K-neighbor power line recognition method using comprehensive features

Abstract

The application discloses a K-nearest neighbor power line identification method using comprehensive features, which comprises the following steps: inputting radar target echo to signal detection to obtain strong scattering points, correlating multiple frames of data to obtain a pitch-distance two-dimensional graph and an X-Y two-dimensional graph, identifying a power tower to obtain a power tower position, obtaining power line straight lines of a tower topology, detecting the power line straight lines, extracting power line features, obtaining power line straight line feature vector samples and noise straight line feature vector samples, identifying unknown straight line feature vectors according to an expert knowledge base, and obtaining a power line identification result. The application is based on power line Bragg scattering echo peaks, combined with closely related power tower target identification, extracts comprehensive features, performs power line matching on the power line straight lines obtained through straight line detection and the power line straight lines obtained through the tower topology, and realizes accurate identification of the power line based on a K-nearest neighbor algorithm, so that the correct identification probability is high, the target identification range is wide, and the self-adaptation capability is strong.

Description

Technical Field

[0001] This invention relates to the field of power line identification technology, and in particular to a K-nearest neighbor power line identification method utilizing comprehensive features. Background Technology

[0002] Compared to traditional fixed-wing aircraft, helicopters possess superior maneuverability, featuring low-altitude, low-speed, vertical takeoff and landing capabilities, and stable attitude, making them widely used in emergency command, disaster relief, geological exploration, and security patrols. Due to their unparalleled low-altitude flight advantages, helicopters are frequently required to perform missions in low-altitude environments. However, due to the complex low-altitude ground environment, the accident rate for helicopters is significantly higher than that for fixed-wing aircraft. Statistical analysis shows that the vast majority of flight accidents are caused by helicopters colliding with obstacles. Among various obstacles, power lines are a significant factor affecting the safety of helicopters flying at low altitudes, especially in environments with low visibility such as rain and fog. To address the threat avoidance problem in low-altitude flight scenarios, helicopter collision avoidance systems play a crucial role in identifying and warning of targets such as power lines. Early methods for helicopter collision avoidance utilized laser and infrared sensors to detect and identify power line targets. While laser and infrared sensor systems are very lightweight and have high distance resolution, they can only detect relatively close power line targets and are significantly affected by weather and other environmental factors. LiDAR may not function properly in rain or fog; infrared imaging methods have poor contrast, making accurate identification of power lines difficult. These methods have limited effectiveness for power line detection and early warning. Therefore, in practical scenarios, millimeter-wave radar is primarily considered for identifying power line targets and facilitating collision avoidance. Compared to microwave radar, millimeter-wave radar features narrow beamwidth, large bandwidth, high range resolution, low power consumption, and small size, making it ideal for applications like helicopter collision avoidance. Furthermore, it can operate normally in adverse weather conditions such as rain and fog, is less affected by environmental factors, and has strong resistance to clutter interference. Applying millimeter-wave radar to helicopter collision avoidance will significantly improve helicopter flight safety and enable all-weather operation. Research on millimeter-wave radar for helicopter collision avoidance remains a hot topic with promising application prospects.

[0003] Electric power lines, with their unique material composition and relatively complex geometry, are typical low-profile, small targets with small radar cross-sections. Studying the millimeter-wave radar echo signals of electric power lines and extracting their electromagnetic scattering characteristics—distinguishing them from other targets in terms of amplitude, phase, and polarization—is crucial for electric power line identification. Among these characteristics, the Bragg scattering properties of electric power lines differ significantly from those of ordinary linear targets and can serve as an important feature for identification. Electric power towers, as stationary targets with large radar cross-sections, are mixed among numerous complex ground stationary targets, requiring differentiation of their target attributes from a vast array of single-point patterns. First, the height information of the power tower is used to identify individual suspected targets, distinguishing them from numerous natural objects. Second, the direction of the electric power lines is confirmed through group information, eliminating false targets. However, relying solely on electric power line or power tower features is insufficient to achieve a high probability of accurate electric power line identification. Summary of the Invention

[0004] The main objective of this invention is to provide a K-nearest neighbor power line identification method that utilizes comprehensive features, aiming to solve the technical problem of low identification accuracy in current power line identification methods.

[0005] To achieve the above objectives, the present invention provides a method for identifying K-nearest neighbor power lines using comprehensive features, the method comprising the following steps:

[0006] S1: Perform target detection processing on the echo of the uniform illumination beam, and obtain a one-dimensional range profile of the target through constant false alarm threshold decision, so as to determine the range information of each strong scattering point of the target;

[0007] S2: Cluster the strong scattering points obtained at different pitch angles, azimuth angles, and distances to obtain scatter plots with different cross-sections;

[0008] S3: Construct a two-dimensional elevation-distance map based on strong scattering points with the same azimuth angle, different elevation angles, and different distances, and extract the height characteristics of the power tower from the elevation-distance map;

[0009] S4: Determine whether a power tower exists in the height dimension and obtain the location of the power tower;

[0010] S5: Connect the power towers in pairs to obtain straight power lines after the topology between the towers is changed;

[0011] S6: Construct an XY two-dimensional graph, and compare the point sequence intensity of each straight line obtained by line detection in the XY two-dimensional graph to identify whether the power tower exists, and obtain the straight line of the power line after line detection;

[0012] S7: When a straight line belongs to both the power line obtained from the inter-tower topology and the power line obtained from the line detection, the match is successful and the matched power line is obtained.

[0013] S8: For a straight line with power towers, extract power line features based on the point sequence on the line;

[0014] S9: Establish an expert knowledge base composed of feature vector samples of different types of power line straight lines and feature vector samples of noise straight lines;

[0015] S10: Using unknown straight line feature vector samples as input, the system performs classification and judgment using an expert knowledge base to obtain the power line identification result.

[0016] Optionally, after step S1, the method further includes estimating the length of the detected target and eliminating cluttered areas.

[0017] Optionally, in step S2, strong scattering points obtained at different elevation angles, different azimuth angles, and different distances are clustered. Specifically, an associated region is set according to the coverage range of the radar antenna beam, and target points falling within the associated region are clustered.

[0018] Optionally, in step S3, extracting the height features of the power tower specifically involves: in the pitch-distance two-dimensional map, based on... Extracting the height characteristics of power towers; where, Let R be the pitch angle vector, R be the distance vector, max[·] be the maximum value, and min[·] be the minimum value. It corresponds one-to-one with the elements in R.

[0019] Optionally, in step S4, determining whether a power tower exists specifically involves: in the height dimension, when H∈H1, determining that a power tower exists; where H1 is the set of power tower heights, typically ranging from 20m to 50m.

[0020] Optionally, in step S4, the expression for the location of the power tower is specifically as follows:

[0021]

[0022]

[0023] In the formula, mean[·] represents the average value.

[0024] Optionally, in step S6, constructing an XY two-dimensional map specifically involves: constructing an azimuth-range two-dimensional map based on strong scattering points at different azimuth angles and distances at an elevation angle that scans a full circle from left to right or from right to left, and converting the azimuth-range two-dimensional map into an XY two-dimensional map.

[0025] Optionally, in step S6, line detection is performed on the XY two-dimensional graph, specifically by using Hough transform to detect lines in the XY two-dimensional graph.

[0026] Optionally, in step S8, the characteristics of the power line include the mean strength, the variance of strength, the number of points, and the point interval.

[0027] Optionally, step S10 specifically includes:

[0028] Distance calculation: For the input unknown straight line feature vector, calculate (M+N) distances with the power line straight line feature vector samples and noise straight line feature vector samples in the expert knowledge base respectively;

[0029] To determine the category of the K smallest distance samples: Among (M+N) distances, determine the category of the K smallest distance samples, where K is an odd number. J of these samples belong to the electric field line samples, and KJ of these samples belong to the noise line samples.

[0030] Weighting: Based on the number of samples of different categories in the expert knowledge base, calculate the power line straight line weighting coefficient N / (M+N) and the noise straight line weighting coefficient M / (M+N) to obtain the power line straight line decision factor J*N / (M+N) and the noise straight line decision factor (KJ)*M / (M+N);

[0031] Binary classification decision: When the decision factor for a straight power line is J*N / (M+N) is greater than the decision factor for a noisy line (KJ)*M / (M+N), the unknown input line is determined to be a straight power line; otherwise, it is a noisy line.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] High probability of correct identification. Relying solely on power tower features for identification is susceptible to topological limitations, leading to a low probability of correct identification. Conversely, relying solely on power line features is easily affected by noise, also resulting in a low probability of correct identification. A K-nearest neighbor power line identification method utilizing the combined features of power towers and power lines overcomes the shortcomings of single-feature identification by extracting comprehensive features composed of the power tower height, power line length, and Bragg scattering characteristics of the power line, achieving a high probability of correct identification.

[0034] The target identification range is wide. Different power line types exist under different transmission technology specifications, resulting in diverse power line characteristics. A K-nearest neighbor power line identification method utilizing the combined characteristics of power towers and power lines flexibly employs Euclidean distance, Manhattan distance, Chebyshev distance, and Mahalanobis distance to reflect the similarity between different power line samples, enabling the identification of various types of power lines and achieving a wide target identification range.

[0035] It exhibits strong adaptability. Since it is impossible to obtain the features of all power line types in advance and establish a complete expert knowledge base, a K-nearest neighbor power line identification method that utilizes the comprehensive features of power towers and power lines updates the expert knowledge base in real time by iterating the results of each identification. This allows the method to adapt to the current application environment and exhibits strong adaptability. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the K-nearest neighbor power line identification method utilizing comprehensive features in this invention.

[0037] Figure 2 This is a schematic diagram of the power identification process in this invention.

[0038] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0039] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0040] This invention provides a method for identifying K-nearest neighbor electric lines using comprehensive features, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the K-nearest neighbor power line identification method utilizing comprehensive features of the present invention.

[0041] In this embodiment, the K-nearest neighbor power line identification method utilizing comprehensive features includes the following steps:

[0042] 1. Signal detection

[0043] Target detection processing is performed on the echoes from the same illumination beam. A one-dimensional range profile of the target is obtained through constant false alarm threshold (CFAT) decision, yielding range information for each strong scattering point of the target. The length of the detected target is estimated, and large areas of clutter are removed.

[0044] 2. Multi-frame data association

[0045] Strong scattering points obtained at different elevation angles, azimuth angles, and distances after clutter removal are clustered. Specifically, association regions are set based on the radar antenna beam coverage, and target points falling within these regions are clustered. Scatter plots with different cross-sections are then obtained from the clustered strong scattering points.

[0046] 3. Feature Extraction of Power Towers

[0047] At the same azimuth angle α0, different pitch angles A pitch-range two-dimensional map is constructed using strong scattering points at different distances R. Within this pitch-range two-dimensional map, based on... Extract the height characteristics of the power tower. In the formula, R is the pitch angle vector, R is the distance vector, max[·] is the maximum value, and min[·] is the minimum value. It corresponds one-to-one with the elements in R.

[0048] 4. Power tower identification

[0049] In the height dimension, when H∈H1, it is determined that a power tower exists, and the location of the power tower (X0,Y0) is obtained through coordinate transformation. H1 is the set of power tower heights, with typical values ​​ranging from 20m to 50m.

[0050]

[0051]

[0052] In the formula, mean[·] represents the average value.

[0053] 5. Inter-tower topology

[0054] Connecting each power tower in pairs, when L∈L1, yields straight power lines with different topologies between the towers. L1 is a set of power line lengths, typically ranging from 150m to 1000m.

[0055] 6. Line detection

[0056] A two-dimensional azimuth-range map is constructed by scanning a full circle of elevation angles from left to right or right to left, with strong scattering points at different azimuth angles and distances. This azimuth-range map is then transformed into an XY map. Hough transform is used to detect straight lines in the XY map. For each detected straight line, a point sequence intensity comparison is performed, comparing the intensity of the line's starting and ending points with points in the middle of the line to identify the presence of power towers. If a power tower is present, the detected power line is identified.

[0057] 7. Power line matching

[0058] When a straight line belongs to both the power line obtained from the inter-tower topology and the power line obtained from the line detection, the match is successful, and the matched power line is obtained.

[0059] 8. Electric Power Line Feature Extraction

[0060] For a straight line containing power towers, four features are extracted based on the point sequence on the line: mean intensity, variance intensity, number of points, and point interval.

[0061] 9. Expert Knowledge Base

[0062] An expert knowledge base is established, consisting of power line straight feature vector samples and noise straight feature vector samples of different types, where the number of power line straight samples is M and the number of noise straight samples is N.

[0063] 10. Power Line Identification

[0064] By using unknown straight line feature vector samples as input, and employing an expert knowledge base for classification and judgment, the power line identification result is obtained.

[0065] In a preferred embodiment. See also Figure 2 The power line identification module, based on the K-nearest neighbor algorithm, divides the power line identification process into four steps: distance calculation, counting the categories of the K smallest distance samples, weighted summation, and binary classification decision, as shown below:

[0066] Step 1: Distance Calculation. For the input unknown straight line feature vector, calculate (M+N) distances with the power line straight line feature vector samples and noise straight line feature vector samples in the expert knowledge base.

[0067] Step 2: Calculate the category of the K smallest distance samples. Among (M+N) distances, calculate the category of the K smallest distance samples, where K is an odd number. J samples belong to the electric field line samples, and KJ samples belong to the noise line samples.

[0068] Step 3: Weighting. Based on the number of samples in different categories in the expert knowledge base, calculate the power line straightness weighting coefficient N / (M+N) and the noise straightness weighting coefficient M / (M+N), to obtain the power line straightness decision factor J*N / (M+N) and the noise straightness decision factor (KJ)*M / (M+N).

[0069] Step 4: Binary classification decision. When the decision factor for a straight power line is J*N / (M+N) is greater than the decision factor for a noisy line is (KJ)*M / (M+N), the input unknown line is determined to be a straight power line; otherwise, it is a noisy line.

[0070] This embodiment proposes a K-nearest neighbor power line identification method based on comprehensive features. By using the Bragg scattering echo peak of the power line as a basis and combining it with the identification of closely related power tower targets, comprehensive features are extracted. Power line matching is performed on the power line straight lines obtained by straight line detection and the power line straight lines obtained by inter-tower topology. Based on the K-nearest neighbor algorithm, accurate identification of power lines is achieved with a high probability of correct identification, a wide target identification range, and strong adaptability.

[0071] The above are merely preferred embodiments of the invention and do not limit the patent scope of the invention. Any equivalent structural or procedural changes made using the contents of the invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the invention.

Claims

1. A K-Nearest Neighbor power line identification method using integrated features, characterized by, The method comprises the following steps: S1: performing target detection processing on echoes of a unified illumination beam, obtaining a one-dimensional range image of a target through a constant false alarm rate decision, and determining distance information of each strong scattering point of the target; S2: clustering strong scattering points obtained at different elevation angles, different azimuth angles and different distances to obtain scatter point graphs of different sections; S3: constructing an elevation-range two-dimensional graph based on strong scattering points at the same azimuth angle, different elevation angles and different distances, and extracting a height feature of a power tower in the elevation-range two-dimensional graph; S4: judging whether the power tower exists in the height dimension and obtaining a position of the power tower; S5: connecting the power towers two by two to obtain power line straight lines after tower topology; S6: constructing an X-Y two-dimensional graph, comparing point sequence intensities of each straight line obtained through straight line detection in the X-Y two-dimensional graph, identifying whether the power tower exists, and obtaining power line straight lines after straight line detection; S7: when the straight line belongs to both the power line straight lines obtained through tower topology and the power line straight lines obtained through straight line detection, the matching is successful, and matched power line straight lines are obtained; S8: extracting power line features based on point sequences on the straight line for the straight line in which the power tower exists; S9: establishing an expert knowledge base composed of different types of power line straight line feature vector samples and noise straight line feature vector samples; S10: taking unknown straight line feature vector samples as input, performing classification decision by using the expert knowledge base, and obtaining a power line recognition result.

2. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, After the step S1, the method further comprises estimating a length of a detected target and eliminating a patchy clutter area.

3. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S2, the strong scattering points obtained at different elevation angles, different azimuth angles and different distances are clustered, specifically, target clustering is performed on target points falling in an associated region according to a radar antenna beam coverage range.

4. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S3, height features of the power tower are extracted, specifically, in the pitch-distance two-dimensional graph, according to extracting height features of the power tower; in the formula, is a pitch angle vector, R is a distance vector, max[·] is a maximum value, min[·] is a minimum value, and the elements in R are one-to-one corresponding.

5. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S4, whether the power tower exists is judged, specifically, when H∈H1 in the height dimension, it is judged that the power tower exists; H1 is a set of power tower heights, and a typical value is 20 m to 50 m.

6. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S4, an expression of the position of the power tower is specifically: In the formula, mean[·] is an average value.

7. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S6, the X-Y two-dimensional graph is constructed, specifically, a position-distance two-dimensional graph is constructed based on the strong scattering points at different azimuth angles and different distances in the elevation angle scanned from left to right or from right to left, and the position-distance two-dimensional graph is converted into the X-Y two-dimensional graph.

8. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S6, the straight line detection is performed in the X-Y two-dimensional graph, specifically, Hough transform is used to detect the straight line in the X-Y two-dimensional graph.

9. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S8, the power line features include an intensity mean value, an intensity variance, a point number and a point interval.

10. The K-Nearest Neighbor power line recognition method using integrated features according to claim 1, wherein, In the step S10, the following is specifically performed: Distance calculation: for the unknown straight line feature vector input, (M+N) distances are calculated with the power line straight line feature vector samples and the noise straight line feature vector samples in the expert knowledge base respectively. Counting the classes of the K smallest distance samples: Among the (M+N) distances, count the classes of the K smallest distance samples, K is an odd number, in which J belongs to the power line straight line samples, K-J belongs to the noise straight line samples; Weighting: According to the number of different classes of samples in the expert knowledge base, calculate the power line straight line weighting coefficient N / (M+N) and the noise straight line weighting coefficient M / (M+N), obtain the power line straight line decision factor J*N / (M+N) and the noise straight line decision factor (K-J)*M / (M+N); Binary classification decision: When the power line straight line decision factor J*N / (M+N) is greater than the noise straight line decision factor (K-J)*M / (M+N), it is judged that the input unknown straight line is a power line straight line, otherwise it is a noise straight line.

Images