted. The absolute value of the regression coefficient of each candidate parameter in the fitted multiple linear regression model is used as the health influence weight of each candidate parameter.

[0032] Furthermore, the multiple linear regression model for the target anomaly grid is as follows:

[0033]

[0034] in, This represents the dependent variable, namely the local health coefficient of the sampling points of the target abnormal grid; Indicates the first The nth independent variable, i.e., the nth target anomaly grid Detection data for each candidate parameter; The term to be solved is the first... The regression coefficients of the independent variables; Represents the intercept term; Indicates the error term; This indicates the number of candidate parameters for the target anomaly mesh.

[0035] Furthermore, the dominant influencing factors for obtaining the target anomaly mesh include:

[0036] In the target anomaly grid, the candidate parameter corresponding to the maximum value of the health impact weight is taken as the dominant influencing factor of the target anomaly grid.

[0037] The present invention also proposes a farmland health monitoring system based on the collaboration of unmanned aerial vehicles (UAVs) and quadruped robots. The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the steps of a farmland health monitoring method based on the collaboration of unmanned aerial vehicles (UAVs) and quadruped robots.

[0038] The present invention has the following beneficial effects:

[0039] This invention addresses the limitation of existing technologies in simultaneously achieving high efficiency and high precision in farmland health monitoring. Therefore, it combines drones and quadruped robots. First, leveraging the drone's high efficiency and wide field of view, it rapidly acquires multispectral images of the farmland under test. The acquired vegetation health index reflects the health status of crops at each pixel location in the image. Simultaneously, considering the varying health conditions in different areas of the farmland, the area is divided into grids. Based on the vegetation health index of each pixel within each grid, abnormal grids with health problems are quickly identified. Subsequent monitoring tasks are then focused on these abnormal grids, thus improving the efficiency of farmland monitoring. To further enhance the precision of farmland health monitoring, this invention first targets the abnormal grids... Multiple sampling points are evenly set up, and a quadruped robot is used to accurately collect the location of each sampling point in the target abnormal grid and the detection data of various factors affecting crop health, i.e., parameter detection data. The obtained local health coefficient reflects the health status of the local area around each sampling point, and some influencing factors related to the crop health of the target abnormal grid are initially screened out, i.e., candidate parameters, reducing the number of parameters in subsequent regression analysis and improving computational efficiency. The obtained health influence weight reflects the degree of influence of each candidate parameter on the health of the target abnormal grid area, thereby screening out the dominant influencing factors of the target abnormal grid. Subsequently, the health problems of the target abnormal grid area can be quickly and accurately improved by targeting the dominant influencing factors, thereby improving the efficiency and accuracy of farmland health monitoring. Attached Figure Description

[0040] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart illustrating a method for monitoring farmland health based on the collaboration of a drone and a quadruped robot, provided in one embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of a grid and the distribution of sampling points in an abnormal grid, provided as an embodiment of the present invention. Detailed Implementation

[0043] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a farmland health monitoring system and method based on the collaboration of a drone and a quadruped robot proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0045] The following description, in conjunction with the accompanying drawings, details the specific scheme of a farmland health monitoring system and method based on the collaboration of an unmanned aerial vehicle (UAV) and a quadruped robot provided by the present invention.

[0046] Please see Figure 1 The diagram illustrates a flowchart of a farmland health monitoring method based on the collaboration of a drone and a quadruped robot, according to an embodiment of the present invention. The method includes:

[0047] Step S1: Use a drone to collect multispectral images of the farmland to be tested.

[0048] In the process of using drones and quadruped robots in a collaborative manner to monitor the health of farmland, drones can conduct large-scale patrols to quickly obtain the overall situation of the farmland and use multispectral and other sensors to identify potential "abnormal areas". At the same time, based on the abnormal area information provided by the drones, quadruped robots can move to designated locations to collect detailed data from close range and multiple angles, verify anomalies and obtain detailed diagnostic information, thereby achieving rapid and accurate health monitoring of farmland.

[0049] Therefore, in this embodiment of the invention, a drone equipped with a multispectral camera is first used to take pictures of the farmland to be tested according to a planned cruise path, thereby acquiring multispectral images of the farmland to be tested. The pixel of the multispectral image is a data vector containing multiple values ​​(usually more than 3), each value corresponding to the reflectance value in a specific and narrower wavelength range (called a "band" or "channel"). For example, a pixel in a multispectral image simultaneously includes the reflectance values ​​of the green light band, red light band, and near-infrared band. Subsequently, the pixel information in the spectral image can be used to analyze the health status of the crops in the farmland to be tested.

[0050] Step S2: Obtain the vegetation health index of the pixels based on the pixel values ​​in the multispectral image; divide the farmland to be tested into grids, and filter out abnormal grids based on the vegetation health index of the pixels in the grids in the multispectral image.

[0051] The pixel values ​​in a multispectral image contain information about the health status of the farmland to be tested. Therefore, the vegetation health index of each pixel is obtained based on the pixel value of each pixel in the multispectral image. The vegetation health index reflects the health of the crops in the farmland to be tested at each pixel location in the image.

[0052] Preferably, in one embodiment of the present invention, the method for obtaining the vegetation health index of each pixel specifically includes:

[0053] The difference between the near-infrared band reflectance and the red band reflectance of each pixel is used as the numerator, and the sum of the near-infrared band reflectance and the red band reflectance of each pixel is used as the denominator. The ratio is used as the normalized differential vegetation index for each pixel. The normalized differential vegetation index mainly reflects the vegetation coverage. The larger the value, the denser and healthier the vegetation or crops at that pixel location are.

[0054] The difference between the near-infrared band reflectance value and the green band reflectance value of each pixel is used as the numerator, and the sum of the near-infrared band reflectance value and the green band reflectance value of each pixel is used as the denominator. The ratio is used as the normalized difference vegetation index (NDVI) of greenness for each pixel. The normalized difference vegetation index mainly reflects the chlorophyll concentration at the canopy level. The larger the value, the higher the chlorophyll content and the healthier the crops.

[0055] The difference between the green light band reflectance and the near-infrared band reflectance of each pixel is used as the numerator, and the sum of the near-infrared band reflectance and the green light band reflectance of each pixel is used as the denominator. The ratio is used as the normalized differential water index for each pixel. The normalized differential water index is used to reflect the water content of plant leaves and canopy.

[0056] The vegetation health index of each pixel is obtained by weighted summing of the normalized difference vegetation index, the normalized difference vegetation index of greenness, and the normalized difference water index, where the weight of the normalized difference water index is negative.

[0057] In one embodiment of the present invention, the expression for the vegetation health index of each pixel can be, for example, as follows:

[0058]

[0059] in, Indicates the first The vegetation health index per pixel; Indicates the first Normalized differential vegetation index for each pixel; Indicates the first Greenness normalized difference vegetation index per pixel; Indicates the first Normalized differential water index for each pixel; , and These represent three weights, where, and The range of values ​​is , The range of values ​​is In one embodiment of the present invention, the following is used: Set to 0.7, Set it to 0.5, Set to -0.2, , and Adjustments can be made dynamically based on the actual situation, and no restrictions are set here. For example, when the focus is on nutrition and healthy growth, and It should be set relatively large. The absolute value should be set relatively small, especially when the focus is on the health of water stress. and It should be set relatively small. The absolute value should be set relatively large.

[0060] in, It was set to a negative number because Larger water content does not necessarily equate to "healthier." It is necessary to distinguish between "normal physiological water content" and "pathological water content changes caused by stress" so that the system can accurately detect these abnormalities with "inconsistent" spectral characteristics caused by diseases, avoiding masking disease symptoms and causing missed detection.

[0061] Considering the varying health conditions in different areas of the farmland to be tested, it is necessary to divide the farmland into grids. The farmland is divided into multiple grids evenly. In an embodiment of the present invention, the grid shape is square, and the side length of each side is usually 20 to 50 meters. In one embodiment of the present invention, the side length of the grid is set to 20 meters, but this is not limited here. Then, based on the vegetation health index of each pixel in each grid, abnormal grids with crop health problems are quickly screened out, so that subsequent monitoring tasks are only carried out on abnormal grids, thereby improving the efficiency of farmland monitoring.

[0062] Preferably, in one embodiment of the present invention, the method for obtaining abnormal meshes specifically includes:

[0063] The average vegetation health index of all pixels in each grid of the multispectral image is normalized, and the calculation results are limited to... This allows us to obtain the vegetation health assessment value for each grid. The vegetation health assessment value reflects the overall health status of the vegetation or crops in each grid area. The higher the vegetation health assessment value of a grid, the healthier the vegetation or crops in that grid area are. Conversely, it indicates that there is a greater possibility of health abnormalities in that grid area. Therefore, based on the vegetation health assessment value, abnormal grids can be screened out from all grids.

[0064] Preferably, in one embodiment of the present invention, the vegetation health assessment values ​​of all grids can be input into the Otsu threshold segmentation algorithm, and the Otsu threshold segmentation algorithm outputs the optimal threshold. Grids with vegetation health assessment values ​​less than the optimal threshold are regarded as abnormal grids. The Otsu threshold segmentation algorithm is a technical means well known to those skilled in the art and will not be described in detail here.

[0065] In one embodiment of the present invention, the normalization process can be specifically, for example, maximum and minimum value normalization. Furthermore, the normalization in subsequent steps can all adopt maximum and minimum value normalization. In other embodiments of the present invention, other normalization methods can be selected according to the specific range of values, or activation functions and hyperbolic tangent functions can be used to implement the normalization process. These will not be elaborated or limited further.

[0066] Step S3: Take any abnormal grid as the target abnormal grid, and use a quadruped robot to collect detection data of different parameters at each sampling point of the target abnormal grid. Based on the vegetation health index of the pixels in the preset neighborhood of the sampling point of the target abnormal grid in the multispectral image, obtain the local health coefficient of the sampling point; analyze the correlation between the local health coefficient of all sampling points and the detection data of each parameter, and screen out candidate parameters; perform regression analysis on the local health coefficient of all sampling points and the detection data of each candidate parameter to obtain the health influence weight of each candidate parameter.

[0067] Once the abnormal grids containing health issues are obtained, each abnormal grid can be analyzed. First, any one abnormal grid is selected as the target abnormal grid, and multiple sampling points are evenly distributed within the target abnormal grid. (See [link to relevant documentation]). Figure 2 This illustration shows a grid and a schematic diagram of the distribution of sampling points in an abnormal grid provided by an embodiment of the present invention. Then, a quadruped robot is used to accurately collect detection data of various factors affecting crop health, i.e., parameter detection data, at close range and from multiple angles at each sampling point of the target abnormal grid. The quadruped robot can be equipped with a high-definition camera and multi-parameter sensors to collect detection data of various parameters. Specific parameters may include, for example, the proportion of lesions, soil compaction, soil moisture, and the number of insect eggs on the back of leaves. Generally, in order to monitor farmland health in a comprehensive manner, a large number of parameters need to be collected. The number and types of parameters are usually between 10 and 30. The types and numbers of parameters can be set by the implementer according to the actual scenario and are not limited here.

[0068] In order to accurately identify the main factors affecting crop health in the target abnormal grid area, this embodiment of the invention first obtains the local health coefficient of each sampling point based on the vegetation health index of pixels in the preset neighborhood of each sampling point in the target abnormal grid. The local health coefficient reflects the health status of vegetation or crops in the local area around each sampling point. Subsequently, by combining the local health coefficient and parameter detection data of each sampling point, parameters that have a more important impact on crop health in the target abnormal grid area can be accurately identified.

[0069] Preferably, in one embodiment of the present invention, the method for obtaining the local health coefficient of each sampling point specifically includes:

[0070] The average vegetation health index of all pixels in the preset neighborhood of each sampling point in the target anomaly grid of the multispectral image is used as the local health coefficient of each sampling point. The preset neighborhood of each sampling point can be specifically an 8-neighborhood or a 24-neighborhood centered on each sampling point, etc., which is not limited here.

[0071] Among the numerous parameters, there are some that are unrelated to the crop health status of the target abnormal grid area. Therefore, this embodiment of the invention needs to analyze the correlation between the detection data of each parameter of the sampling points of the target abnormal grid and the local health coefficient, and preliminarily screen out some influencing factors related to the crop health of the target abnormal grid, namely candidate parameters, so as to reduce the number of parameters in the subsequent regression analysis and improve the computational efficiency.

[0072] Preferably, in one embodiment of the present invention, the method for obtaining candidate parameters of the target anomaly mesh specifically includes:

[0073] The sequence of local health coefficients of all sampling points of the target abnormal grid is taken as the local health coefficient sequence of the target abnormal grid.

[0074] Using any parameter as the target parameter, the sequence of detection data of the target parameter at all sampling points of the target anomaly grid is taken as the detection data sequence of the target anomaly grid with respect to the target parameter. In this sequence, the values ​​at the same position in the local health coefficient sequence and the detection data sequence correspond to the same sampling point.

[0075] The absolute value of the Pearson correlation coefficient between the local health coefficient sequence of the target anomaly grid and the detection data sequence about the target parameter is used as the correlation parameter of the target parameter in the target anomaly grid. In other embodiments of the present invention, Spearman's rank correlation coefficient or Kendall's rank correlation coefficient may also be used, and there is no limitation here.

[0076] The correlation parameter for each parameter can be obtained using the same method described above. A higher correlation parameter indicates a stronger correlation between that parameter and the health status of crops in the target anomaly grid area, meaning it is more likely to influence crop health. Therefore, in the target anomaly network, parameters with correlation parameters greater than a preset correlation threshold can be used as candidate parameters. The preset correlation threshold ranges from [value missing]. In one embodiment of the present invention, the preset correlation threshold is set to 0.5. The preset correlation threshold can also be set by the implementer according to the specific implementation scenario, and is not limited here.

[0077] In the process of screening candidate parameters using correlation, some parameters with low correlation to the health status of crops in the target abnormal grid area were removed. Therefore, regression analysis can be further performed on the local health coefficients of all sampling points of the target abnormal grid and the detection data of each candidate parameter to obtain the health impact weight of each candidate parameter in the target abnormal grid. The health impact weight reflects the degree of influence of each candidate parameter on the health of the target abnormal grid area. Subsequently, based on the health impact weight, the main factors affecting the health of crops in the target abnormal grid area can be screened from the candidate parameters.

[0078] Preferably, in one embodiment of the present invention, the method for obtaining the health impact weight of each candidate parameter of the target anomaly grid specifically includes:

[0079] First, a multiple linear regression model for the target anomaly grid is constructed, wherein the multiple linear regression model is as follows:

[0080]

[0081] in, This represents the dependent variable, namely the local health coefficient of the sampling points of the target abnormal grid; Indicates the first The nth independent variable, i.e., the nth target anomaly grid Detection data for each candidate parameter; The term to be solved is the first... The regression coefficients of the independent variables; Represents the intercept term; Indicates the error term; This indicates the number of candidate parameters for the target anomaly mesh.

[0082] Then, using the least squares method, the local health coefficients of all sampling points of the target abnormal grid and the detection data of each candidate parameter are used as the sample set, and the multiple linear regression model is fitted. The absolute value of the regression coefficient of each candidate parameter in the fitted multiple linear regression model is used as the health influence weight of each candidate parameter. The use of the least squares method for multiple linear regression analysis is a well-known technique in the art and will not be elaborated here.

[0083] Step S4: Based on the health impact weights, obtain the dominant influencing factors of the target abnormal grid.

[0084] For a target anomalous grid, the greater the health impact weight of a candidate parameter, the more likely the change in the value of that candidate parameter is to affect the health status of crops in the target anomalous grid area. Therefore, based on the health impact weight of each candidate parameter, the dominant influencing factor of the target anomalous grid can be obtained, thereby enabling a rapid and accurate investigation of the health influencing factors of crops in the target anomalous grid area.

[0085] Preferably, in one embodiment of the present invention, the method for obtaining the dominant influencing factors of the target abnormal grid specifically includes:

[0086] In the target anomaly grid, the candidate parameter corresponding to the maximum value of the health impact weight is taken as the dominant influencing factor of the target anomaly grid. For example, after the above steps, it was found that soil moisture is the dominant influencing factor of the target anomaly grid, indicating that the main reason for the crop health problem in the target anomaly grid area is the lack of water in the area.

[0087] By using the same method described above, the dominant influencing factors of each abnormal grid in the farmland under test can be obtained. Then, by targeting the dominant influencing factors, the health problems of the abnormal grid area can be improved quickly and accurately, thereby improving the efficiency and accuracy of farmland health monitoring.

[0088] In other embodiments of the present invention, the health impact weights of each candidate parameter of the abnormal grid can be sorted, so as to focus on the candidate parameters with larger health impact weights according to different priorities or degrees. For example, based on the health impact weights of the candidate parameters, the factors affecting the health of crops in the abnormal grid can be improved according to different resource ratios or intensities.

[0089] One embodiment of the present invention provides a farmland health monitoring system based on the collaboration of a drone and a quadruped robot. The system includes a memory, a processor and a computer program, wherein the memory is used to store the corresponding computer program, the processor is used to run the corresponding computer program, and the computer program can implement the methods described in steps S1 to S4 when running in the processor.

[0090] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for monitoring farmland health based on the collaboration of unmanned aerial vehicles (UAVs) and quadruped robots, characterized in that, The method includes: Using drones to collect multispectral images of farmland under test; The vegetation health index of each pixel is obtained based on the pixel value of the multispectral image; the farmland to be tested is divided into grids, and abnormal grids are selected based on the vegetation health index of the pixels in the grids in the multispectral image. Using any one abnormal grid as the target abnormal grid, a quadruped robot is used to collect detection data of different parameters at each sampling point of the target abnormal grid. Based on the vegetation health index of pixels in the preset neighborhood of the sampling point of the target abnormal grid in the multispectral image, the local health coefficient of the sampling point is obtained. The correlation between the local health coefficient of all sampling points and the detection data of each parameter is analyzed to screen out candidate parameters. Regression analysis is performed on the local health coefficient of all sampling points and the detection data of each candidate parameter to obtain the health influence weight of each candidate parameter. Based on the health impact weights, the dominant influencing factors of the target abnormal grid are obtained.

2. The method for monitoring farmland health based on the collaboration of unmanned aerial vehicles and quadruped robots according to claim 1, characterized in that, The method for obtaining the vegetation health index for each pixel includes: The pixel value of each pixel in the multispectral image includes at least the near-infrared band reflectance value, the red band reflectance value, and the green band reflectance value; The difference between the near-infrared band reflectance value and the red band reflectance value of each pixel is used as the numerator, and the sum of the near-infrared band reflectance value and the red band reflectance value of each pixel is used as the denominator. The ratio is used as the normalized differential vegetation index of each pixel. The difference between the near-infrared band reflectance value and the green band reflectance value of each pixel is used as the numerator, and the sum of the near-infrared band reflectance value and the green band reflectance value of each pixel is used as the denominator. The ratio is used as the greenness normalized differential vegetation index of each pixel. The difference between the green band reflectance value and the near-infrared band reflectance value of each pixel is used as the numerator, and the sum of the near-infrared band reflectance value and the green band reflectance value of each pixel is used as the denominator. The ratio is used as the normalized differential water index of each pixel. The vegetation health index of each pixel is obtained by weighted summing of the normalized differential vegetation index, the normalized differential vegetation greenness index, and the normalized differential water index, wherein the weight of the normalized differential water index is negative.

3. The method for monitoring farmland health based on the collaboration of unmanned aerial vehicles and quadruped robots according to claim 1, characterized in that, The process of filtering out abnormal grids from all grids includes: The average vegetation health index of all pixels in each grid of the multispectral image is normalized to obtain the vegetation health assessment value of each grid. Based on the vegetation health assessment values, abnormal grids are selected from all grids.

4. The method for monitoring farmland health based on the collaboration of unmanned aerial vehicles and quadruped robots according to claim 3, characterized in that, The process of filtering out abnormal grids from all grids based on the vegetation health assessment value includes: The vegetation health assessment values ​​of all grids are input into the Otsu threshold segmentation algorithm, which outputs the optimal threshold. Grids whose vegetation health assessment values ​​are less than the optimal threshold are considered abnormal grids.

5. A method for monitoring farmland health based on the collaboration of a drone and a quadruped robot according to claim 1, characterized in that, The process of obtaining the local health coefficient for each sampling point includes: The average value of the vegetation health index of all pixels in a preset neighborhood of each sampling point in the target anomaly grid of the multispectral image is used as the local health coefficient of each sampling point.

6. A method for monitoring farmland health based on the collaboration of a drone and a quadruped robot according to claim 1, characterized in that, The candidate parameters for selecting the target anomaly grid from all parameters include: The sequence of local health coefficients of all sampling points of the target anomaly grid is taken as the local health coefficient sequence of the target anomaly grid; Using any parameter as the target parameter, the sequence of detection data of the target parameter of all sampling points of the target anomaly grid is taken as the detection data sequence of the target anomaly grid with respect to the target parameter, wherein the values ​​at the same position in the local health coefficient sequence and the detection data sequence correspond to the same sampling point. The absolute value of the Pearson correlation coefficient between the local health coefficient sequence of the target anomaly grid and the detection data sequence with respect to the target parameter is used as the correlation parameter of the target parameter in the target anomaly grid. In the target anomaly network, parameters whose correlation parameters are greater than a preset correlation threshold are selected as candidate parameters.

7. A method for monitoring farmland health based on the collaboration of a drone and a quadruped robot according to claim 1, characterized in that, The health impact weights for each candidate parameter of the obtained target anomaly grid include: Construct a multiple linear regression model for the target anomaly grid; Using the least squares method, the local health coefficients of all sampling points of the target abnormal grid and the detection data of each candidate parameter are used as a sample set, and the multiple linear regression model is fitted. The absolute value of the regression coefficient of each candidate parameter in the fitted multiple linear regression model is used as the health influence weight of each candidate parameter.

8. A method for monitoring farmland health based on the collaboration of a drone and a quadruped robot according to claim 7, characterized in that, The multiple linear regression model for the target anomaly grid is as follows: in, This represents the dependent variable, namely the local health coefficient of the sampling points of the target abnormal grid; Indicates the first The nth independent variable, i.e., the nth target anomaly grid Detection data for each candidate parameter; The term to be solved is the first... The regression coefficients of the independent variables; Represents the intercept term; Indicates the error term; This indicates the number of candidate parameters for the target anomaly mesh.

9. A method for monitoring farmland health based on the collaboration of an unmanned aerial vehicle (UAV) and a quadruped robot according to claim 1, characterized in that, The dominant influencing factors for obtaining the target anomaly grid include: In the target anomaly grid, the candidate parameter corresponding to the maximum value of the health impact weight is taken as the dominant influencing factor of the target anomaly grid.

10. A farmland health monitoring system based on the collaboration of a drone and a quadruped robot, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 9.

Images