An Unmanned Aerial Vehicle Precision Spraying Method and System Based on Multi - source Data Fusion

Abstract

The present invention relates to the technical field of unmanned drug application, in particular to a method and system for precise unmanned aerial vehicle (UAV) drug application based on multi-source data fusion. The UAV is controlled based on preset drug application parameters to apply drugs to the crops to be sprayed, and the actual drug application parameters of the UAV are obtained at multiple preset time nodes. The actual drug application parameters in the parameter repository are clustered to obtain the membership matrix of each actual drug application parameter; and the membership matrix is corrected to obtain the corrected membership matrix; different types of actual drug application parameters in the UAV are obtained according to the corrected membership matrix, and various types of actual drug application parameters are compared with the corresponding types of preset drug application parameters to obtain a comparison result. The actual drug application parameters of the UAV are regulated according to the comparison result. By this method, the deviated drug application parameters can be corrected in time, the drug application accuracy of the UAV can be improved, and the drug application efficiency and effect can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned pesticide application, in particular to a method and system for precise pesticide application of unmanned aerial vehicles based on multi-source data fusion. Background Art

[0002] With the progress of agricultural production technology and the development of unmanned aerial vehicle technology, using unmanned aerial vehicles for precise pesticide application in farmland has become an important means of modern agricultural production. Precise pesticide application by unmanned aerial vehicles means precise pesticide application to farmland crops using unmanned aerial vehicles. By using unmanned aerial vehicles for pesticide application, precise, efficient, and flexible pesticide application operations can be achieved, improving the protection effect of crops and reducing pesticide waste and environmental pollution. However, during the process of pesticide application by unmanned aerial vehicles, due to environmental impacts and equipment continuous working errors (such as motor cumulative errors, etc.), the pesticide application parameters will change, resulting in pesticide application drift, affecting the pesticide application accuracy. To solve these problems, the present application proposes a method for precise pesticide application of unmanned aerial vehicles based on multi-source data fusion. Summary of the Invention

[0003] The present invention overcomes the deficiencies of the prior art and provides a method and system for precise pesticide application of unmanned aerial vehicles based on multi-source data fusion.

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

[0005] The first aspect of the present invention discloses a method for precise pesticide application of unmanned aerial vehicles based on multi-source data fusion, including the following steps:

[0006] Obtain the point cloud data of the crops to be pesticided, construct a three-dimensional model diagram of the crops to be pesticided based on the point cloud data, and generate pre-set pesticide application parameters for pesticiding the crops to be pesticided based on the three-dimensional model diagram of the crops to be pesticided;

[0007] Control the unmanned aerial vehicle to apply pesticides to the crops to be pesticided based on the pre-set pesticide application parameters, obtain the actual pesticide application parameters of the unmanned aerial vehicle at multiple preset time nodes, construct a parameter storage library, and import the actual pesticide application parameters into the parameter storage library;

[0008] Perform clustering processing on the actual pesticide application parameters in the parameter storage library to obtain a membership degree matrix of each actual pesticide application parameter; and correct the membership degree matrix to obtain a corrected membership degree matrix;

[0009] Obtain different types of actual pesticide application parameters in the unmanned aerial vehicle according to the corrected membership degree matrix, compare various types of actual pesticide application parameters with the corresponding type of pre-set pesticide application parameters to obtain a comparison result, and regulate the actual pesticide application parameters of the unmanned aerial vehicle according to the comparison result.

[0010] Further, in a preferred embodiment of the present invention, the point cloud data of the crop to be sprayed is acquired, a three-dimensional model diagram of the crop to be sprayed is constructed based on the point cloud data, and the pre-spraying parameters for spraying the crop to be sprayed are generated based on the three-dimensional model diagram of the crop to be sprayed. Specifically:

[0011] The crop to be sprayed is scanned at multiple angular positions by a laser point cloud device, and the point cloud data reflected by the crop to be sprayed at each angular position is acquired to obtain subsets of point cloud data corresponding to different angular positions;

[0012] The Isolation Forest algorithm is introduced, and the isolation scores of the point cloud data in each subset of power data are calculated by the Isolation Forest algorithm. The point cloud data with an isolation score greater than the preset value is removed from the corresponding subset of point cloud data to obtain a filtered subset of point cloud data;

[0013] A three-dimensional coordinate system is constructed, each subset of point cloud data is imported into the three-dimensional coordinate system, and the point cloud data in each subset of point cloud data is aligned in the three-dimensional coordinate system to obtain a complete point cloud model;

[0014] The relative three-dimensional coordinate values between the point cloud data in the point cloud model are obtained in the three-dimensional coordinate system, and a three-dimensional model diagram of the crop to be sprayed is constructed according to the relative three-dimensional coordinate values; the starting point of the spraying path of the unmanned aerial vehicle is initialized, and the pre-spraying range of the unmanned aerial vehicle is acquired;

[0015] Based on the starting point of the spraying path, the pre-spraying range, and the three-dimensional model diagram of the crop to be sprayed, and in combination with the Ant Colony algorithm, the pre-spraying parameters for spraying the crop to be sprayed are iteratively planned; wherein, the pre-spraying parameters include the spraying path, spraying pressure, spraying speed, spraying amount, spraying height, and spray particle size.

[0016] Further, in a preferred embodiment of the present invention, the point cloud data in each subset of point cloud data is aligned in the three-dimensional coordinate system to obtain a complete point cloud model. Specifically:

[0017] The point cloud features of the point cloud data in each subset of point cloud data are extracted, and feature descriptors of each point cloud data are generated based on the point cloud features to describe the local geometric attributes of each point cloud data through the feature descriptors;

[0018] According to the feature descriptors, the corresponding relationships between the point cloud data are found in combination with the nearest neighbor search method, and the point cloud data with the closest distance in different subsets of point cloud data are feature-matched according to the corresponding relationships between the point cloud data to obtain several pairs of point cloud data point pairs;

[0019] Based on a number of pairs of point cloud data points obtained by matching, the initial transformation matrix between each pair of point cloud data points is estimated by combining the least squares method;

[0020] The initial transformation matrix is refined through a global optimization algorithm until the initial transformation matrix meets the preset requirements, and the aligned point cloud data is fused to generate a complete point cloud model.

[0021] Further, in a preferred embodiment of the present invention, the actual application parameters in the parameter repository are clustered to obtain the membership matrix of each actual application parameter, specifically:

[0022] Each actual application parameter is transformed into an independent cluster, and the cosine similarity between each cluster is calculated by the cosine similarity algorithm, and a similarity matrix is constructed according to the cosine similarity;

[0023] Two clusters with the highest similarity are retrieved according to the similarity matrix, and the two clusters with the highest similarity are merged into a new cluster;

[0024] The similarity matrix is updated according to the merged new cluster to reflect the similarity between the new cluster and the remaining clusters; two clusters with the highest similarity are retrieved again in the updated similarity matrix, and the two clusters with the highest similarity are also merged into a new cluster;

[0025] Repeat the above steps until the desired number of clusters is reached to generate a preliminary clustering result, and the membership between each actual application parameter in each cluster and its clustering center is calculated according to the preliminary clustering result, and a membership matrix is generated according to the membership.

[0026] Further, in a preferred embodiment of the present invention, the membership matrix is corrected to obtain a corrected membership matrix, specifically:

[0027] The membership corresponding to each actual application parameter in the membership matrix is obtained, and the actual application parameters with membership greater than the preset threshold are mapped to 1, and the actual application parameters with membership not greater than the preset threshold are mapped to 0 to obtain a unit matrix;

[0028] A set of basis vectors is formed according to the column vectors of the unit matrix, and the vectors and subspaces in the high-dimensional space are represented and operated by using the set of basis vectors to construct an N-dimensional space;

[0029] The membership matrix is imported into the N-dimensional space, the row vectors in the membership matrix are obtained, the Jaccard similarity algorithm is introduced, and the row vectors in the preset position are used as the reference vectors, and the similarity between the reference vectors and the remaining row vectors is calculated by the Jaccard similarity algorithm;

[0030] If the similarity between a row vector and a reference vector is greater than a preset similarity, obtain the actual application parameters corresponding to the row vector, and calculate the Manhattan distance between all the actual application parameters in the row vector and the cluster centers of the remaining clusters;

[0031] Judge whether there is a situation where the Manhattan distance between the actual application parameters in the row vector and the cluster centers of the remaining clusters is less than a preset distance. If so, re-cluster the corresponding actual application parameters in the row vector into the cluster where the Manhattan distance is less than the preset distance, and update the membership matrix to obtain a corrected membership matrix.

[0032] Further, in a preferred embodiment of the present invention, different types of actual application parameters in the unmanned aerial vehicle are obtained according to the corrected membership matrix, and various types of actual application parameters are compared with the corresponding types of preset application parameters to obtain a comparison result. According to the comparison result, the actual application parameters of the unmanned aerial vehicle are adjusted, specifically:

[0033] Obtain different types of actual application parameters in the unmanned aerial vehicle according to the corrected membership matrix, calculate the hash values between various types of actual application parameters and the corresponding types of preset application parameters through the gray hash algorithm; and determine the coincidence degree between various types of actual application parameters and the corresponding types of preset application parameters according to the hash values;

[0034] Compare the coincidence degree with a preset coincidence degree. If the coincidence degree is greater than the preset coincidence degree, mark the actual application parameters of this type as normal application parameters;

[0035] If the coincidence degree is not greater than the preset coincidence degree, compare the actual application parameters of this type with the corresponding type of preset application parameters to obtain an application parameter deviation value, and adjust the actual application parameters of this type according to the application parameter deviation value.

[0036] On the other hand, the present invention discloses a precise application system for unmanned aerial vehicles based on multi-source data fusion. The precise application system for unmanned aerial vehicles includes a memory and a processor. A precise application method program for unmanned aerial vehicles is stored in the memory. When the precise application method program for unmanned aerial vehicles is executed by the processor, the following steps are implemented:

[0037] Obtain the point cloud data of the crops to be applied with pesticides, construct a three-dimensional model diagram of the crops to be applied with pesticides based on the point cloud data, and generate preset application parameters for applying pesticides to the crops to be applied with pesticides based on the three-dimensional model diagram of the crops to be applied with pesticides;

[0038] Based on the pre-set application parameters, control the drone to apply pesticides to the crops to be treated, and obtain the actual application parameters of the drone at multiple preset time nodes, construct a parameter repository, and import the actual application parameters into the parameter repository;

[0039] Perform clustering processing on the actual application parameters in the parameter repository to obtain the membership matrix of each actual application parameter; and correct the membership matrix to obtain the corrected membership matrix;

[0040] Obtain different types of actual application parameters in the drone according to the corrected membership matrix, compare various types of actual application parameters with the corresponding types of pre-set application parameters to obtain a comparison result, and adjust the actual application parameters of the drone according to the comparison result.

[0041] Further, in a preferred embodiment of the present invention, performing clustering processing on the actual application parameters in the parameter repository to obtain the membership matrix of each actual application parameter specifically includes:

[0042] Convert each actual application parameter into an independent cluster, calculate the cosine similarity between each cluster through the cosine similarity algorithm, and construct a similarity matrix according to the cosine similarity;

[0043] Retrieve the two clusters with the highest similarity according to the similarity matrix, and merge the two clusters with the highest similarity into a new cluster;

[0044] Update the similarity matrix according to the merged new cluster to reflect the similarity between the new cluster and the remaining clusters; retrieve the two clusters with the highest similarity again in the updated similarity matrix, and also merge the two clusters with the highest similarity into a new cluster;

[0045] Repeat the above steps until the desired number of clusters is reached, generate a preliminary clustering result, and calculate the membership between each actual application parameter in each cluster and its clustering center according to the preliminary clustering result, and generate a membership matrix according to the membership.

[0046] Further, in a preferred embodiment of the present invention, correcting the membership matrix to obtain the corrected membership matrix specifically includes:

[0047] Obtain the membership corresponding to each actual application parameter in the membership matrix, map the actual application parameters with membership greater than the preset threshold to 1, and map the actual application parameters with membership not greater than the preset threshold to 0 to obtain a unit matrix;

[0048] Form a basis vector group according to the column vectors of the unit matrix, and use the basis vector group to represent and operate on vectors and subspaces in a high-dimensional space to construct an N-dimensional space;

[0049] Import the membership matrix into the N-dimensional space, obtain the row vectors in the membership matrix, introduce the Jaccard similarity algorithm, and use the row vectors in the preset positions as reference vectors to calculate the similarity between the reference vectors and the remaining row vectors through the Jaccard similarity algorithm;

[0050] If the similarity between a certain row vector and the reference vector is greater than the preset similarity, obtain the actual application parameters corresponding to this row vector, and calculate the Manhattan distance between all the actual application parameters in this row vector and the cluster centers of the remaining clusters;

[0051] Judge whether there is a situation where the Manhattan distance between the actual application parameters in this row vector and the cluster centers of the remaining clusters is less than the preset distance. If so, re-cluster the corresponding actual application parameters in this row vector into the cluster with a Manhattan distance less than the preset distance, and update the membership matrix to obtain a corrected membership matrix.

[0052] Furthermore, in a preferred embodiment of the present invention, different types of actual application parameters in the unmanned aerial vehicle are obtained according to the corrected membership matrix, and various types of actual application parameters are compared with the corresponding types of preset application parameters to obtain a comparison result, and the actual application parameters of the unmanned aerial vehicle are regulated according to the comparison result. Specifically:

[0053] Obtain different types of actual application parameters in the unmanned aerial vehicle according to the corrected membership matrix, calculate the hash values between various types of actual application parameters and the corresponding types of preset application parameters through the gray hash algorithm; and determine the coincidence degree between various types of actual application parameters and the corresponding types of preset application parameters according to the hash values;

[0054] Compare the coincidence degree with the preset coincidence degree. If the coincidence degree is greater than the preset coincidence degree, mark the actual application parameters of this type as normal application parameters;

[0055] If the coincidence degree is not greater than the preset coincidence degree, compare the actual application parameters of this type with the corresponding types of preset application parameters to obtain an application parameter deviation value, and regulate the actual application parameters of this type according to the application parameter deviation value.

[0056] The present invention solves the technical defects existing in the background art and has the following beneficial effects: obtaining the point cloud data of the crop to be sprayed, constructing a three-dimensional model diagram of the crop to be sprayed based on the point cloud data, and generating pre-spraying parameters for spraying the crop to be sprayed based on the three-dimensional model diagram of the crop to be sprayed; controlling a drone to spray the crop to be sprayed based on the pre-spraying parameters, obtaining the actual spraying parameters of the drone at multiple preset time nodes, constructing a parameter storage library, and importing the actual spraying parameters into the parameter storage library; performing clustering processing on the actual spraying parameters in the parameter storage library to obtain a membership matrix of each actual spraying parameter; and correcting the membership matrix to obtain a corrected membership matrix; obtaining different types of actual spraying parameters in the drone according to the corrected membership matrix, comparing various types of actual spraying parameters with the corresponding types of pre-spraying parameters to obtain a comparison result, and regulating the actual spraying parameters of the drone according to the comparison result. By this method, the spraying parameters with deviations can be corrected in a timely manner, the spraying accuracy of the drone can be improved, the spraying efficiency and effect can be improved, and the intelligent regulation of the spraying parameters of the drone can be realized through the application of data processing, algorithm optimization and automatic control technologies, so as to improve the spraying effect and the level of farm crop protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain the drawings of other embodiments without creative efforts.

[0058] Figure 1 FIG. 1 is a first method flow chart of a precise spraying method for drones based on multi-source data fusion;

[0059] Figure 2 FIG. 2 is a second method flow chart of a precise spraying method for drones based on multi-source data fusion;

[0060] Figure 3 FIG. 3 is a system block diagram of a precise spraying system for drones based on multi-source data fusion. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] In order to be able to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0062] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways than those specifically described herein, and thus, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0063] As Figure 1 shown, a first aspect of the present invention discloses a method for precise pesticide application of an unmanned aerial vehicle based on multi-source data fusion, including the following steps:

[0064] S102: Obtain the point cloud data of the crop to be pesticided, construct a three-dimensional model diagram of the crop to be pesticided based on the point cloud data, and generate pre-set pesticide application parameters for pesticiding the crop to be pesticided based on the three-dimensional model diagram of the crop to be pesticided;

[0065] S104: Control the unmanned aerial vehicle to apply pesticides to the crop to be pesticided based on the pre-set pesticide application parameters, obtain the actual pesticide application parameters of the unmanned aerial vehicle at multiple preset time nodes, construct a parameter storage library, and import the actual pesticide application parameters into the parameter storage library;

[0066] S106: Perform clustering processing on the actual pesticide application parameters in the parameter storage library to obtain the membership degree matrix of each actual pesticide application parameter; and correct the membership degree matrix to obtain the corrected membership degree matrix;

[0067] S108: Obtain different types of actual pesticide application parameters in the unmanned aerial vehicle according to the corrected membership degree matrix, compare various types of actual pesticide application parameters with the corresponding types of pre-set pesticide application parameters to obtain a comparison result, and regulate the actual pesticide application parameters of the unmanned aerial vehicle according to the comparison result.

[0068] Further, in a preferred embodiment of the present invention, obtaining the point cloud data of the crop to be pesticided, constructing a three-dimensional model diagram of the crop to be pesticided based on the point cloud data, and generating pre-set pesticide application parameters for pesticiding the crop to be pesticided based on the three-dimensional model diagram of the crop to be pesticided are specifically as follows:

[0069] Scan the crop to be pesticided at multiple angular positions through a laser point cloud device, and obtain the point cloud data reflected by the crop to be pesticided at each angular position to obtain subsets of point cloud data corresponding to different angular positions;

[0070] Introduce the isolation forest algorithm, and calculate the isolation scores of each point cloud data in each subset of power data through the isolation forest algorithm, and eliminate the point cloud data with an isolation score greater than the preset value in the corresponding subset of point cloud data to obtain a filtered subset of point cloud data;

[0071] Construct a three-dimensional coordinate system, import each subset of point cloud data into the three-dimensional coordinate system, and align the point cloud data in each subset of point cloud data in the three-dimensional coordinate system to obtain a complete point cloud model;

[0072] Obtain the relative three-dimensional coordinate values between the point cloud data in the point cloud model in the three-dimensional coordinate system, and construct a three-dimensional model diagram of the crop to be sprayed according to the relative three-dimensional coordinate values; Initialize the starting point of the spraying path of the drone, and obtain the preset spraying range of the drone;

[0073] Based on the starting point of the spraying path, the preset spraying range, and the three-dimensional model diagram of the crop to be sprayed, and combined with the ant colony algorithm, iteratively plan the preset spraying parameters when spraying the crop to be sprayed; Among them, the preset spraying parameters include spraying path, spraying pressure, spraying speed, spraying amount, spraying height, and spray particle size.

[0074] It should be noted that after obtaining the point cloud data of multiple regions of the crop to be sprayed, first clean these point cloud data through the isolation forest algorithm to screen out the outlier point cloud data and improve the reliability. Then splice the point cloud data of each region to obtain a complete and continuous point cloud model. Then obtain the relative three-dimensional coordinates between the point cloud data in the point cloud model, and import the relative three-dimensional coordinates into modeling software such as CAD and SolidWorks to obtain a three-dimensional model diagram of the crop to be sprayed. Then, after iteratively constructing the starting point of the spraying path, the preset spraying range, and the three-dimensional model diagram of the crop to be sprayed through the ant colony algorithm, the preset spraying parameters such as the spraying path, spraying pressure, spraying speed, spraying amount, spraying height, and spray particle size when spraying the crop under ideal conditions are obtained.

[0075] As Figure 2 shown, further, in a preferred embodiment of the present invention, aligning the point cloud data in each subset of point cloud data in the three-dimensional coordinate system to obtain a complete point cloud model is specifically as follows:

[0076] S202: Extract the point cloud features of the point cloud data in each subset of point cloud data, generate feature descriptors for each point cloud data based on the point cloud features, and use the feature descriptors to describe the local geometric attributes of each point cloud data;

[0077] S204: According to the feature descriptors, find the corresponding relationship between the point cloud data by combining the nearest neighbor search method, and perform feature matching on the point cloud data with the closest distance in different subsets of point cloud data according to the corresponding relationship between the point cloud data to obtain several pairs of point cloud data point pairs;

[0078] S206: Based on a number of pairs of point cloud data points obtained by matching, combine the least squares method to estimate the initial transformation matrix between each pair of point cloud data points;

[0079] S208: Refine the initial transformation matrix through a global optimization algorithm until the initial transformation matrix meets the preset requirements, and fuse the aligned point cloud data to generate a complete point cloud model.

[0080] It should be noted that the point cloud feature descriptor is a vector or descriptor used to represent the key feature information of each point in the point cloud data. For each feature point, calculate the feature information of other points within its neighborhood to construct the point cloud feature descriptor. The neighborhood features include the point cloud mean, normal direction, curvature, normal histogram, etc. within the neighborhood. Through this method, the point cloud data obtained from each position area can be quickly stitched together to obtain a complete point cloud model, which can improve the model reconstruction speed and accuracy.

[0081] Furthermore, in a preferred embodiment of the present invention, cluster the actual application parameters in the parameter repository to obtain the membership matrix of each actual application parameter, specifically:

[0082] Convert each actual application parameter into an independent cluster, and calculate the cosine similarity between each cluster through the cosine similarity algorithm, and construct a similarity matrix according to the cosine similarity;

[0083] Retrieve the two clusters with the highest similarity according to the similarity matrix, and merge the two clusters with the highest similarity into a new cluster;

[0084] Update the similarity matrix according to the merged new cluster to reflect the similarity between the new cluster and the remaining clusters; retrieve the two clusters with the highest similarity again in the updated similarity matrix, and also merge the two clusters with the highest similarity into a new cluster;

[0085] Repeat the above steps until the desired number of clusters is reached, generate a preliminary clustering result, and calculate the membership between each actual application parameter in each cluster and its clustering center according to the preliminary clustering result, and generate a membership matrix according to the membership.

[0086] It should be noted that by combining the methods of fuzzy clustering and hierarchical clustering to cluster a large number of actual application parameters in the parameter repository to initially distinguish the types of each actual application parameter, such as quickly clustering a large number of actual application parameters in the parameter repository, it can improve the data processing speed, and by combining the methods of fuzzy clustering and hierarchical clustering, it can avoid the phenomenon of local optimal solutions in the clustering process, reduce the clustering error phenomenon, and improve the reliability.

[0087] Further, in a preferred embodiment of the present invention, the membership matrix is corrected to obtain a corrected membership matrix, specifically as follows:

[0088] Obtain the membership degrees corresponding to each actual application parameter in the membership matrix, map the actual application parameters with membership degrees greater than a preset threshold to 1, and map the actual application parameters with membership degrees not greater than the preset threshold to 0 to obtain an identity matrix;

[0089] Form a basis vector group according to the column vectors of the identity matrix, and use the basis vector group to represent and operate vectors and subspaces in a high-dimensional space to construct an N-dimensional space;

[0090] Import the membership matrix into the N-dimensional space, obtain the row vectors in the membership matrix, introduce the Jaccard similarity algorithm, and use the row vectors in the preset positions as reference vectors to calculate the similarity between the reference vectors and the remaining row vectors through the Jaccard similarity algorithm;

[0091] If the similarity between a certain row vector and the reference vector is greater than the preset similarity, obtain the actual application parameter corresponding to this row vector, and calculate the Manhattan distance between all the actual application parameters in this row vector and the cluster centers of the remaining clusters;

[0092] Judge whether there is a situation where the Manhattan distance between the actual application parameters in this row vector and the cluster centers of the remaining clusters is less than the preset distance. If so, re-cluster the corresponding actual application parameters in this row vector into the cluster with a Manhattan distance less than the preset distance, and update the membership matrix to obtain a corrected membership matrix.

[0093] It should be noted that the actual application parameters with membership degrees greater than the preset threshold are mapped to 1, and the actual application parameters with membership degrees not greater than the preset threshold are mapped to 0 to obtain an identity matrix. Each row in the identity matrix represents the actual application parameters contained in a cluster. And a high-dimensional space (N-dimensional space) is constructed according to the identity matrix, and then the similarity between the reference vector and the remaining row vectors in the matrix is calculated through the Jaccard similarity algorithm. If the similarity between a certain row vector and the reference vector is greater than the preset similarity, it indicates that there is a clustering error phenomenon in the corresponding cluster. At this time, the actual application parameters in the cluster with the clustering error are corrected. Through this method, the actual application parameters with clustering errors can be corrected, thereby improving the clustering data accuracy, and further improving the evaluation accuracy of the application parameters, so as to further improve the application accuracy.

[0094] Further, in a preferred embodiment of the present invention, actual application parameters of different types in the unmanned aerial vehicle are obtained according to the corrected membership degree matrix, and the actual application parameters of various types are compared with the corresponding pre-set application parameters to obtain a comparison result, and the actual application parameters of the unmanned aerial vehicle are adjusted according to the comparison result, specifically:

[0095] Actual application parameters of different types in the unmanned aerial vehicle are obtained according to the corrected membership degree matrix, and the hash values between the actual application parameters of various types and the corresponding pre-set application parameters are calculated through the gray hash algorithm; and the coincidence degree between the actual application parameters of various types and the corresponding pre-set application parameters is determined according to the hash values;

[0096] The coincidence degree is compared with a pre-set coincidence degree. If the coincidence degree is greater than the pre-set coincidence degree, the actual application parameters of this type are marked as normal application parameters;

[0097] If the coincidence degree is not greater than the pre-set coincidence degree, the actual application parameters of this type are compared with the corresponding pre-set application parameters to obtain an application parameter deviation value, and the actual application parameters of this type are adjusted according to the application parameter deviation value.

[0098] It should be noted that after the clustering process of the collected actual application parameters is completed, actual application parameters of different types in the unmanned aerial vehicle are obtained according to the corrected membership degree matrix, such as which are application pressure data, which are application speed data, which are spraying amount data, etc., and then the actual application parameters of various types are compared with the corresponding pre-set application parameters, so as to obtain the coincidence degree between the actual application parameters of the same type and the pre-set application parameters, such as the coincidence degree between the actual application pressure of the unmanned aerial vehicle and the pre-set application pressure. If the coincidence degree is not greater than the pre-set coincidence degree, the actual application parameters of this type are compared with the corresponding pre-set application parameters to obtain an application parameter deviation value, and the actual application parameters of this type are adjusted according to the application parameter deviation value. Through this method, the deviated application parameters can be corrected in time, the application accuracy of the unmanned aerial vehicle can be improved, and the application efficiency and effect can be improved.

[0099] In addition, the method further includes the following steps:

[0100] Obtain the application offset under various pre-set working environment factor combination conditions through the big data network, construct a knowledge graph, and import the application offset under various pre-set working environment factor combination conditions into the knowledge graph;

[0101] Obtain the real-time environment factors during the application process of the unmanned aerial vehicle, import the real-time environment factors into the knowledge graph, and calculate the attention scores between the real-time environment factors and various pre-set working environment factors through the local sensitive attention mechanism;

[0102] Obtain the coincidence rate between the real-time environmental factors and various preset working environmental factors according to the attention scores, and obtain multiple coincidence rates; construct a size sorting table, import the multiple coincidence rates into the size sorting table for sorting, so as to sort out the maximum coincidence rate;

[0103] Obtain the preset working environmental factor corresponding to the maximum coincidence rate, construct a retrieval label according to the preset working environmental factor corresponding to the maximum coincidence rate, and retrieve the knowledge graph based on the retrieval label to obtain the spraying offset of the drone during the spraying process;

[0104] Obtain the preset spraying parameters of the drone, and obtain the current spraying time node of the drone. Determine the current spraying focus position of the nozzle according to the current spraying time node and the preset spraying parameters, and correct the current spraying focus position according to the spraying offset.

[0105] It should be noted that the spraying drift phenomenon refers to the position deviation generated by the drone during the spraying process. This deviation may be due to the influence of environmental factors, resulting in the instability of the drone's flight trajectory. For example, when the wind speed is large, the influence of the wind on the drone will cause the position of the sprayer to shift, so the spraying liquid will not be evenly sprayed on the crops. Similarly, changes in air pressure and temperature will also affect the aerodynamic behavior of the drone body, resulting in the spraying drift phenomenon. Through this method, the spraying drift phenomenon of the drone can be corrected according to the actual environmental factors, and the spraying accuracy can be improved.

[0106] As Figure 3 shown, on the other hand, the present invention discloses a precise spraying system for drones based on multi-source data fusion. The precise spraying system for drones includes a memory 11 and a processor 20. A precise spraying method program for drones is stored in the memory 11. When the precise spraying method program for drones is executed by the processor 20, the following steps are implemented:

[0107] Obtain the point cloud data of the crops to be sprayed, construct a three-dimensional model diagram of the crops to be sprayed based on the point cloud data, and generate preset spraying parameters for spraying the crops to be sprayed based on the three-dimensional model diagram of the crops to be sprayed;

[0108] Control the drone to spray the crops to be sprayed based on the preset spraying parameters, obtain the actual spraying parameters of the drone at multiple preset time nodes, construct a parameter storage library, and import the actual spraying parameters into the parameter storage library;

[0109] Perform clustering processing on the actual spraying parameters in the parameter storage library to obtain the membership matrix of each actual spraying parameter; and correct the membership matrix to obtain the corrected membership matrix;

[0110] Based on the corrected membership matrix, obtain the actual pesticide application parameters of different types in the unmanned aerial vehicle, compare the actual pesticide application parameters of various types with the preset pesticide application parameters of the corresponding types to obtain a comparison result, and regulate the actual pesticide application parameters of the unmanned aerial vehicle according to the comparison result.

[0111] Furthermore, in a preferred embodiment of the present invention, perform clustering processing on the actual pesticide application parameters in the parameter repository to obtain the membership matrix of each actual pesticide application parameter, specifically:

[0112] Convert each actual pesticide application parameter into an independent cluster, calculate the cosine similarity between each cluster through the cosine similarity algorithm, and construct a similarity matrix according to the cosine similarity;

[0113] Retrieve the two clusters with the highest similarity according to the similarity matrix, and merge the two clusters with the highest similarity into a new cluster;

[0114] Update the similarity matrix according to the merged new cluster to reflect the similarity between the new cluster and the remaining clusters; retrieve the two clusters with the highest similarity again in the updated similarity matrix, and also merge the two clusters with the highest similarity into a new cluster;

[0115] Repeat the above steps until the desired number of clusters is reached to generate a preliminary clustering result, and calculate the membership degree between each actual pesticide application parameter in each cluster and its clustering center according to the preliminary clustering result, and generate a membership matrix according to the membership degree.

[0116] Furthermore, in a preferred embodiment of the present invention, correct the membership matrix to obtain a corrected membership matrix, specifically:

[0117] Obtain the membership degree corresponding to each actual pesticide application parameter in the membership matrix, map the actual pesticide application parameters with membership degrees greater than the preset threshold to 1, and map the actual pesticide application parameters with membership degrees not greater than the preset threshold to 0 to obtain a unit matrix;

[0118] Form a basis vector group according to the column vectors of the unit matrix, and use the basis vector group to represent and operate on vectors and subspaces in a high-dimensional space to construct an N-dimensional space;

[0119] Import the membership matrix into the N-dimensional space, obtain the row vectors in the membership matrix, introduce the Jaccard similarity algorithm, and use the row vectors in the preset position as the reference vector, and calculate the similarity between the reference vector and the remaining row vectors through the Jaccard similarity algorithm;

[0120] If the similarity between a certain row vector and the reference vector is greater than the preset similarity, obtain the actual application parameters corresponding to this row vector, and calculate the Manhattan distance between all the actual application parameters in this row vector and the cluster centers of the remaining clusters;

[0121] Determine whether there is a case where the Manhattan distance between the actual application parameters in this row vector and the cluster centers of the remaining clusters is less than the preset distance. If so, re-cluster the corresponding actual application parameters in this row vector into the cluster with a Manhattan distance less than the preset distance, and update the membership matrix to obtain the corrected membership matrix.

[0122] Further, in a preferred embodiment of the present invention, obtain different types of actual application parameters in the unmanned aerial vehicle according to the corrected membership matrix, compare various types of actual application parameters with the corresponding type of preset application parameters to obtain a comparison result, and regulate the actual application parameters of the unmanned aerial vehicle according to the comparison result. Specifically:

[0123] Obtain different types of actual application parameters in the unmanned aerial vehicle according to the corrected membership matrix, calculate the hash values between various types of actual application parameters and the corresponding type of preset application parameters through the gray hash algorithm; and determine the coincidence degree between various types of actual application parameters and the corresponding type of preset application parameters according to the hash values;

[0124] Compare the coincidence degree with the preset coincidence degree. If the coincidence degree is greater than the preset coincidence degree, mark the actual application parameters of this type as normal application parameters;

[0125] If the coincidence degree is not greater than the preset coincidence degree, compare the actual application parameters of this type with the corresponding type of preset application parameters to obtain a deviation value of the application parameters, and regulate the actual application parameters of this type according to the deviation value of the application parameters.

[0126] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the couplings, direct couplings, or communication connections between the various components shown or discussed with each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be electrical, mechanical, or other forms.

[0127] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] In addition, in each embodiment of the present invention, each functional unit may be fully integrated into one processing unit, or each unit may be separately regarded as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated unit may be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0129] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks and other various media that can store program codes.

[0130] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the methods of the various embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks, or optical disks and other various media that can store program codes.

[0131] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, and all of them should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for precise pesticide application of unmanned aerial vehicle based on multi-source data fusion, characterized in that Including the following steps: Obtain the point cloud data of the crop to be sprayed, construct a three-dimensional model diagram of the crop to be sprayed based on the point cloud data, and generate the pre-spraying parameters for spraying the crop to be sprayed based on the three-dimensional model diagram of the crop to be sprayed; Control the drone to spray the crop to be sprayed based on the pre-spraying parameters, obtain the actual spraying parameters of the drone at multiple preset time nodes, construct a parameter repository, and import the actual spraying parameters into the parameter repository; Perform clustering processing on the actual spraying parameters in the parameter repository to obtain the membership matrix of each actual spraying parameter; and correct the membership matrix to obtain the corrected membership matrix; Obtain different types of actual spraying parameters in the drone according to the corrected membership matrix, compare various types of actual spraying parameters with the corresponding types of pre-spraying parameters to obtain a comparison result, and adjust the actual spraying parameters of the drone according to the comparison result; Among them, performing clustering processing on the actual spraying parameters in the parameter repository to obtain the membership matrix of each actual spraying parameter is specifically: Convert each actual spraying parameter into an independent cluster, calculate the cosine similarity between each cluster through the cosine similarity algorithm, and construct a similarity matrix according to the cosine similarity; Retrieve the two clusters with the highest similarity according to the similarity matrix, and merge the two clusters with the highest similarity into a new cluster; Update the similarity matrix according to the merged new cluster to reflect the similarity between the new cluster and the remaining clusters; retrieve the two clusters with the highest similarity again in the updated similarity matrix, and also merge the two clusters with the highest similarity into a new cluster; Repeat the above steps until the desired number of clusters is reached, generate a preliminary clustering result, and calculate the membership between each actual spraying parameter in each cluster and its clustering center according to the preliminary clustering result, and generate a membership matrix according to the membership; Among them, correcting the membership matrix to obtain the corrected membership matrix is specifically: Obtain the membership corresponding to each actual spraying parameter in the membership matrix, map the actual spraying parameters with membership greater than the preset threshold to 1, and map the actual spraying parameters with membership not greater than the preset threshold to 0 to obtain an identity matrix; Form a base vector group according to the column vectors of the identity matrix, and use the base vector group to represent and operate vectors and subspaces in a high-dimensional space to construct an N-dimensional space; Import the membership matrix into the N-dimensional space, obtain the row vectors in the membership matrix, introduce the Jaccard similarity algorithm, and use the row vectors in the preset position as the reference vector, and calculate the similarity between the reference vector and the remaining row vectors through the Jaccard similarity algorithm; If the similarity between a certain row vector and the reference vector is greater than the preset similarity, obtain the actual spraying parameter corresponding to the row vector, and calculate the Manhattan distance between all the actual spraying parameters in the row vector and the clustering centers of the remaining clusters; Determine whether there is a Manhattan distance between the actual application parameters in this row vector and the cluster centers of the remaining clusters that is less than the preset distance. If so, re-cluster the corresponding actual application parameters in this row vector into the cluster with a Manhattan distance less than the preset distance, and update the membership matrix to obtain a corrected membership matrix.

2. The method for precise pesticide application of an unmanned aerial vehicle based on multi-source data fusion according to claim 1, characterized in that Obtain the point cloud data of the crop to be sprayed, construct a three-dimensional model diagram of the crop to be sprayed based on the point cloud data, and generate the preset application parameters for spraying the crop to be sprayed based on the three-dimensional model diagram of the crop to be sprayed. Specifically: Scan the crop to be sprayed at multiple angular positions through a laser point cloud device, and obtain the point cloud data fed back by the crop to be sprayed at each angular position, obtaining subsets of point cloud data corresponding to different angular positions; Introduce the Isolation Forest algorithm, and calculate the isolation scores of the point cloud data in each subset of power data through the Isolation Forest algorithm. Remove the point cloud data with an isolation score greater than the preset value from the corresponding subset of point cloud data, obtaining a filtered subset of point cloud data; Construct a three-dimensional coordinate system, import each subset of point cloud data into the three-dimensional coordinate system, and perform alignment processing on the point cloud data in each subset of point cloud data in the three-dimensional coordinate system to obtain a complete point cloud model; Obtain the relative three-dimensional coordinate values between the point cloud data in the point cloud model in the three-dimensional coordinate system, and construct a three-dimensional model diagram of the crop to be sprayed based on the relative three-dimensional coordinate values; Initialize the starting point of the spraying path of the drone and obtain the preset spraying range of the drone; Based on the starting point of the spraying path, the preset spraying range, and the three-dimensional model diagram of the crop to be sprayed, and in combination with the Ant Colony algorithm, iteratively plan to obtain the preset application parameters for spraying the crop to be sprayed; Among them, the preset application parameters include the spraying path, spraying pressure, spraying speed, spraying volume, spraying height, and spray particle size.

3. The method for precise pesticide application of an unmanned aerial vehicle based on multi-source data fusion according to claim 2, wherein, Perform alignment processing on the point cloud data in each subset of point cloud data in the three-dimensional coordinate system to obtain a complete point cloud model. Specifically: Extract the point cloud features of the point cloud data in each subset of point cloud data, generate feature descriptors for each point cloud data based on the point cloud features, so as to describe the local geometric attributes of each point cloud data through the feature descriptors; According to the feature descriptors, combine the nearest neighbor search method to find the corresponding relationship between the point cloud data, and perform feature matching on the point cloud data with the closest distance in different subsets of point cloud data according to the corresponding relationship between the point cloud data, obtaining several pairs of point cloud data point pairs; Based on the several pairs of point cloud data point pairs obtained by matching, combine the least squares method to estimate the initial transformation matrix between each pair of point cloud data point pairs; Refine the initial transformation matrix through a global optimization algorithm until the initial transformation matrix meets the preset requirements, and fuse the aligned point cloud data to generate a complete point cloud model.

4. The method for precise pesticide application of an unmanned aerial vehicle based on multi-source data fusion according to claim 1, wherein Obtain the actual application parameters of different types in the UAV according to the corrected membership degree matrix, compare the actual application parameters of various types with the pre-set application parameters of the corresponding types to obtain a comparison result, and regulate the actual application parameters of the UAV according to the comparison result. Specifically: Obtain the actual application parameters of different types in the UAV according to the corrected membership degree matrix, and calculate the hash values between the actual application parameters of various types and the pre-set application parameters of the corresponding types through the gray hash algorithm; And determine the coincidence degree between the actual application parameters of various types and the pre-set application parameters of the corresponding types according to the hash values; Compare the coincidence degree with the pre-set coincidence degree. If the coincidence degree is greater than the pre-set coincidence degree, mark the actual application parameters of this type as normal application parameters; If the coincidence degree is not greater than the pre-set coincidence degree, compare the actual application parameters of this type with the pre-set application parameters of the corresponding type to obtain an application parameter deviation value, and regulate the actual application parameters of this type according to the application parameter deviation value.

5. A precise drone spraying system based on multi-source data fusion, characterized in that, The UAV precise application system includes a memory and a processor. A UAV precise application method program is stored in the memory. When the UAV precise application method program is executed by the processor, the following steps are implemented: Obtain the point cloud data of the crop to be applied with pesticides, construct a three-dimensional model diagram of the crop to be applied with pesticides based on the point cloud data, and generate pre-set application parameters for applying pesticides to the crop to be applied with pesticides based on the three-dimensional model diagram of the crop to be applied with pesticides; Control the UAV to apply pesticides to the crop to be applied with pesticides based on the pre-set application parameters, obtain the actual application parameters of the UAV at multiple preset time nodes, construct a parameter storage library, and import the actual application parameters into the parameter storage library; Perform clustering processing on the actual application parameters in the parameter storage library to obtain a membership degree matrix of each actual application parameter; and correct the membership degree matrix to obtain a corrected membership degree matrix; Obtain the actual application parameters of different types in the UAV according to the corrected membership degree matrix, compare the actual application parameters of various types with the pre-set application parameters of the corresponding types to obtain a comparison result, and regulate the actual application parameters of the UAV according to the comparison result; Among them, performing clustering processing on the actual application parameters in the parameter storage library to obtain a membership degree matrix of each actual application parameter is specifically: Convert each actual application parameter into an independent cluster, calculate the cosine similarity between each cluster through the cosine similarity algorithm, and construct a similarity matrix according to the cosine similarity; Retrieve the two clusters with the highest similarity according to the similarity matrix, and merge the two clusters with the highest similarity into a new cluster; Update the similarity matrix according to the merged new cluster to reflect the similarity between the new cluster and the remaining clusters; retrieve the two clusters with the highest similarity again in the updated similarity matrix, and also merge the two clusters with the highest similarity into a new cluster; Repeat the above steps until the desired number of clusters is reached, generate a preliminary clustering result, and calculate the membership degree between each actual application parameter and its clustering center in each cluster according to the preliminary clustering result, and generate a membership matrix according to the membership degree; Among them, the membership matrix is corrected to obtain a corrected membership matrix, specifically: Obtain the membership degree corresponding to each actual application parameter in the membership matrix, map the actual application parameters with membership degrees greater than the preset threshold to 1, and map the actual application parameters with membership degrees not greater than the preset threshold to 0 to obtain an identity matrix; Form a set of basis vectors according to the column vectors of the identity matrix, and use the set of basis vectors to represent and operate on vectors and subspaces in a high-dimensional space to construct an N-dimensional space; Import the membership matrix into the N-dimensional space, obtain the row vectors in the membership matrix, introduce the Jaccard similarity algorithm, and use the row vectors in the preset position as the reference vectors, and calculate the similarity between the reference vectors and the remaining row vectors through the Jaccard similarity algorithm; If the similarity between a certain row vector and the reference vector is greater than the preset similarity, obtain the actual application parameter corresponding to the row vector, and calculate the Manhattan distance between all the actual application parameters in the row vector and the clustering centers of the remaining clusters; Judge whether there is a situation where the Manhattan distance between the actual application parameters in the row vector and the clustering centers of the remaining clusters is less than the preset distance. If so, re-cluster the corresponding actual application parameters in the row vector into the cluster with a Manhattan distance less than the preset distance, and update the membership matrix to obtain a corrected membership matrix.

6. The precision pesticide application system for unmanned aerial vehicles based on multi-source data fusion according to claim 5, characterized in that, Obtain different types of actual application parameters in the drone according to the corrected membership matrix, compare various types of actual application parameters with the corresponding type of preset application parameters to obtain a comparison result, and regulate the actual application parameters of the drone according to the comparison result, specifically: Obtain different types of actual application parameters in the drone according to the corrected membership matrix, and calculate the hash values between various types of actual application parameters and the corresponding type of preset application parameters through the gray hash algorithm; And determine the coincidence degree between various types of actual application parameters and the corresponding type of preset application parameters according to the hash value; Compare the coincidence degree with the preset coincidence degree. If the coincidence degree is greater than the preset coincidence degree, mark the actual application parameters of this type as normal application parameters; If the coincidence degree is not greater than the preset coincidence degree, compare the actual application parameters of this type with the corresponding type of preset application parameters to obtain a deviation value of the application parameters, and regulate the actual application parameters of this type according to the deviation value of the application parameters.

Images