A method of orthogonal data extraction for real-time multi-target classification and recognition

Abstract

The present invention belongs to the technical field of radar overall design, and discloses an orthogonal data extraction method suitable for real-time multi-target classification and identification in the field of pitch simultaneous multi-beam three-coordinate radar. The present invention designs a cache airspace in an information processing extension according to the distance segment where the target is located, the azimuth / elevation beam coverage associated with the target distance and the maximum speed / maximum altitude boundary conditions, to avoid caching and transmitting a large amount of invalid data in the entire range / all directions / all elevation angles. After completing the track filtering, the pitch gate range is determined according to the target distance segment, and the target distance gate is determined according to the amplitude distribution characteristics of the distance unit, with the target point track as the center, so as to accurately extract the relevant orthogonal data of the target from the cache and reduce the amount of data transmitted to the target classification and identification extension. This method can stably and real-timely realize the real-time classification and identification of targets in the entire airspace under the conditions of the original hardware cache and data transmission capabilities.

Description

Technical Field

[0001] The invention belongs to the technical field of radar overall design. Background Art

[0002] In order to ensure high sensitivity for target classification and recognition, radar target classification and recognition needs to extract features from the phase information of the target echo, and the target classification and recognition extension must input the target's orthogonal data. The amount of orthogonal data of radar echo is very large. How to extract target-related orthogonal data from the orthogonal data of radar echo determines the real-time, accuracy, recognition capacity, implementation cost and system stability of target classification and recognition.

[0003] The first existing method is that the information processing extension sends all the orthogonal data of the radar echo to the target classification and recognition extension. The target classification and recognition extension caches the orthogonal data of the radar echo for a scanning cycle. After receiving the target point data sent by the information processing extension, the orthogonal data is extracted and the target is classified and recognized based on the point data. The advantage of this method is that it can ensure the real-time classification and recognition of targets in the entire airspace, but the disadvantages are also obvious - the information processing extension and the target classification and recognition extension must have high-speed data transmission capabilities, and the target classification and recognition extension must also have a large-capacity cache and powerful data processing capabilities. This method requires a high hardware cost, and due to its complex transmission and processing, the stability of the system will also be affected.

[0004] In order not to increase the hardware cost and ensure the stability of the system, another existing method is: first track the target, and according to the target position in the next scanning cycle, the information processing extension provides orthogonal data of the relevant area centered on the extrapolated position of the target. This method extracts orthogonal data in a certain space after the next scanning cycle arrives, which is used for target identification, with the extrapolated position of the target as the center. The disadvantage of this method is that it takes at least one scanning cycle to output the target classification and identification result, which may be acceptable for high-speed scanning fire control radars (radar scanning cycles generally do not exceed 2s), but for reconnaissance and surveillance radars (radar scanning cycles are generally not less than 10s), the lag time is unacceptable. The error of the extrapolated position is much larger than that of the associated point traces, and the amount of data that needs to be extracted and transmitted will also be relatively large, and the accuracy of target type identification will be reduced.

[0005] In order to ensure real-time performance and the accuracy of target classification and recognition, the third existing method is to establish a three-dimensional wave gate based on the extrapolated position of a target track in the previous antenna cycle as the center, cache the radar echo data in the wave gate, and after the target track is generated, extract the orthogonal data of the target from the cached data with the target point track as the center and send it to the target classification and recognition extension for target classification and recognition. Compared with method 2, this method can realize real-time classification and recognition of targets in the current cycle, and the recognition accuracy is relatively high. The disadvantage of this method is that it can only recognize a single batch of targets at the same time, and after issuing the target classification and recognition command, it takes at least one scanning cycle to output the target classification and recognition results. Summary of the invention

[0006] In order to solve the problem that the real-time classification and identification function of all airspace targets cannot be realized based on the existing hardware conditions, the present invention provides an orthogonal data extraction method for real-time multi-target classification and identification, which can not only stably and real-time realize the classification and identification of all airspace targets and ensure the accuracy of target identification, but also is based on the original hardware caching and data transmission capabilities.

[0007] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0008] Step 1: Use the orthogonal data after pulse compression as the input data for target recognition, extract the target-related orthogonal data based on the target point trace, and realize the classification and recognition of the full airspace targets in real time;

[0009] The digital orthogonal data after pulse compression is divided into two paths, one of which is used for normal signal processing and data processing, and then outputs the track and point track; the other is used for caching, and selective caching is performed according to different distance segments, azimuth sectors and elevation ranges; according to the location information of the associated point track and the target feature information, the target-related orthogonal data is extracted from the buffer, and the necessary orthogonal data is output to the target classification and recognition extension for classification and recognition. The schematic diagram is shown in the figure below. Figure 1 and Figure 2 shown.

[0010] Step 2: Cache and extract target related orthogonal data in the information processing subunit;

[0011] In order to avoid large-scale data transmission, the target-related orthogonal data is cached and extracted in the information processing unit.

[0012] Step 3: Cache target-related orthogonal data based on target features, and reduce the cached data volume while ensuring that all airspace target-related orthogonal data is cached;

[0013] In the information processing extension, the orthogonal data of radar echoes related to targets in the entire airspace are cached. The range of data that needs to be cached is reasonably planned based on the target motion characteristics, target point extraction principles, and radar beam coverage. This greatly reduces unnecessary cached data, reduces the occupancy rate of system resources, and ensures the stability and real-time performance of the system.

[0014] Step 4: Accurately extract target-related orthogonal data from the cache with the target point as the center for classification and recognition.

[0015] After the target's current cycle track is generated, the target point track (the current scanning cycle track associated point track) is taken as the center, and the relevant orthogonal data of the target is accurately extracted from the cached data according to the target's electromagnetic reflection characteristics and the target feature information contained in the target point track, which is used as the input of the target classification and recognition extension to improve the accuracy of the data and reduce the amount of data transmitted to the target classification and recognition extension.

[0016] Furthermore, the step 3 includes: designing a cache airspace according to the distance segment where the target is located, and only caching data of the airspace where the target has main energy distribution.

[0017] In order to ensure the real-time performance of target type identification, the radar echo orthogonal data of the current scanning cycle must be cached. Since the amount of radar echo orthogonal data is huge, if the radar echo orthogonal data is cached in the target type identification extension, the system needs to have the ability to transmit large-scale data in real time; the present invention designs the cache airspace according to the distance segment where the target is located in the information processing extension, only caches the data of the target's main energy distribution airspace, and simplifies the cache target-related orthogonal data. The scheme for reducing cache data is described in detail below.

[0018] Cache all echo orthogonal data of a scanning cycle. The data volume calculation formula is:

[0019] DataL=Amp×2×R n ×P m ×E t ……………………(1)

[0020] Amp: the real or imaginary part of the quadrature amplitude data of the unit echo, usually 2 bytes;

[0021] R n : The number of distance units for one pulse. The test range is 80km and the distance resolution unit is 10m. The value is 80,000÷10=8,000.

[0022] P m : The total number of pulses in one scanning cycle (antenna cycle), the value is 10,000,000μs÷860μs=11,627.9;

[0023] E t : The number of elevation angle layers. The test conditions are that the elevation beam width is 6°, the elevation range is 60°, and the overlapping area is 20%. Therefore, the total number of elevation angle layers is 60°÷(6°×80%)=12.

[0024] According to formula (1), the amount of data that needs to be cached in the test environment is 2×2×8,000×11,627×12=4,464,768,000B, which is about 4.46GB. This amount of data is not large for general computing environments (using electronic disks or hard disks), but it is huge if real-time performance is required. The access speed of the data is very high, especially the data recording speed. The read and write speed of the electronic disk is 250MB / S, which is 7 to 8 times the read and write speed of the hard disk. The read and write speed of the memory is 10GB / S. If the memory cache is used to store the data for one scanning cycle, the storage time required is ideally 446ms, and the time required for each pulse is 38.36μs, which is too large for the repetition period of 860μs. If the full-range and omnidirectional radar echo orthogonal data storage is used, other processing (such as signal processing, data processing, and target-related data extraction, etc.) cannot be performed. For signal processing with a pulse data update speed of microseconds, this time is obviously too long. Based on the above analysis, it is concluded that real-time signal processing and data processing of echo data cannot be achieved if full airspace data caching is used.

[0025] In order to ensure the real-time performance of the information processing branch without increasing the cache space of the information processing branch, it is necessary to reduce the amount of cached data on a large scale. Since the target's flight altitude is generally below 10,000m, the pitch range of the target with a distance greater than 10km is significantly smaller than the pitch range of the near zone. See the schematic diagram of the radar beam coverage for details. Figure 3 The near zone uses all elevation layer data cache, and the far zone uses part of the elevation layer cache. According to the target motion characteristics and the principle of target point extraction, the target data update azimuth span in the overhead area (near zone) will be very large. Based on this feature, the range of cached data can be reduced in the pitch dimension.

[0026] The classic and commonly used target point extraction method for radars is to update it in sectors. This method generally takes the azimuth width of the largest target at the minimum distance plus the radar azimuth beam width as the minimum value of the sector. When the scan line reaches the third sector, the point information of the first sector is output. The azimuth cache must cache at least two sectors of echo orthogonal data to ensure that the corresponding orthogonal data can be found in the cached data when the track-related points are output. In order to ensure that the orthogonal data of the target is found and identified according to the track-related points, at least three detection sectors need to be cached. The experimental data calculation is shown in the implementation process.

[0027] This method reasonably plans and cuts the cache data according to the target motion characteristics, target point extraction principles, and radar beam coverage, reducing unnecessary cache data and reducing the occupied cache space and data processing resources.

[0028] Furthermore, the step 4 includes: determining the target pitch gate and range gate sizes according to different target distance segments and target range unit amplitude distribution characteristics, and accurately extracting target related orthogonal data from the cache.

[0029] The sizes of target range gate, elevation gate and azimuth gate are determined according to the amplitude distribution characteristics of target range unit, the elevation correlation range and azimuth beam broadening in different distance segments, and the target related orthogonal data is accurately extracted from the cache to further reduce the amount of data transmitted to the target classification and identification extension.

[0030] The present invention extracts relevant orthogonal data with the target point trace position of the current scanning cycle as the center. The target is divided into point targets and non-point targets according to the distance extension value of the target in the target point trace information, and the length of the relevant data of the extracted target in the distance dimension is determined accordingly. Since the target position is accurately determined, according to the reflection characteristics and distribution characteristics of the target, the target-related data in the elevation dimension are all concentrated in three elevation layers centered on the point trace, and the relevant data of the far-field target is distributed in at most two elevation layers. In the azimuth dimension, the full airspace detection often updates the target orthogonal data in sectors. The size of each sector can fully accommodate the widening of the target in azimuth. Taking into account the special situation that the azimuth span of cross-sector targets and near-field targets is relatively large, the target extracts relevant data of at most one sector in the azimuth dimension. The scheme for reducing the extracted data is described in detail below.

[0031] According to the characteristics of the identified target, the data structure of the orthogonal data associated with the target point trace is determined. Compared with the 10m distance unit, small unmanned aerial vehicles or flying birds belong to point targets. The data related to point targets are expanded from three dimensions: one related distance unit in the distance dimension and several distance units in the front and back; for large targets, it is necessary to expand the classification processing in the distance according to the size of the target. If the size of the target is larger than the distance resolution unit of the target echo, it is necessary to expand the distance. In the elevation range, in the far area (greater than 3km), the target has relevant data in up to 3 elevation layers. In the near area (not more than 3km), due to the serious crossover of the elevation beam, the target echo data will appear in multiple elevation layers. The actual target recognition only needs the data in one layer for normal recognition. Considering that there is a situation where the target is in the boundary area of ​​​​two elevation layers and it is not easy to determine which elevation layer has a larger signal-to-noise ratio of the target echo, in order to ensure complete relevant data, the data of the two elevation layers are extracted for recognition. The number of relevant pulses calculated by the azimuth beam width and the azimuth scanning speed in the azimuth dimension;

[0032] The beneficial effects of the present invention are as follows: the cache space required by the method of the present invention is significantly smaller than the cache space required for the full airspace radar echo orthogonal data cache, and the accurate extraction of target-related orthogonal data from the cache with the target point track as the center further reduces the amount of data sent to the target classification and identification extension, and the complexity of the software required for caching and extracting data is significantly reduced. The method of the present invention can realize real-time classification and identification of targets in the entire airspace, and real-time classification and identification can be performed from the moment the target track is generated. Under the condition of keeping the interface communication capability and data processing capability of the information processing extension and the target classification and identification extension unchanged, the classification and identification of targets tracked in the entire airspace can be realized in real time and stably, and has a wider adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Schematic diagram of orthogonal data caching and extraction method for real-time multi-target classification and recognition;

[0034] Figure 2 Accurately extract relevant orthogonal data schematics from cache data;

[0035] Figure 3 Schematic diagram of radar power coverage. The horizontal axis is distance (km) and the vertical axis is height (m). DETAILED DESCRIPTION

[0036] The radar digital orthogonal data after pulse compression is divided into two paths, one of which is used for normal signal processing, data processing and output of track and associated point traces; the second path is used for caching, extraction and target identification, and caches are distinguished according to different distance segments, azimuth sectors and elevation detection range characteristics; according to the location information of the associated point traces and the target size feature information, the target-related orthogonal data is extracted from the buffer and output to the target classification and identification extension for identification. The schematic diagram is shown in the figure. Figure 1 and Figure 2 shown.

[0037] In order to reduce the resources occupied by cached data and ensure the real-time and correctness of target identification, only the orthogonal data of the area where the target to be identified in the entire airspace is located is cached; the orthogonal data related to the target in this scanning cycle is extracted from the cached data according to the target point trace position information. In order to reduce the resources occupied by data transmission between the information processing extension and the target identification extension and improve the real-time performance of target type identification, the amount of data transmitted from the information processing extension to the target classification and identification extension is minimized on the basis of ensuring the integrity of the relevant data.

[0038] The test targets are DJI Phantom 4, carrier pigeons, civil aircraft and large container ships. The implementation process is described in detail below:

[0039] Firstly, in the information processing unit, the data structure of the cache is designed in sections from the azimuth dimension, the elevation dimension and the distance dimension, and the cache space is opened up to cache the orthogonal data of the radar echo;

[0040] 1) The range of target orthogonal data that needs to be cached in the azimuth dimension is designed as follows: based on the maximum speed of the target not exceeding 300m / s, assuming that it flies over the radar, since the radar beam scanning period generally does not exceed 10s, as long as the target within 3km is cached in all directions. For large targets with a maximum length of 400m, the azimuth width spanned at 3km is not greater than 7.66°. Considering that the azimuth beam width generally does not exceed 2°, the azimuth detection sector width is not less than 9.66 degrees. The azimuth detection sector of the test environment radar is set to 12°; when detecting the entire airspace, the point track data of the first sector is often output when the scan line crosses the third sector. Here, the azimuth cache has 48° of data, and the redundant 2 sectors are sufficient to ensure that the orthogonal data related to the current target point track can be extracted in real time after the track is output. Update the target orthogonal data in sectors.

[0041] 2) Elevation dimension According to the needs of the detection airspace and the design of radar beamforming, the near area needs to cache more elevation beam layers, and the far area needs fewer elevation beam layers. In this experiment, the area not larger than 10km needs to cover -20° to 60°, and the area larger than 10km needs to cover -1° to 8°, which is enough to cover the entire target-related area. The elevation dimension caches data in units of the receiving beam elevation layer, and the range of the cached data is determined by the coverage range of the elevation layer. That is, the bottom two elevation layers cache echo orthogonal data according to the maximum detection range of the radar; the remaining elevation layers only cache echo orthogonal data within a range of no more than 10km.

[0042] 3) The data is cached in three distance segments using different strategies. Within 3 km, the azimuth and elevation are cached in the entire airspace. According to formula (1), the maximum amount of data cached in this distance segment is: 2×2×300×11,627×12=167,428,800B; in the 3km to 10km distance segment, the azimuth is cached within a range of 48°, and the elevation dimension is cached for all elevation layers. According to formula (1), the data volume is: 2×2×700×11,627÷7.5×12=52,088,960B; beyond 10km, the azimuth span is 48°, and the elevation span is 2 elevation layers. According to formula (1), the data volume is: 2×2×7,000×11,627÷7.5×2=86,814,933bit. A total of 306,332,693 bits, about 306.33MB, is only 6.86% of the original 4.46GB of full-range cache. The amount of data required to record each pulse is: 2×2×1,000×12+2×2×7,000×2=104000 bits. The cache time required for each pulse is 10.4μs, which is acceptable for the 860μs repetition period. Through experiments, it is verified that under the ordinary software and hardware environment under the experimental conditions, if the data cached in a scan cycle exceeds 800MB, there will be system stability problems, and if it exceeds 1,000MB, it will crash. If it is less than 600MB, the system is stable and has no abnormalities. The amount of data required to be cached by this method is 306.33MB, which does not affect the stability and real-time performance of the system.

[0043] Then, for each batch of targets, after the target track is filtered, a functional unit for extracting orthogonal data based on the target track is triggered; in the process of extracting associated data, the data structure and data range of the extracted orthogonal data are determined by the characteristics of the target track.

[0044] 1) Determine the scale of data to be extracted based on the characteristic information of the target point traces. Relative to the distance unit of 10m, DJI Phantom 4 and carrier pigeons are both point targets, and the distance is extracted according to 20 distance units; large container ships with a maximum length of 400m extract 50 distance units. Point targets and large targets can be determined based on the number of distance-related units in the point trace information. Point targets with less than 5 distance-related units are determined, and others are determined as large targets. According to the test environment, the number of azimuth-related pulses of the radar for point targets is 32, and the azimuth gate span is set to 40 pulses. In the pitch dimension, 3 elevation layers are extracted in areas less than 10km, and 2 elevation layer data are extracted in other areas.

[0045] 2) If the amplitude of each distance unit is represented by 2 bytes, then the orthogonal data related to each large target in the near area is 50×40×3×2×2=24,000bit≈24kB according to formula (1). The orthogonal data related to each point target in the far area is 20×40×2×2×2=6,400bit≈6.4kB according to formula (1). The target track data content does not exceed 50B. Assuming that the upper limit of the maximum processing target is 1000 batches of targets, the amount of data that needs to be sent to the target classification and identification extension in each scanning cycle ranges from 6.45MB to 24.05MB. A 1000M network interface is used between the information processing extension and the target classification and identification extension in the environment. Through experiments, it is verified that in the test environment, if the amount of data sent to the target identification extension in a concentrated manner (within two sectors, 24°) in a scanning cycle exceeds 250MB, the system stability problem will occur, and if it exceeds 300MB, the transmission channel will be blocked. If it is less than 100MB, the system is stable without abnormalities. In the extreme case, the amount of data that needs to be centrally transmitted by this method is 24.05MB, which does not affect the stability and real-time performance of the system.

[0046] Output the target-related orthogonal data and track information to the target classification and identification sub-unit. The target identification sub-unit processes and outputs the target track information with the target identification result to the radar display and control sub-unit, the command and control system or the reconnaissance and surveillance system in real time.

Claims

1. A real-time multi-target classification and recognition orthogonal data extraction method, characterized by: Step 1: Use the orthogonal data after pulse compression as the input data for target identification, extract the target-related orthogonal data based on the target point trace, and realize the classification and identification of the full airspace targets in real time; divide the digital orthogonal data after pulse compression into two paths, one of which is used for normal signal processing and data processing, and then outputs the track and point trace; the other is used for caching, and selective caching is performed according to different distance segments, azimuth sectors and elevation angle ranges; Extract target-related orthogonal data in the buffer according to the location information of the associated point traces and the target feature information, and output the necessary orthogonal data to the target classification and recognition extension for classification and recognition; Step 2: Cache and extract target related orthogonal data in the information processing subunit; Step 3: Cache the target-related orthogonal data based on the target characteristics, and reduce the cached data volume on the basis of ensuring that the orthogonal data related to the full airspace targets are cached; cache the orthogonal data of the radar echo related to the full airspace targets in the information processing extension, plan the range of data that needs to be cached based on the target motion characteristics, target point extraction principle, and radar beam coverage, and reduce unnecessary cached data; Step 4: With the target point track as the center, accurately extract the target-related orthogonal data from the cache for classification and identification; after the target's current cycle track is generated, take the target point track as the center, and the target point track is the current scanning cycle track-related point track; according to the electromagnetic reflection characteristics of the target and the target feature information contained in the target point track, accurately extract the target's related orthogonal data from the cached data as the input of the target classification and identification extension to improve the accuracy of the data and reduce the amount of data transmitted to the target classification and identification extension.

2. The orthogonal data extraction method for real-time multi-target classification and recognition according to claim 1 is characterized in that: In the step 3: the cache airspace is designed according to the distance segment where the target is located, and only the data of the airspace where the target's main energy is distributed is cached.

3. The orthogonal data extraction method for real-time multi-target classification and recognition according to claim 1 is characterized in that: The step 4 determines the size of the target pitch gate and the range gate according to different target distance segments and target range unit amplitude distribution characteristics, and accurately extracts target related orthogonal data from the cache.

Images