Method and System for Monitoring Land Planning Resources Based on UAV Remote Sensing

Abstract

The present invention relates to the field of image recognition technology, and more particularly to a method and system for land planning resource monitoring based on unmanned aerial vehicle remote sensing. The method comprises: selecting a plurality of preferred initial clustering centers from all pixels based on the grayscale difference and distance between each pixel in each connected domain and other pixels, and the grayscale difference and distance between each pixel in each connected domain and pixels in adjacent connected domains of each connected domain; obtaining a plurality of clusters in a land remote sensing image, merging some adjacent clusters based on the difference between the grayscale distributions of all pixels in two clusters, and recording the merged clusters as marked clusters; and performing land resource monitoring and planning based on the plurality of marked clusters in the land remote sensing image. The present invention reduces the number of clustering iterations, improves clustering efficiency, and also improves the accuracy of land resource planning.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a land planning resource monitoring method and system based on unmanned aerial vehicle (UAV) remote sensing. Background Art

[0002] Land planning and resource monitoring refers to the process of monitoring and evaluating land use and resource conditions using various technical means to support land planning, resource management, and environmental protection. Land planning and resource monitoring can provide scientific evidence and support for decision makers, promoting the rational use and sustainable development of land resources. One method for land planning and resource monitoring is to acquire remote sensing images of land using drone technology.

[0003] When land remote sensing images are used for land resource planning and monitoring, the K-means clustering algorithm can be used to divide land remote sensing images into different categories, and land planning can be carried out by monitoring different categories of land. However, when the K-means clustering algorithm is used to divide land remote sensing images into different categories of land, since the initial cluster centers are randomly selected, some initial cluster centers require more clustering to obtain the optimal clustering results, which increases the amount of computer calculations and the time required for calculation. In addition, the initial cluster centers are sensitive to the cluster structure, and there are great differences in the clustering results corresponding to different initial cluster centers. Therefore, the selection of the initial cluster centers affects the final clustering results, resulting in a reduction in the accuracy of land resource planning. Summary of the Invention

[0004] The present invention provides a land planning resource monitoring method and system based on UAV remote sensing to solve existing problems.

[0005] The land planning resource monitoring method and system based on UAV remote sensing of the present invention adopts the following technical solutions:

[0006] One embodiment of the present invention provides a land planning resource monitoring method based on UAV remote sensing, the method comprising the following steps:

[0007] Acquisition of land remote sensing images;

[0008] Connected domain detection is performed on the land remote sensing image to obtain several connected domains. Based on the grayscale difference and distance between each pixel in each connected domain and other pixels, and the grayscale difference and distance between each pixel in each connected domain and the pixels in the connected domains adjacent to each connected domain, several optimal initial clustering centers are screened from all pixels.

[0009] All pixels in the land remote sensing image are clustered according to several preferred initial clustering centers to obtain several clusters. Based on the difference between the grayscale distributions of all pixels in two clusters, some adjacent clusters are merged and the merged clusters are recorded as marked clusters.

[0010] The area ratio of each type of land is obtained based on several labeled clusters in land remote sensing images, and land resource monitoring and planning are carried out based on the area ratio of each type of land.

[0011] Furthermore, the method of selecting a plurality of preferred initial cluster centers from all pixels based on the grayscale difference and distance between each pixel in each connected domain and other pixels, and the grayscale difference and distance between each pixel in each connected domain and pixels in adjacent connected domains of each connected domain, includes the following specific steps:

[0012] According to the grayscale difference between each pixel in each connected domain and all other pixels, the distance between each pixel in each connected domain and the center of the connected domain, and the grayscale difference between each pixel in each connected domain and all pixels in adjacent connected domains, the possibility of each pixel in each connected domain being the initial cluster center is obtained;

[0013] When the probability that each pixel point in each connected domain is an initial cluster center point is greater than or equal to a preset first threshold ε, the pixel point is recorded as a candidate initial cluster center point;

[0014] According to the distance between each candidate initial cluster center point in each connected domain and all candidate initial cluster center points in the connected domains adjacent to each connected domain, and the possibility of each pixel point in each connected domain being the initial cluster center point, the possibility of each candidate initial cluster center point in each connected domain being the preferred initial cluster center point is obtained;

[0015] The candidate initial cluster center point with the greatest possibility among all candidate initial cluster center points in each connected domain is selected as the preferred initial cluster center point, and is recorded as the preferred initial cluster center point.

[0016] Furthermore, the possibility of each pixel in each connected domain being the initial cluster center is obtained based on the grayscale difference between each pixel in each connected domain and all other pixels, the distance between each pixel in each connected domain and the center of the connected domain, and the grayscale difference between each pixel in each connected domain and all pixels in adjacent connected domains. The specific calculation method is as follows:

[0017]

[0018] Where H i,jRepresents the gray value of the jth pixel in the i-th connected component, H i，r represents the gray value of the rth pixel in the i-th connected domain, n represents the number of all pixels in each connected domain, d i,j H represents the distance between the jth pixel in the i-th connected domain and the center point of the connected domain, i，e represents the mean grayscale value of all pixels in the e-th adjacent connected domain of the i-th connected domain, m represents the number of all adjacent connected domains of each connected domain, || is the absolute value symbol, norm() represents the linear normalization function, Q i,j Indicates the possibility that the jth pixel in the i-th connected component is the initial cluster center.

[0019] Furthermore, the possibility of each candidate initial cluster center point in each connected domain being the preferred initial cluster center point is obtained based on the distance between each candidate initial cluster center point in each connected domain and all candidate initial cluster center points in connected domains adjacent to each connected domain, and the possibility of each pixel point in each connected domain being the initial cluster center point. The specific calculation method includes the following:

[0020]

[0021] Where, d i，c，e，v represents the distance between the cth candidate initial cluster center point in the ith connected domain and the vth candidate initial cluster center point in the eth adjacent connected domain, M represents the number of all candidate initial cluster centers in any connected domain adjacent to each connected domain, N represents the number of all connected domains adjacent to each connected domain, Q i，j Indicates the possibility that the jth candidate initial cluster center point in the i-th connected domain is the initial cluster center point, Q i，min Indicates the minimum value of the possibility that all initial cluster centers in the i-th connected domain are the initial cluster centers, Q i，max W represents the maximum value of the possibility that all initial cluster centers in the i-th connected domain are the initial cluster centers. i，c Indicates the possibility of the cth candidate initial cluster center point in the i-th connected component being the preferred initial cluster center point.

[0022] Furthermore, clustering all pixels in the land remote sensing image according to several preferred initial clustering center points to obtain several clusters includes the following specific steps:

[0023] According to all the preferred initial cluster center points, all the pixels in the land remote sensing image are clustered by the K-means clustering algorithm to obtain several clusters.

[0024] Furthermore, the method of merging some adjacent clusters according to the difference between the grayscale distributions of all pixels in the two clusters and recording the merged cluster as a marked cluster includes the following specific steps:

[0025] According to the difference between the grayscale distribution of all pixels in the two clusters and the grayscale difference of all pixels in the two clusters, the possibility of merging the two clusters is obtained;

[0026] Adjacent clusters whose merging probability between the two clusters is greater than a preset second threshold β are merged, and all clusters after the merger are recorded as marked clusters.

[0027] Furthermore, the method of obtaining the area ratio of each type of land based on a plurality of labeled clusters in the land remote sensing image and performing land resource monitoring and planning based on the area ratio of each type of land includes the following specific steps:

[0028] Collect a large number of land remote sensing images, obtain the areas corresponding to several labeled clusters in each land remote sensing image, record the images containing the areas corresponding to several labeled clusters in each land remote sensing image as each reference image, use a large number of reference images to train the CNN neural network, and obtain a trained CNN neural network, where the loss function is the cross entropy loss function;

[0029] Through the trained CNN neural network, the images of the areas corresponding to several labeled clusters in the land remote sensing image are input, and the output is the land of different categories;

[0030] The specific calculation method for the area ratio of each type of land is as follows:

[0031]

[0032] Where, MJ represents the total area of land remote sensing image, MJ u，L represents the area of the region corresponding to the Lth labeled cluster of the uth type of land, G represents the number of all labeled clusters corresponding to each type of land, μ u represents the area proportion of the u-th category of land;

[0033] Land planning resources are monitored based on the area proportion of each type of land and combined with the GeoNode platform.

[0034] Furthermore, the connected domain detection is performed on the land remote sensing image to obtain a plurality of connected domains, and the specific steps include the following:

[0035] By detecting connected domains in land remote sensing images based on a depth-first search algorithm, several connected domains in the land remote sensing images are obtained.

[0036] The present invention also provides a land planning resource monitoring system based on UAV remote sensing, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the system implements any one of the steps of the above-mentioned land planning resource monitoring method based on UAV remote sensing.

[0037] The beneficial effects of the technical solution of the present invention are as follows: the present invention selects several preferred initial cluster center points from all pixel points based on the grayscale difference and distance between each pixel point in each connected domain and other pixel points, and the grayscale difference and distance between each pixel point in each connected domain and the pixel points in the connected domain adjacent to each connected domain, thereby improving the accuracy of selecting the initial cluster center points; clusters all pixel points in the land remote sensing image according to the several preferred initial cluster center points to obtain several clusters, merges some adjacent clusters according to the difference between the grayscale distributions of all pixel points in two clusters, and records the merged clusters as marked clusters, thereby improving the accuracy of clustering; obtains the area proportion of each category of land according to the several marked clusters in the land remote sensing image, and performs land resource monitoring and planning according to the area proportion of each category of land, thereby reducing the number of clustering iterations, improving the efficiency of clustering, and also improving the accuracy of land resource planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 This is a flow chart of the steps of the land planning resource monitoring method based on UAV remote sensing of the present invention;

[0040] Figure 2 Flowchart for land planning resource monitoring using UAV remote sensing. DETAILED DESCRIPTION

[0041] To further illustrate the technical means and effects of the present invention to achieve the intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effects of the land planning resource monitoring method and system based on drone remote sensing proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable form.

[0042] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0043] The specific scheme of the land planning resource monitoring method and system based on UAV remote sensing provided by the present invention is described in detail below with reference to the accompanying drawings.

[0044] See also Figure 1 , which shows a flowchart of a land planning resource monitoring method based on UAV remote sensing provided by an embodiment of the present invention, the method includes the following steps:

[0045] Step S001: Collect land remote sensing images.

[0046] It should be noted that since land resources include cultivated land, gardens, woodlands, grasslands and construction land, it is necessary to obtain land remote sensing images based on drones, and use land remote sensing images to divide different categories of land resources, so as to complete the monitoring of changes in different categories of land resources.

[0047] Specifically, a panoramic camera is installed on a UAV to collect images of land resources. The images of land resources are preprocessed by grayscale conversion and Gaussian filtering to remove noise, and the preprocessed images are recorded as land remote sensing images.

[0048] Among them, grayscale conversion and Gaussian filtering denoising are both well-known technologies and will not be described in detail here.

[0049] At this point, the land remote sensing image is obtained.

[0050] Step S002: Connected domain detection is performed on the land remote sensing image to obtain several connected domains. Based on the grayscale difference and distance between each pixel point in each connected domain and other pixel points, and the grayscale difference and distance between each pixel point in each connected domain and the pixel points in the connected domain adjacent to each connected domain, several preferred initial cluster center points are screened out from all pixel points.

[0051] It should be noted that since land remote sensing images contain multiple different types of land resource categories, that is, the same land resource category may be one area in the land remote sensing image, further analysis is performed by obtaining the connected domain in the land remote sensing image.

[0052] Specifically, connected domains in the land remote sensing image are detected by using a depth-first search (DFS) algorithm to obtain several connected domains. The depth-first search (DFS) algorithm is a well-known technology and will not be described in detail here.

[0053] It should be further explained that since land remote sensing images contain several connected domains, each connected domain has at least one initial cluster center; therefore, the greater the probability that the pixels in each connected domain or the surrounding pixels are clustered into one category. The smaller the grayscale difference between each pixel in each connected domain and all other pixels, the greater the probability that each pixel in the connected domain is the initial cluster center. Conversely, the smaller the grayscale difference between each pixel in the connected domain and all other pixels, the smaller the probability that each pixel in the connected domain is the initial cluster center. In order to distinguish between different categories, the grayscale value of the pixels in each cluster must be significantly different from the grayscale values of the pixels in other clusters. Therefore, further analysis is performed by comparing the grayscale value of each pixel in each connected domain with the grayscale values of the pixels in the surrounding connected domains.

[0054] Specifically, a reference coordinate system is constructed with the lower-left pixel of the land remote sensing image as the origin, the horizontal axis pointing rightward, and the vertical axis pointing upward. The coordinates of the center point of each connected domain are obtained by summing the horizontal and vertical coordinates of all pixels in the domain and dividing by the number of pixels.

[0055] Based on the difference between the grayscale value of each pixel in each connected domain and the grayscale values of all other pixels, the distance between each pixel in each connected domain and the center point of the connected domain, and the difference between the grayscale value of each pixel in each connected domain and the average grayscale value of all pixels in adjacent connected domains, the possibility of each pixel in each connected domain being the initial cluster center is obtained. As an embodiment, the specific calculation method is:

[0056]

[0057] Where H i,j Represents the gray value of the jth pixel in the i-th connected component, H i，r represents the gray value of the rth pixel in the i-th connected domain, n represents the number of all pixels in each connected domain, d i,j H represents the distance between the jth pixel in the i-th connected domain and the center point of the connected domain, i，e represents the mean grayscale value of all pixels in the e-th adjacent connected domain of the i-th connected domain, m represents the number of all adjacent connected domains of each connected domain, || is the absolute value symbol, norm() represents the linear normalization function, Q i，j Indicates the possibility that the jth pixel in the i-th connected component is the initial cluster center. In this embodiment, the distance is the Euclidean distance.

[0058] in, Indicates the difference between the grayscale value of each pixel in each connected domain and the grayscale values of all other pixels. The larger the difference, the smaller the possibility that the pixel is the initial cluster center. Conversely, the greater the possibility that the pixel is the initial cluster center. The larger the distance between each pixel in each connected domain and the connected domain center, the smaller the possibility that the pixel is the initial cluster center. Conversely, the greater the possibility that the pixel is the initial cluster center. It represents the difference between the gray value of each pixel in each connected domain and the mean gray value of all pixels in the connected domains adjacent to each connected domain. The smaller the difference, the less likely the pixel is to be the initial cluster center. On the contrary, the greater the possibility that the pixel is to be the initial cluster center.

[0059] A first threshold ε is preset, wherein this embodiment uses ε=0.88 as an example for description. This embodiment does not impose any specific limitation, and ε may be determined according to specific implementation circumstances. When the probability that each pixel point in each connected domain is an initial cluster center point is greater than or equal to the preset first threshold ε, the pixel point is recorded as a candidate initial cluster center point.

[0060] It should be noted that the greater the distance between each candidate initial cluster center point and other candidate initial cluster center points, the greater the possibility that the candidate initial cluster center point is the initial cluster center point of a cluster. Conversely, the smaller the distance between each candidate initial cluster center point and other candidate initial cluster center points, it means that there are many candidate initial cluster center points around the candidate initial cluster center point, and the possibility that the candidate initial cluster center point is the initial cluster center point of a cluster is smaller.

[0061] Specifically, based on the distance between each candidate initial cluster center point in each connected domain and all candidate initial cluster center points in connected domains adjacent to each connected domain, and the possibility of each pixel point in each connected domain being an initial cluster center point, the possibility of each candidate initial cluster center point in each connected domain being a preferred initial cluster center point is obtained. As an embodiment, the specific calculation method is:

[0062]

[0063] Where, d i，c，e，v represents the distance between the cth candidate initial cluster center point in the ith connected domain and the vth candidate initial cluster center point in the eth adjacent connected domain, M represents the number of all candidate initial cluster centers in any connected domain adjacent to each connected domain, N represents the number of all connected domains adjacent to each connected domain, Q i，jIndicates the possibility that the jth candidate initial cluster center point in the i-th connected domain is the initial cluster center point, Q i，min Indicates the minimum value of the possibility that all initial cluster centers in the i-th connected domain are the initial cluster centers, Q i，max W represents the maximum value of the possibility that all initial cluster centers in the i-th connected domain are the initial cluster centers. i，c Indicates the possibility of the cth candidate initial cluster center point in the i-th connected component being the preferred initial cluster center point.

[0064] The greater the distance between each candidate initial cluster center point in each connected domain and all candidate initial cluster center points in the connected domains adjacent to each connected domain, the greater the possibility that the candidate initial cluster center point is selected as the preferred initial cluster center point. Conversely, the smaller the possibility that the candidate initial cluster center point is selected as the preferred initial cluster center point. The greater the possibility that each initial cluster center point in each connected domain is selected as the initial cluster center point, the greater the possibility that the candidate initial cluster center point is selected as the preferred initial cluster center point. Conversely, the smaller the possibility that the candidate initial cluster center point is selected as the preferred initial cluster center point.

[0065] The candidate initial cluster center point with the greatest possibility among all candidate initial cluster center points in each connected domain is selected as the preferred initial cluster center point, and is recorded as the preferred initial cluster center point.

[0066] At this point, all the optimal initial cluster center points are obtained.

[0067] Step S003: Cluster all pixels in the land remote sensing image according to several preferred initial clustering center points to obtain several clusters. Merge some adjacent clusters according to the difference between the grayscale distributions of all pixels in two clusters, and record the merged clusters as marked clusters.

[0068] It should be noted that, when analyzing the connected domain, the land resources of the same cultivated land may be divided into two or more areas due to the presence of a small river in the middle. Therefore, clustering is performed by selecting the preferred initial cluster center point, and the similar clusters are merged according to the grayscale difference of the pixels in different clusters after clustering and the difference between the grayscale distribution of all pixels in different clusters.

[0069] Specifically, all pixels in the land remote sensing image are clustered using the K-means clustering algorithm based on all the preferred initial cluster centers to obtain a number of clusters. The K-means clustering algorithm is a well-known technique and will not be described in detail here.

[0070] According to the difference between the grayscale distribution of all pixels in the two clusters and the grayscale difference of all pixels in the two clusters, the possibility of merging the two clusters is obtained.

[0071] A second threshold β is preset, wherein this embodiment is described by taking β=0.9 as an example, and this embodiment does not make any specific limitation, wherein β may be determined according to specific implementation conditions.

[0072] Adjacent clusters whose merging probability between the two clusters is greater than a preset second threshold β are merged, and all clusters after the merger are recorded as marked clusters.

[0073] At this point, several labeled clusters are obtained.

[0074] Step S004: land resource monitoring and planning is performed based on a number of labeled clusters in the land remote sensing image.

[0075] It should be noted that in practice, some lands of the same category cannot all be together and may be separated by other categories of land. In this way, the same category of land resources in several labeled clusters may correspond to multiple labeled clusters. Therefore, the gradient resource category division of several labeled clusters is realized through neural networks.

[0076] Specifically, a large number of land remote sensing images are collected, and regions corresponding to several labeled clusters are obtained from each land remote sensing image. Images containing regions corresponding to the labeled clusters in each land remote sensing image are recorded as reference images. A CNN neural network is trained using these reference images to obtain a trained CNN neural network, where the loss function is a cross-entropy loss function. The CNN neural network structure is well known in the art and will not be described in detail here.

[0077] Through the trained CNN neural network, the images of areas corresponding to several labeled clusters in the land remote sensing image are input, and the output is land of different categories.

[0078] According to the occupied area of each type of land and the total area of the land remote sensing image, the area ratio of each type of land is obtained. As an embodiment, the specific calculation method is as follows:

[0079]

[0080] Where, MJ represents the total area of land remote sensing image, MJ u,L represents the area of the region corresponding to the Lth labeled cluster of the uth type of land, G represents the number of all labeled clusters corresponding to each type of land, μ u Indicates the area proportion of the u-th category of land.

[0081] At this point, the area proportion of each type of land is obtained.

[0082] Land planning resources monitoring is carried out based on the area ratio of each type of land and combined with the GeoNode platform. The flow chart of land planning resources monitoring using drone remote sensing is as follows: Figure 2 shown.

[0083] It should be noted that the exp(-x) model used in this embodiment is only used to indicate a negative correlation and to constrain the model output to be within the interval (0, 1). In specific implementation, it can be replaced with other models with the same purpose. This embodiment only uses the exp(-x) model as an example for description without making specific limitations on it, where x refers to the input of the model.

[0084] This embodiment provides a land planning resource monitoring system based on drone remote sensing, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the land planning resource monitoring method based on drone remote sensing in steps S001 to S004 is implemented.

[0085] At this point, this embodiment is completed.

[0086] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A land planning resource monitoring method based on UAV remote sensing, characterized in that: The method comprises the following steps: Acquisition of land remote sensing images; Connected domain detection is performed on the land remote sensing image to obtain several connected domains. Based on the grayscale difference and distance between each pixel in each connected domain and other pixels, and the grayscale difference and distance between each pixel in each connected domain and the pixels in the connected domains adjacent to each connected domain, several optimal initial clustering centers are screened from all pixels. All pixels in the land remote sensing image are clustered according to several preferred initial clustering centers to obtain several clusters. Based on the difference between the grayscale distributions of all pixels in two clusters, some adjacent clusters are merged and the merged clusters are recorded as marked clusters. The area ratio of each type of land is obtained based on several labeled clusters in land remote sensing images, and land resource monitoring and planning are carried out based on the area ratio of each type of land; According to the grayscale difference and distance between each pixel in each connected domain and other pixels, and the grayscale difference and distance between each pixel in each connected domain and the pixels in the adjacent connected domains of each connected domain, several preferred initial cluster centers are selected from all pixels. The specific steps include the following: According to the grayscale difference between each pixel in each connected domain and all other pixels, the distance between each pixel in each connected domain and the center of the connected domain, and the grayscale difference between each pixel in each connected domain and all pixels in adjacent connected domains, the possibility of each pixel in each connected domain being the initial cluster center is obtained; When the probability that each pixel point in each connected domain is the initial cluster center point is greater than or equal to the preset first threshold ε, the pixel point is recorded as a candidate initial cluster center point; According to the distance between each candidate initial cluster center point in each connected domain and all candidate initial cluster center points in the connected domains adjacent to each connected domain, and the possibility of each pixel point in each connected domain being the initial cluster center point, the possibility of each candidate initial cluster center point in each connected domain being the preferred initial cluster center point is obtained; Select all candidate initial cluster center points in each connected domain as the candidate initial cluster center point with the greatest possibility of being the preferred initial cluster center point, and record it as the preferred initial cluster center point; According to the grayscale difference between each pixel in each connected domain and all other pixels, the distance between each pixel in each connected domain and the center point of the connected domain, and the grayscale difference between each pixel in each connected domain and all pixels in adjacent connected domains, the possibility of each pixel in each connected domain being the initial cluster center point is obtained. The specific calculation method is as follows: Where H i,j Represents the gray value of the jth pixel in the i-th connected component, H i,r represents the gray value of the rth pixel in the i-th connected domain, n represents the number of all pixels in each connected domain, d i,j H represents the distance between the jth pixel in the i-th connected domain and the center point of the connected domain, i,e represents the mean grayscale value of all pixels in the e-th adjacent connected domain of the i-th connected domain, m represents the number of all adjacent connected domains of each connected domain, || is the absolute value symbol, norm() represents the linear normalization function, Q i,j Indicates the possibility that the jth pixel in the i-th connected domain is the initial cluster center; Based on the distance between each candidate initial cluster center point in each connected domain and all candidate initial cluster center points in the connected domains adjacent to each connected domain, and the possibility of each pixel point in each connected domain being the initial cluster center point, the possibility of each candidate initial cluster center point in each connected domain being the preferred initial cluster center point is obtained. The specific calculation method is as follows: Where, d i,c,e,v represents the distance between the cth candidate initial cluster center point in the ith connected domain and the vth candidate initial cluster center point in the eth adjacent connected domain, M represents the number of all candidate initial cluster centers in any connected domain adjacent to each connected domain, N represents the number of all connected domains adjacent to each connected domain, Q i,j Indicates the possibility that the jth candidate initial cluster center point in the i-th connected domain is the initial cluster center point, Q i,min Indicates the minimum value of the possibility that all initial cluster centers in the i-th connected domain are the initial cluster centers, Q i,max W represents the maximum value of the possibility that all initial cluster centers in the i-th connected domain are the initial cluster centers. i,c Indicates the possibility of the cth candidate initial cluster center point in the i-th connected component being the preferred initial cluster center point.

2. The land planning resource monitoring method based on UAV remote sensing according to claim 1 is characterized in that: All pixels in the land remote sensing image are clustered according to several selected initial clustering centers to obtain several clusters. The specific steps are as follows: According to all the preferred initial cluster center points, all the pixels in the land remote sensing image are clustered by the K-means clustering algorithm to obtain several clusters.

3. The land planning resource monitoring method based on UAV remote sensing according to claim 1 is characterized in that: Based on the difference between the grayscale distributions of all pixels in two clusters, some adjacent clusters are merged and the merged cluster is recorded as the marked cluster. The specific steps are as follows: According to the difference between the grayscale distribution of all pixels in the two clusters and the grayscale difference of all pixels in the two clusters, the possibility of merging the two clusters is obtained; Adjacent clusters whose merging probability between the two clusters is greater than a preset second threshold β are merged, and all clusters after the merger are recorded as marked clusters.

4. The land planning resource monitoring method based on UAV remote sensing according to claim 1 is characterized in that: The area ratio of each type of land is obtained based on several labeled clusters in the land remote sensing image. The land resource monitoring and planning are carried out based on the area ratio of each type of land. The specific steps include the following: Collect a large number of land remote sensing images, obtain the areas corresponding to several labeled clusters in each land remote sensing image, record the images containing the areas corresponding to several labeled clusters in each land remote sensing image as each reference image, use a large number of reference images to train the CNN neural network, and obtain a trained CNN neural network, where the loss function is the cross entropy loss function; Through the trained CNN neural network, the images of the areas corresponding to several labeled clusters in the land remote sensing image are input, and the output is the land of different categories; The specific calculation method for the area ratio of each type of land is as follows: Where, MJ represents the total area of land remote sensing image, MJ u,L represents the area of the region corresponding to the Lth labeled cluster of the uth type of land, G represents the number of all labeled clusters corresponding to each type of land, μ u represents the area proportion of the u-th category of land; Land planning resources are monitored based on the area proportion of each type of land and combined with the GeoNode platform.

5. The land planning resource monitoring method based on UAV remote sensing according to claim 1 is characterized in that: Connected domain detection is performed on land remote sensing images to obtain several connected domains. The specific steps are as follows: By detecting connected domains in land remote sensing images based on a depth-first search algorithm, several connected domains in the land remote sensing images are obtained.

6. A land planning resource monitoring system based on UAV remote sensing, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the land planning resource monitoring method based on drone remote sensing as described in any one of claims 1 to 5 are implemented.

Images