A fixed double runway FOD detection method based on multi-source circumferential scanning

Abstract

This invention discloses a fixed dual-runway FOD detection method based on multi-source perimeter scanning, belonging to the field of data fusion detection technology. Specifically, it includes: synchronously acquiring multi-polarization echo data of the two runways using a fixed perimeter scanning radar; combining an adaptive clutter cancellation method with dynamic background template matching and a runway material feature library for time-varying clutter suppression; generating a time-frequency domain three-dimensional feature matrix through time-frequency analysis; synchronously acquiring and preprocessing image data; marking target areas using a constant false alarm rate algorithm and image detection; establishing a multi-source data geographic coordinate mapping model; completing data registration and cross-validation in the runway physical coordinate system; and achieving real-time tracking of real FOD targets based on radar and image fusion data through multi-dimensional confirmation using radar polarization / micro-motion features and image visual features. This invention improves the accuracy of FOD detection and location recognition through multi-source data fusion, effectively reducing the false alarm rate and providing assurance for airport safety operations.

Description

Technical Field

[0001] This invention belongs to the field of data fusion detection technology, specifically a fixed dual-runway FOD detection method based on multi-source cyclic scanning. Background Technology

[0002] Foreign Object Debris (FOD) on airport runways can impact the safe takeoff and landing of aircraft. Millimeter-wave radar, with its advantages of high range resolution, immunity to lighting conditions, and excellent all-weather operation, has become a crucial technology for FOD detection. However, in practical applications, millimeter-wave radar faces numerous challenges in FOD detection.

[0003] Clutter interference is a major problem. Airport environments are complex, with various buildings, vehicles, personnel, and vegetation around the runway. These all reflect millimeter-wave radar signals, creating clutter. Even with conventional clutter cancellation processing, it is difficult to completely eliminate interference reflections from strongly scattering targets or non-uniform clutter, resulting in a high false alarm probability and affecting the accuracy of FOD detection.

[0004] Furthermore, non-metallic FODs have weak reflectivity to millimeter waves. Many common non-metallic FODs, such as plastic fragments, rubber particles, and paper, have small radar cross-sections, resulting in very weak signals reflected back to the radar. Under the cover of strong background clutter, traditional FOD detection methods have difficulty detecting and distinguishing them, reducing the probability of detecting such FODs.

[0005] Current FOD detection methods are insufficient in addressing these challenges and cannot meet the high accuracy and reliability requirements of airports for FOD detection. Therefore, developing a FOD detection method that can effectively overcome these problems is of significant practical importance. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a fixed dual-runway FOD detection method based on multi-source perimeter scanning. This method simultaneously acquires multi-polarization echo data from both runways using a fixed perimeter scanning radar. It combines an adaptive clutter cancellation method with dynamic background template matching and a runway material feature library for time-varying clutter suppression, generating a time-frequency domain three-dimensional feature matrix through time-frequency analysis. Simultaneously, image data is acquired and preprocessed, and target areas are marked using a constant false alarm rate algorithm and image detection. A multi-source data geographic coordinate mapping model is established, and data registration and cross-validation are completed in the runway physical coordinate system. Through multi-dimensional confirmation using radar polarization / micro-motion features and image visual features, real-time tracking of true FOD targets is achieved based on radar and image fusion data. This invention improves the accuracy of FOD detection and location recognition through multi-source data fusion, effectively reducing the false alarm rate and ensuring safe airport operations.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A fixed dual-runway FOD detection method based on multi-source peripheral scanning includes:

[0009] S1: Simultaneously acquire multi-polarization echo data of the two runways using a fixed perimeter-scanning radar. Employ an adaptive clutter cancellation method with dynamic background template matching, combined with a pre-established runway material feature library, to suppress time-varying clutter in the multi-polarization echo data. Perform time-frequency analysis on the clutter-suppressed multi-polarization echo data to generate a three-dimensional feature matrix in the time-frequency domain containing amplitude-frequency-polarization features. Simultaneously trigger the runway monitoring camera to acquire image data of the corresponding area to complete image preprocessing.

[0010] S2: Input the three-dimensional feature matrix in the time-frequency domain into the constant false alarm rate detection algorithm to initially detect areas with FOD targets in the entire dual-runway area and mark the target points in the radar coordinate system. At the same time, perform target detection on the preprocessed image data and mark the suspected foreign object areas in the image coordinate system.

[0011] S3: Establish a multi-source data geographic coordinate mapping model, and at the same time, perform data registration and cross-validation based on the runway physical coordinate system;

[0012] S4: For the initially matched candidate targets, multi-dimensional features are introduced for confirmation, and the confirmed real FOD targets are tracked in real time based on fused data; the fused data is multi-source fusion information formed by registering and associating radar echo data and image data.

[0013] Specifically, the method of synchronously acquiring multi-polarization echo data of the two runways using a fixed perimeter-scan radar includes:

[0014] The radar scanning cycle is set to T seconds based on the length, width, and detection accuracy requirements of the two runways.

[0015] The radar is configured with multiple polarization transmission and reception modes, and uses horizontal polarization and vertical polarization to alternately transmit and receive signals to obtain echo data under different polarization modes;

[0016] The acquired multipolar echo data undergoes preliminary digital processing to obtain multipolar echo data.

[0017] Specifically, the adaptive clutter cancellation method using dynamic background template matching, combined with a pre-established runway material feature library, performs time-varying clutter suppression on multi-polarization echo data, including:

[0018] Real-time acquisition of environmental data surrounding the radar to construct an initial dynamic background template; the environmental data surrounding the radar includes background echo data when there is no target.

[0019] A sliding window approach is used to statistically analyze the background echo data within the window, calculate its mean, and adjust the parameters of the background template based on the mean.

[0020] Acquire multipolar echo data and correct the multipolar echo data by combining it with a pre-established runway material feature library;

[0021] The error between the corrected echo data and the dynamic background template is calculated point by point using the minimum mean square error criterion. Clutter is then canceled according to the magnitude of the error to obtain multi-polarized echo data after clutter suppression.

[0022] Specifically, the step of performing time-frequency analysis on the clutter-suppressed multi-polarization echo data to generate a three-dimensional time-frequency domain feature matrix containing amplitude-frequency-polarization characteristics includes:

[0023] The short-time Fourier transform method is used to perform time-frequency analysis on the clutter-suppressed multipolar echo data. During the time-frequency analysis, the clutter-suppressed echo data is segmented according to the preset window function type and window length parameters.

[0024] For each echo data segment, its amplitude value at different frequency points is calculated, and combined with multi-polarization information, the amplitude-frequency characteristics corresponding to different polarization modes are obtained respectively.

[0025] The amplitude-frequency features under different polarization modes are integrated to form a three-dimensional data set containing amplitude-frequency-polarization features;

[0026] The three-dimensional data set is normalized to generate the final time-frequency domain three-dimensional feature matrix.

[0027] Specifically, the steps of S2 include:

[0028] Obtain the three-dimensional feature matrix in the time-frequency domain, and preprocess the three-dimensional feature matrix in the time-frequency domain by normalizing each element in the matrix;

[0029] Within the entire dual-track domain, a sliding window approach is used to calculate the local statistics for each data point in the normalized time-frequency domain three-dimensional feature matrix; the local statistics include the mean and variance.

[0030] The detection threshold is calculated using local statistics based on the preset false alarm probability.

[0031] Compare the amplitude value of each data point in the window with the detection threshold value. If the amplitude value exceeds the threshold value, it is marked as a target point in the radar coordinate system.

[0032] Preprocessed image data is acquired, and a deep learning-based object detection algorithm is used to identify foreign objects. The bounding box coordinates of suspected foreign object regions in the preprocessed image are extracted and marked as suspected foreign object regions in the image coordinate system. At the same time, the confidence score of each suspected foreign object region is recorded.

[0033] Specifically, establishing the multi-source data geographic coordinate mapping model includes:

[0034] Based on the latitude and longitude coordinates and scanning parameters of the radar installation location, a transformation relationship between the radar coordinate system and the geographic coordinate system is established; the transformation relationship is described by a coordinate transformation matrix.

[0035] Based on the camera intrinsic parameter calibration results and extrinsic parameters, a mapping relationship between the image coordinate system and the geographic coordinate system is established;

[0036] A physical coordinate system for the runway is established with any endpoint of the runway as the origin and the runway direction as the coordinate axis. Mapping models for radar-physical coordinates and image-physical coordinates are constructed respectively.

[0037] The radar target point and the suspected foreign object area are converted to the runway physical coordinate system, and the spatial position deviation value is calculated.

[0038] Specifically, the data registration and cross-validation based on the runway physical coordinate system includes:

[0039] Calculate the Euclidean distance between the radar target point and the centroid of the suspected foreign object area. When the Euclidean distance is less than or equal to a preset threshold, it is determined to be a candidate matching target.

[0040] Cluster analysis is performed on candidate matching targets based on prior information of runway pavement zoning to initially distinguish between real and false targets; the prior information of runway pavement zoning includes centerline region, edge region, and taxiway region.

[0041] Radar target points without image matching are marked as pending verification, and suspected foreign object areas without radar response are marked as low-confidence targets.

[0042] Specifically, the process of introducing multi-dimensional feature verification for the initially matched candidate targets includes:

[0043] At the radar data level, the target polarization scattering matrix features are extracted through multi-polarization processing and compared with a typical FOD polarization feature library to calculate the radar feature matching degree.

[0044] At the image data level, visual features of suspected foreign object areas are extracted and matched with a typical FOD visual feature library to calculate the image feature matching degree.

[0045] The fusion decision-making process confirms the true FOD target when both the radar feature matching degree and the image feature matching degree are greater than or equal to a preset radar feature matching degree threshold and a preset image feature matching degree threshold.

[0046] Specifically, target characteristics are confirmed through multi-polarization processing and micro-motion feature analysis, including:

[0047] For the initially identified real FOD targets, obtain their corresponding multi-polarization echo data, including amplitude and phase information under horizontal and vertical polarization modes;

[0048] Polarization decomposition is performed on multi-polarization echo data to calculate the polarization scattering matrix characteristics of the target. Based on the element values ​​of the polarization scattering matrix, the polarization characteristics of the target are analyzed. The polarization characteristics of the target include polarizability and polarization angle.

[0049] The polarization characteristics of the target are compared with the pre-established polarization feature library of typical FOD targets and false targets, and the similarity is used to determine whether the target is a real FOD target.

[0050] Meanwhile, for the initially identified real FOD targets, their echo data are acquired in multiple consecutive scanning cycles to obtain the target's echo data;

[0051] Short-time Fourier transform is used to process the target's echo data to extract the target's micro-motion frequency and micro-motion amplitude characteristics;

[0052] Based on the differences between the typical micro-movement patterns of FOD targets and the micro-movement characteristics of false targets, determine whether a target is a real FOD target.

[0053] Specifically, the real-time tracking of confirmed real FOD targets based on fused data includes:

[0054] The Kalman filter algorithm is used to track confirmed real FOD targets in real time, and the state equation and observation equation of the target are established. The state equation describes the changes of the target's position and velocity state over time. The observation equation describes the radar's observation process of the target's state.

[0055] Within each scanning cycle, the target's current state estimate and radar observations are used to update the state and predict the target's state at the next moment.

[0056] Based on the prediction results, the target's trajectory information is updated in real time, and the tracking results are fed back.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1. This invention proposes a fixed dual-runway FOD detection method based on multi-source perimeter scanning. This method synchronously acquires multi-polarization echo data of the two runways using a fixed perimeter scanning radar, and uses an adaptive clutter cancellation method with dynamic background template matching combined with a runway material feature library for time-varying clutter suppression. Then, time-frequency analysis is performed to generate a three-dimensional feature matrix in the time-frequency domain containing amplitude-frequency-polarization features. This series of operations effectively improves the quality of the echo data, enables more accurate extraction of feature information related to FOD targets, provides a reliable data foundation for subsequent detection, enhances the accuracy and sensitivity of detection, and reduces false detections and missed detections caused by clutter interference.

[0059] 2. This invention proposes a fixed dual-runway FOD detection method based on multi-source perimeter scanning. After initially marking target points by inputting a time-frequency domain three-dimensional feature matrix into a constant false alarm rate (CFAR) detection algorithm, a geographic coordinate mapping model is established for coordinate transformation. Combined with prior information on runway pavement zoning, target points are clustered to initially distinguish between real and false targets. Finally, real FOD targets are confirmed and tracked in real time through multi-polarization processing and micro-motion feature analysis. This process can accurately determine the target location, effectively eliminate interference from false targets, and achieve accurate identification and real-time monitoring of real FOD targets, providing strong support for airport safety operations. It also helps to promptly detect and handle FOD on the runway and reduce flight safety hazards. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of a fixed dual-runway FOD detection method based on multi-source peripheral scanning according to the present invention;

[0061] Figure 2 This is a flowchart illustrating the principle of a fixed dual-runway FOD detection method based on multi-source peripheral scanning according to the present invention.

[0062] Figure 3 This is a flowchart of the radar target point marking process of the present invention. Detailed Implementation

[0063] Example 1

[0064] Please see Figure 1 and Figure 2 The present invention provides an embodiment of a fixed dual-runway FOD detection method based on multi-source peripheral scanning, comprising the following steps:

[0065] S1: Simultaneously acquire multi-polarization echo data of the two runways using a fixed perimeter-scanning radar. Employ an adaptive clutter cancellation method with dynamic background template matching, combined with a pre-established runway material feature library, to suppress time-varying clutter in the multi-polarization echo data. Perform time-frequency analysis on the clutter-suppressed multi-polarization echo data to generate a three-dimensional feature matrix in the time-frequency domain containing amplitude-frequency-polarization features. Simultaneously trigger the runway monitoring camera to acquire image data of the corresponding area to complete image preprocessing.

[0066] Image preprocessing includes denoising, distortion correction, and region of interest cropping.

[0067] S2: Input the three-dimensional feature matrix in the time-frequency domain into the constant false alarm rate detection algorithm to initially detect areas with FOD targets in the entire dual-runway area and mark the target points in the radar coordinate system. At the same time, perform target detection on the preprocessed image data and mark the suspected foreign object areas in the image coordinate system.

[0068] It should be noted that the constant false alarm rate (CFAR) detection algorithm processes multi-dimensional data, including time, frequency, and amplitude information. For a three-dimensional feature matrix in the time-frequency domain containing amplitude-frequency-polarization features, the CFAR detection algorithm can be input and processed in the following ways:

[0069] (1) The constant false alarm rate detection algorithm treats each layer of the three-dimensional feature matrix, for example, each pair of frequency and polarization combinations, as a two-dimensional slice for processing;

[0070] (2) For each two-dimensional slice, the constant false alarm rate detection algorithm applies a sliding window on the slice, calculates the statistical characteristics of the local background noise, and determines the detection threshold based on these characteristics;

[0071] (3) For the position of each pixel in the three-dimensional feature matrix in the time-frequency domain, the constant false alarm rate detection algorithm compares its amplitude value with the threshold value estimated by the background noise.

[0072] (4) If the amplitude value of any pixel exceeds the threshold, it is considered that the target signal may exist at the corresponding position;

[0073] (5) The constant false alarm rate detection algorithm will output a binary matrix with the same shape as the input matrix, marking all pixels that exceed the threshold value as the detected target area.

[0074] S3: Establish a multi-source data geographic coordinate mapping model, and at the same time, perform data registration and cross-validation based on the runway physical coordinate system;

[0075] S4: For the initially matched candidate targets, multi-dimensional features are introduced for confirmation, and the confirmed real FOD targets are tracked in real time based on fused data; the fused data is multi-source fusion information formed by registering and associating radar echo data and image data.

[0076] It should be explained that FOD detection, or foreign object debris detection, refers to the process of identifying, locating, and issuing early warnings about foreign objects that may threaten aviation safety through technical means. Its core objective is to promptly detect foreign objects in areas such as airport runways, taxiways, and aprons, and to prevent these objects from being sucked into aircraft engines or colliding with aircraft landing gear and other components, thereby preventing mechanical failures, flight delays, or even flight accidents.

[0077] Example 2

[0078] Please see Figure 3 In this embodiment, the simultaneous acquisition of multi-polarization echo data of two runways using a fixed perimeter-scanning radar includes:

[0079] A1: Set the radar scanning cycle. Based on the length, width, and detection accuracy requirements of the two runways, determine the scanning cycle to be T seconds.

[0080] It is important to emphasize that when setting the radar scanning cycle, the airport's flight traffic and weather conditions should be further considered. When the flight traffic is high or the weather conditions are poor, the scanning cycle T seconds should be shortened to improve the timeliness of detection.

[0081] A2: Configure the radar with multiple polarization transmission and reception modes, and use horizontal polarization and vertical polarization to alternately transmit and receive signals to obtain echo data under different polarization modes;

[0082] Furthermore, the specific steps of A2 include:

[0083] (1) Two polarization methods are adopted: horizontal polarization and vertical polarization. Horizontal polarization means that when the radar transmits and receives signals, the direction of the electric field vector of the electromagnetic wave is horizontal; vertical polarization means that the direction of the electric field vector of the electromagnetic wave is vertical. The choice of these two polarization methods is based on the different reflection characteristics of different ground objects to different polarization signals. By alternately transmitting and receiving these two polarization signals, richer target information can be obtained.

[0084] (2) Determine the alternating transmission and reception order of horizontally polarized and vertically polarized signals. For example, first transmit the horizontally polarized signal and receive its echo, then transmit the vertically polarized signal and receive its echo, and so on in a cyclical alternation. This alternation method can ensure that echo data under both polarization modes can be obtained in each scanning cycle.

[0085] (3) Configure the antenna. The radar antenna needs to be able to transmit and receive horizontally polarized and vertically polarized signals. This is achieved by setting different polarizers in the antenna system. For example, use an orthogonal mode coupler to separate and combine horizontally polarized and vertically polarized signals to ensure that the antenna can accurately transmit and receive signals of the corresponding polarization.

[0086] (4) Debug the radar transmitter and receiver to enable them to stably transmit and receive horizontally polarized and vertically polarized signals. At the same time, optimize the receiver’s sensitivity, bandwidth and dynamic range to improve the ability to receive weak echo signals.

[0087] (5) According to the set alternating transmission strategy, the radar transmitter transmits horizontally polarized and vertically polarized signals in sequence. During the transmission process, it is necessary to accurately control parameters such as the transmission time, pulse width and repetition frequency of the signal to ensure the stability and consistency of the signal.

[0088] (6) The radar antenna receives the echo signal reflected back from the target and transmits it to the receiver. The receiver amplifies, filters and digitizes the echo signal to convert it into a digital signal suitable for subsequent processing. During the reception process, it is necessary to distinguish the echo signals under different polarization modes to avoid signal confusion.

[0089] (7) Store the acquired echo data under different polarization modes. High-speed data storage devices, such as solid-state drives or disk arrays, can be used to ensure data security and fast storage.

[0090] (8) Establish a data management system to classify, label and index the stored echo data. For example, classify and store the data according to information such as scanning time, polarization mode and target area to facilitate subsequent query and processing.

[0091] A3: Perform preliminary digital processing on the acquired multipolar echo data to obtain multipolar echo data.

[0092] The initial digital processing of the acquired multipolar echo data mainly involves analog-to-digital conversion.

[0093] The adaptive clutter cancellation method employing dynamic background template matching, combined with a pre-established runway material feature library, performs time-varying clutter suppression on multi-polarization echo data, including:

[0094] B1: Real-time acquisition of environmental data around the radar to construct an initial dynamic background template; the environmental data around the radar includes background echo data when there is no target.

[0095] Furthermore, constructing the initial dynamic background template includes: using statistical methods to construct the initial dynamic background template, for example, using the mean method to calculate the average value of image pixels over a period of time as the background template.

[0096] B2: Using a sliding window approach, statistical analysis is performed on the background echo data within the window, its mean is calculated, and the parameters of the background template are adjusted based on the mean.

[0097] It should be noted that when updating the dynamic background template, the statistical parameters of the background echo data are updated using an exponentially weighted moving average method, giving more weight to recent data to better adapt to environmental changes. The exponentially weighted moving average method is existing technology in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0098] B3: Acquire multipolar echo data and correct the multipolar echo data by combining it with a pre-established runway material feature library;

[0099] Furthermore, the specific steps for B3 include:

[0100] (1) Acquire multipolar echo data;

[0101] (2) Collect material samples in different areas of the runway, such as the runway surface, runway edge, and seams. When sampling, ensure that the samples are representative and cover various materials and conditions of the runway.

[0102] (3) Bring the collected samples to the laboratory and use professional instruments and equipment, such as a vector network analyzer, to analyze the electromagnetic properties of the samples, including dielectric constant and conductivity parameters;

[0103] (4) Organize the data obtained from laboratory analysis, classify them according to different areas and material types of the runway, and use the database management system MySQL to establish a runway material feature library. Store the organized data in the runway material feature library and add corresponding tags to each data record.

[0104] (5) Based on the location information of the radar scan, the collected multi-polarized echo data is correlated with the data in the runway material feature library. For example, if the radar scans any specified location on the runway, the material characteristic data corresponding to that location is searched from the runway material feature library.

[0105] In particular, because the scattering characteristics of runway materials to electromagnetic waves may be different under different polarization modes, it is necessary to ensure that the polarization mode of the echo data matches the polarization mode corresponding to the material characteristic data recorded in the feature library.

[0106] (6) Select a scattering model based on the runway material and radar working principle. For example, for a surface with a certain roughness, such as a runway, use the Kirchhoff approximation model to describe the electromagnetic wave scattering process. The Kirchhoff approximation model is existing technology in this field and is not an inventive solution of this application. It will not be elaborated here.

[0107] (7) Using the parameters in the runway material feature library and the selected scattering model, the multipolar echo data is corrected. The purpose of the correction is to eliminate the influence of the runway material on the echo signal, so that the echo data can better reflect the characteristics of the target itself.

[0108] B4: The error between the corrected echo data and the dynamic background template is calculated point by point using the minimum mean square error criterion. Clutter is canceled according to the magnitude of the error to obtain the multi-polarized echo data after clutter suppression.

[0109] Furthermore, the specific steps for clutter cancellation based on the magnitude of the error include:

[0110] (1) Set the error threshold according to the actual application scenario and requirements;

[0111] (2) Obtain the error results between the corrected echo data and the dynamic background template;

[0112] When the absolute value of the error result is less than the set error threshold, the corresponding point is considered to belong to background clutter, and the corresponding background template value is subtracted from the corrected echo data, that is, clutter cancellation is performed.

[0113] The process of performing time-frequency analysis on the clutter-suppressed multi-polarization echo data to generate a three-dimensional time-frequency domain feature matrix containing amplitude-frequency-polarization characteristics includes:

[0114] C1: The short-time Fourier transform method is used to perform time-frequency analysis on the clutter-suppressed multipolar echo data. During the time-frequency analysis, the clutter-suppressed echo data is segmented according to the preset window function type and window length parameters. The short-time Fourier transform method is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0115] Furthermore, the preset window function type and window length parameters are determined in the following manner:

[0116] (1) Analyze the characteristics of echo signals in a dual-runway environment, including the signal bandwidth, time-varying characteristics, and Doppler frequency shift range;

[0117] (2) Through simulation experiments, the effects of different window function types and window length parameters on time-frequency analysis results were compared. In this invention, the window function types selected are Hanning window, Hamming window, and rectangular window.

[0118] (3) Based on the simulation results, select the window function type and window length parameter that can achieve the optimal time-frequency resolution, wherein the time-frequency resolution is evaluated by the time-frequency aggregation index.

[0119] C2: For each echo data segment, calculate its amplitude value at different frequency points, and combine it with multi-polarization information to obtain the corresponding amplitude-frequency characteristics under different polarization modes;

[0120] Furthermore, the specific steps of C2 include:

[0121] (1) Acquire each echo data segment, and perform noise reduction and normalization on each echo data segment to obtain the preprocessed echo data segment;

[0122] (2) Perform a fast Fourier transform on each preprocessed echo data segment to convert the time domain signal into a frequency domain signal, obtain the spectrum of each data segment, and extract the amplitude spectrum from the spectrum of each preprocessed echo data segment. The fast Fourier transform is a prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0123] (3) Select the frequency point. For each preprocessed echo data segment, record its amplitude value at the selected frequency point. For multi-polarization data, the amplitude value under different polarization modes needs to be recorded separately.

[0124] (4) Organize the amplitude-frequency features under different polarization modes into a feature matrix. The rows of the feature matrix represent different frequency points, the columns represent different polarization modes, and the matrix elements represent the amplitude values ​​under the corresponding frequency points and polarization modes.

[0125] C3: Integrate the amplitude-frequency features under different polarization modes to form a three-dimensional data set containing amplitude-frequency-polarization features;

[0126] C4: Normalizes the three-dimensional data set to generate the final time-frequency domain three-dimensional feature matrix.

[0127] The specific steps of S2 include:

[0128] D1: Obtain the three-dimensional feature matrix in the time-frequency domain, and preprocess the three-dimensional feature matrix in the time-frequency domain by normalizing each element in the matrix;

[0129] D2: Within the entire dual-track domain, a sliding window method is used to calculate the local statistics for each data point in the normalized time-frequency domain three-dimensional feature matrix; the local statistics include the mean and variance.

[0130] D3: Calculate the detection threshold using local statistics based on the preset false alarm probability;

[0131] In calculating the detection threshold, the noise characteristics of the radar system are taken into account, and the false alarm probability is adjusted in an adaptive manner. The detection threshold is dynamically optimized according to the actual detection situation to improve the detection accuracy.

[0132] Furthermore, the specific steps of D3 include:

[0133] (1) Set the false alarm probability, that is, the probability of mistakenly judging the background signal as the target signal;

[0134] (2) Obtain local statistics, and calculate the detection threshold based on the distribution characteristics and false alarm probability of the local statistics. The detection threshold is equal to the sum of the product of the standard deviation of the local statistics and the first component plus the mean. The first component is the inverse cumulative distribution function of the standard normal distribution. The independent variable of the inverse cumulative distribution function of the standard normal distribution is the difference between 1 and the false alarm probability.

[0135] D4: Compare the amplitude value of each data point in the window with the detection threshold. If the amplitude value exceeds the threshold, it is marked as a target point in the radar coordinate system.

[0136] Preprocessed image data is acquired, and a deep learning-based object detection algorithm is used to identify foreign objects. The bounding box coordinates of suspected foreign object regions in the preprocessed image are extracted and marked as suspected foreign object regions in the image coordinate system. At the same time, the confidence score of each suspected foreign object region is recorded.

[0137] The establishment of the multi-source data geographic coordinate mapping model specifically includes:

[0138] E1: Based on the latitude and longitude coordinates and scanning parameters of the radar installation location, establish the transformation relationship between the radar coordinate system and the geographic coordinate system; the transformation relationship is described by a coordinate transformation matrix;

[0139] The installation location of the radar is obtained through the Global Positioning System.

[0140] E2: Based on the camera intrinsic parameter calibration results and extrinsic parameters, establish the mapping relationship between the image coordinate system and the geographic coordinate system;

[0141] E3: Establish a physical coordinate system for the runway with any endpoint of the runway as the origin and the runway direction as the coordinate axis, and construct mapping models for radar-physical coordinates and image-physical coordinates respectively.

[0142] E4: Convert the radar target point and the suspected foreign object area to the runway physical coordinate system and calculate the spatial position deviation value.

[0143] Furthermore, the radar target point is transformed from the radar coordinate system to the runway physical coordinate system, including:

[0144] (1) For each target point marked in the radar coordinate system, obtain its coordinate value;

[0145] (2) Based on the dual-runway geographic coordinate mapping model, the target point coordinates in the radar coordinate system are converted into the geographic coordinate system using the coordinate transformation matrix. Then, based on the mapping relationship between the runway physical coordinate system and the geographic coordinate system, the target point coordinates in the geographic coordinate system are converted into the runway physical coordinate system.

[0146] This involves obtaining the geographic information of the two runways from the airport's geographic information system database, and then cleaning and preprocessing the obtained geographic information to ensure the accuracy and integrity of the data.

[0147] The data registration and cross-validation based on the runway physical coordinate system includes:

[0148] F1: Calculate the Euclidean distance between the radar target point and the centroid of the suspected foreign object area. When the Euclidean distance is less than or equal to a preset threshold, it is determined as a candidate matching target.

[0149] F2: Cluster analysis is performed on candidate matching targets based on prior information of runway pavement partitions to initially distinguish between real and false targets; the prior information of runway pavement partitions includes centerline region, edge region, and taxiway region.

[0150] Furthermore, a preliminary distinction is made between genuine FOD targets and spoof targets, including:

[0151] (1) Read the prior information of runway pavement partitioning, divide the runway into different areas according to the prior information of runway pavement partitioning, including the runway centerline area, runway edge area, and taxiway area, and record the characteristics and functions of each area;

[0152] Among them, runway pavement zoning prior information refers to the knowledge about the division of different areas of the runway pavement and their related characteristics that is obtained in advance based on the runway's design planning, actual use, functional requirements, and historical experience before target detection, status monitoring, and foreign object identification. Using runway pavement zoning prior information, key areas of concern can be quickly located. For example, in target detection, if it is known that the takeoff and landing area is where foreign objects are likely to appear and have the greatest impact on flight safety, then detailed detection of this area can be prioritized, reducing unnecessary calculations and resource waste.

[0153] (2): For each region, different clustering parameters are set; the clustering parameters include the cluster radius and the minimum number of cluster points;

[0154] Furthermore, setting different clustering parameters includes:

[0155] (1) Analyze the historical FOD target distribution data of different runway areas, and statistically analyze the average size, distribution density and other information of real FOD targets in different areas;

[0156] (2) Based on the statistical results and combined with the prior information of the runway pavement zoning, set an appropriate clustering radius and minimum number of clustering points for each region to ensure that the real FOD targets can be accurately clustered, while avoiding misjudging false targets as real targets.

[0157] (3) A density-based clustering algorithm is used to perform cluster analysis on the target points in the physical coordinate system of the runway. The density-based clustering algorithm is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0158] (4) Based on the clustering results and combined with the prior information of the runway pavement partition, determine whether each cluster conforms to the distribution characteristics of the real FOD target. If it does, the target point in the cluster is determined to be the real FOD target; otherwise, it is determined to be a false target.

[0159] F3: Radar target points without image matching are marked as pending verification, and suspected foreign object areas without radar response are marked as low-confidence targets.

[0160] The preliminary matching of candidate targets is confirmed by introducing multi-dimensional features, including:

[0161] G1: At the radar data level, the target polarization scattering matrix features are extracted through multi-polarization processing and compared with a typical FOD polarization feature library to calculate the radar feature matching degree.

[0162] G2: At the image data level, visual features of suspected foreign object areas are extracted and matched with a typical FOD visual feature library to calculate the image feature matching degree.

[0163] G3: Fusion decision: When the radar feature matching degree is greater than or equal to the preset radar feature matching degree threshold and the image feature matching degree is greater than or equal to the preset image feature matching degree threshold, the real FOD target is confirmed.

[0164] The process of confirming target characteristics through multi-polarization processing and micro-motion feature analysis includes:

[0165] H1: For the initially identified real FOD targets, obtain their corresponding multi-polarization echo data, including amplitude and phase information under horizontal and vertical polarization modes;

[0166] H2: Perform polarization decomposition on the multi-polarization echo data, calculate the polarization scattering matrix characteristics of the target, and analyze the polarization characteristics of the target based on the element values ​​of the polarization scattering matrix; the polarization characteristics of the target include polarizability and polarization angle.

[0167] Furthermore, the process of polarization decomposition of the multipolar echo data and calculation of the target's polarization scattering matrix characteristics includes:

[0168] (1) The Pauli decomposition method is used to decompose the multipolar echo data. The Pauli decomposition method is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0169] (2) Based on the decomposition results, calculate the target's characteristic parameters such as polarization amplitude, polarization phase, and polarization correlation coefficient;

[0170] (3) Normalize the calculated feature parameters to eliminate the differences between different polarization channels and improve the comparability of features.

[0171] H3: Compare the polarization characteristics of the target with a pre-established polarization feature library of typical FOD targets and false targets, and determine whether the target is a real FOD target based on the similarity.

[0172] Furthermore, the specific steps of H3 include:

[0173] (1) Obtain echo data of the target under different polarization modes. The echo data of the target under different polarization modes contains polarization characteristic information of the target, such as amplitude and phase information under different polarization channels;

[0174] (2) The original echo data collected is subjected to wavelet denoising processing to remove background noise and interference signals. Wavelet denoising is a prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0175] (3) Extract polarization features from the denoised polarized echo data; the polarization features include polarization scattering matrix, polarizability, and polarization ratio;

[0176] (4) Collect common FOD targets on the runway, such as metal fragments, plastic bottles and stones. Use a multi-polarization radar system to measure these targets, obtain their echo data under different polarization modes, extract polarization features, and construct a polarization feature library of typical FOD targets.

[0177] (5) For each FOD target, record its characteristic parameters such as polarization scattering matrix and polarizability, and classify and store them, for example, classify them according to materials such as metal and non-metal.

[0178] (6) False targets include echoes generated by uneven runway surfaces, vegetation, building edges, etc. The same multipolar radar system is used to measure these false targets, extract polarization features, and construct a polarization feature library of false targets.

[0179] (7) Construct a target feature vector from the polarization features of the target to be detected;

[0180] (8) Select a typical target's polarization feature from the typical FOD target polarization feature library and construct a typical target feature vector;

[0181] (9) Select a polarization feature of a false target from the polarization feature library of false targets and construct a false target feature vector;

[0182] (10) Cosine similarity is used to measure the similarity between the feature vector of the target to be detected and all typical target feature vectors in the polarization feature library of typical FOD targets, and multiple first-class similarity values ​​are obtained. Cosine similarity is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0183] (11) Cosine similarity is used to measure the similarity between the feature vector of the target to be detected and all the feature vectors of false targets in the polarization feature library of false targets, and multiple second-class similarity values ​​are obtained;

[0184] (12) Set a similarity threshold. If the first type of similarity value is greater than the similarity threshold and the second type of similarity value is less than the similarity threshold, then the target is judged to be a real FOD target; otherwise, it is judged to be a false target.

[0185] H4: At the same time, for the initially identified real FOD targets, their echo data are acquired in multiple consecutive scanning cycles to obtain the target's echo data;

[0186] H5: Short-time Fourier transform is used to process the echo data of the target to extract the micro-motion frequency characteristics and micro-motion amplitude characteristics of the target;

[0187] Furthermore, the specific steps of H5 include:

[0188] (1) Perform a short-time Fourier transform on the target echo signal to obtain the time-frequency distribution of the signal;

[0189] (2) In the time-frequency distribution, analyze the frequency changes of the signal and extract the micro-Doppler frequency shift information;

[0190] (3) Calculate the micro-motion frequency and micro-motion amplitude of the target based on the micro-Doppler frequency shift information; the micro-motion frequency is determined by the periodic change of the micro-Doppler frequency shift; the micro-motion amplitude is determined by the difference between the maximum and minimum values ​​of the micro-Doppler frequency shift.

[0191] H6: Determine whether a target is a real FOD target based on the differences between the typical micro-movement patterns of FOD targets and the micro-movement characteristics of false targets.

[0192] The real-time tracking of confirmed real FOD targets based on fused data includes:

[0193] Q1: The Kalman filter algorithm is used to track the confirmed real FOD target in real time, and the state equation and observation equation of the target are established. The state equation describes the change law of the target's position and velocity state over time. The observation equation describes the radar's observation process of the target state. The Kalman filter algorithm is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0194] Q2: In each scanning cycle, based on the target's current state estimate and radar observations, the Kalman filter algorithm is used to update the state and predict the target's state at the next moment.

[0195] Q3: Based on the prediction results, update the target's trajectory information in real time and provide feedback on the tracking results.

[0196] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A fixed dual-runway FOD detection method based on multi-source peripheral scanning, characterized in that, include: S1: Simultaneously acquire multi-polarization echo data of the two runways using a fixed perimeter-scanning radar. Employ an adaptive clutter cancellation method with dynamic background template matching, combined with a pre-established runway material feature library, to suppress time-varying clutter in the multi-polarization echo data. Perform time-frequency analysis on the clutter-suppressed multi-polarization echo data to generate a three-dimensional feature matrix in the time-frequency domain containing amplitude-frequency-polarization features. Simultaneously trigger the runway monitoring camera to acquire image data of the corresponding area to complete image preprocessing. S2: Input the time-frequency domain three-dimensional feature matrix into the constant false alarm rate detection algorithm to initially detect areas with FOD targets in the entire dual-runway area and mark the target points in the radar coordinate system. At the same time, perform target detection on the preprocessed image data and mark the suspected foreign object areas in the image coordinate system. S3: Establish a multi-source data geographic coordinate mapping model, and at the same time, perform data registration and cross-validation based on the runway physical coordinate system; S4: For the initially matched candidate targets, multi-dimensional features are introduced for confirmation, and the confirmed real FOD targets are tracked in real time based on fused data; the fused data is multi-source fusion information formed by registering and associating radar echo data and image data; The adaptive clutter cancellation method employing dynamic background template matching, combined with a pre-established runway material feature library, performs time-varying clutter suppression on multi-polarization echo data, including: Real-time acquisition of environmental data surrounding the radar to construct an initial dynamic background template; the environmental data surrounding the radar includes background echo data when there is no target. A sliding window approach is used to statistically analyze the background echo data within the window, calculate its mean, and adjust the parameters of the background template based on the mean. Acquire multipolar echo data and correct the multipolar echo data by combining it with a pre-established runway material feature library; The error between the corrected echo data and the dynamic background template is calculated point by point using the minimum mean square error criterion. Clutter is then canceled according to the magnitude of the error to obtain multi-polarized echo data after clutter suppression. The process of performing time-frequency analysis on the clutter-suppressed multi-polarization echo data to generate a three-dimensional time-frequency domain feature matrix containing amplitude-frequency-polarization characteristics includes: The short-time Fourier transform method is used to perform time-frequency analysis on the clutter-suppressed multipolar echo data. During the time-frequency analysis, the clutter-suppressed echo data is segmented according to the preset window function type and window length parameters. For each echo data segment, its amplitude value at different frequency points is calculated, and combined with multi-polarization information, the amplitude-frequency characteristics corresponding to different polarization modes are obtained respectively. The amplitude-frequency features under different polarization modes are integrated to form a three-dimensional data set containing amplitude-frequency-polarization features; The three-dimensional data set is normalized to generate the final time-frequency domain three-dimensional feature matrix.

2. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 1, characterized in that, The method of synchronously acquiring multi-polarization echo data of the two runways using a fixed perimeter-scan radar includes: The radar scanning cycle is set to T seconds based on the length, width, and detection accuracy requirements of the two runways. The radar is configured with multiple polarization transmission and reception modes, and uses horizontal polarization and vertical polarization to alternately transmit and receive signals to obtain echo data under different polarization modes; The acquired multipolar echo data undergoes preliminary digital processing to obtain multipolar echo data.

3. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 2, characterized in that, The specific steps of S2 include: Obtain the three-dimensional feature matrix in the time-frequency domain, and preprocess the three-dimensional feature matrix in the time-frequency domain by normalizing each element in the matrix; Within the entire dual-track domain, a sliding window approach is used to calculate the local statistics for each data point in the normalized time-frequency domain three-dimensional feature matrix; the local statistics include the mean and variance. The detection threshold is calculated using local statistics based on the preset false alarm probability. Compare the amplitude value of each data point in the window with the detection threshold value. If the amplitude value exceeds the threshold value, it is marked as a target point in the radar coordinate system. Preprocessed image data is acquired, and a deep learning-based object detection algorithm is used to identify foreign objects. The bounding box coordinates of suspected foreign object regions in the preprocessed image are extracted and marked as suspected foreign object regions in the image coordinate system. At the same time, the confidence score of each suspected foreign object region is recorded.

4. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 3, characterized in that, The establishment of the multi-source data geographic coordinate mapping model specifically includes: Based on the latitude and longitude coordinates and scanning parameters of the radar installation location, a transformation relationship between the radar coordinate system and the geographic coordinate system is established; the transformation relationship is described by a coordinate transformation matrix. Based on the camera intrinsic parameter calibration results and extrinsic parameters, a mapping relationship between the image coordinate system and the geographic coordinate system is established; A physical coordinate system for the runway is established with any endpoint of the runway as the origin and the runway direction as the coordinate axis. Mapping models for radar-physical coordinates and image-physical coordinates are constructed respectively. The radar target point and the suspected foreign object area are converted to the runway physical coordinate system, and the spatial position deviation value is calculated.

5. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 4, characterized in that, The data registration and cross-validation based on the runway physical coordinate system includes: Calculate the Euclidean distance between the radar target point and the centroid of the suspected foreign object area. When the Euclidean distance is less than or equal to a preset threshold, it is determined to be a candidate matching target. Cluster analysis is performed on candidate matching targets based on prior information of runway pavement zoning to initially distinguish between real and false targets; the prior information of runway pavement zoning includes centerline region, edge region, and taxiway region. Radar target points without image matching are marked as pending verification, and suspected foreign object areas without radar response are marked as low-confidence targets.

6. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 5, characterized in that, The preliminary matching of candidate targets is confirmed by introducing multi-dimensional features, including: At the radar data level, the target polarization scattering matrix features are extracted through multi-polarization processing and compared with a typical FOD polarization feature library to calculate the radar feature matching degree. At the image data level, visual features of suspected foreign object areas are extracted and matched with a typical FOD visual feature library to calculate the image feature matching degree. The fusion decision-making process confirms the true FOD target when both the radar feature matching degree and the image feature matching degree are greater than or equal to a preset radar feature matching degree threshold and a preset image feature matching degree threshold.

7. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 6, characterized in that, The process of confirming target characteristics through multi-polarization processing and micro-motion feature analysis includes: For the initially identified real FOD targets, obtain their corresponding multi-polarization echo data, including amplitude and phase information under horizontal and vertical polarization modes; Polarization decomposition is performed on multi-polarization echo data to calculate the polarization scattering matrix characteristics of the target. Based on the element values ​​of the polarization scattering matrix, the polarization characteristics of the target are analyzed. The polarization characteristics of the target include polarizability and polarization angle. The polarization characteristics of the target are compared with the pre-established polarization feature library of typical FOD targets and false targets, and the similarity is used to determine whether the target is a real FOD target. Meanwhile, for the initially identified real FOD targets, their echo data are acquired in multiple consecutive scanning cycles to obtain the target's echo data; Short-time Fourier transform is used to process the target's echo data to extract the target's micro-motion frequency and micro-motion amplitude characteristics; Based on the differences between the typical micro-movement patterns of FOD targets and the micro-movement characteristics of false targets, determine whether a target is a real FOD target.

8. The fixed dual-runway FOD detection method based on multi-source peripheral scanning as described in claim 7, characterized in that, The real-time tracking of confirmed real FOD targets based on fused data includes: The Kalman filter algorithm is used to track confirmed real FOD targets in real time, and the state equation and observation equation of the target are established. The state equation describes the changes of the target's position and velocity state over time. The observation equation describes the radar's observation process of the target's state. Within each scanning cycle, the target's current state estimate and radar observations are used to update the state and predict the target's state at the next moment. Based on the prediction results, the target's trajectory information is updated in real time, and the tracking results are fed back.

Images