Low-altitude radar information processing system based on big data analysis

Abstract

The invention relates to the technical field of data processing, and discloses a low-altitude radar information processing system based on big data analysis. The system comprises a data fusion module, a false alarm suppression module, a measurement target identification module and a grading early warning mechanism establishment module. Firstly, low-altitude radar signals are collected, space-time reference synchronization is carried out, original low-altitude radar data are obtained, and then data association matching is carried out by using a clustering fusion algorithm; secondly, optimizing the threshold value by using a multi-strategy fusion particle swarm algorithm, and performing false alarm suppression; establishing a three-path feature network model based on a multi-modal neural network architecture, carrying out target recognition, and outputting a low-altitude radar measurement target recognition result; and finally, according to a low-altitude radar measurement target identification result, carrying out environment constraint integration, generating a radar target motion track, and establishing a grading early warning mechanism. According to the method, low-altitude radar data are processed and modeled, the purpose of low-altitude radar information processing is achieved, and the method is accurate and objective.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly to a low-altitude radar information processing system based on big data analysis. BACKGROUND

[0002] With the promotion of low-altitude airspace opening policy and the rapid development of low-altitude targets such as unmanned aerial vehicles and navigable aircraft, the demand for low-altitude monitoring in the fields of low-altitude safety control, urban security and traffic management is increasingly urgent. As the core sensor for low-altitude target detection, the information processing capability of low-altitude radar is directly related to the accuracy and timeliness of situation awareness. Currently, low-altitude radar information processing is facing severe challenges: the low-altitude environment is complex and changeable, with a large amount of strong clutter interference such as ground buildings, mountains and weather clutter; the target types are diverse, and have the characteristics of low, small, slow, hovering and high mobility, with a huge difference in RCS (radar cross section). The existing low-altitude radar information processing system still has a large room for improvement in processing efficiency, target recognition accuracy, adaptability to complex environment and depth of multi-source information fusion, and it is urgent to introduce more advanced technical means.

[0003] The core process of the traditional low-altitude radar information processing method mainly includes signal preprocessing to suppress clutter and preliminarily detect targets; point aggregation and correlation to aggregate the original detection points into effective point tracks and correlate them with existing tracks; target tracking and filtering to estimate and predict the target state using Kalman filtering, particle filtering, etc., to form stable tracks; and track management, including track initiation, maintenance, termination and track quality assessment.

[0004] The traditional low-altitude radar information processing method has bottlenecks in processing efficiency when facing massive and rapidly growing low-altitude radar data, and cannot meet the real-time requirements of high data throughput and low processing delay, and cannot effectively cope with the complex and changeable low-altitude environment such as low-altitude building groups and bird groups, and has weak self-adaptability. SUMMARY

[0005] In view of the problems in the related art, the present application provides a low-altitude radar information processing system based on big data analysis to overcome the technical problems existing in the prior art.

[0006] To solve the technical problems, the present application is realized by the following technical scheme:

[0007] The present application is a low-altitude radar information processing system based on big data analysis, specifically comprising a data fusion module, a false alarm suppression module, a measurement target identification module and a hierarchical early warning mechanism establishment module.

[0008] The data fusion module is configured to collect low-altitude radar signals, perform time-space reference synchronization, use a clustering fusion algorithm to perform data correlation matching, generate low-altitude radar measurement target trajectories, and output a low-altitude radar fusion data set.

[0009] The false alarm suppression module is configured to use a multi-strategy fusion particle swarm algorithm to optimize the threshold, suppress false alarms in the low-altitude radar fusion data set, and generate a processed low-altitude radar data set.

[0010] The measurement target identification module is configured to establish a three-path feature network model based on a multi-modal neural network architecture, perform target identification based on the processed low-altitude radar data set, and output a low-altitude radar measurement target identification result.

[0011] The hierarchical early warning mechanism establishment module is configured to integrate environmental constraints based on the low-altitude radar measurement target identification result, calculate a trajectory threat index, and establish a hierarchical early warning mechanism.

[0012] Preferably, the collection of low-altitude radar signals and the time-space reference synchronization include:

[0013] The low-altitude radar is connected to a Beidou time service module, low-altitude radar signals are collected, and are unified to a coordinated universal time reference. The low-altitude radar coordinate system is converted to a geodetic coordinate system to obtain original low-altitude radar data.

[0014] Preferably, the generation of low-altitude radar measurement target trajectories includes:

[0015] The original low-altitude radar data is converted into low-altitude radar data points, the low-altitude radar data points are divided into adjacent data point pairs, the adjacent data point pairs are compared and merged into candidate clusters, and the candidate clusters are associated as continuous trajectories when the spatial distance between the candidate clusters is less than the predicted displacement to obtain a low-altitude radar measurement target trajectory set.

[0016] The candidate cluster positions and candidate cluster speeds are calculated to obtain a predicted displacement, and when the spatial distance between the candidate clusters is less than the predicted displacement, the two candidate clusters are associated as continuous trajectories to obtain a low-altitude radar measurement target trajectory set.

[0017] Preferably, the use of a multi-strategy fusion particle swarm algorithm to optimize the threshold includes:

[0018] A signal strength detection threshold and a clutter correlation distance threshold are set, the false alarm rate, the low-altitude radar measurement target detection probability, and the detection delay are calculated, and an optimization objective function is established based on the minimum error.

[0019] A leader bidirectional search strategy and a follower unidirectional search strategy are introduced to improve the particle swarm algorithm to obtain a multi-strategy fusion particle swarm algorithm.

[0020] The optimization objective function is used as the fitness function, a threshold initial range is set, a search space is generated in the threshold initial range, a particle swarm exists in the search space, each particle position in the particle swarm represents a signal strength detection threshold and a clutter correlation distance threshold, and an updating process, i.e., an optimization process, is performed on the particle swarm;

[0021] The particle swarm is initialized, the leader bidirectional search strategy and the follower unidirectional search strategy are used to update the particle velocity, and the particle position is generated; until the current iteration number reaches the maximum iteration number, the iteration is stopped, the final particle swarm is generated, the particle corresponding to the best fitness function value is found in the final particle swarm, and the optimized signal strength detection threshold and the optimized clutter correlation distance threshold are obtained.

[0022] Preferably, the false alarm suppression on the low-altitude radar fusion data set comprises:

[0023] When the signal strength of the original low-altitude radar data is less than the optimized signal strength detection threshold, the corresponding original low-altitude radar data is deleted; when the distance of the low-altitude radar data point in the low-altitude radar fusion data set is not greater than the optimized clutter correlation distance threshold, the corresponding low-altitude radar data point is deleted, and a processed low-altitude radar data set is generated.

[0024] Preferably, the three-path feature network model is established based on a multi-modal neural network architecture, which comprises:

[0025] According to the processed low-altitude radar data set, spatial features, spectral features, and time sequence features are generated;

[0026] The three-path feature network model comprises a spatial feature channel, a spectral feature channel, and a motion feature channel, the spatial feature channel receives the spatial features, and then processes them through a convolution kernel, a maximum pooling, and a convolution kernel to output a spatial feature vector; the spectral feature channel receives the spectral features, and then processes them through a bidirectional long short-term memory neural network to output a spectral feature vector; the motion feature channel receives the time sequence features, and then processes them through a full connection layer and an attention layer to output a motion feature vector; the spatial feature vector, the spectral feature vector, and the motion feature vector are dimensionally concatenated, and then dimensionally reduced through a full connection layer to output a model recognition result, thereby establishing the three-path feature network model.

[0027] Preferably, the output low-altitude radar measurement target recognition result comprises:

[0028] The original low-altitude radar data is re-acquired, the low-altitude radar measurement target is labeled, and low-altitude radar data clutter is added to obtain a target recognition sample set; after the three-path feature network model is trained, a final three-path feature network model is obtained.

[0029] The spatial feature, the spectral feature and the time sequence feature are input into a final three-path feature network model, and a low-altitude radar measurement target recognition result is output.

[0030] Preferably, the environment constraint integration according to the low-altitude radar measurement target recognition result comprises:

[0031] According to the low-altitude radar measurement target recognition result, a low-altitude radar measurement target is labeled, an environment constraint database is set, and environment constraint integration is performed according to the environment constraint database to obtain a radar target motion trajectory.

[0032] Preferably, the establishment of the hierarchical early warning mechanism comprises:

[0033] The deviation of the radar target motion trajectory and the environment constraint database is calculated to obtain an approach distance, a speed deviation and a height deviation, a trajectory threat index is calculated, a threat level is set, the trajectory threat index is corresponded to the threat level, and a real-time response measure is set to establish the hierarchical early warning mechanism.

[0034] The present application has the following advantages:

[0035] 1. The present application solves the problem of inconsistent coordinates and time by collecting low-altitude radar signals and performing time and space reference synchronization, eliminates time and space reference errors, and then uses a clustering fusion algorithm for data association matching to generate a low-altitude radar measurement target trajectory, output a low-altitude radar fusion data set, effectively processes data association, reduces data processing complexity, reduces data processing delay, and accurately separates real moving targets.

[0036] 2. The present application uses a multi-strategy fusion particle swarm algorithm to optimize the signal strength detection threshold and the clutter association distance threshold, overcomes the problems of insufficient information utilization, low search efficiency and easy premature convergence of traditional algorithms, has high search quality, dynamically optimizes the results obtained by the double thresholds to approach the best threshold, and then performs false alarm suppression to achieve accurate target separation in a complex low-altitude environment, can effectively eliminate weak signal interference and adjacent trajectory clutter, and is suitable for low-altitude complex and variable environments.

[0037] 3. The present application establishes a three-path feature network model based on a multi-modal neural network architecture, fuses multi-modal feature deep fusion and lightweight network, uses three-path feature complementation: spatial positioning + motion mode + behavior feature, covers target physical properties, motion mode and behavior intention, solves the misjudgment problem of low-altitude similar targets, and has small model space occupation, fast processing speed, can reduce cloud transmission bandwidth pressure, is suitable for border defense, mountainous area scenes without network coverage, and has strong adaptive ability.

[0038] 4. The application integrates the environment-constrained results of low-altitude radar target identification, solves the trajectory distortion problem caused by radar signal attenuation and offset in complex urban environments, generates radar target motion trajectories, establishes a hierarchical warning mechanism, quantifies threat indexes by introducing a weighted model, avoids excessive prevention or missed risk, and realizes low-altitude radar information processing.

[0039] Of course, implementing any product of the present application does not necessarily require all the advantages described above to be achieved at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0041] Figure 1 The flowchart of the low-altitude radar information processing system based on big data analysis is provided for the application;

[0042] Figure 2 The flowchart of the low-altitude radar information processing method based on big data analysis is provided for the application;

[0043] Figure 3 The flowchart of the threshold optimization process provided by the application is provided. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the application will be described clearly and completely in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0045] The traditional low-altitude radar information processing method has a bottleneck in processing efficiency when facing massive and rapidly growing low-altitude radar data, and cannot meet the real-time requirements of high data throughput and low processing delay, and cannot effectively cope with low-altitude complex and variable environments such as low-altitude building groups and bird groups, and has weak adaptive ability.

[0046] To solve the above technical problems, the application provides a low-altitude radar information processing system based on big data analysis, which comprises a data preprocessing module, a feature extraction module, a target identification module, a trajectory generation module, a threat index calculation module, a threat index quantification module, a threat index weighting module, a threat index classification module, a threat index alarm module, and a threat index output module. Figure 1As shown, the embodiment of the present application provides a low-altitude radar information processing system based on big data analysis, which specifically comprises: a data fusion module, a false alarm suppression module, a measured target identification module and a hierarchical early warning mechanism establishment module; the data fusion module is used for collecting low-altitude radar signals and performing time-space reference synchronization, then using a clustering fusion algorithm for data correlation matching to generate low-altitude radar measured target trajectory and output low-altitude radar fusion data set; the false alarm suppression module is used for using a multi-strategy fusion particle swarm algorithm to optimize threshold, then performing false alarm suppression on the low-altitude radar fusion data set to generate processed low-altitude radar data set; the measured target identification module is used for establishing a three-way feature network model based on a multi-modal neural network architecture, performing target identification according to the processed low-altitude radar data set, and outputting low-altitude radar measured target identification result; the hierarchical early warning mechanism establishment module is used for performing environment constraint integration according to the low-altitude radar measured target identification result, calculating trajectory threat index, and establishing a hierarchical early warning mechanism.

[0047] Specific embodiments are that a low-altitude radar system is deployed in a certain suburban area of a city to monitor unmanned aerial vehicle activities, the radar is located at a fixed site (longitude: 118.50°E, latitude: 32.05°N, height: 50 meters); the radar coverage radius is 5 kilometers, mainly detects low-altitude targets (such as unmanned aerial vehicles), the flight height is between 100-500 meters, several unmanned aerial vehicle targets fly in the radar coverage area, the trajectory is roughly a curve motion (from west to east), the radar system has accessed a Beidou timing module, providing a good scene for the embodiment of the present application.

[0048] In the implementation of the above embodiment, first, the low-altitude radar signal is collected and time-space reference synchronization is performed, then a clustering fusion algorithm is used for data association matching to generate a low-altitude radar measurement target trajectory and output a low-altitude radar fusion data set; this method solves the problem of inconsistent coordinates and time by time-space reference synchronization, eliminates time-space reference errors, and effectively processes data association (reduces false positives and fragmented trajectories) by using a clustering fusion algorithm, thereby reducing data processing complexity, reducing data processing delay, and accurately separating real moving targets; second, an optimization objective function is established based on minimum error, a multi-strategy fusion particle swarm algorithm is used to optimize the signal strength detection threshold and clutter association distance threshold, and then false alarm suppression is performed; in this process, the multi-strategy fusion particle swarm algorithm overcomes the problems of insufficient information utilization, low search efficiency, and easy premature convergence of traditional algorithms, has high search quality, and dynamically optimizes the double thresholds to obtain results close to the optimal thresholds, thereby achieving accurate target separation in complex low-altitude environments (bird flocks, atmospheric clutter), effectively removing weak signal interference and adjacent trajectory clutter, and being suitable for low-altitude complex and variable environments; third, a three-path feature network model is established based on a multi-modal neural network architecture, target recognition is performed after training, and low-altitude radar measurement target recognition results are output; this model fuses multi-modal feature deep fusion and lightweight network, uses three-path complementary features: spatial positioning + motion pattern + behavior features, covers target physical properties, motion patterns, and behavior intentions, solves the misjudgment problem of low-altitude similar targets, has small model space occupation and fast processing speed, can reduce cloud transmission bandwidth pressure, is suitable for border defense and mountainous areas without network coverage, and has strong adaptive ability; finally, the low-altitude radar measurement target recognition results are integrated according to environmental constraints to generate a radar target motion trajectory and establish a hierarchical early warning mechanism; this method considers environmental constraints, solves the problem of trajectory distortion caused by radar signal attenuation and deviation in complex urban environments, introduces a weighted model to quantify threat indexes, avoids excessive prevention or missed judgment risks, and realizes low-altitude radar information processing.

[0049] Further, in order to better introduce the technical scheme of the embodiment of the present application, as shown in Figure 2 , a low-altitude radar information processing system based on big data analysis is described in detail, which specifically includes the following contents:

[0050] S1, collect low-altitude radar signals and perform time-space reference synchronization to obtain original low-altitude radar data, then use a clustering fusion algorithm for data association matching to generate a low-altitude radar measurement target trajectory and output a low-altitude radar fusion data set;

[0051] The S1 includes the following steps:

[0052] S11, access the low-altitude radar to the Beidou timing module, collect the low-altitude radar signal, compress the radar pulse width, eliminate static object clutter through the moving target display filter, and unify the low-altitude radar signal to the coordinated universal time reference; obtain the low-altitude radar longitude and latitude coordinates and the low-altitude radar measured target azimuth, calculate the sum of the low-altitude radar longitude and latitude coordinates and the low-altitude radar measured target azimuth, convert the low-altitude radar coordinate system into the geodetic coordinate system, complete the time and space reference synchronization, and obtain the original low-altitude radar data;

[0053] S12, use a clustering fusion algorithm to perform data correlation matching on the original low-altitude radar data to generate a low-altitude radar measured target trajectory, and the specific steps are as follows:

[0054] S121, convert the original low-altitude radar data into low-altitude radar data points, the low-altitude radar data points include the longitude, latitude, height, speed, time stamp, signal strength, etc. of the low-altitude radar measured target, and obtain a low-altitude radar data point set; traverse the low-altitude radar data point set, set a spatial distance threshold and a speed difference threshold, when the spatial distance between the low-altitude radar data points is less than the spatial distance threshold and the speed difference is less than the speed difference threshold, mark the corresponding two low-altitude radar data points as adjacent data point pairs; then calculate the weighted average value of the positions of the low-altitude radar data points in the adjacent data point pairs and the average value of the speeds, as the position of the adjacent data point pair and the speed of the adjacent data point pair, compare the adjacent data point pairs and merge them into candidate clusters;

[0055] S122, set a time window, the time window is the maximum difference of the time stamps of the adjacent data points in the candidate cluster, select a time window k, judge the continuity of the time window k and the time window k+1; calculate the weighted average value of the positions of the adjacent data point pairs in the candidate cluster and the average value of the speeds, to obtain the position of the candidate cluster and the speed of the candidate cluster; calculate the product of the speed of the candidate cluster and the size of the time window, to obtain the predicted displacement; when the spatial distance between the candidate clusters is less than the predicted displacement, associate the two candidate clusters as a continuous trajectory, that is, merge the time window k and the time window k+1; merge the time windows in turn to obtain a low-altitude radar measured target trajectory set;

[0056] S13, according to the low-altitude radar measured target trajectory set, calculate the longitude average value, latitude average value, maximum height value and speed average value of the low-altitude radar measured target in the low-altitude radar measured target trajectory set, and mark the time window end moment, as the longitude, latitude, height, speed and time stamp of the low-altitude radar measured target trajectory respectively, to obtain a low-altitude radar fusion data set;

[0057] In this embodiment, low-altitude radar signals are collected and time-space reference synchronization is performed, then a clustering fusion algorithm is used for data association matching to generate a low-altitude radar measurement target trajectory and output a low-altitude radar fusion data set; this method solves the problem of inconsistent coordinates and time through time-space reference synchronization, eliminates time-space reference errors, and effectively processes data association (reduces false positives and fragmented trajectories) through the clustering fusion algorithm, reducing data processing complexity and delay and accurately separating real moving targets; specifically, for example, starting at UTC time 2024-10-01 10:00:00, low-altitude radar signals are collected, static object clutter is eliminated through a moving target indication filter, time is synchronized to the UTC reference, and all timestamps are ensured to be accurate to the millisecond level; continuous measurement (time interval of about 0.5 seconds) of a UAV yields original low-altitude radar data with timestamps of 2024-10-01 10:00:00.000 to 2024-10-01 10:00:10.000; comparing the data points with timestamps of 10:00:00.000 (longitude 118.500, latitude 32.050, height 100 meters, speed 8 m / s, signal strength 70 dB) and 10:00:00.500 (longitude 118.502, latitude 32.052, height 110 meters, speed 9 m / s, signal strength 75 dB), through statistical analysis of previous year's original low-altitude radar data, it is found that when the distance is less than 150 meters and the speed is less than 5 m / s, the data points are considered to be continuous points in space, and a spatial distance threshold of 150 meters and a speed difference threshold of 5 m / s are set, and the two data points are recorded as adjacent data point pairs; the weighted average of the adjacent data point pairs is calculated based on the signal strength: longitude = (118.500 x 70 + 118.502 x 75) / (70 + 75) ≈ 118.501 E, latitude = (32.050 x 70 + 32.052 x 75) / (70 + 75) ≈ 32.051 N, average speed = (8 + 9) / 2 = 8.5 m / s, forming candidate cluster A, and candidate clusters B and C are repeatedly obtained; a time window size of 5 seconds is set, and the average speed of candidate cluster A (8.5 m / s) multiplied by the time window size yields a predicted displacement of 42.5 meters, the distance between clusters A and B is less than the predicted displacement, forming a partial trajectory, which is then associated with candidate cluster C to form a continuous trajectory, i.e., a low-altitude radar measurement target trajectory, and the statistical characteristics of each trajectory are calculated to obtain a low-altitude radar fusion data set;

[0058] S2, based on the minimum error, an optimization objective function is established, a multi-strategy fusion particle swarm algorithm is used to optimize the signal strength detection threshold and the clutter association distance threshold, and then false alarm suppression is performed on the low-altitude radar fusion data set to generate a processed low-altitude radar data set;

[0059] The S2 includes the following steps:

[0060] S21, divide the airspace monitored by the low-altitude radar into three-dimensional grid units, mark the low-altitude radar fusion data set in the three-dimensional grid units, and generate a low-altitude radar measured target trajectory; set a signal strength detection threshold and a clutter correlation distance threshold, and when the signal strength of the original low-altitude radar data is not greater than the signal strength detection threshold, mark the corresponding original low-altitude radar data as original low-altitude radar data clutter;

[0061] Further calculate the distance between the set of low-altitude radar measured target trajectories and the original low-altitude radar data clutter, denoted as a clutter correlation distance, and when the clutter correlation distance is not greater than the clutter correlation distance threshold, the original low-altitude radar data clutter is attributed to the low-altitude radar measured target trajectory; at this time, calculate the proportion of original low-altitude radar data clutter identified as a low-altitude radar measured target, i.e. the false alarm rate, calculate the low-altitude radar measured target detection probability, and record the detection delay, and establish an optimization objective function based on the minimum error wherein α, β and δ represent weights, A represents the false alarm rate, B represents the low-altitude radar measured target detection probability, c represents the detection delay, and C represents the maximum detection delay;

[0062] S22, introduce a leader bidirectional search strategy and a follower unidirectional search strategy to improve the particle swarm algorithm, obtain a multi-strategy fusion particle swarm algorithm, and further optimize the signal strength detection threshold and the clutter correlation distance threshold to obtain an optimized signal strength detection threshold and an optimized clutter correlation distance threshold, as shown in the threshold optimization process, and the specific steps are as follows: Figure 3

[0063] S221, take the optimization objective function as a fitness function, set a threshold initial range, generate a search space in the threshold initial range, and there are particle swarms in the search space, the number of particle swarms is p, the dimension of particle swarms is q, and each particle position in the particle swarms represents the signal strength detection threshold and the clutter correlation distance threshold, and the particle swarms are updated, i.e. the optimization process;

[0064] S222, initialize the particle swarms to obtain particle initial velocities and particle initial positions, use the fitness function value to measure the pros and cons of the particle positions, the best fitness function value corresponds to the best particle position; set the current iteration number as t, the maximum iteration number as T, the i-th particle position at the t-th iteration as W i (t), the i-th particle velocity at the t-th iteration as V i (t), the maximum inertia weight as ε, r1 and r2 represent random numbers between the interval [0, 1], and the best particle position at the t-th iteration is denoted as W″ best (t), the best particle position before the t-th iteration is denoted as W′ best (t);

[0065] ​Set d1 represents the i-th particle to the t-th iteration before the best particle position step, d2 represents the i-th particle to the t-th iteration when the best particle position step, using the leader bidirectional search strategy to update the particle velocity, get the t+1 iteration i-th particle velocity V i (t+1) = ε·V i (t) + r1·d1(W best (t) - W i (t)) + r2·d2(W best (t) - W i (t));

[0066] S223, using the follower unidirectional search strategy to update the particle velocity again, calculate the current fitness function value, select the top K particles in the current fitness function value and calculate the weighted average value, get the group guide coefficient, set the attenuation coefficient representing a constant between the interval [0, 1], the i-th particle velocity at the t+1 iteration Wherein The group guide coefficient is calculated; at this time, the i-th particle position W i (t+1) = W i (t) + V i (t+1);

[0067] Complete this iteration, enter the next iteration, until the current iteration reaches the maximum iteration, stop iteration, generate the final particle swarm, find the particle corresponding to the best fitness function value in the final particle swarm, get the optimized signal strength detection threshold and the optimized clutter correlation distance threshold;

[0068] S23, when the signal strength of the original low altitude radar data is less than the optimized signal strength detection threshold, delete the corresponding original low altitude radar data, update the low altitude radar fusion data set; when the distance of the low altitude radar data point in the low altitude radar fusion data set is not greater than the optimized clutter correlation distance threshold, delete the corresponding low altitude radar data point, update the low altitude radar fusion data set again, generate the processed low altitude radar data set;

[0069] In this embodiment, an optimization objective function is established based on minimum error, a multi-strategy fusion particle swarm algorithm is used to optimize the signal strength detection threshold and clutter correlation distance threshold, and then false alarm suppression is performed; in this process, the multi-strategy fusion particle swarm algorithm overcomes the problems of insufficient information utilization, low search efficiency and easy premature convergence of traditional algorithms, has high search quality, and dynamically optimizes the double thresholds to obtain results close to the optimal thresholds, realizes accurate target separation in a complex low-altitude environment (bird flock, atmospheric clutter), can effectively remove weak signal interference and adjacent track clutter, and is suitable for a low-altitude complex and variable environment; specifically, for example, the monitored airspace is divided into 100m*100m*50m three-dimensional grid units, bird flock data and atmospheric clutter are marked, a low-altitude radar fusion data set includes 200 points (including 180 clutter points + 20 real target points), 22 clutter points (10 bird flock points + 12 atmospheric clutter points) are wrongly classified into tracks, at this time, the false alarm rate is 22 / 200 = 11%; 18 target points are correctly identified, the detection probability is 18 / 20 = 90%, the detection delay is 0.4 seconds, the maximum allowed delay is set to 1.0 second; the influence of the false alarm rate, the target detection probability and the detection delay on the final detection result is determined based on expert experience, an initial weight group is obtained, 10,000 real UAV intrusion events (including false positives) are obtained, and the probability of successful interception of the UAV is calculated according to the initial weight group The initial weight corresponding to the highest interception probability is obtained, which is recorded as the weight in the optimization objective function, the weights are set to 0.6, 0.3 and 0.1, and the initial fitness function value is calculated to be 0.104; the fitness function value is minimized by the target through the multi-strategy fusion particle swarm algorithm; the number of particles is set to 50, the dimension is 2 (corresponding to two thresholds), the signal strength is [50, 80] dB, the clutter distance is [100, 500] meters, the leader bidirectional search strategy and the follower unidirectional search strategy are used for particle position updating, and when the maximum iteration number is reached, the optimal particle position [68.5, 175.3] is obtained, and the minimum fitness is 0.062 (40% lower than 0.104); at this time, all original data ≤ 68.5 dB are deleted, and the clutter points within 175.3 meters of the track are deleted, a total of 3 false alarm points (originally 22) are left, the target points are retained: 19 (originally 20), the detection delay is: 0.2 seconds (originally 0.4 seconds), and the processed low-altitude radar data set (118.504°E, 32.054°N, 140 meters, 10.2 m / s) is output;

[0070] S3, a three-path feature network model is established based on a multi-modal neural network architecture, after training, target recognition is performed according to the processed low-altitude radar data set, and a low-altitude radar measurement target recognition result is output;

[0071] The S3 includes the following steps:

[0072] S31, convert the position of the low-altitude radar measurement target trajectory into grid coordinates according to the processed low-altitude radar data set, map the signal strength of the original low-altitude radar data to color depth to obtain spatial features; then construct a Doppler velocity spectrum graph, which includes the direction angle and velocity of the low-altitude radar measurement target trajectory, map the number of lattices in the Doppler velocity spectrum graph to image color depth to obtain spectral features; generate time sequence features according to the time stamp of the processed low-altitude radar data set, which includes position offset, height change and velocity change;

[0073] S32, establish a three-path feature network model based on a multi-modal neural network architecture, and obtain a final three-path feature network model after training, the specific steps are as follows:

[0074] S321, set the three-path feature network model to include a spatial feature channel, a spectral feature channel and a motion feature channel, after the spatial feature channel receives the spatial features, it is processed through a convolution kernel, a maximum pooling and a convolution kernel respectively, and outputs a spatial feature vector; after the spectral feature channel receives the spectral features, it outputs a spectral feature vector through a bidirectional long short-term memory neural network; after the motion feature channel receives the time sequence features, it outputs a motion feature vector through a fully connected layer and an attention layer respectively; the spatial feature vector, the spectral feature vector and the motion feature vector are dimensionally concatenated, and then dimensionally reduced through a fully connected layer to output a model recognition result, and a three-path feature network model is established;

[0075] S322, reacquire the original low-altitude radar data, label the low-altitude radar measurement target, add low-altitude radar data clutter, and extract spatial features, spectral features and time sequence features to obtain a target recognition sample set, and divide the target recognition sample set into a sample training set and a sample test set;

[0076] input the sample training set into the three-path feature network model for three-path joint training, set the maximum number of training rounds, stop training when the maximum number of training rounds is reached, and obtain a trained three-path feature network model; then input the sample test set into the trained three-path feature network model, and use the early stopping mechanism to obtain the final three-path feature network model when the output result accuracy does not improve for 10 consecutive rounds, otherwise adjust the weight until the output result accuracy does not improve for 10 consecutive rounds;

[0077] S33, input the spatial features, spectral features and time sequence features into the final three-path feature network model to output a low-altitude radar measurement target recognition result;

[0078] In this embodiment, a three-path feature network model is established based on a multi-modal neural network architecture, target recognition is performed after training, and a low-altitude radar measurement target recognition result is output; the model fuses multi-modal feature deep fusion and lightweight network, uses three-path feature complementation: spatial positioning + motion mode + behavior feature, covers target physical properties, motion mode and behavior intention, solves the misjudgment problem of low-altitude similar targets, and has small model space occupation, fast processing speed, can reduce cloud transmission bandwidth pressure, is suitable for network coverage-free border defense and mountainous area scenes, and has strong adaptive ability; specifically, for example, spatial feature generation: convert the target position (118.504°E, 32.054°N) into a 100m*100m grid coordinate, and linearly map the original intensity 70-90dB to the gray value 150-255; spectral feature generation: Doppler velocity spectrum graph (velocity range: -20 to 20m / s, divided into 40 intervals), target velocity 10.2m / s→falls into the 24th velocity interval (corresponding to 8-10.5m / s), grid number mapping: target interval count 5; time sequence feature generation: position offset: eastward displacement 0.008° (about 800 meters), height change: 100 meters→140 meters (climbing rate 20m / s), velocity change: 8→12m / s (acceleration 1m / s 2 ); three-path feature network model: spatial feature channel adopts 3*3 convolution kernel (extracts basic texture feature)→outputs 32 feature maps, 2*2 maximum pooling (compresses dimension), 5*5 convolution kernel (recognizes aggregation mode); the spectral feature channel adopts a bidirectional LSTM unit to capture spectral time sequence dependence, the motion feature channel adopts a fully connected layer to extract velocity / height mutation features, an attention layer: assigns high weight to abnormal motion, and splices the three-path feature vectors and then reduces them to 128-dimensional fusion features through a fully connected layer, and outputs the model recognition result; 1000 samples (including rotary-wing / fixed-wing unmanned aerial vehicles, birds, kites) are used, 5% noise is added, 800 samples (80%) are used as a sample training set, and 200 samples (20%) are used as a sample test set, and a final three-path feature network model is trained; input spatial features, spectral features and time sequence features, output rotary-wing unmanned aerial vehicle probability: 97.3%, fixed-wing unmanned aerial vehicle probability: 2.1%, bird probability: 0.6%, and a low-altitude radar measurement target recognition result is obtained as a rotary-wing unmanned aerial vehicle;

[0079] S4, according to the low-altitude radar measurement target recognition result, environment constraint integration is performed, a radar target motion trajectory is generated, a trajectory threat index is calculated, and a hierarchical early warning mechanism is established;

[0080] The S4 includes the following steps:

[0081] S41, according to the low-altitude radar measurement target recognition result, the low-altitude radar measurement target is marked, the final low-altitude radar data set is obtained in combination with the processed low-altitude radar data set, an environment constraint database is set, the environment constraint database includes the no-fly zone boundary, the terrain shielding area, the real-time weather data and the like, the environment constraint integration is carried out according to the environment constraint database, and the radar target motion trajectory is obtained;

[0082] S42, the deviation of the radar target motion trajectory and the environment constraint database is calculated, the approach distance, the speed deviation and the height deviation are obtained, the approach distance, the speed deviation and the height deviation are respectively given threat weights, the trajectory threat index is calculated, the threat level one, the threat level two and the threat level three are set, the trajectory threat index is corresponded to the threat level, the real-time response measures are set, and the hierarchical early warning mechanism is established;

[0083] In the embodiment, the environment constraint integration is carried out according to the low-altitude radar measurement target recognition result, the radar target motion trajectory is generated, and the hierarchical early warning mechanism is established; the method considers the environment constraint, solves the trajectory distortion problem caused by the radar signal attenuation and deviation in the complex urban environment, introduces the weighted model to quantify the threat index, and avoids the excessive prevention and control or the risk of missing judgment; specifically, for example, the current trajectory point: longitude 118.504°E, latitude 32.054°N, height 140 meters, speed 12 m / s, continuously approaching the government building (distance shortening rate 20 m / s); the unmanned aerial vehicle control airspace is inquired from the official airspace management platform of the Civil Aviation Administration of China, the no-fly zone boundary (the government building as the center with a radius of 500 meters circular area) and the terrain shielding area (east high-rise building group) are set, the weather data (northwest wind, wind speed 8 m / s, visibility 5 kilometers) is collected, the shortest distance from the no-fly zone boundary is 150 meters, the speed limit is 10 m / s, and the safe height is 120 meters; at this time, the trajectory threat index is calculated, 10000 real unmanned aerial vehicle intrusion events (including false alarms) are obtained, the reasons are that the approach to the no-fly zone accounts for about 50%, the high-speed penetration accounts for about 30%, and the ultra-low altitude flight accounts for about 20%, so the threat weights of the approach distance, the speed deviation and the height deviation are set to 0.5, 0.3 and 0.2, and the trajectory threat index is calculated The hierarchical early warning mechanism is established, the index range is set: threat level one (>1.5), threat level two (0.5-1.5) and threat level three (<0.5), and the low-altitude radar information processing is realized.

[0084] In the description of the specification, reference to "one embodiment", "an example", "a specific example" or the like means that a particular feature, structure, material or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. The appearances of the phrases "in one embodiment", "an example", "a specific example" or the like in various places in the specification are not necessarily referring to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0085] The preferred embodiments of the application disclosed above are only to help explain the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of the specification. The specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and use the application.

Claims

1. A low-altitude radar information processing system based on big data analysis, characterized in that, include: The data fusion module is used to collect low-altitude radar signals and synchronize them with spatiotemporal references. Then, it uses a clustering fusion algorithm to perform data association and matching, generate the trajectory of the low-altitude radar measured target, and output the low-altitude radar fusion data set. The false alarm suppression module is used to optimize the threshold using a multi-strategy fusion particle swarm algorithm, and then to suppress false alarms on the low-altitude radar fusion dataset to generate a processed low-altitude radar dataset. The target recognition module is used to establish a three-way feature network model based on a multimodal neural network architecture, perform target recognition based on the processed low-altitude radar data set, and output the low-altitude radar target recognition results. The tiered early warning mechanism module is used to integrate environmental constraints based on the target identification results measured by low-altitude radar, calculate the trajectory threat index, and establish a tiered early warning mechanism.

2. The low altitude radar information processing system based on big data analysis as claimed in claim 1, wherein, The process of acquiring low-altitude radar signals and synchronizing spatiotemporal references includes: The low-altitude radar is connected to the BeiDou timing module to collect low-altitude radar signals and unify them to the Coordinated Universal Time (UTC) reference. The radar coordinate system is then transformed into the geodetic coordinate system to obtain the raw low-altitude radar data.

3. The low altitude radar information processing system based on big data analysis as claimed in claim 2, wherein, The generation of the low-altitude radar-measured target trajectory includes: The original low-altitude radar data is converted into low-altitude radar data points, and the low-altitude radar data points are divided into neighboring data point pairs. The neighboring data point pairs are compared and merged into candidate clusters. The positions and velocities of candidate clusters are calculated to obtain the predicted displacement. When the spatial distance between candidate clusters is less than the predicted displacement, the two candidate clusters are associated as a continuous trajectory to obtain a set of low-altitude radar measurement target trajectories.

4. The low altitude radar information processing system based on big data analytics as claimed in claim 3, wherein, The optimization of the threshold using a multi-strategy fusion particle swarm optimization algorithm includes: Set signal strength detection threshold and clutter correlation distance threshold, calculate false alarm rate and low-altitude radar target detection probability, record detection delay, and establish an optimization objective function based on minimum error; The particle swarm optimization algorithm is improved by introducing a bidirectional search strategy for leaders and a unidirectional search strategy for followers, resulting in a multi-strategy fusion particle swarm optimization algorithm. The optimization objective function is used as the fitness function. An initial range of thresholds is set, and a search space is generated within the initial range of thresholds. A particle swarm exists in the search space. The position of each particle in the particle swarm represents the signal strength detection threshold and the clutter correlation distance threshold. The process of updating the particle swarm is the optimization process. Initialize the particle swarm, update the particle velocity using a leader bidirectional search strategy and a follower unidirectional search strategy, and generate particle positions; stop iterating when the current iteration count reaches the maximum iteration count, generate the final particle swarm, find the particle corresponding to the best fitness function value in the final particle swarm, and obtain the optimized signal strength detection threshold and the optimized clutter correlation distance threshold.

5. The low altitude radar information processing system based on big data analytics as claimed in claim 4, wherein, The false alarm suppression of the low-altitude radar fusion data set includes: When the signal strength of the original low-altitude radar data is less than the optimized signal strength detection threshold, the corresponding original low-altitude radar data is deleted; when the distance between low-altitude radar data points in the low-altitude radar fusion data set is not greater than the optimized clutter association distance threshold, the corresponding low-altitude radar data points are deleted, and a processed low-altitude radar data set is generated.

6. The low altitude radar information processing system based on big data analysis as claimed in claim 5, wherein, The three-path feature network model is established based on a multi-modal neural network architecture, and the three-path feature network model includes: According to the processed low-altitude radar data set, spatial features, spectral features and time sequence features are generated; The three-path feature network model includes a spatial feature channel, a spectral feature channel and a motion feature channel, the spatial feature channel receives spatial features, and outputs a spatial feature vector after being processed by a convolution kernel, a maximum pooling and a convolution kernel, respectively; the spectral feature channel receives spectral features, and outputs a spectral feature vector after being processed by a bidirectional long short-term memory neural network; the motion feature channel receives time sequence features, and outputs a motion feature vector after being processed by a full connection layer and an attention layer, respectively; the spatial feature vector, the spectral feature vector and the motion feature vector are dimensionally concatenated, and then are dimensionally reduced by a full connection layer to output a model recognition result, thereby establishing the three-path feature network model.

7. The low altitude radar information processing system based on big data analytics as claimed in claim 6, wherein, The output low-altitude radar measurement target recognition result includes: The original low-altitude radar data is reacquired, the low-altitude radar measurement target is labeled, and low-altitude radar data clutter is added to obtain a target recognition sample set; after training, the three-path feature network model is obtained as a final three-path feature network model; The spatial features, the spectral features and the time sequence features are input into the final three-path feature network model to output the low-altitude radar measurement target recognition result.

8. The low altitude radar information processing system based on big data analysis as claimed in claim 7, wherein, According to the low-altitude radar measurement target recognition result, the environment constraint integration includes: According to the low-altitude radar measurement target recognition result, the low-altitude radar measurement target is labeled, an environment constraint database is set, and the environment constraint integration is performed according to the environment constraint database to obtain a radar target motion trajectory.

9. The low altitude radar information processing system based on big data analytics as claimed in claim 8, wherein, The establishment of the hierarchical early warning mechanism includes: The deviation of the radar target motion trajectory and the environment constraint database is calculated to obtain an approach distance, a speed deviation and a height deviation, and a trajectory threat index is calculated; a threat level is set, the trajectory threat index is corresponded to the threat level, and a real-time response measure is set to establish the hierarchical early warning mechanism.

10. A low-altitude radar information processing method based on big data analysis, characterized in that, Specifically, it includes: S1, collecting low-altitude radar signals and performing time-space reference synchronization to obtain original low-altitude radar data, and then using a clustering fusion algorithm to perform data association matching to generate a low-altitude radar measurement target trajectory, and outputting a low-altitude radar fusion data set; S2, establishing an optimization objective function based on a minimum error, using a multi-strategy fusion particle swarm algorithm to optimize a signal strength detection threshold and a clutter association distance threshold, and then performing false alarm suppression on the low-altitude radar fusion data set to generate a processed low-altitude radar data set; S3, establishing a three-path feature network model based on a multi-modal neural network architecture, and after training, performing target recognition according to the processed low-altitude radar data set to output a low-altitude radar measurement target recognition result; S4, performing environment constraint integration according to the low-altitude radar measurement target recognition result to generate a radar target motion trajectory, calculating a trajectory threat index, and establishing a hierarchical early warning mechanism.

Images