A UAV echo characteristics modeling method based on clustering algorithm

Abstract

The present invention relates to the field of data processing, and in particular to a method for modeling the echo characteristics of unmanned aerial vehicles (UAVs) based on a clustering algorithm. The method comprises: extracting characteristic parameters from the original UAV radar echo signal and mapping them into two-dimensional data points; dividing the two-dimensional data points into regional blocks and generating connected blocks; when the number of connected blocks meets a first preset condition, calculating the centroid of each connected block, and using the data point closest to the centroid as the initial center point of the connected block; when the number of connected blocks meets a second preset condition, selecting a connected block for splitting until the number of all connected blocks meets the first preset condition, and finally determining the initial center point of the connected block; clustering the two-dimensional data points using the initial center point, and establishing a radar cross-section model based on the center point of each cluster. The present invention dynamically adjusts the number of connected blocks and the selection of the initial center point to ensure that the selection of the initial center point is more reasonable and improves the clustering effect.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more particularly to a method for modeling echo characteristics of unmanned aerial vehicle (UAV) based on a clustering algorithm. Background Art

[0002] At present, the application of drones in fields such as geographic surveying and mapping and aerial photography for film and television has become quite mature. At the same time, they are also playing an increasingly critical role in emerging scenarios such as emergency rescue. However, with the rapid popularization of drones, the phenomena of "illegal flying" and "indiscriminate flying" have become frequent, posing a serious threat to public safety.

[0003] Traditional drone detection relies primarily on a combination of radar detection and guidance and optoelectronic confirmation. While radar offers advantages such as long detection range and high efficiency, it is susceptible to interference from factors such as ground clutter, multipath effects, and birds, leading to increased false alarm rates.

[0004] Drone echo data is extremely complex, containing a large number of sampling points. Using the K-means clustering algorithm can simplify this complex data into a few clusters, making the data structure clearer and facilitating subsequent analysis and modeling. However, the traditional K-means clustering algorithm randomly selects its initial center points. Considering the characteristics of drone echoes, if both initial center points are located in low-intensity echo areas and are close to each other, some medium-intensity echo data may be incorrectly divided into these two clusters, failing to form independent clusters. Furthermore, if the initial center point is chosen at the edge of the data distribution or near an outlier, distorted clustering results will result. The clustering results serve as input data for modeling. Incorrect clustering results in the model's inability to distinguish between normal and abnormal echoes, impacting subsequent analysis of the drone's flight status. Summary of the Invention

[0005] In order to solve the technical problem that the K-means clustering algorithm mentioned above is prone to incorrect clustering when processing drone echo data due to the random selection of initial center points, which affects subsequent modeling, the present invention provides the following technical solution.

[0006] A UAV echo characteristic modeling method based on clustering algorithm includes:

[0007] Extract characteristic parameters from the original radar echo signal of the UAV and map the characteristic parameters into two-dimensional data points;

[0008] Divide two-dimensional data points into regional blocks and generate connected blocks;

[0009] When the number of connected blocks meets the first preset condition, the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block; when the number of connected blocks meets the second preset condition, connected blocks are selected for splitting until the number of all connected blocks meets the first preset condition, then the splitting is stopped, and the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block;

[0010] The two-dimensional data points are clustered using the initial center point to obtain multiple clusters, and a radar cross-section model is established based on the center point of each cluster.

[0011] Preferably, the characteristic parameters are signal amplitude and phase.

[0012] Preferably, the connected block generation includes:

[0013] For each block, count the number of data points that fall within the block, and perform binary classification on all blocks. The block with the largest average number of points is marked as 1, and the other blocks are marked as 0.

[0014] The area blocks marked as 1 are merged into a connected block according to the spatial neighborhood relationship, and then all independent connected blocks are obtained.

[0015] Preferably, the areas of all connected blocks are calculated, and the connected block with the largest area is selected for splitting.

[0016] Preferably, if there are multiple connected blocks with the same area value and the largest area, the priorities corresponding to these connected blocks with the largest area are calculated, and the connected block with the largest priority is selected for splitting.

[0017] Preferably, the splitting process includes:

[0018] The connected block selected for splitting is mapped to a point set, that is, the centroid of each regional block contained in the connected block is taken as a point, and the point set is binary classified, where the two points with the farthest distance in the point set are selected as the initial center points of the binary classification; after the binary classification is completed, the point difference between the two categories is calculated. If the point difference is greater than or equal to the preset value, for each point in the category with more points, its distance to the center point of this category is calculated, and the distance from the point in this category to the center point of another category is calculated; for each point, the difference between its distance to the center point of this category and the distance to the center point of another category is calculated, and the calculated differences are sorted from large to small, and the points ranked last with the same number of point differences are reallocated to another category to complete the connected block splitting.

[0019] Preferably, the priority acquisition process includes:

[0020] The aspect ratio, fullness and corner point connectivity of each connected block among all connected blocks with the same and largest area values ​​are calculated, and the product of the aspect ratio, fullness and corner point connectivity is used as the priority.

[0021] Preferably, it also includes:

[0022] When the number of connected blocks meets the third preset condition, a clustering algorithm is used to cluster the centroids of the connected blocks to obtain k virtual connected blocks, and then to obtain k initial center points.

[0023] The beneficial effects of the present invention are:

[0024] The present invention first retains the key information of the echo signal through dimensionality reduction, focuses on the core features of the signal, and provides intuitive input for clustering; then filters out noise and sparse areas through density segmentation, preliminarily locates potential cluster centers, and avoids the deviation of traditional clustering algorithms caused by noise or outliers. Further, different operations are performed according to whether the number of connected blocks meets the preset conditions, and the number of connected blocks and the selection of initial center points are dynamically adjusted to meet the requirements of subsequent clustering operations, ensuring that the selection of initial center points is more reasonable and improving the clustering effect.

[0025] A radar cross-section model is built based on the center point of each cluster. The center point of each cluster represents the average characteristics of the data points within that cluster. The radar cross-section model built based on these center points can reflect the different patterns of drone radar echo characteristics, achieving effective modeling of drone echo characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a method flow chart of steps S1 to S4 in a method for modeling echo characteristics of a drone based on a clustering algorithm in an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the regional block labeling results in a method for modeling UAV echo characteristics based on a clustering algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0029] Reference Figure 1 A method for modeling UAV echo characteristics based on clustering algorithm includes steps S1 to S4, which are as follows:

[0030] S1: Extract characteristic parameters from the original radar echo signal of the UAV and map the characteristic parameters into two-dimensional data points.

[0031] In one embodiment, a radar device is used to detect the drone and receive the reflected original signal (i.e., the echo data of the electromagnetic wave). Key features such as signal amplitude, phase, frequency, and delay are further extracted from the original signal. The extracted feature parameters are divided into two groups, each with two parameters, to form two two-dimensional coordinate spaces.

[0032] For example, space 1: the horizontal axis is signal amplitude, and the vertical axis is phase, which describes the reflection characteristics of the target; space 2: the horizontal axis is frequency, and the vertical axis is delay, which describes the motion state and distance of the target.

[0033] For example, the amplitude extracted from the original radar echo signal of a drone in a certain measurement is , the phase is , then draw the point in space 1 ; The frequency is , the delay is , then draw the point in space 2 .

[0034] Through the above processing, the originally complex radar echo raw signal can be converted into clusterable two-dimensional data points, which is convenient for analyzing and processing the drone radar echo model.

[0035] S2: Divide the two-dimensional data points into regional blocks and generate connected blocks.

[0036] To further analyze the distribution of data points in S1, let's take space 1 in S1 as an example. In one embodiment, assuming that the data points need to be classified into k categories, the entire space is divided into k×k small square blocks. The size of each block is determined by the data range (coordinate range) and the value of k.

[0037] Then, for each area block, the number of data points falling into the area block is counted, and the K-means clustering algorithm is used to divide all area blocks into two categories according to the number of their data points. One category represents high-density area blocks (i.e., the number of points is large), and the other category represents low-density area blocks (i.e., the number of points is small). Then, the clustering result is obtained, that is, each area block is assigned to a category.

[0038] Next, for the above clustering results, the area block with the largest average number of points is marked as 1, and the other area blocks are marked as 0. The area blocks marked as 1 are merged into larger connected areas according to the spatial neighborhood relationship. The marking results are as follows Figure 2 shown.

[0039] For example, each area block marked as 1 is checked. If its four neighbors (four adjacent blocks above, below, left, and right) are also 1, these blocks are merged into a connected block, and finally multiple independent connected blocks are obtained.

[0040] The raw data may contain a large number of discrete points, and directly clustering these points would be computationally intensive. By dividing the space into blocks and classifying and merging the blocks, the number of data units that need to be processed can be significantly reduced, thereby lowering computational complexity.

[0041] S3: When the number of connected blocks meets the first preset condition, the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block; when the number of connected blocks meets the second preset condition, the connected blocks are selected for splitting until the number of all connected blocks meets the first preset condition, then the splitting is stopped, and the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block.

[0042] The traditional K-means clustering algorithm requires pre-specifying the number of cluster categories k and depends on the position of the initial center point. The selection of the initial center point directly affects the quality of the clustering results and the convergence speed of the algorithm.

[0043] If the number of initial center points (i.e., the number of connected blocks) is less than k, K-means clustering cannot be started. Conversely, if the number of initial center points (i.e., the number of connected blocks) is greater than k, additional processing is required to select k representative center points.

[0044] In one embodiment, different processing methods are formulated based on the number of connected blocks. When the number of connected blocks meets the first preset condition (i.e., the number of connected blocks is equal to k), the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block.

[0045] When the number of connected blocks meets the second preset condition (i.e., the number of connected blocks is less than k), directly selecting the centroids of the connected blocks will result in an insufficient number of initial center points. Therefore, new initial center points can be generated by splitting the connected blocks to increase diversity.

[0046] For example, the areas of all connected blocks are calculated, and the connected block with the largest area is selected for splitting. The splitting operation is as follows:

[0047] The connected block selected for splitting is mapped to a point set, that is, the centroid of each regional block contained in the connected block is taken as a point, and the point set is classified into two categories, where the two points with the farthest distance in the point set are selected as the initial center points of the two categories.

[0048] After the binary classification is completed, the difference in the number of points between the two categories is calculated, marked as d. If d is greater than or equal to 2, for each point in the category with more points, its distance to the center point of this category is calculated. At the same time, the distance from the point to the center point of the other category is calculated. Then, for each point, the difference between the distance to the center point of this category and the distance to the center point of the other category is calculated. The calculated differences are sorted from large to small, and the last d points are reallocated to the other category to balance the number of points in the two categories. Finally, the original connected block is split into two new connected blocks (one new connected block for each class).

[0049] If d is less than 2, the binary classification results are sufficiently balanced and no further adjustments are needed. Simply divide the region corresponding to the point into two new connected blocks based on the results.

[0050] After the split is completed, the number of connected blocks is recounted. If the number of connected blocks still meets the second preset condition, the split operation is repeated until the number of connected blocks meets the first preset condition.

[0051] It should be noted that if there are multiple connected blocks with the same area value and the largest area, the priorities corresponding to these connected blocks with the largest area are calculated, and the connected block with the largest priority is selected for splitting.

[0052] The calculation formula for the above priority is:

[0053]

[0054] Where, is the priority of the connected block, 、 are the length and width of the circumscribed rectangle of the connected block, is the area of ​​the connected block, is the number of corner points shared by the connected block with other connected blocks, is the number of corner points on the outermost boundary of the connected block.

[0055] in, Indicates the aspect ratio of the connected block. In clustering, elongated connected blocks are more difficult to handle, so it is also necessary to adjust the contribution direction of the aspect ratio, that is, use , and then when The smaller, The larger the priority The larger it is, the more slender connected blocks are split first; Indicates the fullness of the connected block, that is, the compactness of the connected block in the circumscribed rectangle. The compact connected block (i.e. Large) usually do not need to be split first because their shapes are more regular. ,when The smaller, The larger the priority The larger it is, the more sparsely populated connected blocks are split first; Indicates the degree of connection between the connected block and other connected blocks at the corner point. This indicates that the connected block has more connections with other connected blocks at the corner points. The larger the value, the higher the priority. The larger it is, the more it needs to be split first.

[0056] It should be noted that it is also considered that when the number of connected blocks meets the third preset condition (that is, the number of connected blocks is greater than k), in this case, the centroid coordinates of each connected block are used as data points, and the traditional K-means clustering algorithm is used to cluster these centroid points into k classes. Each clustering result corresponds to a "virtual connected block" whose center is the centroid point of the class, and finally k connected blocks are obtained, which meets the requirement of the initial number of center points.

[0057] According to the above operations, all connected blocks are finally obtained, and the number of connected blocks is equal to k. The centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block.

[0058] According to the above operations of S2 and S3, all the initial center points in space 2 can be obtained in the same way.

[0059] S4: Use the initial center point to cluster the two-dimensional data points to obtain multiple clusters, and establish a radar scattering cross-section model based on the center point of each cluster.

[0060] Based on all the initial center points obtained above, these initial center points are used as the initial center points for the K-means clustering algorithm to generate k clusters. The center point of each cluster is calculated to represent the typical characteristics of the cluster. Based on the characteristics of the cluster center, an appropriate mathematical model (such as a polynomial or exponential function) is selected to fit the relationship between the radar cross section (RCS) and the input parameters. A corresponding RCS model is established for each cluster to predict the echo characteristics of different types of drones under different observation angles and radar parameters.

[0061] In summary, the present invention first partitions two-dimensional data points into regional blocks and generates connected blocks. The initial center point selection is then dynamically adjusted based on the number of connected blocks. When the number of connected blocks meets a first pre-determined condition, the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point. When the number of connected blocks meets a second pre-determined condition, connected blocks are selected for splitting until the first pre-determined condition is met, ultimately determining the initial center point. This approach avoids the blind random selection of initial center points in the traditional K-means algorithm, making the selection of initial center points more rational.

[0062] During the connected block splitting process, features such as area, aspect ratio, fullness, and corner connectivity are considered, prioritizing the connected blocks with the largest area or the highest priority for splitting. This splitting method based on data distribution characteristics can more rationally adjust the distribution of connected blocks, ensuring that the initial center point covers the main distribution area of ​​the data and avoiding distorted clustering results.

[0063] By dynamically adjusting the number of connected blocks and the selection of the initial center point, the present invention improves the clustering effect. Reasonable selection of the initial center point makes the clustering results more accurate and can better reflect the inherent laws and patterns of the drone's echo characteristics.

[0064] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A UAV echo characteristics modeling method based on clustering algorithm, characterized in that: include: Extract characteristic parameters from the original radar echo signal of the UAV and map the characteristic parameters into two-dimensional data points; Divide two-dimensional data points into regional blocks and generate connected blocks; When the number of connected blocks meets the first preset condition, the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block; When the number of connected blocks meets the second preset condition, a connected block is selected for splitting, the areas of all connected blocks are calculated, and the connected block with the largest area is selected for splitting; if there are multiple connected blocks with the same and largest area values, the priorities corresponding to these connected blocks with the largest areas are calculated, and the connected block with the largest priority is selected for splitting. The splitting process includes: mapping the connected block selected for splitting into a point set, that is, taking the centroid of each area block contained in the connected block as a point, and performing binary classification on the point set, wherein the two points with the farthest distance in the point set are selected as the initial center points of the binary classification; after the binary classification is completed, the difference in the number of points of the two categories is calculated. If the difference in the number of points is greater than or equal to the preset value, for each point in the category with more points, its distance to the current category is calculated. The distance between the center point of the category and the distance between the point in this category and the center point of another category is calculated; for each point, the difference between its distance to the center point of this category and the distance to the center point of another category is calculated, the calculated differences are sorted from large to small, and the points ranked last with the same number of points difference are reallocated to another category to complete the connected block splitting; the aspect ratio, fullness and corner point connectivity of each connected block among all connected blocks with the same area value and the largest area value are calculated, and the product of the aspect ratio, fullness and corner point connectivity is used as the priority; until the number of all connected blocks meets the first preset condition, the splitting is stopped, and the centroid of each connected block is calculated, and the data point closest to the centroid is used as the initial center point corresponding to the connected block; The two-dimensional data points are clustered using the initial center point to obtain multiple clusters, and a radar cross-section model is established based on the center point of each cluster.

2. The method for modeling UAV echo characteristics based on clustering algorithm according to claim 1, characterized in that: The characteristic parameters are signal amplitude and phase.

3. The method for modeling UAV echo characteristics based on clustering algorithm according to claim 2, characterized in that: The connected block generation includes: For each block, count the number of data points that fall within the block, and perform binary classification on all blocks. The block with the largest average number of points is marked as 1, and the other blocks are marked as 0. The area blocks marked as 1 are merged into a connected block according to the spatial neighborhood relationship, and then all independent connected blocks are obtained.

4. The method for modeling UAV echo characteristics based on clustering algorithm according to claim 1, characterized in that: Also includes: When the number of connected blocks meets the third preset condition, a clustering algorithm is used to cluster the centroids of the connected blocks to obtain k virtual connected blocks, and then to obtain k initial center points.

Images