Anti-collision radar target detection and identification method based on remote sensing data knowledge assistance

Abstract

The application provides a collision avoidance radar target detection and identification method based on remote sensing data knowledge assistance, relates to the low-altitude collision avoidance technology field of helicopters / unmanned aerial vehicles, and solves the detection limitation problem of existing collision avoidance radars for small targets such as power lines. The method first constructs a power tower position auxiliary knowledge base based on remote sensing satellite image data, loads a real-time processing database according to the position data of the carrier inertial navigation system during the flight task; after the radar works, the echo signal is received and threshold detection is performed to obtain a list of isolated targets, and then tracking processing and initialization of attribute labels are performed to form a tracking target list; the tracking target list is matched with the real-time processing database, and if the matching is successful, the power tower target is obtained, and then the unmatched target is identified to determine whether it is a power line target, and after Hough detection confirmation, the detection and identification process is completed. The application not only guarantees the detection and identification ability of the radar itself, but also improves the detection and identification ability for power towers / lines.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of low-altitude collision avoidance of helicopters / unmanned aerial vehicles, and particularly relates to a collision avoidance radar target detection and identification method based on remote sensing data knowledge assistance. BACKGROUND

[0002] When a helicopter performs a low-altitude task, especially in critical scenarios such as emergency rescue, small obstacles such as power lines are often difficult for the pilot to detect in time. Such obstacles are small in size and hidden in position, and in poor visibility or unfamiliar flight environment, they are extremely easy to be ignored, posing a serious threat to flight safety. Factors such as visibility restrictions, environmental changes, and time pressure exacerbate the problem, making traditional visual methods insufficient, and a reliable automatic obstacle avoidance system is urgently needed to ensure task safety. In the low-altitude flight environment, power lines and isolated obstacles are widely distributed, especially in complex terrain or urban areas, and their detection difficulty is further increased, and any negligence can lead to a catastrophic accident. Therefore, developing efficient and accurate obstacle detection technology has become a core requirement for the safe operation of aircraft such as helicopters.

[0003] To address this challenge, a variety of obstacle avoidance sensor technologies have been widely applied to helicopter platforms, including electromagnetic field detectors, infrared detectors, laser radars, and millimeter wave radars. Among them, millimeter wave radars have become the mainstream choice due to their small size, light weight, low power consumption, and all-weather working characteristics. These sensors detect obstacles by emitting and receiving electromagnetic waves, but due to the volume, weight, and cost constraints of the helicopter platform, only a single sensor system can usually be deployed. Millimeter wave radars perform stably in bad weather or light conditions, but their detection capability is still insufficient when facing small targets, especially on low reflectivity obstacles such as power lines, which are prone to missed detection or false positives.

[0004] In the specific application of millimeter wave radars, early methods are based on the Bragg scattering characteristics of power lines in the millimeter wave frequency band, combined with the Hough straight line transformation algorithm to identify power line targets. This method attempts to use the characteristics of scattered signals to locate straight-line structures, but in actual flight tests, it is found that the radar scattering cross-section of power lines has very weak echo intensity, and only appears at specific incident angles, resulting in a significant reduction in detection probability. Its application effect is not good, the false alarm rate is high, and it is difficult to meet the real-time needs in complex backgrounds. In addition, the Hough transform is sensitive to noise and background interference, and is prone to failure in cluttered environments, further limiting its reliability.

[0005] In recent years, the rise of machine learning technology has brought new ideas to obstacle detection. Algorithms such as decision trees, support vector machines, and convolutional neural networks have been introduced into the field of power line identification. These methods have improved detection accuracy by training data models and have performed well in specific scenarios. However, machine learning models face the bottleneck of insufficient generalization ability, relying too much on training data, and performing poorly on unseen flight environments or new obstacle types. In practical engineering applications, models are difficult to adapt to changing terrain, weather, and obstacle distribution, leading to large fluctuations in detection performance and difficulties in deployment. This limitation not only increases development costs but also reduces the robustness of the system, making it difficult to ensure stable operation in unknown scenarios.

[0006] Overall, existing obstacle avoidance technologies still have significant defects in detecting small and high-risk targets such as power lines, including low detection probability, high false alarm rate, and weak generalization ability. In complex background environments, the system's recognition ability for group targets or isolated obstacles is limited, and it cannot balance the accuracy and real-time requirements. These shortcomings not only affect flight safety but also restrict the efficiency of emergency rescue and other tasks, making it urgent to develop adaptive and improved solutions to improve overall performance. SUMMARY

[0007] The purpose of the present application is to solve the problem of low detection probability and high false alarm rate of existing anti-collision radars for small targets such as power lines. Therefore, a radar target detection and recognition method based on remote sensing data knowledge assistance is proposed. Without increasing the original hardware resources of helicopters and other aircraft, the present application introduces third-party remote sensing satellite images to generate power tower positions as prior auxiliary knowledge base, ensuring the detection and recognition ability of the radar itself while improving the detection and recognition ability of high-risk targets such as power towers and lines.

[0008] The present application adopts the following technical solutions to achieve the purpose:

[0009] A radar target detection and recognition method based on remote sensing data knowledge assistance, comprising the following steps:

[0010] S1, acquire remote sensing satellite image data, adopt a machine learning target recognition algorithm, and combine geographic location information to pre-construct a power tower position auxiliary knowledge base;

[0011] S2, in a flight mission, load the power tower target position data within the preset range of the corresponding flight area from the power tower position auxiliary knowledge base according to the aircraft inertial navigation position data, forming a real-time processing database for the aircraft anti-collision radar;

[0012] S3, the carrier collision avoidance radar works, receives the echo signal and pre-processes, executes threshold detection, and obtains the isolated target list corresponding to the flight area; each target in the isolated target list is tracked and processed, and the attribute label type is initialized to form a tracking target list;

[0013] S4, the tracking target list is matched with the power tower target in the real-time processing database, and if the matching is successful, the attribute label type of the corresponding target is marked as a power tower, and a power tower target is obtained;

[0014] S5, after the matching is completed, each target in the tracking target list that is not successfully matched is identified as a power line, and a power line target is obtained; finally, the obtained power tower target and power line target are subjected to Hough straight line detection, and if the detection is passed, the corresponding power tower / line target is confirmed, and the target detection and identification is completed.

[0015] Specifically, in step S1, the power tower image in the remote sensing satellite image data is marked according to the known power tower position, and is provided to the YOLO algorithm for target identification; for the identified power tower target, the position conversion is performed in combination with the geographic information in the remote sensing satellite image data, and a power tower position auxiliary knowledge base containing the longitude and latitude of each power tower target is generated;

[0016] The position conversion process is completed by using the geographic information built-in the image metadata in the remote sensing satellite image data, and converting in combination with the pixel position of the power tower target on the image.

[0017] Preferably, in step S2, according to the carrier inertial navigation position data and the carrier flight speed, the preset longitude and latitude range at the current position of the carrier is periodically updated and selected as the flight area, the power tower target position data in the flight area is loaded, and the longitude and latitude coordinates of the power tower target therein are converted to the geocentric rectangular coordinate system, thereby generating a real-time processing database of the carrier collision avoidance radar.

[0018] Specifically, in step S3, after the carrier collision avoidance radar receives the echo signal, the echo signal is pre-processed first, a one-dimensional range is generated by FFT transformation, and then the one-dimensional range is Doppler compensated, thereby completing the pre-processing process.

[0019] Specifically, in step S3, after the echo signal is pre-processed, a range-azimuth two-dimensional image is formed, and threshold detection is performed thereon; first, noise estimation is performed to obtain the image noise ; then the amplitudes of the image points in the range-azimuth two-dimensional image with amplitudes lower than the image noise are set to 0, and a first-level threshold detection result is obtained; and then the amplitudes of the image points in the range-azimuth two-dimensional image with amplitudes higher than the image noise A preset number of image points are used to record the distance and orientation of these image points, thus obtaining the secondary threshold detection results. Based on the primary and secondary threshold detection results, a list of isolated targets is obtained after removing group targets.

[0020] Preferably, when removing group targets, for the first-level threshold detection results, binary processing is performed, and the amplitude of image points with amplitude greater than 0 is set to 1; for the second-level threshold detection results, based on the target list it contains, combined with the first-level threshold detection results after binary processing, the connected regions of each target are extracted; if the number of pixels in the connected region of any target is greater than a preset value, the target is considered to be a group target, and it is output and removed from the target list; after completing the traversal processing of all targets in the target list, an isolated target list is obtained.

[0021] Preferably, in step S3, after obtaining the list of isolated targets, the position coordinates of each target are transformed from the radar polar coordinate system to the geocentric rectangular coordinate system before tracking processing; the tracking processing procedure is as follows:

[0022] First, determine if a non-empty tracking target list already exists. If the tracking target list is still empty, i.e. there are no targets in it, then the tracking target list is initialized directly based on the isolated target list. Each target in the isolated target list is assigned a tracking target number, and the attribute label type of each target is initialized to isolated target, the number of times each target was detected is 1, and finally the time when each target was detected is set.

[0023] If the target tracking list is not empty, it is fused with the obtained isolated target list. During the fusion process, each target in the isolated target list is compared with each target in the target tracking list. If the preset fusion criteria are met, the fusion is performed, the detection count of the target under the corresponding label is incremented by 1, and its last detection time is updated. If a target in the isolated target list does not meet the preset fusion criteria with any of the targets in the target tracking list, the target is added to the end of the target tracking list as a new target, and its initialization is performed accordingly.

[0024] Specifically, in step S4, during matching, any target in the target list will be tracked. The position in the geocentric rectangular coordinate system is denoted as It will process any target in the database in real time. The position in the geocentric rectangular coordinate system is denoted as If the distance between the two satisfies the following formula:

[0025]

[0026] Then the two are considered to match; in the formula, Matched judgment threshold value is determined according to radar ranging and angle measurement error synthesis.

[0027] Specifically, in step S5, when identifying the power line of each target that is not successfully matched, points meeting the Bragg scattering characteristics are detected and extracted from each target, so as to determine whether the corresponding target is a power line target; according to the Bragg scattering principle of electromagnetic waves, the radar signal incidence angle The occurrence of the peak value and the valley value in the corresponding backscattering cross-section area is used to determine whether the corresponding target is a power line target.

[0028] Specifically, in step S5, the obtained power tower target and power line target are converted to a Hough plane through Hough transformation, and then detection and confirmation of the corresponding power tower / line target are realized.

[0029] As described above, since the technical solution is adopted, the present application has the following beneficial effects:

[0030] The present application improves the overall performance of the collision avoidance radar system in a complex low-altitude environment, especially in dealing with high-risk targets such as power towers / lines; the method effectively integrates external geographic information data resources without increasing any additional hardware burden on the helicopter, and builds an auxiliary knowledge base, providing important prior information support for real-time detection of the radar.

[0031] The present application greatly enhances the detection and recognition ability of the power tower / line target; by using the power tower position information provided by the auxiliary knowledge base, the detection and analysis of the radar can be more accurately guided and focused, significantly improving the discovery probability and confirmation reliability of the associated power line target. At the same time, the present application reduces the false alarm signals that may be generated by the system under complex background interference, and the probability of incorrectly identifying other targets as power towers / lines, thereby greatly improving the accuracy and reliability of the alarm.

[0032] The present application maintains the system's conventional detection capabilities for group targets and isolated obstacles, while optimizing the comprehensive response capability for high-risk and small targets such as power lines; its core advantage lies in not relying on expensive hardware upgrades or complex multi-sensor fusion, but only through the integration of software and data layers, which can effectively improve the performance. The present application is easy to implement, has high cost-effectiveness, and has high practical value and promotion potential, and is particularly suitable for helicopters and unmanned aerial vehicles that are sensitive to weight, volume and power consumption. BRIEF DESCRIPTION OF DRAWINGS

[0033] The present application is further described in detail by the following drawings, which specifically include 5 figures, as follows:

[0034] Figure 1 It is a detailed flowchart of the target detection and recognition method of the present application.

[0035] Figure 2 The flow chart for constructing the power tower position auxiliary knowledge base in the method of the present application;

[0036] Figure 3 The image marking of the known position of the power tower in the method of the present application;

[0037] Figure 4 The neighborhood in the extraction of the connected region in the method of the present application;

[0038] Figure 5 The schematic diagram for applying the Bragg scattering principle in the method of the present application. DETAILED DESCRIPTION

[0039] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0040] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0041] A collision avoidance radar target detection and recognition method based on remote sensing data knowledge assistance, comprising the following steps:

[0042] S1, acquiring remote sensing satellite image data, using a machine learning target recognition algorithm, combining geographic position information, and pre-constructing a power tower position auxiliary knowledge base;

[0043] S2, in the flight task, according to the inertial navigation position data of the carrier aircraft, loading the power tower target position data within the preset range of the corresponding flight area from the power tower position auxiliary knowledge base, forming a real-time processing database of the carrier aircraft collision avoidance radar;

[0044] S3, the carrier aircraft collision avoidance radar works, receives the echo signal and pre-processes, performs threshold detection, and obtains the isolated target list of the corresponding flight area; each target in the isolated target list is tracked and processed, and after the attribute label type is initialized, a tracked target list is formed;

[0045] S4. Match the list of tracked targets with the power tower targets in the real-time processing database. If a match is found, mark the attribute tag type of the corresponding target as power tower to obtain the power tower target.

[0046] S5. After matching is completed, power line identification is performed on each target in the tracking target list that did not match successfully to obtain power line targets; finally, Hough line detection is performed on the obtained power tower targets and power line targets. If the detection is successful, the corresponding power tower / line target is confirmed, and the target detection and identification is completed.

[0047] Figure 1 The detailed process of the method is shown and can be viewed simultaneously; this embodiment will describe the details of each step in the above order.

[0048] First, in step S1, as follows Figure 2 As shown, based on the known locations of power towers, the images of power towers in remote sensing satellite image data are marked and provided to the YOLO algorithm for target recognition. For the identified power tower targets, the location is transformed by combining the geographical information in the remote sensing satellite image data to generate a power tower location auxiliary knowledge base containing the latitude and longitude of each power tower target.

[0049] In this embodiment, high-resolution remote sensing satellite image data can be obtained from various sources, such as the China Center for Resources Satellite Data and Google Maps. The image source resolution used in this embodiment is 1m. Subsequently, based on some pre-known locations of power towers, the power tower images in the acquired high-resolution remote sensing satellite image data are marked, as can be seen in [reference needed]. Figure 3 The illustration shows that when labeling target images, images of power towers in different terrains and landforms should be selected as much as possible to improve the generalization and adaptability of machine learning.

[0050] When using machine learning image recognition algorithms to identify power tower targets in remote sensing satellite image data, this implementation uses the YOLO v10 algorithm as the machine learning image recognition algorithm. YOLO is a single-stage target detection algorithm based on CNN. It predicts all target categories and bounding boxes in the image through one forward pass. The model input is the labeled power tower target image, and the output is each power tower target in the remote sensing satellite image.

[0051] Subsequently, the identified power tower targets were combined with remote sensing data and geographic information to perform location conversion, generating a power tower location database. The database format consists of the longitude and latitude coordinates of each power tower target.

[0052] In this embodiment, the location conversion process utilizes the geographic information embedded in the image metadata of the remote sensing satellite image data, combined with the pixel location of the power tower target on the image, to complete the conversion. For example, in a certain remote sensing satellite image, the pixel of the power tower target is located at... , The target horizontal pixels (from left to right, starting from 0). The target's geographical location is represented by pixels in the vertical direction (from top to bottom, starting from 0). for:

[0053]

[0054] in, 、 This is the starting point of the image in the geographic coordinate system, typically the geographic coordinates of the top-left pixel of the image. Eastward coordinates The pixel scale for the corresponding direction, in meters per pixel, is usually a positive number. North coordinates The pixel scale for the corresponding direction, in meters per pixel, is usually a negative number. For the horizontal direction relative to the east coordinate The impact, Vertical direction relative to north coordinates The impact.

[0055] In step S2, after the radar is powered on, the pre-generated power tower target location data can be loaded into the radar signal processing work area. Based on the aircraft's inertial navigation position data and flight speed, a preset latitude and longitude range at the aircraft's current position is periodically updated and selected as the flight area. The power tower target location data within this flight area is loaded, and the latitude and longitude coordinates of the power tower targets are converted to a geocentric rectangular coordinate system, thereby generating a real-time processing database for the aircraft collision avoidance radar.

[0056] This implementation method selects the vicinity of the current location of the carrier aircraft. The system identifies power tower targets within a specified latitude and longitude range and converts their coordinates to a geocentric rectangular coordinate system to create a real-time database, which is updated every half hour.

[0057] The method for converting latitude and longitude to the geocentric rectangular coordinate system is as follows:

[0058]

[0059] In the formula, Latitude Longitude; The altitude is set to 0 here; The colatitude curvature radius of the ellipsoid is as follows:

[0060]

[0061] In the formula, is the long semi-axis of the earth, and is 6378137 m, is the first eccentricity of the earth, and is 0.00669437999.

[0062] In step S3, after the carrier anti-collision radar receives the echo signal, the echo signal is first preprocessed, a one-dimensional distance direction is generated through FFT transformation, and then the one-dimensional distance direction is Doppler compensated, so as to complete the preprocessing process; the compensated Doppler frequency is as follows:

[0063]

[0064] In the formula, is the northward velocity of the carrier, is the eastward velocity of the carrier; is the azimuth beam pointing angle, is the elevation beam pointing angle; is the wavelength of the carrier.

[0065] Subsequently, the echo signal is preprocessed to form a distance-azimuth two-dimensional image, and threshold detection is performed on the image; first, noise estimation is performed, and the amplitude of the first point in the distance-azimuth two-dimensional image is recorded as The amplitudes of each point are sorted, and the points with high amplitude values on the high side in the sorting are removed, that is, the image noise is as follows:

[0066]

[0067]

[0068]

[0069] In the formula, is the total number of pixel points in the image, is the number of points with high amplitude values on the high side that are removed; is the average value of the remaining data, is the standard deviation of the remaining data; represents sorting in ascending order.

[0070] Subsequently, the amplitudes of the image points in the distance-azimuth two-dimensional image with amplitudes lower than the image noise are set to 0, that is, a first-level threshold detection result is obtained; and the amplitudes of the image points in the distance-azimuth two-dimensional image with amplitudes higher than the image noise The image points of the preset multiple, the distance number and the azimuth number where the image points are located are recorded, i.e. a secondary threshold detection result is obtained; based on the primary threshold detection result and the secondary threshold detection result, a group target is removed to obtain an isolated target list.

[0071] As a preferred embodiment of the present embodiment, when the group target is removed, for the primary threshold detection result, binary processing is performed, and the amplitudes of the image points with amplitudes greater than 0 are set to 1; for the secondary threshold detection result, based on the target list contained therein, the connected region extraction of each target is performed in combination with the binary-processed primary threshold detection result; if the pixel number of the connected region extracted for any target is greater than a preset value, the target is considered to be a group target, and the target is output and removed from the target list; after the traversal processing of all targets in the target list is completed, an isolated target list is obtained.

[0072] In the extraction of the connected region, the target obtained by the secondary threshold detection is taken as a seed, a new label is assigned, and the seed is placed at the bottom of a stack; it is determined whether the stack is empty, if not, the element is marked with the label assigned in the foregoing, the elements in the stack are taken out, and the eight pixels in the neighborhood of the element in the primary threshold detection result are sequentially accessed, as shown in the neighborhood of the pixel, it is determined whether the pixel value is 1, if 1, the pixel is pushed into the stack; the process is repeated until the stack is empty; subsequently, the number of elements taken out from the stack is counted, if greater than or equal to a preset value, the target is a group target, and the target is output as a group target; if less than the preset value, the target is an isolated target, and the target is placed in the isolated target list. Figure 4

[0073] In step S3, after the isolated target list is obtained, the position coordinates of each target in the isolated target list are converted from the radar polar coordinate system to the geocentric rectangular coordinate system, and then tracking processing is performed. The process of converting the target from the radar polar coordinate system to the geocentric rectangular coordinate system is as follows:

[0074] Firstly, the radar polar coordinate system is converted to the aircraft rectangular coordinate system, the purpose of the conversion is to convert the target from the radar polar coordinate position to the rectangular coordinate, wherein and respectively represent the radar distance, the radar azimuth angle and the radar pitch angle; the conversion formula is:

[0075]

[0076] Then, the aircraft rectangular coordinate system is converted to the aircraft geographic coordinate system, which is the East-North-Sky coordinate (ENU) of the geographic position of the aircraft, and the conversion formula is:

[0077] ​​​​​​​​

[0078] wherein, is the yaw angle of the carrier platform, is the pitch angle of the carrier platform, is the roll angle of the carrier platform.

[0079] Then, the carrier geographic coordinate system is converted to the geocentric rectangular coordinate system , and the conversion formula is:

[0080]

[0081] wherein, is the geodetic longitude of the radar platform, is the geodetic latitude of the radar platform, , , is the position of the radar platform in the geocentric rectangular coordinate system.

[0082] In step S3, the tracking process is as follows:

[0083] First, it is determined whether there is a non-empty tracking target list. If the tracking target list is still empty, i.e., there is no target in it, the initialization of the tracking target list is directly performed based on the isolated target list, each target in the isolated target list is assigned a tracking target serial number, and the attribute label type of each target is initialized to isolated target, the number of times each target is detected is 1, and the last time each target is detected.

[0084] If the tracking target list is not empty, it is fused with the obtained isolated target list. When the fusion is performed, each target in the isolated target list is compared with the targets in the tracking target list one by one. If the preset fusion condition standard is met, the fusion is performed and the number of times the target under the corresponding label is detected is increased by 1, and the last detection time is updated. If the preset fusion condition standard is not met between a target in the isolated target list and all targets in the tracking target list, the target is added to the end of the tracking target list as a new target, and the initialization thereof is performed correspondingly.

[0085] In this embodiment, the preset fusion condition standard here is the same as the matching method principle to be introduced in step S4, i.e., the geocentric rectangular coordinate system position of any target in the isolated target list is recorded as ; the geocentric rectangular coordinate system position of any target in the tracking target list is recorded as ; if the distance between the two satisfies the following formula:

[0086]

[0087] ​​Then the two are considered to match; in the formula, The matching judgment threshold is determined based on the combined errors of radar ranging and angle measurement, and is preferably set to 100m in this embodiment.

[0088] After traversing all isolated targets in the target list, the latest target list is obtained. In addition, the detection time of all targets in the target list is checked. If the detection time of a target exceeds a preset value and the target is not updated, the target is deleted.

[0089] In step S4, the matching process is the same as that used in isolated target list tracking; any target in the target list will be matched. The position in the geocentric rectangular coordinate system is denoted as It will process any target in the database in real time. The position in the geocentric rectangular coordinate system is denoted as If the distance between the two satisfies the following formula:

[0090]

[0091] Then the two are considered to match; in the formula, In this embodiment, the matching judgment threshold, which is determined based on the combined errors of radar ranging and angle measurement, is preferably set to 100m.

[0092] In step S5, when identifying electric field lines for each unmatched target, points satisfying the Bragg scattering property are detected and extracted from each target to determine whether the corresponding target is an electric field line target. Electric field lines have Bragg scattering properties, which can be quickly separated from other targets. Therefore, this property can be used to extract points that satisfy the Bragg scattering property from the target.

[0093] See also Figure 5 The indication, Figure 5 middle The horizontal spacing of the stranded power lines represents the overall diameter after the power lines are twisted together. The operating wavelength of radar signals Then, based on the Bragg scattering principle of electromagnetic waves, if the incident angle of the radar signal... The following relationship must be satisfied:

[0094]

[0095] This represents the energy of the two parallel incident radar waves at this time ( Figure 5 The two parallel energy waves (labeled 1 and 2) are in the same direction after reflection, meaning that the backscattering cross-section (RCS) of the electric field lines reaches its peak at this point; where... For measurement ordinal numbers; conversely, if the radar signal incident angle... satisfy the following relationship:

[0096]

[0097] Then it represents that the energy of the two radar waves which are parallel incident at this time is reversed after reflection, that is, the backscattering cross section of the power line at this time appears a valley value; thus according to the incident angle of the radar signal According to the occurrence of the peak value and the valley value of the corresponding backscattering cross section, it is judged whether the corresponding target is the power line target.

[0098] In the embodiment, when the radar works at 34GHz, the peak angle interval of the backscattering cross section of the power line is shown in the following table 1.

[0099] Table 1 Relationship table between the distance between the splices and the peak angle of the RCS

[0100]

[0101] Finally, in step S5, the obtained power tower target and power line target are converted to the Hough plane by the Hough transform, and then the detection and confirmation of the corresponding power tower / line target are realized; the specific mode is as follows:

[0102] For any point of the plane rectangular coordinate system , the expression of all straight lines passing through the point is as follows:

[0103]

[0104] In the formula, is the polar radius, which represents the shortest distance from the straight line to the coordinate origin; is the polar angle, which represents the included angle between the normal line of the straight line and the positive direction of the x axis; the parameters of these straight lines are drawn in - plane, and a sinusoidal curve is obtained; the - plane is the Hough plane, and multiple points in the - plane will produce multiple different sinusoidal curves;

[0105] According to the above principle, when the power tower target and the power line target are converted to the Hough plane, the number of curves in which the intersection points are located is counted to detect the straight line, and the greater the number is, the more points on the straight line there are; by presetting a number threshold, when the number of curves in which a certain intersection point is located is greater than the number threshold, it is considered that the power tower target and the power line target corresponding to the curve focusing on the intersection point are detected and confirmed, and are the correct power tower / line target; in the embodiment, the preset number threshold is 5.

Claims

1. A method for collision avoidance radar target detection and identification based on remote sensing data knowledge assistance, characterized in that, The method comprises the following steps: S1, acquiring remote sensing satellite image data, adopting a machine learning target recognition algorithm, and combining geographic position information to pre-construct an electric power tower position auxiliary knowledge base; S2, in a flight task, loading electric power tower target position data within a preset range of a corresponding flight area from the electric power tower position auxiliary knowledge base according to aircraft inertial navigation position data to form a real-time processing database of the aircraft collision avoidance radar; S3, the aircraft collision avoidance radar works, receives a return signal and pre-processes the return signal, performs threshold detection, and obtains an isolated target list of the corresponding flight area; tracking processing is performed on each target in the isolated target list, and an attribute label type is initialized to form a tracking target list; S4, the tracking target list is matched with the electric power tower target in the real-time processing database, and if the matching is successful, the attribute label type of the corresponding target is marked as an electric power tower to obtain an electric power tower target; S5, after the matching is completed, electric power line identification is performed on each target in the tracking target list that is not successfully matched to obtain an electric power line target; finally, Hough straight line detection is performed on the obtained electric power tower target and electric power line target, and if the detection is passed, the corresponding electric power tower / line target is confirmed, and target detection and recognition are completed; In step S1, the electric power tower image in the remote sensing satellite image data is marked according to the known electric power tower position, and is provided to a YOLO algorithm for target recognition; the position of the electric power tower target is converted in combination with the geographic information in the remote sensing satellite image data to generate an electric power tower position auxiliary knowledge base containing the longitude and latitude of each electric power tower target; The position conversion process is completed after conversion in combination with the pixel position of the electric power tower target on the image by using the geographic information built in the image metadata in the remote sensing satellite image data; In step S2, the electric power tower target position data in a preset longitude and latitude range at the current position of the aircraft is loaded according to the aircraft inertial navigation position data and the flight speed of the aircraft, and the longitude and latitude coordinates of the electric power tower target in the electric power tower target position data are converted to the geocentric rectangular coordinate system to generate the real-time processing database of the aircraft collision avoidance radar; In step S4, when matching, the position of any target in the tracking target list in the geocentric rectangular coordinate system is recorded as ; the position of any target in the real-time processing database in the geocentric rectangular coordinate system is recorded as ; if the distance between the two satisfies the following formula: , the target is determined to be a tracking target.​ If the following condition is satisfied, it is considered that the two are matched; in the formula, is the matching judgment threshold determined according to the radar ranging and angle measurement error synthesis; In step S3, after the isolated target list is obtained, the position coordinates of each target in the isolated target list are converted from the radar polar coordinate system to the geocentric rectangular coordinate system, and then tracking processing is performed; the tracking processing process is as follows: Firstly, it is judged whether there is a non-empty tracking target list, if the tracking target list is still empty, that is, there is no target in the tracking target list, then the initialization of the tracking target list is directly performed based on the isolated target list, each target in the isolated target list is assigned a tracking target serial number, and the attribute label type of each target is initialized as an isolated target, the number of times of detecting each target is 1, and the time of last detecting each target is initialized; if the tracking target list is not empty, the tracking fusion is performed between the tracking target list and the obtained isolated target list; each target in the isolated target list is compared with the targets in the tracking target list one by one during the tracking fusion, if a preset fusion condition standard is met, the fusion is performed and the number of times of detecting the target under the corresponding label is increased by 1, and the last detection time is updated; If a target in the isolated target list does not meet the preset fusion condition standard with all targets in the tracking target list, the target is added to the end of the tracking target list as a new target, and initialization of the target is performed.

2. The method according to claim 1, wherein: In step S3, after the carrier anti-collision radar receives the echo signal, the echo signal is first pre-processed, a one-dimensional distance direction is generated through FFT transformation, and then the one-dimensional distance direction is Doppler compensated, so as to complete the pre-processing process; the compensated Doppler frequency The following formula: wherein Vn is the north velocity of the platform, Ve is the east velocity of the platform; is the azimuth beam pointing angle, is the elevation beam pointing angle; is the wavelength of the carrier.

3. The method according to claim 1, wherein: In step S3, the echo signal is pre-processed to form a range-azimuth two-dimensional image, and threshold detection is performed on the image. First, noise estimation is performed, and the amplitudes of the first points in the range-azimuth two-dimensional image are recorded as The amplitudes of each point are sorted, and the points with high amplitude values on the high side of the sort are removed, i.e. image noise The following formula is used: In the formula, is the total number of pixels in the image, is the number of points on the high side of the amplitude value that are removed; is the average value of the remaining data, is the standard deviation of the remaining data; represents sorting in ascending order; Subsequently, the amplitude of the range-azimuth 2D image is lower than the image noise. The amplitude of the image points is set to 0, thus obtaining the first-level threshold detection result; then, the amplitude of the points in the range-azimuth two-dimensional image that is higher than the image noise is extracted. A preset number of image points are used to record the distance and orientation of these image points, thus obtaining the secondary threshold detection results. Based on the primary and secondary threshold detection results, a list of isolated targets is obtained after removing group targets.

4. The anti-collision radar target detection and recognition method according to claim 3, characterized in that: When the group target is removed, for the first threshold detection result, binary processing is performed, and amplitudes of image points with amplitudes greater than 0 are set to 1; For the second threshold detection result, based on a target list contained in the second threshold detection result, the first threshold detection result after binary processing is combined, and a connected region of each target is extracted; if a pixel number of the connected region of any target extracted is greater than a preset value, the target is considered as a group target, the target is output, and the target is removed from the target list; After traversal processing of all targets in the target list is completed, an isolated target list is obtained.

5. The method of claim 1, wherein: In step S5, when performing power line identification on each target that is not successfully matched, a point satisfying the Bragg scattering characteristic is detected and extracted from each target, so as to determine whether the corresponding target is a power line target; a twisted level spacing of the power line is acquired in advance , a working wavelength of the radar signal After that, based on the Bragg scattering principle of electromagnetic wave, if the incidence angle of the radar signal satisfies the following relationship: represents the time when the two radar wave energies incident in parallel are reflected in the same direction, i.e. the time when the backscattering cross section of the power line reaches its maximum; where is the ordinal number; on the contrary, if the radar signal incidence angle satisfies the following relation: represents the two beams of radar wave energy at this time parallel incident after reflection is reversed, that is, the backscattering cross section of the power line at this time appears a valley; thus according to the radar signal incidence angle According to the occurrence of the peak and valley of the corresponding backscattering cross section, it is judged whether the corresponding target is a power line target.

6. The method of claim 1, wherein: In step S5, the obtained power tower target and power line target are converted to a Hough plane by Hough transformation, and detection and confirmation of the corresponding power tower / line target are realized; a specific mode is as follows: For any point of the plane rectangular coordinate system , all straight lines passing through the point are expressed as follows: wherein, is the polar radius, representing the shortest distance from the straight line to the coordinate origin; is the polar angle, representing the included angle between the normal line of the straight line and the positive direction of the x-axis, ; the parameters of these straight lines are drawn in - the plane, obtaining a sinusoidal curve; the - plane is the Hough plane, and multiple points in - the plane will generate multiple different sinusoidal curves; According to the detection and confirmation process of the corresponding power tower / line target, after the power tower target and the power line target are converted to the Hough plane, a number of curves in which each intersection point in the Hough plane is counted to detect a straight line, and the greater the number is, the more points on the straight line are; through a preset number threshold, when the number of curves in which a certain intersection point is greater than the number threshold, it is considered that the power tower target and the power line target corresponding to the curve focusing on the intersection point are detected and confirmed, and are correct power tower / line targets.

Images