Method and system for detecting damage caused by important tobacco pests based on drone patrol

Abstract

The present invention relates to the technical field of tobacco pest detection, in particular to a method and system for detecting damage caused by important tobacco pests based on drone cruise. The parallax map of each target tobacco planting area is obtained by drone cruise, a global three-dimensional model is constructed based on the parallax map of each target tobacco planting area, the global three-dimensional model is analyzed in the field of matrix grid registration, and the optimal cruise plan of the drone is planned in combination with the growth height difference and the terrain height difference; the notch contour Fourier descriptor of the surface of all tobacco plants in each sub-field is obtained by executing the drone cruise plan, and the notch contour Fourier descriptor is imported into the notch contour index array mapping table for index matching, so as to draw the dynamics of the damage symptoms of the distribution map of important tobacco pests and evaluate them. The present invention can perform fast and efficient damage degree analysis on important tobacco pests based on drone cruise, so as to improve the damage detection rate and accuracy of important tobacco pests to the target tobacco planting area.

Description

Technical Field

[0001] The present invention relates to the technical field of tobacco pest detection, and in particular to a method and system for detecting damage caused by important tobacco pests based on unmanned aerial vehicle cruise. Background Art

[0002] The main types of pests in the tobacco planting process are tobacco budworms and fall armyworms. Both will eat tobacco leaves, rhizomes and buds at different feeding rates at different growth stages, resulting in irregular holes and notches on the tobacco surface, reducing the photosynthetic area of ​​the leaves, reducing the overall health of the tobacco and photosynthetic efficiency, and thus hindering yield growth. The traditional drone cruise hazard detection method usually maintains a uniform height for shooting tobacco plants, making it impossible to maintain consistent resolution of surface notch images, reducing the quality of image analysis; and the traditional hazard detection method can only detect the distribution or quantity of important tobacco pests, making it difficult to achieve simultaneous detection of quantity and distribution, and unable to analyze the hazard by continuously detecting the pests' tolerance to pesticides, resulting in low accuracy and efficiency of important tobacco pest hazard detection, and large errors in the detection results, which cannot meet the actual needs of important tobacco pest hazard detection. Summary of the invention

[0003] The present invention overcomes the shortcomings of the prior art and provides a method and system for detecting damage caused by important tobacco pests based on drone patrol.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] The first aspect of the present invention provides a method for detecting tobacco major pests based on drone patrol, comprising the following steps:

[0006] The disparity map of each target tobacco planting area is obtained through drone cruising, and a global three-dimensional model is constructed based on the disparity map of each target tobacco planting area. After fitting the global three-dimensional model in the matrix grid registration field, the mean shift clustering algorithm and the growth height difference are combined to determine the growth stage of tobacco plants in each sub-field, so as to plan the drone's cruising behavior and obtain the drone's optimal cruising plan.

[0007] The drone cruise scheme is executed by the drone to obtain image data of the surfaces of all tobacco plants in each sub-field and perform feature description, so as to obtain the notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field, and the notch contour Fourier descriptors are imported into the notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and perform evaluation, so as to obtain a first hazard detection result;

[0008] Based on the first hazard detection result, the remote sensing camera of the drone is controlled to perform cruise shooting to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain a second hazard detection result;

[0009] By spraying the preset pesticide after deployment in the target tobacco planting area and then controlling the drone to conduct a second cruise shooting, several second tobacco plant surface notch time-series images in each sub-field within the second target time period are obtained. The area value of the area after the preset pesticide is deployed is analyzed by combining the corresponding first tobacco plant surface notch time-series images, and the third hazard detection result is obtained.

[0010] Furthermore, in a preferred embodiment of the present invention, the disparity map of each target tobacco planting area is obtained by cruising the drone, a global three-dimensional model is constructed based on the disparity map of each target tobacco planting area, and the growth stage of the tobacco plants in each sub-area is determined by combining the mean shift clustering algorithm and the growth height difference after fitting the global three-dimensional model in the matrix grid registration field, so as to plan the cruising behavior of the drone and obtain the optimal cruising plan of the drone, which specifically includes the following steps:

[0011] Obtaining a preset cruise mission of the drone, extracting one or more target tobacco planting areas that the drone needs to cruise through the preset cruise mission, and controlling the drone's depth camera to automatically cruise and photograph the target tobacco planting areas to obtain three-dimensional point cloud data of each target tobacco planting area;

[0012] The SGM dense matching algorithm is introduced, in which the three-dimensional point cloud data is calculated to extract M groups of multi-view feature points, and the disparity matching operation is performed on the M groups of multi-view feature points to generate the disparity map of each target tobacco planting area;

[0013] Aggregating the global path disparity of the disparity map of each target tobacco planting area under different disparity factors, converting the disparity map of each target tobacco planting area into a depth map based on the global path disparity, and reconstructing the model using the depth map to obtain a global three-dimensional model of each target tobacco planting area;

[0014] Constructing a matrix grid registration domain, obtaining tobacco plants planted in a target tobacco planting area, obtaining a maximum growth height of the tobacco plants in an optimal growth stage in a big data network based on the tobacco plants, and adjusting a domain size of the three-dimensional matrix grid registration domain based on the maximum growth height to generate an initial matrix grid registration domain of the tobacco plants in an optimal growth stage;

[0015] Importing the global three-dimensional model of each target tobacco planting area into the initial matrix grid registration field for alignment and registration, after the alignment and registration is completed, removing the aligned and registered area of ​​the global three-dimensional model in the initial matrix grid registration field to obtain an unaligned and registered initial matrix grid registration field;

[0016] Several growth height differences of tobacco plants under different preset growth stage conditions are obtained, and N sub-fields in the initial matrix grid registration field that are not properly registered are analyzed through each growth height difference, and a growth height difference distribution model for each sub-field is generated. The growth stage of tobacco plants in each sub-field is determined by combining the mean shift clustering algorithm and the growth height difference distribution model, so that the cruising behavior of the UAV is planned according to the growth stage of the tobacco plants in each sub-field, and the optimal cruising plan of the UAV is obtained.

[0017] Furthermore, in a preferred embodiment of the present invention, the method of obtaining several growth height differences of tobacco plants under different preset growth stage conditions, analyzing N sub-fields in the initial matrix grid registration field that is not aligned with the registration through each growth height difference, generating a growth height difference distribution model for each sub-field, and combining the mean shift clustering algorithm and the growth height difference distribution model to determine the growth stage of the tobacco plants in each sub-field, so that the cruising behavior of the UAV is planned according to the growth stage of the tobacco plants in each sub-field, and obtaining the optimal cruising plan of the UAV, specifically includes the following steps:

[0018] Based on the big data network, the growth height reference interval of the tobacco plant under different preset growth stage conditions is obtained, and the growth height difference between the optimal growth stage of the tobacco plant and each preset growth stage is calculated by combining the growth height reference interval and the maximum growth height, so as to obtain a plurality of growth height difference values;

[0019] Obtaining a preset planting interval between adjacent tobacco plants and a cruise visual range of the UAV, and dividing the initial matrix grid registration area that is not aligned into N sub-areas according to the preset planting interval and the cruise visual range;

[0020] In each sub-domain, a plurality of growth height difference values ​​corresponding to each preset growth stage are fitted and embedded, so that appropriate growth height difference values ​​are displayed in the initial matrix grid registration domain that is not properly registered, and a growth height difference distribution model of each sub-domain is generated;

[0021] The mean shift clustering algorithm is introduced, and multiple initial clustering centers are constructed in the mean shift clustering algorithm based on different preset growth stages, and the kernel density estimation value of several growth height differences corresponding to each preset growth stage at each initial clustering center is calculated to determine the mean shift vector;

[0022] The current position of each initial cluster center is updated and iterated according to the mean shift vector. If the position change amplitude of each initial cluster center decreases, the update iteration is stopped, and several growth height differences are assigned to the nearest centroid cluster to obtain the clustering results of each growth height difference in the corresponding preset growth stage;

[0023] Obtaining an image resolution threshold required for drone cruising, identifying a growth height difference distribution model for each subfield through clustering results of each growth height difference at a corresponding preset growth stage, and determining the growth stage of tobacco plants in each subfield;

[0024] The growth height difference corresponding to the growth stage of tobacco plants in each sub-field compared to the optimal growth stage is extracted, and the growth height difference is adjusted based on the image resolution threshold to obtain the optimal cruising height of the UAV for the tobacco plants in each sub-field. The cruising height and cruising angle of the UAV are planned according to the optimal cruising height to generate the optimal cruising plan of the UAV.

[0025] Furthermore, in a preferred embodiment of the present invention, the drone cruise scheme is executed by the drone to obtain image data of the surface of all tobacco plants in each sub-field and perform feature description, obtain the notch contour Fourier descriptor of the surface of all tobacco plants in each sub-field, import the notch contour Fourier descriptor into the notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and evaluate it, and obtain the first hazard detection result, which specifically includes the following steps:

[0026] The drone is set to execute the drone cruise scheme to automatically cruise and shoot the target tobacco planting area, so as to obtain image data of the surface of all tobacco plants in each sub-field;

[0027] A Laplace edge detection operator is introduced to calculate the second-order derivatives in the image data of the surfaces of all tobacco plants in each sub-field, and the notch contour feature vectors of the surfaces of all tobacco plants in each sub-field are extracted according to the second-order derivatives, and uniform discrete contour point coordinates are collected and marked on the notch contour feature vectors of the surfaces of all tobacco plants in each sub-field;

[0028] Constructing a complex sequence space, generating a real part and an imaginary part of each discrete contour point based on uniform discrete contour point coordinates, calculating and transforming the real part and the imaginary part of each discrete contour point in the complex sequence space to generate a complex sequence of discrete contour point coordinates, and performing Fourier transformation on the complex sequence of discrete contour point degrees to obtain the Fourier descriptors of the notched contours of all tobacco plant surfaces in each subfield;

[0029] Obtaining the chewing mouthparts hazard profiles of different important tobacco pest species at different preset growth stages through a big data network, and simultaneously obtaining the chewing notch profile Fourier descriptors formed by each chewing mouthparts hazard profile on the surface of the tobacco plant; wherein the different important tobacco pest species include tobacco budworm and Spodoptera litura;

[0030] K groups of index queues are established according to different types of important tobacco pests, and the index array quantity threshold of each group of index queues is preset according to the chewing mouthparts hazard profiles of the different important tobacco pests at different growth stages, and the corresponding index queue is divided into a plurality of index array spaces based on the index array quantity threshold;

[0031] Each of the chewing notch contour Fourier descriptors is correspondingly embedded in the space of the index array to obtain a notch contour index array mapping table, and the degree of fit between each notch contour Fourier descriptor and each chewing notch contour Fourier descriptor is calculated one by one. If the degree of fit is greater than a preset degree of fit, the chewing notch contour Fourier descriptor corresponding to the degree of fit greater than the preset degree of fit is extracted;

[0032] The chewing notch profile Fourier descriptors corresponding to the notch profiles with a matching degree greater than a preset matching degree are traversed and indexed through a notch profile index array mapping table, so as to output and mark the important tobacco pest species corresponding to the notch profile Fourier descriptors on the surfaces of all tobacco plants in each sub-field, and draw a distribution map of important tobacco pests;

[0033] The preset harm coefficient of each important tobacco pest species to tobacco plants is obtained, and the distribution map of important tobacco pests is evaluated based on the preset harm coefficient to obtain a first harm detection result.

[0034] Furthermore, in a preferred embodiment of the present invention, the remote sensing camera of the drone is controlled to perform cruise shooting based on the first hazard detection result to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain the second hazard detection result, which specifically includes the following steps:

[0035] A first target time period is preset, and based on the distribution map of important tobacco pest species, the remote sensing camera of the drone is controlled to cruise and shoot each sub-field, and a plurality of tobacco plant surface notch time-series images in each sub-field within the first target time period are obtained, which are marked as first tobacco surface notch time-series images;

[0036] The Horn-Schunck algorithm is introduced, and the partial differential optical flow equation is constructed in the Horn-Schunck algorithm based on the time-series images of each first tobacco plant surface notch in each sub-field within the first target time period, and the pixel continuity of the partial differential optical flow equation is analyzed to obtain the energy function;

[0037] The neighboring average value of the current optical flow in the time-series image of the first tobacco plant surface notches is calculated by an energy function, and the partial differential optical flow equation is iteratively solved based on the neighboring average value to continuously update the optical flow of the optical flow field until a preset optical flow field convergence threshold is reached, and finally a plurality of first tobacco plant surface notch time-series change models of each sub-field within the first target time period are output;

[0038] Obtain the actual growth stage and actual environmental factors of each sub-field in the first target time period, and obtain the feeding rate of each important tobacco pest species under the combination of the actual growth stage and actual environmental factors based on the big data network;

[0039] Constructing a naive Bayesian network, importing the temporal change models of the surface notches of the first tobacco plants into the naive Bayesian network for inference, obtaining the temporal change rates of the surface notches of the first tobacco plants, and calculating the ratio of the feeding rate to the temporal change rate of each surface notch of the first tobacco plants;

[0040] According to the ratio, the number of important tobacco pest species that cause the time series change model of notches on the surface of tobacco plants in each sub-field within the first target time period is determined, and the number of important tobacco pest species is evaluated based on the preset hazard coefficient to obtain a second hazard detection result.

[0041] Furthermore, in a preferred embodiment of the present invention, the drone is controlled to perform secondary cruise shooting after spraying the preset pesticide after deployment in the target tobacco planting area, and a plurality of second tobacco plant surface notch time-series images in each sub-field within the second target time period are obtained, and the area value of the area after the preset pesticide after deployment is analyzed in combination with the corresponding first tobacco plant surface notch time-series images to obtain the third hazard detection result, which specifically includes the following steps:

[0042] Obtain pesticide spraying records of the target tobacco planting area, and extract the historical killing rate of the preset pesticide on the number of important tobacco pest species in each sub-field within the first target time period based on the pesticide spraying records; if the historical killing rate is less than the preset killing rate, calculate the initial tolerance index of the important tobacco pest species to the preset pesticide in combination with the number of important tobacco pest species and the historical killing rate;

[0043] A second target time period is preset, and the formula and dosage of the preset pesticide are readjusted in combination with the first hazard detection result and the second hazard detection result to obtain the prepared preset pesticide, and the drone is controlled to spray the prepared preset pesticide onto the surface of the tobacco plants in each sub-field;

[0044] After the spraying is completed, the remote sensing camera of the UAV is used to conduct a second cruise shooting of each sub-field to obtain a number of tobacco surface notch time-series images in each sub-field within the second target time period, which are marked as the second tobacco plant surface notch time-series images;

[0045] Defining each of the first tobacco plant surface notch time-series images as a redundant region, and removing the redundant region from the corresponding second tobacco plant surface notch time-series image to obtain a second tobacco plant surface notch time-series image region after removal;

[0046] Divide the second target time period into a plurality of uniform target time stamps, obtain the area parameter of the time-sequential image region of the second tobacco plant surface notches after being removed on each uniform target time stamp, and calculate the ratio between the area parameters of the time-sequential image region of the second tobacco plant surface notches after being removed on adjacent uniform target time stamps;

[0047] The ratio between the area parameters of the removed second tobacco plant surface notch time-series image area on adjacent uniform target timestamps is fitted into a Bezier curve, and the slope of the Bezier curve is obtained. The adjusted preset pesticide tolerance index is calculated by combining the slope of the Bezier curve with the initial tolerance index, and the adjusted preset pesticide tolerance index is evaluated based on the preset hazard coefficient to obtain a third hazard detection result.

[0048] A second aspect of the present invention provides a tobacco major pest hazard detection system based on drone cruise, the tobacco major pest hazard detection system based on drone cruise comprising a memory and a processor, the memory storing a tobacco major pest hazard detection method program based on drone cruise, and when the tobacco major pest hazard detection method program based on drone cruise is executed by the processor, the following steps are implemented:

[0049] The disparity map of each target tobacco planting area is obtained through drone cruising, and a global three-dimensional model is constructed based on the disparity map of each target tobacco planting area. After fitting the global three-dimensional model in the matrix grid registration field, the mean shift clustering algorithm and the growth height difference are combined to determine the growth stage of tobacco plants in each sub-field, so as to plan the drone's cruising behavior and obtain the drone's optimal cruising plan.

[0050] The drone cruise scheme is executed by the drone to obtain image data of the surfaces of all tobacco plants in each sub-field and perform feature description, so as to obtain the notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field, and the notch contour Fourier descriptors are imported into the notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and perform evaluation, so as to obtain a first hazard detection result;

[0051] Based on the first hazard detection result, the remote sensing camera of the drone is controlled to perform cruise shooting to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain a second hazard detection result;

[0052] By spraying the preset pesticide after deployment in the target tobacco planting area and then controlling the drone to conduct a second cruise shooting, several second tobacco plant surface notch time-series images in each sub-field within the second target time period are obtained. The area value of the area after the preset pesticide is deployed is analyzed by combining the corresponding first tobacco plant surface notch time-series images, and the third hazard detection result is obtained.

[0053] The present invention solves the technical defects existing in the background technology, and the beneficial technical effects of the present invention are:

[0054] Obtain the disparity map of each target tobacco planting area through drone cruising, build a global three-dimensional model based on the disparity map of each target tobacco planting area, analyze the global three-dimensional model in the field of matrix grid registration and plan the drone's cruising behavior in combination with the growth height difference, and obtain the best cruise plan of the drone; obtain the notch contour Fourier descriptor of all tobacco surfaces in each sub-field through the drone executing the drone cruising plan, import the notch contour Fourier descriptor into the notch contour index array mapping table for index matching, so as to draw the distribution map of important tobacco pests and evaluate it, and obtain the first hazard detection result; control the remote sensing camera of the drone to cruise and shoot based on the first hazard detection result, so as to construct several first tobacco surface notch time series change models in each sub-field within the first target time period, and determine the number of important tobacco pest species in each sub-field that cause the tobacco surface notch time series change model by analyzing the first tobacco surface notch time series change model and evaluate it, and obtain the second hazard detection result. The present invention can quickly and efficiently analyze the degree of harm of important tobacco pests based on drone cruising, so as to improve the detection rate and accuracy of the harm of important tobacco pests to the target tobacco planting area. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, drawings of other embodiments can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A first method flow chart of a method for detecting tobacco major pests based on drone patrol is shown;

[0057] Figure 2 A second method flow chart of a method for detecting tobacco major pests based on drone patrol is shown;

[0058] Figure 3 A third method flow chart of a method for detecting tobacco major pests based on drone patrol is shown;

[0059] Figure 4 The system framework diagram of the tobacco major pest damage detection system based on drone patrol is shown. DETAILED DESCRIPTION

[0060] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0061] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0062] The first aspect of the present invention provides a method for detecting tobacco major pests based on drone patrol, such as Figure 1 As shown, the following steps are included:

[0063] S102: Obtain a disparity map of each target tobacco planting area through drone cruising, construct a global three-dimensional model based on the disparity map of each target tobacco planting area, and after fitting the global three-dimensional model in the matrix grid registration field, combine the mean shift clustering algorithm and the growth height difference to determine the growth stage of the tobacco plants in each sub-field, so as to plan the drone's cruising behavior and obtain the drone's optimal cruising plan;

[0064] S104: acquiring image data of the surfaces of all tobacco plants in each sub-field by executing the drone cruising scheme through the drone and performing feature description, obtaining notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field, importing the notch contour Fourier descriptors into a notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and perform evaluation, thereby obtaining a first hazard detection result;

[0065] S106: Based on the first hazard detection result, the remote sensing camera of the drone is controlled to perform cruise shooting to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain a second hazard detection result;

[0066] S108: After spraying the preset pesticide after deployment in the target tobacco planting area, the drone is controlled to conduct a second cruise shooting to obtain a number of second tobacco plant surface notch time-series images in each sub-field within the second target time period, and the control effect of the preset pesticide after deployment is analyzed by combining the area value after the corresponding first tobacco plant surface notch time-series image is eliminated to obtain the third hazard detection result.

[0067] Furthermore, in a preferred embodiment of the present invention, the disparity map of each target tobacco planting area is obtained by cruising the drone, a global three-dimensional model is constructed based on the disparity map of each target tobacco planting area, and the growth stage of the tobacco plants in each sub-area is determined by combining the mean shift clustering algorithm and the growth height difference after fitting the global three-dimensional model in the matrix grid registration field, so as to plan the cruising behavior of the drone and obtain the optimal cruising plan of the drone, which specifically includes the following steps:

[0068] Obtaining a preset cruise mission of the drone, extracting one or more target tobacco planting areas that the drone needs to cruise through the preset cruise mission, and controlling the drone's depth camera to automatically cruise and photograph the target tobacco planting areas to obtain three-dimensional point cloud data of each target tobacco planting area;

[0069] The SGM dense matching algorithm is introduced, in which the three-dimensional point cloud data is calculated to extract M groups of multi-view feature points, and the disparity matching operation is performed on the M groups of multi-view feature points to generate the disparity map of each target tobacco planting area;

[0070] Aggregating the global path disparity of the disparity map of each target tobacco planting area under different disparity factors, converting the disparity map of each target tobacco planting area into a depth map based on the global path disparity, and reconstructing the model using the depth map to obtain a global three-dimensional model of each target tobacco planting area;

[0071] Constructing a matrix grid registration domain, obtaining tobacco plants planted in a target tobacco planting area, obtaining a maximum growth height of the tobacco plants in an optimal growth stage in a big data network based on the tobacco plants, and adjusting a domain size of the three-dimensional matrix grid registration domain based on the maximum growth height to generate an initial matrix grid registration domain of the tobacco plants in an optimal growth stage;

[0072] Importing the global three-dimensional model of each target tobacco planting area into the initial matrix grid registration field for alignment and registration, after the alignment and registration is completed, removing the aligned and registered area of ​​the global three-dimensional model in the initial matrix grid registration field to obtain an unaligned and registered initial matrix grid registration field;

[0073] Several growth height differences of tobacco plants under different preset growth stage conditions are obtained, and N sub-fields in the initial matrix grid registration field that are not properly registered are analyzed through each growth height difference, and a growth height difference distribution model for each sub-field is generated. The growth stage of tobacco plants in each sub-field is determined by combining the mean shift clustering algorithm and the growth height difference distribution model, so that the cruising behavior of the UAV is planned according to the growth stage of the tobacco plants in each sub-field, and the optimal cruising plan of the UAV is obtained.

[0074] It should be noted that usually drones need to fly above the tobacco to take dynamic photos for patrol inspection of tobacco plants, but the heights of tobacco plants at different growth stages are different. For example, mature tobacco plants are significantly taller than young tobacco plants. In addition, due to the differences in temperature, humidity, light and other factors at different locations in the tobacco planting area, the growth stages of tobacco plants at different locations are also unevenly distributed, so the growth heights are uneven. If traditional drones maintain a unified cruising altitude and angle at different locations in the tobacco planting area, the resolution of the images of the shorter young tobacco plants may be unclear and the camera may have difficulty in focusing accurately. The poor quality of the cruise photography leads to errors in the damage analysis of important tobacco pests. This method obtains three-dimensional point cloud data of the target tobacco planting area through unmanned aerial vehicle cruising and constructs a global three-dimensional model of the target tobacco planting area in the SGM dense matching algorithm; wherein the SGM dense matching algorithm is a stereo matching algorithm that can comprehensively consider the matching costs in multiple directions to optimize the disparity estimation, thereby ensuring the accuracy and efficiency of the three-dimensional model construction; then the global three-dimensional model is fitted and registered in the matrix grid registration field. Since tobacco will not continue to grow after it grows to a certain optimal growth stage, the matrix grid registration field is a three-dimensional grid rectangular registration space based on the maximum growth height of the optimal growth stage, which can accommodate the global three-dimensional model of all different heights of the target tobacco planting area; and the area that cannot be fitted and registered can express the height difference of each tobacco plant that has not reached the optimal growth stage compared to the optimal growth stage. Therefore, the unmanned aerial vehicle can plan the cruising height and angle of each tobacco plant according to this height difference, so that the resolution of the captured image remains consistent and clear, thereby improving the accuracy of the detection of important tobacco pests, reducing the occurrence of poor image analysis quality caused by cruising, and having high reliability.

[0075] Furthermore, in a preferred embodiment of the present invention, the method of obtaining several growth height differences of tobacco plants under different preset growth stage conditions, analyzing N sub-fields in the initial matrix grid registration field that is not aligned with the registration through each growth height difference, generating a growth height difference distribution model for each sub-field, and combining the mean shift clustering algorithm and the growth height difference distribution model to determine the growth stage of the tobacco plants in each sub-field, so that the cruising behavior of the UAV is planned according to the growth stage of the tobacco plants in each sub-field, and obtaining the optimal cruising plan of the UAV, specifically includes the following steps:

[0076] Based on the big data network, the growth height reference interval of the tobacco plant under different preset growth stage conditions is obtained, and the growth height difference between the optimal growth stage of the tobacco plant and each preset growth stage is calculated by combining the growth height reference interval and the maximum growth height, so as to obtain a plurality of growth height difference values;

[0077] Obtaining a preset planting interval between adjacent tobacco plants and a cruise visual range of the UAV, and dividing the initial matrix grid registration area that is not aligned into N sub-areas according to the preset planting interval and the cruise visual range;

[0078] In each sub-domain, a plurality of growth height difference values ​​corresponding to each preset growth stage are fitted and embedded, so that appropriate growth height difference values ​​are displayed in the initial matrix grid registration domain that is not properly registered, and a growth height difference distribution model of each sub-domain is generated;

[0079] The mean shift clustering algorithm is introduced, and multiple initial clustering centers are constructed in the mean shift clustering algorithm based on different preset growth stages, and the kernel density estimation value of several growth height differences corresponding to each preset growth stage at each initial clustering center is calculated to determine the mean shift vector;

[0080] The current position of each initial cluster center is updated and iterated according to the mean shift vector. If the position change amplitude of each initial cluster center decreases, the update iteration is stopped, and several growth height differences are assigned to the nearest centroid cluster to obtain the clustering results of each growth height difference in the corresponding preset growth stage;

[0081] Obtaining an image resolution threshold required for drone cruising, identifying a growth height difference distribution model for each subfield through clustering results of each growth height difference at a corresponding preset growth stage, and determining the growth stage of tobacco plants in each subfield;

[0082] The growth height difference corresponding to the growth stage of tobacco plants in each sub-field compared to the optimal growth stage is extracted, and the growth height difference is adjusted based on the image resolution threshold to obtain the optimal cruising height of the UAV for the tobacco plants in each sub-field. The cruising height and cruising angle of the UAV are planned according to the optimal cruising height to generate the optimal cruising plan of the UAV.

[0083] It should be noted that in order to ensure that the resolution of the cruise images taken by the drone for each tobacco plant meets the requirements, it is necessary to further determine the growth stage of each tobacco plant in the target tobacco planting area. The various cruise behaviors of the drone can be accurately planned based on the growth height of the specific growth stage combined with the image resolution requirements; after eliminating the initial matrix grid registration field that cannot be matched in the matrix grid registration field, first calculate the growth height difference between the optimal growth stage of the tobacco variety and each preset growth stage. Since the cruise vision of the drone camera has a certain range limitation and there is a certain planting interval between each tobacco plant in different fields, there will be differences when the drone shoots tobacco plants within the cruise vision range in different fields, resulting in the tobacco plants at the edge of the cruise vision range being segmented, which reduces the image quality. Therefore, it is necessary to The cruising visual range and the preset planting interval between each tobacco plant divide the initial matrix grid registration area that is not aligned into N sub-areas to ensure the integrity of the tobacco plant features in the captured image; then, by clustering each growth height difference to the corresponding preset growth stage, and identifying the growth height difference distribution model of each sub-area according to the clustering result, the growth stage of each tobacco variety in each sub-area can be determined, and the corresponding growth height difference is displayed in the initial matrix grid registration area that is not aligned according to the growth stage of each tobacco variety, so as to further plan the optimal cruising height and angle of the drone for each tobacco variety, greatly improve the resolution quality of image shooting, so that the image quality of each tobacco plant when the drone cruises remains uniform, and provide clear, accurate and reliable image feature basis for subsequent important tobacco pest damage detection.

[0084] Furthermore, in a preferred embodiment of the present invention, the drone cruise scheme is executed by the drone to obtain image data of the surface of all tobacco plants in each sub-field and perform feature description, obtain the notch contour Fourier descriptor of the surface of all tobacco plants in each sub-field, import the notch contour Fourier descriptor into the notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and evaluate it, and obtain the first hazard detection result, which specifically includes the following steps:

[0085] The drone is set to execute the drone cruise scheme to automatically cruise and shoot the target tobacco planting area, so as to obtain image data of the surface of all tobacco plants in each sub-field;

[0086] A Laplace edge detection operator is introduced to calculate the second-order derivatives in the image data of the surfaces of all tobacco plants in each sub-field, and the notch contour feature vectors of the surfaces of all tobacco plants in each sub-field are extracted according to the second-order derivatives, and uniform discrete contour point coordinates are collected and marked on the notch contour feature vectors of the surfaces of all tobacco plants in each sub-field;

[0087] Constructing a complex sequence space, generating a real part and an imaginary part of each discrete contour point based on uniform discrete contour point coordinates, calculating and transforming the real part and the imaginary part of each discrete contour point in the complex sequence space to generate a complex sequence of discrete contour point coordinates, and performing Fourier transformation on the complex sequence of discrete contour point degrees to obtain the Fourier descriptors of the notched contours of all tobacco plant surfaces in each subfield;

[0088] Obtaining the chewing mouthparts hazard profiles of different important tobacco pest species at different preset growth stages through a big data network, and simultaneously obtaining the chewing notch profile Fourier descriptors formed by each chewing mouthparts hazard profile on the surface of the tobacco plant; wherein the different important tobacco pest species include tobacco budworm and Spodoptera litura;

[0089] K groups of index queues are established according to different types of important tobacco pests, and the index array quantity threshold of each group of index queues is preset according to the chewing mouthparts hazard profiles of the different important tobacco pests at different growth stages, and the corresponding index queue is divided into a plurality of index array spaces based on the index array quantity threshold;

[0090] Each of the chewing notch contour Fourier descriptors is correspondingly embedded in the space of the index array to obtain a notch contour index array mapping table, and the degree of fit between each notch contour Fourier descriptor and each chewing notch contour Fourier descriptor is calculated one by one. If the degree of fit is greater than a preset degree of fit, the chewing notch contour Fourier descriptor corresponding to the degree of fit greater than the preset degree of fit is extracted;

[0091] The chewing notch profile Fourier descriptors corresponding to the notch profiles with a matching degree greater than a preset matching degree are traversed and indexed through a notch profile index array mapping table, so as to output and mark the important tobacco pest species corresponding to the notch profile Fourier descriptors on the surfaces of all tobacco plants in each sub-field, and draw a distribution map of important tobacco pests;

[0092] The preset harm coefficient of each important tobacco pest species to tobacco plants is obtained, and the distribution map of important tobacco pests is evaluated based on the preset harm coefficient to obtain a first harm detection result.

[0093] It should be noted that in tobacco planting, tobacco budworm and fall armyworm are two common pests that cause harm. Both of them mainly feed on the rhizomes and leaves of tobacco when they are in the juvenile and mature stages, causing the appearance of chewing notch characteristics on the surface of the tobacco rhizomes and leaves, resulting in damage to the leaves, buds and rhizomes, and even causing secondary infections, reducing the survival rate, growth and yield of tobacco planting, which is not conducive to tobacco planting. An important indicator of tobacco pests for tobacco planting is the specific distribution of pests in tobacco planting areas. The method converts the image feature contour of the tobacco surface into a Fourier descriptor in a complex sequence space to obtain the notch contour Fourier descriptors of all tobacco surfaces in each sub-field, and constructs a notch contour index mapping table that maps the chewing notch contour Fourier descriptors formed on the tobacco surface by the chewing mouthparts damage contour of the important tobacco pests, imports the converted notch contour Fourier descriptors of the tobacco surface into the notch contour index mapping table for index matching, and calculates them one by one. The degree of fit between each notch contour Fourier descriptor and each chewing notch contour Fourier descriptor, if the degree of fit is greater than the preset degree of fit, then it means that the notch contour on the tobacco surface at the current index is highly consistent with the chewing mouthpart hazard contour of important tobacco pests, indicating that this notch contour is produced by the chewing of important tobacco pests. Therefore, the type of important tobacco pests that produces this notch contour can be further matched in the notch contour index mapping table. After traversing the indexes of all notch contours, a distribution map of important tobacco pests in the target tobacco planting area can be finally formed, and a distribution map can be evaluated based on the preset hazard coefficient of the important tobacco pest types for tobacco varieties. According to the evaluation results, the hazard of the currently distributed important tobacco pests to the tobacco planting area can be detected, and the overall hazard level of important tobacco pests is preliminarily detected, providing accurate and reliable detection and evaluation basis for the formulation of overall protective measures for tobacco planting. The detection method through index matching is efficient and fast, which improves the efficiency of important tobacco pest hazard detection.

[0094] Furthermore, in a preferred embodiment of the present invention, the remote sensing camera of the drone is controlled to perform cruise shooting based on the first hazard detection result to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain a second hazard detection result, such as Figure 2 As shown, the specific steps include:

[0095] S202: Preset a first target time period, control the remote sensing camera of the drone to cruise and shoot each sub-field based on the distribution map of important tobacco pest species, and obtain a number of tobacco plant surface notch time-series images in each sub-field within the first target time period, which are marked as first tobacco plant surface notch time-series images;

[0096] S204: introducing the Horn-Schunck algorithm, constructing a partial differential optical flow equation in the Horn-Schunck algorithm based on each first tobacco plant surface notch time-series image in each sub-field within the first target time period, and analyzing the pixel continuity of the partial differential optical flow equation to obtain an energy function;

[0097] S206: calculating the neighboring average value of the current optical flow in the first tobacco plant surface notch time-series image through the energy function, iteratively solving the partial differential optical flow equation based on the neighboring average value, so as to continuously update the optical flow of the optical flow field until a preset optical flow field convergence threshold is reached, and finally outputting a plurality of first tobacco plant surface notch time-series change models in each sub-field within the first target time period;

[0098] S208: Obtaining the actual growth stage and actual environmental factors of each sub-field in the first target time period, and obtaining the feeding rate of each important tobacco pest species under the combination of the actual growth stage and actual environmental factors based on the big data network;

[0099] S210: constructing a naive Bayesian network, importing the temporal change models of the plurality of first tobacco plant surface notches into the naive Bayesian network for inference, obtaining a plurality of first tobacco plant surface notch temporal change rates, and calculating the ratio of the feeding rate to the temporal change rate of each first tobacco plant surface notch;

[0100] S212: Determine the number of important tobacco pest species that cause the time series change model of notches on the surface of tobacco plants in each sub-field within the first target time period according to the ratio, and evaluate the number of important tobacco pest species based on the preset hazard coefficient to obtain a second hazard detection result.

[0101] It should be noted that important indicators for detecting the harm of major tobacco pests to tobacco planting also include the number of pest distribution. The notch contour feature on the tobacco surface can be produced by the gradual chewing of a major tobacco pest, or it can be formed by the continuous expansion of multiple major tobacco pests chewing at the same time. Therefore, this method obtains multiple continuous tobacco surface notch time-series images within a preset first target time period through UAV cruising, and calculates multiple continuous tobacco surface notch time-series images through the Horn-Schunck algorithm to construct a tobacco surface notch time-series change model. The tobacco surface notch time-series change model can intuitively express the notch contour changing over time. The expansion change can be further calculated according to the expansion change rate of the notch contour combined with the ratio between the feeding rate of each important tobacco pest species in the actual growth stage and the actual environmental factor conditions, and the number distribution information of important tobacco pests that gnaw on the notch contour of each tobacco surface can be accurately determined. The error rate of notch changes in traditional frame-by-frame image analysis can be greatly reduced by the method of temporal change model analysis. Compared with the traditional frame-by-frame image analysis method, the analysis rate is improved, the interference of redundant feature noise in the image is reduced, and the detection efficiency and quality of the number distribution of important tobacco pests are improved. Among them, the Chinese name of the Horn-Schunck algorithm is the Horn-Schunck optical flow algorithm, which is used to estimate the movement of pixels in the image and extract dynamic image features. The Horn-Schunck algorithm can maintain the optical flow field of the global consistency of the image, and through the optical flow field smoothing term, it has a certain robustness to random noise in the image, effectively reducing the influence of local noise on the results, further improving the display accuracy and reliability of the temporal change model, and ensuring that the hazard detection results are accurate.

[0102] Furthermore, in a preferred embodiment of the present invention, the drone is controlled to perform secondary cruise shooting after spraying the preset pesticide after deployment on the target tobacco planting area, and a plurality of second tobacco plant surface notch time-series images in each sub-field within the second target time period are obtained, and the area value of the area after the preset pesticide after deployment is analyzed in combination with the corresponding first tobacco plant surface notch time-series images to obtain a third hazard detection result, such as Figure 3 As shown, the specific steps include:

[0103] S302: Obtain pesticide spraying records of the target tobacco planting area, extract the historical killing rate of the preset pesticide on the number of important tobacco pest species in each sub-field within the first target time period based on the pesticide spraying records, and if the historical killing rate is less than the preset killing rate, calculate the initial tolerance index of the important tobacco pest species to the preset pesticide in combination with the number of important tobacco pest species and the historical killing rate;

[0104] S304: Preset a second target time period, and readjust the formula and dosage of the preset pesticide in combination with the first hazard detection result and the second hazard detection result to obtain the prepared preset pesticide, and then control the drone to spray the prepared preset pesticide on the surface of the tobacco plants in each sub-field;

[0105] S306: After the spraying is completed, the remote sensing camera of the UAV performs a second cruise shooting of each sub-field to obtain a number of tobacco plant surface notch time-series images in each sub-field within the second target time period, which are marked as second tobacco plant surface notch time-series images;

[0106] S308: defining each of the first tobacco plant surface notch time-series images as a redundant region, and removing the redundant region from the corresponding second tobacco plant surface notch time-series image to obtain a second tobacco plant surface notch time-series image region after removal;

[0107] S310: dividing the second target time period into a plurality of uniform target time stamps, obtaining the area parameter of the removed second tobacco plant surface notch time-series image region on each uniform target time stamp, and calculating the ratio between the area parameters of the removed second tobacco plant surface notch time-series image regions on adjacent uniform target time stamps;

[0108] S312: Fit the ratios of the area parameters of the removed second tobacco plant surface notch time-series image regions on adjacent uniform target timestamps into a Bezier curve, and obtain the slope of the Bezier curve, calculate the adjusted preset pesticide tolerance index by combining the slope of the Bezier curve with the initial tolerance index, and evaluate the adjusted preset pesticide tolerance index based on the preset hazard coefficient to obtain a third hazard detection result.

[0109] It should be noted that pesticide spraying is usually used to control major tobacco pests. However, over time, major tobacco pests will gradually develop adaptability and tolerance to the preset pesticides, making it difficult for the sprayed preset pesticides to kill the major tobacco pests. This results in the major tobacco pests being able to survive in the preset pesticide spraying environment and the low killing rate of the preset pesticides, further increasing the harm of major tobacco pests to tobacco cultivation. The solution to this problem is usually to readjust the formula or dosage of the preset pesticides to achieve the purpose of resetting the tolerance index. However, traditional hazard detection methods cannot further evaluate the tolerance of the prepared preset pesticides to the major tobacco pests. As a result, the major tobacco pests will quickly adapt to the prepared preset pesticides, reducing the improved quality and killing rate of the pesticides, greatly increasing the pesticide tolerance of the major tobacco pests, and being detrimental to high-quality tobacco cultivation. The present method can adjust and optimize the optimal formula and dosage of the preset pesticides, and spray the prepared preset pesticides by drones. Pesticide and obtain several second tobacco surface notch time-series images of each sub-field within the second target time period. Since the second tobacco surface notch time-series images are taken on the basis of the first tobacco surface notch time-series images, it is only necessary to analyze the notch expansion area of ​​the second tobacco surface notch time-series images compared with the first tobacco surface notch time-series images, that is, the second tobacco surface notch time-series image area after being eliminated. By analyzing the area change trend of the second tobacco surface notch time-series image area after being eliminated on the uniform time stamp, the gnawing number and killing rate of important tobacco pests can be known, and then the tolerance index of the preset pesticide after deployment can be calculated; this method can detect and evaluate the tolerance index of the preset pesticide for important tobacco pests each time it is re-deployed, so as to continuously update the performance of the preset pesticide and the tolerance characteristics of the important tobacco pests, so that a prevention and control plan can be formulated according to the degree of pest tolerance, so as to reduce the harm of important tobacco pests to tobacco planting and improve tobacco growth quality and survival rate.

[0110] In addition, the method for detecting important tobacco pests based on drone patrol also includes the following steps:

[0111] Obtaining an activity index of a preset pesticide for killing important tobacco pests at a preset time point, obtaining multiple activity indices, and combining the multiple activity indices for calculation to obtain an activity index change rate;

[0112] If the activity index change rate is greater than the preset activity index change rate, it means that the activity of the preset pesticide in killing important tobacco pests has decreased, and several climate factors that cause the activity of the preset pesticide to decrease are obtained based on the big data network;

[0113] Introducing the factor analysis method to calculate the factor correlation between the activity index change rate and each climate factor, extracting the climate factor corresponding to the factor correlation greater than the preset factor correlation, defining it as the key climate factor that causes the preset pesticide to produce the said activity change rate, and obtaining one or more key climate factors;

[0114] Obtaining the actual climate factor set of the current target tobacco planting area; if there is at least one key climate factor in the actual climate factor set, obtaining a pesticide system knowledge graph based on a big data network, importing preset pesticides and key climate factors into the pesticide system knowledge graph for identification, and obtaining the inactivation characteristics and inactivation rate of the preset pesticide under the key climate factor;

[0115] Constructing a bubble chart, importing the inactivation characteristics and inactivation rate of the preset pesticide under the key climate factor into the bubble chart for calculation and fitting, and obtaining the inactivation bubble volume value of the preset pesticide under the key climate factor;

[0116] If the deactivated bubble volume value is less than the preset deactivated bubble volume value, a hash algorithm is introduced to calculate the hash value between the deactivated bubble volume value and the preset deactivated bubble volume value, and the preset pesticide is readjusted according to the hash value. At the same time, the cruising frequency and intensity of the drone are increased based on the hash value.

[0117] It should be noted that pesticides have certain pest-killing activity, but under certain specific climatic conditions, the pesticides sprayed on tobacco plants will become inactivated, resulting in a significant decrease in the pest-killing performance of the pesticides, making it impossible to achieve the expected killing effect, and will indirectly increase the pesticide tolerance of important tobacco pests, which is not conducive to high-quality tobacco cultivation. This method can accurately locate and screen out highly correlated climate factors that lead to decreased pesticide activity. If there is at least one key climate factor in the actual climate factor concentration of the current target tobacco planting area, it means that the climate environment of the current target tobacco planting area will cause the pesticide activity to decrease. Therefore, the inactivation bubble volume value can be calculated in the bubble chart by pre-setting the inactivation characteristics and inactivation rate of the pesticide under the key climate factor, and the formula or dosage of the preset pesticide is readjusted according to the inactivation bubble volume value. At the same time, the cruise frequency and intensity of the drone are increased to ensure that the adjusted pesticide will not be inactivated in the current target tobacco planting area, thereby ensuring the efficient killing performance of the pesticide on important tobacco pests. Increasing the cruise frequency and intensity of the drone can continuously update the damage detection results of important tobacco pests, improve the real-time and accuracy of damage detection, and timely display the damage trend of important tobacco pests to tobacco planting to formulate response strategies.

[0118] The second aspect of the present invention provides a tobacco major pest hazard detection system based on drone cruise, the tobacco major pest hazard detection system based on drone cruise includes a memory 41 and a processor 42, the memory 41 stores a tobacco major pest hazard detection method program based on drone cruise, and when the tobacco major pest hazard detection method program based on drone cruise is executed by the processor 42, Figure 4 As shown, implement the following steps:

[0119] The disparity map of each target tobacco planting area is obtained through drone cruising, and a global three-dimensional model is constructed based on the disparity map of each target tobacco planting area. After fitting the global three-dimensional model in the matrix grid registration field, the mean shift clustering algorithm and the growth height difference are combined to determine the growth stage of tobacco plants in each sub-field, so as to plan the drone's cruising behavior and obtain the drone's optimal cruising plan.

[0120] The drone cruise scheme is executed by the drone to obtain image data of the surfaces of all tobacco plants in each sub-field and perform feature description, so as to obtain the notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field, and the notch contour Fourier descriptors are imported into the notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and perform evaluation, so as to obtain a first hazard detection result;

[0121] Based on the first hazard detection result, the remote sensing camera of the drone is controlled to perform cruise shooting to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain a second hazard detection result;

[0122] By spraying the preset pesticide after deployment in the target tobacco planting area and then controlling the drone to conduct a second cruise shooting, several second tobacco plant surface notch time-series images in each sub-field within the second target time period are obtained. The area value of the area after the preset pesticide is deployed is analyzed by combining the corresponding first tobacco plant surface notch time-series images, and the third hazard detection result is obtained.

[0123] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for detecting important tobacco pests based on drone patrol, characterized in that: The following steps are involved: The disparity map of each target tobacco planting area is obtained through drone cruising, and a global three-dimensional model is constructed based on the disparity map of each target tobacco planting area. After fitting the global three-dimensional model in the matrix grid registration field, the mean shift clustering algorithm and the growth height difference are combined to determine the growth stage of tobacco plants in each sub-field, so as to plan the drone's cruising behavior and obtain the drone's optimal cruising plan. The drone cruise scheme is executed by the drone to obtain image data of the surfaces of all tobacco plants in each sub-field and perform feature description, so as to obtain the notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field, and the notch contour Fourier descriptors are imported into the notch contour index array mapping table for index matching, so as to draw a distribution map of important tobacco pests and perform evaluation, so as to obtain a first hazard detection result; Based on the first hazard detection result, the remote sensing camera of the drone is controlled to perform cruise shooting to construct a plurality of first tobacco plant surface notch time series change models in each sub-field within the first target time period, and the number of important tobacco pest species that cause the tobacco plant surface notch time series change model in each sub-field is determined and evaluated by analyzing the first tobacco plant surface notch time series change model to obtain a second hazard detection result; After spraying the preset pesticide after deployment in the target tobacco planting area, the drone is controlled to perform a second cruise shooting, and a plurality of second tobacco plant surface notch time-series images in each sub-area within the second target time period are obtained. The area value of the area after the preset pesticide after deployment is analyzed in combination with the corresponding first tobacco plant surface notch time-series images, and the third hazard detection result is obtained; Among them, after fitting the global three-dimensional model in the matrix grid registration field, the mean shift clustering algorithm and the growth height difference are combined to determine the growth stage of tobacco plants in each sub-field to plan the cruising behavior of the drone and obtain the optimal cruising plan of the drone, which specifically includes the following steps: Constructing a matrix grid registration domain, obtaining tobacco plants planted in a target tobacco planting area, obtaining a maximum growth height of the tobacco plants in an optimal growth stage in a big data network based on the tobacco plants, and adjusting a domain size of the three-dimensional matrix grid registration domain based on the maximum growth height to generate an initial matrix grid registration domain of the tobacco plants in an optimal growth stage; Importing the global three-dimensional model of each target tobacco planting area into the initial matrix grid registration field for alignment and registration, after the alignment and registration is completed, removing the aligned and registered area of ​​the global three-dimensional model in the initial matrix grid registration field to obtain an unaligned and registered initial matrix grid registration field; Obtain several growth height differences of tobacco plants under different preset growth stage conditions, analyze N sub-fields in the initial matrix grid registration field that is not aligned with the registration through each growth height difference, generate a growth height difference distribution model for each sub-field, and determine the growth stage of tobacco plants in each sub-field by combining the mean shift clustering algorithm and the growth height difference distribution model, so as to plan the cruising behavior of the UAV according to the growth stage of the tobacco plants in each sub-field, and obtain the optimal cruising plan of the UAV; Among them, a number of growth height differences of tobacco plants under different preset growth stage conditions are obtained, and N sub-fields in the initial matrix grid registration field that is not aligned are analyzed through each growth height difference, and a growth height difference distribution model for each sub-field is generated. The growth stage of the tobacco plants in each sub-field is determined by combining the mean shift clustering algorithm and the growth height difference distribution model, so that the cruising behavior of the UAV is planned according to the growth stage of the tobacco plants in each sub-field, and the optimal cruising plan of the UAV is obtained, which specifically includes the following steps: Based on the big data network, the growth height reference interval of the tobacco plant under different preset growth stage conditions is obtained, and the growth height difference between the optimal growth stage of the tobacco plant and each preset growth stage is calculated by combining the growth height reference interval and the maximum growth height, so as to obtain a plurality of growth height difference values; Obtaining a preset planting interval between adjacent tobacco plants and a cruise visual range of the UAV, and dividing the initial matrix grid registration area that is not aligned into N sub-areas according to the preset planting interval and the cruise visual range; In each sub-domain, a plurality of growth height difference values ​​corresponding to each preset growth stage are fitted and embedded, so that appropriate growth height difference values ​​are displayed in the initial matrix grid registration domain that is not properly registered, and a growth height difference distribution model of each sub-domain is generated; The mean shift clustering algorithm is introduced, and multiple initial clustering centers are constructed in the mean shift clustering algorithm based on different preset growth stages, and the kernel density estimation value of several growth height differences corresponding to each preset growth stage at each initial clustering center is calculated to determine the mean shift vector; The current position of each initial cluster center is updated and iterated according to the mean shift vector. If the position change amplitude of each initial cluster center decreases, the update iteration is stopped, and several growth height differences are assigned to the nearest centroid cluster to obtain the clustering results of each growth height difference in the corresponding preset growth stage; Obtaining an image resolution threshold required for drone cruising, identifying a growth height difference distribution model for each subfield through clustering results of each growth height difference at a corresponding preset growth stage, and determining the growth stage of tobacco plants in each subfield; Extract the growth height difference value corresponding to the growth stage of tobacco plants in each sub-field compared with the optimal growth stage, adjust the growth height difference value based on the image resolution threshold, obtain the optimal cruising height of the drone for the tobacco plants in each sub-field, plan the cruising height and cruising angle of the drone according to the optimal cruising height, and generate the optimal cruising plan of the drone; The notch contour Fourier descriptor is imported into the notch contour index array mapping table for index matching to draw a distribution map of important tobacco pests and perform an evaluation to obtain a first hazard detection result, which specifically includes the following steps: Obtaining the chewing mouthparts hazard profiles of different important tobacco pest species at different preset growth stages through a big data network, and simultaneously obtaining the chewing notch profile Fourier descriptors formed by each chewing mouthparts hazard profile on the surface of the tobacco plant; wherein the different important tobacco pest species include tobacco budworm and Spodoptera litura; K groups of index queues are established according to different types of important tobacco pests, and the index array quantity threshold of each group of index queues is preset according to the chewing mouthparts hazard profiles of the different important tobacco pests at different growth stages, and the corresponding index queue is divided into a plurality of index array spaces based on the index array quantity threshold; Each of the chewing notch contour Fourier descriptors is correspondingly embedded in the space of the index array to obtain a notch contour index array mapping table, and the degree of fit between each notch contour Fourier descriptor and each chewing notch contour Fourier descriptor is calculated one by one. If the degree of fit is greater than a preset degree of fit, the chewing notch contour Fourier descriptor corresponding to the degree of fit greater than the preset degree of fit is extracted; The chewing notch profile Fourier descriptors corresponding to the notch profiles with a matching degree greater than a preset matching degree are traversed and indexed through a notch profile index array mapping table, so as to output and mark the important tobacco pest species corresponding to the notch profile Fourier descriptors on the surfaces of all tobacco plants in each sub-field, and draw a distribution map of important tobacco pests; Obtaining a preset harm coefficient of each major tobacco pest species to tobacco plants, and evaluating a distribution map of major tobacco pests based on the preset harm coefficient to obtain a first harm detection result; The method of determining the number of important tobacco pest species that cause the tobacco plant surface notch temporal change model in each sub-field by analyzing the first tobacco plant surface notch temporal change model and evaluating the number of important tobacco pest species that cause the tobacco plant surface notch temporal change model to be generated, and obtaining the second hazard detection result specifically includes the following steps: Obtain the actual growth stage and actual environmental factors of each sub-field in the first target time period, and obtain the feeding rate of each important tobacco pest species under the combination of the actual growth stage and actual environmental factors based on the big data network; Constructing a naive Bayesian network, importing the temporal change models of the surface notches of the first tobacco plants into the naive Bayesian network for inference, obtaining the temporal change rates of the surface notches of the first tobacco plants, and calculating the ratio of the feeding rate to the temporal change rate of each surface notch of the first tobacco plants; According to the ratio, the number of important tobacco pest species that cause the time series change model of notches on the surface of tobacco plants in each sub-field within the first target time period is determined, and the number of important tobacco pest species is evaluated based on the preset hazard coefficient to obtain a second hazard detection result.

2. The method for detecting important tobacco pests based on drone patrol according to claim 1 is characterized in that: The method of obtaining the disparity map of each target tobacco planting area by cruising with a drone and constructing a global three-dimensional model based on the disparity map of each target tobacco planting area specifically includes the following steps: Obtaining a preset cruise mission of the drone, extracting one or more target tobacco planting areas that the drone needs to cruise through the preset cruise mission, and controlling the drone's depth camera to automatically cruise and photograph the target tobacco planting areas to obtain three-dimensional point cloud data of each target tobacco planting area; The SGM dense matching algorithm is introduced, in which the three-dimensional point cloud data is calculated to extract M groups of multi-view feature points, and the disparity matching operation is performed on the M groups of multi-view feature points to generate the disparity map of each target tobacco planting area; The global path parallax of the disparity map of each target tobacco planting area under different parallax factors is aggregated, the disparity map of each target tobacco planting area is converted into a depth map based on the global path parallax, and the model is reconstructed using the depth map to obtain a global three-dimensional model of each target tobacco planting area.

3. The method for detecting important tobacco pests based on drone patrol according to claim 1 is characterized in that: The method of acquiring image data of the surfaces of all tobacco plants in each sub-field and performing feature description by executing the drone cruising scheme through the drone to obtain the notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field specifically includes the following steps: The drone is set to execute the drone cruise scheme to automatically cruise and shoot the target tobacco planting area, so as to obtain image data of the surface of all tobacco plants in each sub-field; A Laplace edge detection operator is introduced to calculate the second-order derivatives in the image data of the surfaces of all tobacco plants in each sub-field, and the notch contour feature vectors of the surfaces of all tobacco plants in each sub-field are extracted according to the second-order derivatives, and uniform discrete contour point coordinates are collected and marked on the notch contour feature vectors of the surfaces of all tobacco plants in each sub-field; A complex sequence space is constructed, and the real part and the imaginary part of each discrete contour point are generated based on the uniform discrete contour point coordinates. The real part and the imaginary part of each discrete contour point are calculated and transformed in the complex sequence space to generate a complex sequence of discrete contour point coordinates, and the complex sequence of discrete contour point degrees is Fourier transformed to obtain the notch contour Fourier descriptors of the surfaces of all tobacco plants in each sub-field.

4. The method for detecting important tobacco pests based on drone patrol according to claim 1 is characterized in that: The method of controlling the remote sensing camera of the drone to perform cruise shooting based on the first hazard detection result to construct a time-series change model of several first tobacco plant surface notches in each sub-field within the first target time period specifically includes the following steps: A first target time period is preset, and based on the distribution map of important tobacco pest species, the remote sensing camera of the drone is controlled to cruise and shoot each sub-field, and a plurality of tobacco plant surface notch time-series images in each sub-field within the first target time period are obtained, which are marked as first tobacco surface notch time-series images; The Horn-Schunck algorithm is introduced, and the partial differential optical flow equation is constructed in the Horn-Schunck algorithm based on the time-series images of each first tobacco plant surface notch in each sub-field within the first target time period, and the pixel continuity of the partial differential optical flow equation is analyzed to obtain the energy function; The neighboring average value of the current optical flow in the time-series image of the first tobacco plant surface notches is calculated through the energy function, and the partial differential optical flow equation is iteratively solved based on the neighboring average value to continuously update the optical flow of the optical flow field until a preset optical flow field convergence threshold is reached, and finally, several first tobacco plant surface notch time-series change models of each sub-field within the first target time period are output.

5. The method for detecting important tobacco pests based on drone patrol according to claim 1 is characterized in that: The method comprises the following steps: spraying the preset pesticide after deployment on the target tobacco planting area and then controlling the drone to perform a second cruise shooting, obtaining a plurality of second tobacco plant surface notch time-series images in each sub-field within the second target time period, combining the corresponding first tobacco plant surface notch time-series images to analyze the area value after the preset pesticide after deployment, and obtaining the third hazard detection result. Obtain pesticide spraying records of the target tobacco planting area, and extract the historical killing rate of the preset pesticide on the number of important tobacco pest species in each sub-field within the first target time period based on the pesticide spraying records; if the historical killing rate is less than the preset killing rate, calculate the initial tolerance index of the important tobacco pest species to the preset pesticide in combination with the number of important tobacco pest species and the historical killing rate; A second target time period is preset, and the formula and dosage of the preset pesticide are readjusted in combination with the first hazard detection result and the second hazard detection result to obtain the prepared preset pesticide, and the drone is controlled to spray the prepared preset pesticide onto the surface of the tobacco plants in each sub-field; After the spraying is completed, the remote sensing camera of the UAV is used to conduct a second cruise shooting of each sub-field to obtain a number of tobacco surface notch time-series images in each sub-field within the second target time period, which are marked as the second tobacco plant surface notch time-series images; Defining each of the first tobacco plant surface notch time-series images as a redundant region, and removing the redundant region from the corresponding second tobacco plant surface notch time-series image to obtain a second tobacco plant surface notch time-series image region after removal; Divide the second target time period into a plurality of uniform target time stamps, obtain the area parameter of the time-sequential image region of the second tobacco plant surface notches after being removed on each uniform target time stamp, and calculate the ratio between the area parameters of the time-sequential image region of the second tobacco plant surface notches after being removed on adjacent uniform target time stamps; The ratio between the area parameters of the removed second tobacco plant surface notch time-series image area on adjacent uniform target timestamps is fitted into a Bezier curve, and the slope of the Bezier curve is obtained. The adjusted preset pesticide tolerance index is calculated by combining the slope of the Bezier curve with the initial tolerance index, and the adjusted preset pesticide tolerance index is evaluated based on the preset hazard coefficient to obtain a third hazard detection result.

6. The tobacco major pest damage detection system based on drone patrol is characterized by: The system for detecting important tobacco pests based on drone patrols includes a memory and a processor. The memory stores a program for detecting important tobacco pests based on drone patrols. When the program for detecting important tobacco pests based on drone patrols is executed by the processor, the steps of the method for detecting important tobacco pests based on drone patrols as described in any one of claims 1 to 5 are implemented.

Images