Radar data retrieval feature matching recognition method and system

Abstract

The application discloses a radar data retrieval feature matching identification method and system, relates to the technical field of radar data, and comprises the following steps: obtaining radar echo data to be identified, and preprocessing the radar echo data; target features are extracted from the preprocessed radar echo data; based on a local sensitive hash index, the target features are subjected to similarity matching with standard features in a pre-constructed radar feature database; and according to a matching result, target or scene identification information corresponding to the standard features with the highest similarity to the target features is output. The technical problems of insufficient feature expression capability, low matching efficiency and low identification accuracy of the existing radar target identification method in a complex environment are solved.

Description

Technical Field

[0001] This application relates to the field of radar data technology, specifically to a radar data retrieval-based feature matching and recognition method and system. Background Technology

[0002] Radar technology is increasingly being used in civil aviation, meteorological monitoring, intelligent transportation and other fields, which puts forward higher requirements for the accuracy, real-time performance and robustness of radar target and scene identification.

[0003] However, traditional radar target recognition methods rely heavily on manually designed features and template matching algorithms. When faced with complex electromagnetic environments, multi-target interference, target attitude changes, and massive amounts of radar data, they often suffer from problems such as insufficient feature representation capabilities, low matching efficiency, and low recognition accuracy.

[0004] Meanwhile, with the improvement of radar resolution and the enhancement of data acquisition capabilities, the amount of radar data has exploded. Traditional feature matching methods based on global search have high computational complexity and are difficult to meet the needs of real-time identification. Summary of the Invention

[0005] This application provides a radar data retrieval-based feature matching and recognition method and system, which solves the technical problems of insufficient feature expression ability, low matching efficiency, and low recognition accuracy of existing radar target recognition methods in complex environments.

[0006] The technical solution to the above-mentioned technical problems in this application is as follows: In a first aspect, this application provides a radar data retrieval-based feature matching and recognition method, the method comprising: Acquire the radar echo data to be identified, and preprocess the radar echo data; Extract target features from the preprocessed radar echo data; Based on the locality-sensitive hash index, the target features are matched with standard features in a pre-built radar feature database for similarity. Based on the matching results, the target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature is output.

[0007] Secondly, this application provides a radar data retrieval-based feature matching and recognition system, including: The data acquisition module is used to acquire radar echo data to be identified and to preprocess the radar echo data. The feature extraction module is used to extract target features from the preprocessed radar echo data; The similarity matching module is used to perform similarity matching between the target features and standard features in a pre-built radar feature database based on the local sensitive hash index. The recognition output module is used to output target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature based on the matching result.

[0008] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a radar data retrieval-based feature matching and identification method and system. First, it acquires the radar echo data to be identified and preprocesses it by performing noise reduction, pulse compression, and range Doppler processing to improve data quality. Second, it extracts target features from the preprocessed radar echo data, generating a multi-dimensional target feature vector containing time-domain, frequency-domain, and time-frequency joint features to characterize the target's properties. Third, based on local sensitive hash indexing, it performs similarity matching between the target features and standard features in a pre-built radar feature database, thereby efficiently narrowing the matching range and improving matching speed. Finally, based on the matching results, when the highest similarity coefficient reaches a preset threshold, it outputs the corresponding target type, model, motion parameters, or scene category identification information. If the threshold is not reached, it is determined to be an unknown target / scene and an alarm mechanism is triggered. Relevant data is stored in a temporary database for manual review, and the alarm information is simultaneously uploaded to the monitoring center display terminal, achieving efficient and accurate radar target and scene identification.

[0009] Through the above technical solution, this application effectively solves the technical problems of insufficient feature expression ability, low matching efficiency and low recognition accuracy of existing radar target recognition methods in complex environments, and improves the accuracy, real-time performance and robustness of radar target and scene recognition. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic flowchart of the radar data retrieval-based feature matching and recognition method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the radar data retrieval-based feature matching and recognition system provided in an embodiment of this application.

[0012] The components represented by each number in the attached diagram are explained below: Data acquisition module 11, feature extraction module 12, similarity matching module 13, and recognition output module 14. Detailed Implementation

[0013] This application provides a radar data retrieval-based feature matching and recognition method and system to address the technical problems of insufficient feature expression capabilities, low matching efficiency, and low recognition accuracy in existing radar target recognition methods under complex environments.

[0014] Example 1, as Figure 1 As shown, this application provides a radar data retrieval-based feature matching and recognition method, including: S10: Acquire the radar echo data to be identified and preprocess the radar echo data; In this embodiment, radar echo data to be identified is first acquired. Based on the receiving antenna array of the phased array radar, electromagnetic echo signals reflected in the target space are collected in real time. The electromagnetic echo signals carry information such as the target's distance, azimuth, and velocity. By processing the phase and amplitude of the signals received by the antenna elements at different positions, the antenna array can detect and locate the space target.

[0015] Specifically, step S10 in the method includes: Based on the phased array radar receiving antenna array, the electromagnetic echo signal reflected in the target space is collected in real time, and the electromagnetic echo signal is converted into a digital signal and transmitted to the data processing unit. The radar echo data is preprocessed, including noise reduction, pulse compression, and range Doppler processing. The radar echo data includes multi-dimensional radar data of the target radar at different azimuth and elevation angles.

[0016] In this embodiment of the application, the collected electromagnetic echo signal is first converted from analog to digital by analog-to-digital conversion. The converted digital signal is then transmitted to the data processing unit for further preprocessing.

[0017] Secondly, the radar echo data is denoised using adaptive filtering algorithms, such as the threshold denoising method based on wavelet transform. By decomposing the signal at multiple scales, the high-frequency coefficients where noise energy is concentrated are thresholded, retaining the effective components of the target signal, thereby reducing the interference of noise in the complex electromagnetic environment on subsequent feature extraction.

[0018] Next, pulse compression processing is performed. Matched filtering technology is used to compress the transmitted wide pulse signal into a narrow pulse to improve the radar's range resolution, so that multiple targets that are originally close to each other can be clearly distinguished.

[0019] Subsequently, range-Doppler processing is performed. By performing a two-dimensional Fourier transform on the echo signal in the range and Doppler dimensions, the range-Doppler spectrum of the target is obtained. This spectrum can intuitively reflect the target's range information and radial velocity information.

[0020] Furthermore, the radar echo data covers the echo information of the target at different azimuth and elevation angles, such as an all-around scan from 0° to 360° and an elevation range from -10° to 90°, forming a multi-dimensional radar data cube to ensure that the distribution characteristics of the target in space can be fully captured.

[0021] S20: Extract target features from the preprocessed radar echo data; In this embodiment of the application, after completing the preprocessing of the radar echo data, the target feature extraction stage is entered. First, the preprocessed radar echo data is segmented to reduce the complexity of data processing, and the feature changes of the target under different states are captured. Then, the target feature vector is extracted.

[0022] Specifically, step S20 in the method includes: The preprocessed radar echo data is segmented to obtain radar data fragments; Feature calculations are performed on the radar data segments to generate target features.

[0023] The target features are based on high-resolution distance image sequences and are multi-dimensional target feature vectors that include time-domain, frequency-domain, and time-frequency joint features.

[0024] In this embodiment, the preprocessed radar echo data is first segmented. Specifically, a sliding window-based method can be used, where the window size and sliding step are set to segment the continuous radar echo data stream into several radar data segments that are time-correlated and contain potential targets.

[0025] For example, for dynamic targets, an appropriate window duration can be set according to the target's motion characteristics, such as setting the window duration according to the expected speed range, so that each data segment can fully reflect the target's motion state and scattering characteristics within that time period; for static scenes, regions are divided according to the spatial distribution characteristics of the scene, and the echo data of different regions are treated as independent data segments.

[0026] Secondly, feature calculations are performed on the segmented radar data fragments, and multi-dimensional target feature vectors are extracted based on the high-resolution range image sequence.

[0027] Among them, the time-domain features include statistical properties such as the mean, variance, peak position, and pulse width of the HRRP sequence, which reflect the energy distribution and structural changes of the target in the distance dimension.

[0028] Frequency domain features can be obtained by performing a Fourier transform on the HRRP sequence, including the peak frequency, bandwidth, and spectral entropy of the spectrum, which are used to capture the frequency response characteristics of the target scattering point.

[0029] The time-frequency joint features can be obtained by using methods such as short-time Fourier transform or wavelet transform to convert the HRRP sequence to the time-frequency plane and extract the texture features, energy distribution center, instantaneous frequency change rate, etc. of the time-frequency image to effectively characterize the dynamic scattering characteristics of the target during motion or under complex postures.

[0030] Specifically, target features are generated. Based on the high-resolution range image sequence, statistical characteristics such as the mean, variance, peak position offset, and pulse width change rate of the radar data segments obtained after preprocessing are extracted through time-domain analysis. Frequency-domain analysis is performed through fast Fourier transform to calculate the peak frequency, 3dB bandwidth, spectral entropy, and energy proportion of each harmonic component. A time-frequency matrix is ​​constructed through short-time Fourier transform or wavelet transform to extract the gray-level co-occurrence matrix features of the time-frequency image, the centroid coordinates of the time-frequency energy accumulation region, and the maximum fluctuation amplitude of the instantaneous frequency.

[0031] After normalizing the aforementioned time-domain, frequency-domain, and time-frequency joint features, they are combined in a preset dimensional order to form a multi-dimensional target feature vector of dimension N×1, where N is determined based on the number of features in the specific application scenario, ensuring that the feature vector can comprehensively and without redundancy characterize the physical properties and motion characteristics of the target or scene. By fusing time-domain, frequency-domain, and time-frequency joint features, the resulting multi-dimensional target feature vector can more comprehensively depict the physical properties and motion state of the target.

[0032] S30: Based on the local sensitive hash index, perform similarity matching between the target features and standard features in the pre-built radar feature database; In this embodiment of the application, after the target feature extraction is completed, the similarity matching stage is entered. First, an index structure is constructed on the standard feature vectors in the radar feature database based on the local sensitive hash algorithm.

[0033] Then, similarity matching is performed. The target feature vector extracted in step S20 is input into the LSH index system. For the selected candidate standard feature vectors, the similarity coefficient between the target feature and each candidate standard feature is calculated using measurement methods such as cosine similarity or Euclidean distance, and then sorted.

[0034] The constructed radar feature database includes: Collect radar sample data of known targets under different conditions; The radar sample data is preprocessed and features are extracted to generate standard features; The standard features are associated with and stored with the corresponding target or scene labels to form the radar feature database.

[0035] In this embodiment of the application, firstly, when constructing the radar feature database, radar sample data of known targets under diverse conditions are collected, covering different attitude angles, motion speeds, motion accelerations of the targets, as well as different electromagnetic environment interference intensities and background clutter types.

[0036] For example, the attitude angle azimuth angle ranges from 0° to 360° in intervals of 5°, and the pitch angle ranges from -10° to 30° in intervals of 2°; the motion speed is uniformly distributed and sampled from 0 m / s to 300 m / s; the motion acceleration is stepped from -5 m / s² to 5 m / s², and different electromagnetic interference intensities are included, such as the gradient setting of the signal-to-noise ratio from -10 dB to 20 dB; background clutter types are included, such as typical clutter models such as ground clutter, sea clutter, and meteorological clutter.

[0037] For scene recognition, radar echo data of different scenes such as urban building complexes, mountainous terrain, airport runways, and port terminals are collected at different times and under different lighting conditions. During the collection process, the precise target type, model information, motion parameters, or scene category label corresponding to each sample are recorded simultaneously.

[0038] Secondly, the collected radar sample data undergoes the same preprocessing procedure as the radar echo data to be identified, namely analog-to-digital conversion, adaptive wavelet threshold denoising, pulse compression, and range Doppler processing, to ensure the consistency between the standard features and the features of the target to be identified at the data preprocessing level.

[0039] Subsequently, the preprocessed sample data is segmented using the same feature extraction method as in step S20. The same sliding window size and step size are used, and time-domain features, frequency-domain features, and time-frequency joint features are extracted to generate a standard feature vector that is completely consistent with the dimension N of the feature vector of the target to be identified.

[0040] Finally, the generated standard feature vectors are associated with the corresponding target or scene labels, and metadata is added to each standard feature vector, including environmental parameters during sample collection, radar operating parameters, and data quality assessment metrics.

[0041] Furthermore, the associated data is stored according to a preset data structure to construct a radar feature database containing massive standard features and their corresponding label information. The database supports dynamic updates, allowing for the periodic addition of new target sample data and scene data, and performs version management and redundancy cleanup on existing data to ensure the database's timeliness and accuracy.

[0042] Specifically, step S30 in the method includes: An LSH hash table is constructed for the standard features in the radar feature database; The target features are mapped to hash buckets using a hash function; The similarity coefficient is calculated and sorted within the hash bucket and its neighboring buckets.

[0043] In this embodiment of the application, firstly, an LSH hash table is constructed for the standard features in the radar feature database.

[0044] Specifically, for the dimension N of the multidimensional target feature vector, a suitable LSH family is selected, such as p-stable distribution LSH or cosine distance LSH, and L independent hash tables are constructed according to the preset number of hash functions K and the number of hash tables L. For each standard feature vector, it is mapped to K integer indices through K hash functions to form a hash key, and then the standard feature vector is stored in the hash bucket corresponding to the hash key in the L hash tables.

[0045] Furthermore, during the construction of the hash table, by optimizing the parameters of the hash function, such as the scale parameter of the p-stable distribution and the random hyperplane direction of the cosine LSH, the hash table is ensured to maintain the local similarity of feature vectors. That is, similar standard feature vectors are mapped to the same or neighboring hash buckets with a high probability.

[0046] Secondly, the target feature vector to be identified is mapped using the same K hash functions used when constructing the hash table to generate corresponding hash keys, thereby locating L target hash buckets in the L hash tables. To improve recall, in addition to the target hash bucket, neighboring hash buckets are also retrieved, i.e., buckets corresponding to hash keys with a Hamming distance of 1 or 2, to avoid similar features being mapped to different buckets due to the randomness of the hash functions.

[0047] Then, from all the retrieved candidate hash buckets, the standard feature vectors stored within are extracted as the candidate set for similarity calculation. Cosine similarity is used as the metric to calculate the similarity coefficient between the target feature vector and each candidate standard feature vector. The formula for cosine similarity is: cosθ=(A·B) / (||A|| ||B||), where A is the target feature vector, B is the candidate standard feature vector, A·B is the vector dot product, and ||A|| and ||B|| are the L2 norms of vectors A and B, respectively. After calculating the similarity coefficients of all candidate standard features, they are sorted from highest to lowest coefficient, and the top M standard feature vectors with the highest similarity are selected as the initial candidates for matching results; for example, M can be 20.

[0048] For example, to calculate similarity, suppose the feature vector A of the target to be identified is [0.2, 0.5, 0.3, 0.8, 0.1], and the candidate standard feature vector B1 is [0.3, 0.4, 0.2, 0.7, 0.2]. The inner product of vectors A and B1 is 0.2×0.3+0.5×0.4+0.3×0.2+0.8×0.7+0.1×0.2=0.9; the L2 norm of vector A is... ≈1.0149; the L2 norm of vector B1 is ≈0.9055; then the cosine similarity between A and B1 is 0.9 / (1.0149×0.9055)≈0.979.

[0049] S40: Based on the matching result, output the target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature.

[0050] The identification information of the target or scene includes target type, model, motion parameters, or scene category.

[0051] In this embodiment of the application, after obtaining the top M candidate standard feature vectors after similarity ranking, the target or scene label corresponding to the candidate feature is first statistically analyzed. If the similarity coefficient of the candidate feature ranked first exceeds the preset confidence threshold, the target type, model, motion parameters or scene category corresponding to it is directly output as the final recognition result.

[0052] Specifically, step S40 in the method includes: Set a preset similarity threshold; When the highest similarity coefficient is greater than or equal to the preset similarity threshold, the recognition information of the target or scene corresponding to the highest similarity standard feature is output; When the highest similarity coefficient is less than the preset similarity threshold, the output recognition result is unknown target / scene, and an alarm mechanism is triggered.

[0053] In this embodiment, firstly, a preset similarity threshold is set, which is dynamically adjusted according to the recognition accuracy requirements of the actual application scenario and the sample distribution characteristics of the radar feature database.

[0054] For example, in target recognition scenarios where high accuracy is required, the preset similarity threshold can be set to 0.85; while in civilian scenarios where high coverage is required, the threshold can be appropriately reduced to 0.75. The threshold setting should also refer to the average similarity between feature vectors of the same type of target in the database and the minimum similarity between feature vectors of different types of targets to ensure that different targets are effectively distinguished while reducing the false recognition rate.

[0055] Next, the highest similarity coefficient after sorting in step S30 is compared with the preset similarity threshold. When the highest similarity coefficient is greater than or equal to the preset similarity threshold, it is determined that the target or scene to be identified is successfully matched with the known target or scene corresponding to the standard feature. At this time, the target or scene label information associated with the standard feature is extracted from the radar feature database, including target type, specific model, motion parameters or scene category, and output to the user interface or downstream processing system in a preset format.

[0056] Furthermore, when the highest similarity coefficient is less than the preset similarity threshold, it indicates that the target or scene to be identified has not reached an effective matching degree with the existing standard features in the database. At this time, the identification result of "unknown target / scene" is output.

[0057] Specifically, when the highest similarity coefficient is less than the preset similarity threshold, the output recognition result is an unknown target / scene, and an alarm mechanism is triggered, including: The radar echo data and target features corresponding to the unknown target / scene are automatically stored in a temporary database and marked as data to be manually reviewed. The alarm mechanism includes alarm signal frequency and intensity, and alarm information is synchronously uploaded to the display terminal of the monitoring center.

[0058] In this embodiment, firstly, when the highest similarity coefficient is less than the preset similarity threshold, the original radar echo data of the target to be identified, the preprocessed intermediate data, the extracted target feature vector, and the key parameters in the similarity matching process are automatically and completely stored in the preset temporary database.

[0059] At the same time, metadata information such as timestamps, radar equipment numbers, data collection location coordinates, and environmental parameters at the time are added and uniformly marked as "data to be manually reviewed" so that technicians can trace and analyze it later.

[0060] Secondly, once the alarm mechanism is triggered, the frequency and intensity of the alarm signal can be configured according to actual application needs. For example, in a warning scenario, a high-frequency, bright flashing audible and visual alarm can be set to quickly attract the attention of monitoring personnel; while in a general monitoring scenario, a combination of low-to-medium frequency alert sounds and screen pop-ups can be used.

[0061] Specifically, alarm information will be uploaded to the display terminal of the monitoring center in real time. The terminal interface will highlight the "unknown target / scene" alarm event and display the basic information of the event, such as the time of occurrence, the associated radar equipment, and the approximate location of the target, so that monitoring personnel can grasp the situation in a timely manner and take corresponding measures.

[0062] In summary, compared with existing technologies, this application effectively improves the feature representation capability, matching efficiency, and recognition accuracy of radar target and scene recognition in complex environments by constructing multi-dimensional target feature vectors and combining them with local sensitive hash indexing technology.

[0063] In summary, the embodiments of this application have at least the following technical effects: This application provides a radar data retrieval-based feature matching and identification method. First, it acquires the radar echo data to be identified and preprocesses it by performing noise reduction, pulse compression, and range Doppler processing to improve data quality. Second, it extracts target features from the preprocessed radar echo data, generating a multi-dimensional target feature vector containing time-domain, frequency-domain, and time-frequency joint features to characterize the target's properties. Third, based on a local sensitive hash index, it performs similarity matching between the target features and standard features in a pre-built radar feature database, thereby efficiently narrowing the matching range and improving matching speed. Finally, based on the matching results, when the highest similarity coefficient reaches a preset threshold, it outputs the corresponding target type, model, motion parameters, or scene category identification information. If the threshold is not reached, it is determined to be an unknown target / scene and an alarm mechanism is triggered. Relevant data is stored in a temporary database for manual review, and the alarm information is simultaneously uploaded to the monitoring center display terminal, achieving efficient and accurate radar target and scene identification.

[0064] Through the above technical solution, this application effectively solves the technical problems of insufficient feature expression ability, low matching efficiency and low recognition accuracy of existing radar target recognition methods in complex environments, and improves the accuracy, real-time performance and robustness of radar target and scene recognition.

[0065] Example 2, as Figure 2 As shown, based on the same inventive concept as the radar data retrieval feature matching and recognition method provided in Embodiment 1, this application also provides a radar data retrieval feature matching and recognition system, including: The data acquisition module 11 is used to acquire radar echo data to be identified and to preprocess the radar echo data. Feature extraction module 12 is used to extract target features from the preprocessed radar echo data; The similarity matching module 13 is used to perform similarity matching between the target features and standard features in a pre-built radar feature database based on the local sensitive hash index. The recognition output module 14 is used to output target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature based on the matching result.

[0066] In one embodiment, the data acquisition module 11 is specifically used for: Based on the phased array radar receiving antenna array, the electromagnetic echo signal reflected in the target space is collected in real time, and the electromagnetic echo signal is converted into a digital signal and transmitted to the data processing unit. The radar echo data is preprocessed, including noise reduction, pulse compression, and range Doppler processing. The radar echo data includes multi-dimensional radar data of the target radar at different azimuth and elevation angles.

[0067] In one embodiment, the feature extraction module 12 is specifically used for: The preprocessed radar echo data is segmented to obtain radar data fragments; Feature calculations are performed on the radar data segments to generate target features.

[0068] Furthermore, in one embodiment of the application, the target feature is based on a high-resolution distance image sequence and is a multi-dimensional target feature vector including time-domain, frequency-domain, and time-frequency joint features.

[0069] Furthermore, in one embodiment of the application, the constructed radar feature database includes: Collect radar sample data of known targets under different conditions; The radar sample data is preprocessed and features are extracted to generate standard features; The standard features are associated with and stored with the corresponding target or scene labels to form the radar feature database.

[0070] Furthermore, the similarity matching module 13 is specifically used for: An LSH hash table is constructed for the standard features in the radar feature database; The target features are mapped to hash buckets using a hash function; The similarity coefficient is calculated and sorted within the hash bucket and its neighboring buckets.

[0071] Furthermore, the identification output module 14 is specifically used for: Set a preset similarity threshold; When the highest similarity coefficient is greater than or equal to the preset similarity threshold, the recognition information of the target or scene corresponding to the highest similarity standard feature is output; When the highest similarity coefficient is less than the preset similarity threshold, the output recognition result is unknown target / scene, and an alarm mechanism is triggered.

[0072] Furthermore, the identification information of the target or scene includes target type, model, motion parameters, or scene category.

[0073] Furthermore, in one embodiment, when the highest similarity coefficient is less than the preset similarity threshold, the output recognition result is an unknown target / scene, and an alarm mechanism is triggered, including: The radar echo data and target features corresponding to the unknown target / scene are automatically stored in a temporary database and marked as data to be manually reviewed. The alarm mechanism includes alarm signal frequency and intensity, and alarm information is synchronously uploaded to the display terminal of the monitoring center.

Claims

1. A radar data retrieval-based feature matching and recognition method, characterized in that, The method includes: Acquire the radar echo data to be identified, and preprocess the radar echo data; Extract target features from the preprocessed radar echo data; Based on the locality-sensitive hash index, the target features are matched with standard features in a pre-built radar feature database for similarity. Based on the matching results, the target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature is output.

2. The radar data retrieval-based feature matching and recognition method as described in claim 1, characterized in that, Acquire the radar echo data to be identified, and preprocess the radar echo data, including: Based on the phased array radar receiving antenna array, the electromagnetic echo signal reflected in the target space is collected in real time, and the electromagnetic echo signal is converted into a digital signal and transmitted to the data processing unit. The radar echo data is preprocessed, including noise reduction, pulse compression, and range Doppler processing. The radar echo data includes multi-dimensional radar data of the target radar at different azimuth and elevation angles.

3. The radar data retrieval-based feature matching and recognition method as described in claim 1, characterized in that, Extracting target features from the preprocessed radar echo data includes: The preprocessed radar echo data is segmented to obtain radar data fragments; Feature calculations are performed on the radar data segments to generate target features.

4. The radar data retrieval-based feature matching and recognition method as described in claim 3, characterized in that, The target features are based on high-resolution distance image sequences and are multi-dimensional target feature vectors that include time-domain, frequency-domain, and time-frequency joint features.

5. The radar data retrieval-based feature matching and recognition method as described in claim 1, characterized in that, The constructed radar feature database includes: Collect radar sample data of known targets under different conditions; The radar sample data is preprocessed and features are extracted to generate standard features; The standard features are associated with and stored with the corresponding target or scene labels to form the radar feature database.

6. The radar data retrieval-based feature matching and recognition method as described in claim 1, characterized in that, Based on locality-sensitive hash indexing, the target features are matched with standard features in a pre-built radar feature database based on similarity, including: An LSH hash table is constructed for the standard features in the radar feature database; The target features are mapped to hash buckets using a hash function; The similarity coefficient is calculated and sorted within the hash bucket and its neighboring buckets.

7. The radar data retrieval-based feature matching and recognition method as described in claim 6, characterized in that, Based on the matching results, the target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature is output, including: Set a preset similarity threshold; When the highest similarity coefficient is greater than or equal to the preset similarity threshold, the recognition information of the target or scene corresponding to the highest similarity standard feature is output; When the highest similarity coefficient is less than the preset similarity threshold, the output recognition result is an unknown target / scene, and an alarm mechanism is triggered.

8. The radar data retrieval-based feature matching and recognition method as described in claim 7, characterized in that, The target or scene identification information includes target type, model, motion parameters, or scene category.

9. The radar data retrieval-based feature matching and recognition method as described in claim 7, characterized in that, When the highest similarity coefficient is less than the preset similarity threshold, the output recognition result is "unknown target / scene," and an alarm mechanism is triggered, including: The radar echo data and target features corresponding to the unknown target / scene are automatically stored in a temporary database and marked as data to be manually reviewed. The alarm mechanism includes alarm signal frequency and intensity, and alarm information is synchronously uploaded to the display terminal of the monitoring center.

10. A radar data retrieval-based feature matching and recognition system, characterized in that, The radar data retrieval-based feature matching and identification method according to any one of claims 1-9 includes: The data acquisition module is used to acquire radar echo data to be identified and to preprocess the radar echo data. The feature extraction module is used to extract target features from the preprocessed radar echo data; The similarity matching module is used to perform similarity matching between the target features and standard features in a pre-built radar feature database based on the local sensitive hash index. The recognition output module is used to output target or scene recognition information corresponding to the standard feature with the highest similarity to the target feature based on the matching result.

Images