Moving target detection method and implementation system in strong sea clutter environment

Abstract

The present invention provides a method and implementation system for detecting moving targets in a strong sea clutter environment, comprising the following steps: Step S1: comparing the radar echo data to be tested with a sample library to perform a preliminary separation of clutter and target; Step S2: detecting the target based on this preliminary separation. The present invention effectively detects slow and small targets, and continuous technical improvements have been made in applications. The effectiveness of the method has been demonstrated through multiple practical application tests. Based on radar target performance, technical improvements such as automatic parameter extraction, sub-image convolution, and adaptive dual thresholding have been implemented, achieving the realization of dynamic clutter maps for water targets in strong clutter environments and developing a moving target detection and processing module. Application tests have significantly improved the detection and tracking of targets in waters such as aquaculture areas and bridges over rivers.

Description

Technical Field

[0001] The present invention relates to the technical field of improvement of radar technology application, and in particular to a moving target detection method and implementation system in a strong sea clutter environment. Background Art

[0002] The sea and rivers present a strong clutter environment for radar interference. Clutter is a human-defined term. In reality, we only care about large and small ships and buoys. Other echoes, such as floating objects, waves, tides, rain, snow, typhoons, islands, bird flocks, and reflections from buildings (which may be intertwined and therefore of no concern), are collectively referred to as clutter (echoes that actually reflect radar but are not of interest in the application). Because different types of clutter have distinct characteristics, it's impossible to determine the type of clutter simply by looking at the radar echo. The generation of sea clutter, as described above, depends on many complex factors, primarily the radar's operating state and the ocean environment during detection. Specifically, these factors have a significant impact: the radar signal's angle of incidence, transmit frequency, and wind speed and direction. Processing models such as random processes (echo amplitude distribution, constant false alarm detection, high-order accumulation, correlation detection), chaos models, and fractal theory have yielded improvements in certain applications. While some theories have shown promising results in simulations, actual testing has proven otherwise. Detecting small targets in strong clutter environments is a global technical challenge. The exploration of detecting small, slow-moving targets in strong clutter environments continues unabated. However, years of technological improvements have resulted in slow progress, and existing approaches have clearly reached a bottleneck.

[0003] Chinese invention patent publication number CN113009444A discloses a method and apparatus for target detection in a generalized Gaussian texture sea clutter background. The method comprises the following steps: obtaining received echo information; constructing probability density functions of the sea clutter in the received echo under alternative and null hypothesis conditions based on the received echo information; determining a likelihood ratio detection function based on the probability density function; determining a target detection function based on the likelihood ratio detection function; and performing target detection using the target detection function.

[0004] Regarding the above-mentioned related technologies, the inventors believe that the above-mentioned methods have made slow progress in exploring the detection of small and slow targets in strong clutter environments and have poor effectiveness. Summary of the Invention

[0005] In view of the defects in the prior art, the purpose of the present invention is to provide a moving target detection method and implementation system in a strong sea clutter environment.

[0006] According to the present invention, a method for detecting a moving target in a strong sea clutter environment comprises the following steps:

[0007] Step S1: Compare the radar echo data to be tested with the sample library to perform preliminary separation of clutter and targets;

[0008] Step S2: Detect the target based on the preliminary separation of clutter and target.

[0009] Preferably, in step S1, optical detection VIBE is used to separate radar targets from clutter. Based on the characteristics of radar data and combined with optical detection, the radar data to be tested is compared with the sample library through multi-frame scanning to preliminarily separate the target or update the sample library.

[0010] Preferably, in step S2, the motion and shape characteristics of the target are introduced, and a large-area sub-image convolution statistical threshold is adopted to detect the moving target according to the motion characteristics of the previous and next frames; multiple frames of echo accumulation and a small area are selected, and after accumulation, a sub-image convolution statistical threshold is applied to determine the stationary target;

[0011] Dual-channel processing of moving targets and stationary targets is used to detect moving targets and stationary targets.

[0012] Preferably, the method further comprises step S3: performing statistical analysis based on the clutter background and characteristics of various targets, adaptively adjusting relevant thresholds and parameters, and detecting and tracking various targets.

[0013] Preferably, in step S3, adaptive threshold and parameter processing is performed: parameter statistics are performed on the clutter environment by area, on the moving target by motion characteristics, and on the stationary target by history and echo characteristics, and the discrimination threshold of the relevant target is adjusted and the relevant parameters are set accordingly to detect and track various types of targets.

[0014] According to the present invention, a moving target detection implementation system in a strong sea clutter environment includes the following modules:

[0015] Module M1: compares the radar echo data to be tested with the sample library to perform preliminary separation of clutter and targets;

[0016] Module M2: Target detection based on preliminary separation of clutter and target.

[0017] Preferably, in the module M1, optical detection VIBE is used to separate radar targets from clutter. Based on the characteristics of radar data and combined with optical detection, the radar data to be tested is compared with the sample library through multi-frame scanning to preliminarily separate the target or update the sample library.

[0018] Preferably, in the module M2, the motion and shape characteristics of the target are introduced, and a large-area sub-image convolution statistical threshold is adopted to detect the moving target according to the motion characteristics of the target in the previous and next frames; multiple frames of echo accumulation and a small area are selected, and after accumulation, a sub-image convolution statistical threshold is applied to determine the stationary target;

[0019] Dual-channel processing of moving targets and stationary targets is used to detect moving targets and stationary targets.

[0020] Preferably, the system further comprises a module M3: performing statistical analysis based on the clutter background and characteristics of various targets, adaptively adjusting relevant thresholds and parameters, and detecting and tracking various targets.

[0021] Preferably, in the module M3, adaptive threshold and parameter processing are performed: parameter statistics are performed on the clutter environment by area, on the moving targets by motion characteristics, and on the stationary targets by history and echo characteristics. The discrimination thresholds of the relevant targets are adjusted accordingly and relevant parameters are set to detect and track various types of targets.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. This paper combines radar echo characteristics to improve the VIBE method and subgraph convolution-related methods to achieve effective detection of moving (including slow) targets and stationary (including small) targets. Continuous technical improvements in application and multiple practical application tests have demonstrated the effectiveness of this method.

[0024] 2. Based on radar target performance, the present invention has made technical improvements such as automatic parameter extraction, sub-image convolution, and adaptive dual thresholding. This has achieved the realization of dynamic clutter maps of water targets in strong clutter environments and the development of a moving target detection and processing module. In application tests, the detection and tracking effects of water targets such as sea surface aquaculture areas and river bridges have been greatly improved.

[0025] 3. After preliminary verification, the present invention has been improved many times, and the implemented module has been applied in the multi-element joint perception system, achieving good results in moving target detection and stationary target detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0027] Figure 1 Flowchart for the method and system implementation of the present invention;

[0028] Figure 2 This is a radar echo map of a certain marine aquaculture area;

[0029] Figure 3This is the effect diagram of moving target detection;

[0030] Figure 4 This is a radar echo map near a coastal bridge;

[0031] Figure 5 This is a picture of a river channel and the reflected echoes on both sides that cannot detect small targets;

[0032] Figure 6 This is a comparison chart of the initial (background difference) moving target detection and the effect without application;

[0033] Figure 7 Rendering of a boat crossing the bridge for stable tracking;

[0034] Figure 8 This is a diagram showing the stable tracking effect of a ship in a 10km area;

[0035] Figure 9 To stably track the stationary (hanging) target after movement;

[0036] Figure 10 The effect diagram of stable tracking of slow and small targets. DETAILED DESCRIPTION

[0037] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0038] The embodiment of the present invention discloses a method and system for detecting moving targets in a strong sea clutter environment. Figure 1 As shown, the system includes the following steps: Step S1: Compare the radar echo data to the sample library to perform preliminary separation of clutter and target. The system uses optical detection (VIBE) to achieve radar target and clutter separation. Based on radar data characteristics and combined with optical detection, the radar scans multiple frames and compares the radar data to the sample library to perform preliminary target separation or update the sample library.

[0039] Based on the characteristics of navigation radar echo data, optical detection (VIBE) is used to implement an effective radar target and clutter separation algorithm. Combining the advantages of optical detection, the radar scans multiple frames, and the sample library, which is primarily clutter, is then repeatedly compared with the sample library. If the incoming point data exceeds the sample library data multiple times, the incoming point is identified as a foreground candidate, and the target is initially separated. Otherwise, the incoming point is considered background, and the clutter sample library is updated using a novel clutter update method. Based on the characteristics of radar data (such as bearing lines and range resolution) and echo characteristics, the video image is shaped like a ring (with a smaller inner radius and a larger outer radius). The echo input is a sequence of range sampling points in azimuth angles, and then a sequence of range echoes in azimuth angles is added, with the scan cycle continuing 360 degrees. Using optical detection (VIBE), radar targets are separated from clutter. By scanning multiple radar frames (starting from true north, with each frame measuring 360 degrees), the radar echo data to be measured is compared with a sample library (the minimum data from several previous frames, called background, which can be considered clutter). If the value exceeds a certain threshold, the echo is considered a foreground candidate and preliminarily identified as a target. Otherwise, the sample library (background) is updated. This achieves preliminary target separation or updates the sample library. By utilizing a larger sample library of echo background (smaller echo data) frames, historical knowledge of point location echoes is fully utilized, significantly increasing separation reliability.

[0040] Using navigation radar to detect targets on the sea surface is of great significance. Sea surfaces and rivers are characterized by strong clutter environments. Due to the complex mechanisms of clutter formation, detecting small, slow targets under various clutter distribution assumptions has always been a global technical challenge. Model-based radar methods for detecting small, slow targets need to overcome relevant technical bottlenecks. Considering the optical detection method VIBE (Visual Background Extractor), which has achieved some noise suppression results. Proposed by Dliver Barnich in 2009, VIBE is a background modeling method consisting of three main components: background initialization, foreground target segmentation (determination), and background model updating. Currently, moving target detection algorithms are classified into three categories: optical flow, inter-frame difference, and background difference. The ViBe-based background difference method proposed by Barnich et al. offers advantages such as low computational complexity, high speed, robustness, and noise immunity. While the ViBe algorithm offers good overall performance, its unique background model initialization and update mechanism can produce "ghosting" artifacts, which can interfere with subsequent moving target detection. Optical images differ significantly from navigation radar images. The resolution of optical images is uniform, with one frame of image generated at a time. Due to the significant differences between optical target detection and radar, improvements are required in several areas to achieve optimal target detection. Radar images are scanned images centered on the radar (increasing in azimuth), and their resolution is uneven, decreasing near the center and increasing away from it. One frame of radar image data must be processed sequentially (gradually, in a sequential order, as the azimuth increases from 0 degrees to 360 degrees). For radar target detection applications, the following improvements are required: initialization sample data must conform to the radar's sampling format; data must be gradually expanded and circulated (in 360-degree azimuth). High point resolution is achieved at long distances, and the radar pulse repetition frequency is high, resulting in data interleaving. The resolution is calculated based on the distance between the point and the center (used in the subsequent calculation of the target size); then foreground extraction (RVIFE, the full name of which is Radar Visual Foreground Extractor, is used in Chinese) is performed to achieve preliminary separation of the target and clutter.

[0041] Initial separation of target, clutter and noise in echo: Initialization: The time of one circle (frame) of navigation radar scanning is determined by the antenna speed ω (unit: revolutions per minute, the time required for one circle is ) is determined by the maximum detection distance R max Determine the radar pulse transmission repetition frequency f (unit: Hz; Where c is the speed of light); the emitted pulse width τ basically determines the distance resolution of the target and radar working (the channel for receiving echo signal is closed when pulse is transmitted) blind area The length of the radar antenna L (unit: m) determines the azimuth line resolution of the target Where f0 and λ are the operating frequency and wavelength of the radar respectively.

[0042] Assume that the radar working distance range is R min ≤R≤R max ; R represents the radar working distance; the azimuth range is 0≤β<360°; β represents the azimuth. If ω=24 turns, it is equivalent to a radar scan of one frame in 2.5s, which is about 144 degrees per second. The radar's transmit pulse, such as 2250Hz, has about 15.625 radar transmit pulse cycles per degree in 360-degree azimuth. Due to the uniform rotation of the antenna and the pulse iso-periodic transmission, a frame (360 degrees) can have a maximum of i=5625 main pulses transmitted. Assume that each main pulse samples a maximum of j=4000 points (if R max =20km, equivalent to 5m accuracy, smaller comparable distance resolution), at the position of target echo (i, j), there are n=30 sample library data (from n rounds of target scanning echoes, since each round takes 2.5 seconds, a total of 75 seconds is required, and the startup time is a bit long. The data sampled in the first three rounds and the data around a certain point can be used for average filling, and the subsequent processing gradually replaces the sample library without much impact), and the echo (amplitude) value (radar echo intensity) of the (i, j) position is: h(i,j,k), k=1,2,…n, where k represents the clutter depth and n is the number of sample libraries in the sample library.

[0043] Based on the image segmentation method, the radar echo at a certain point is also divided into foreground and background. The foreground represents the real target, and the background represents clutter and noise. The target echo in the sample library at any position is temporarily considered to be clutter background.

[0044] Step S2: Detect the target based on the preliminary separation of clutter and target. The dual-channel image convolution method is used to effectively detect moving targets and stationary targets. The motion and shape characteristics of the target are introduced, and a large area set according to the motion characteristics of typical targets in the sea area and a sub-image convolution statistical threshold set according to the shape characteristics of typical targets are adopted to detect moving targets based on the motion characteristics of the previous and next frames; multiple frames of echo accumulation and a small area set according to the motion characteristics of slow targets are selected, and after accumulation, the sub-image convolution statistical threshold is used to judge the stationary target, that is, accumulation is first performed and then the sub-image convolution statistical threshold is implemented to judge the stationary target. Dual-channel processing of moving targets and stationary targets is used to detect moving targets and stationary targets, that is, effective detection of moving targets and stationary targets (in specific areas) is achieved.

[0045] In the moving target detection channel, the motion and shape characteristics of typical targets are introduced. Large-area sub-image convolution statistics and threshold discrimination are used to detect moving targets based on the target motion characteristics of the preceding and following frames. This approach primarily uses the target's motion characteristics to select a large area, and the target's shape to select the sub-image shape. The sub-image is then slid within the large area to count the number of foreground candidates. If the number exceeds a certain threshold, the target is considered. The judgment logic here is that the echo of a truly moving target is connected by directional curved motion between the preceding and following frames.

[0046] The issue that needs to be explained here is that stationary targets on the sea surface are not stationary like those on land. Stationary targets on the sea surface (such as anchored ships or buoys) move slowly without any direction (such as left and right, front and back, circling, etc.), and may even move regularly or irregularly within a very small circle. Even some navigation marks (extending from the seabed and not moving) have radar echoes (with random errors) with this irregular factor. The echoes of such moving targets cannot pass through the moving target detection channel, and a new stationary target detection channel needs to be established. Because during the detection of moving targets, the stationary targets are considered to be background and are eliminated. The stationary target detection mentioned above refers to stationary targets outside the radar target shielding area or in a designated area of ​​interest or specific area (such as buoys on the waterway).

[0047] In the stationary target detection channel, multi-frame echo accumulation and small-area sub-image convolution statistical threshold judgment are selected for small targets. This differs from the moving target detection channel in that: a small area that matches the motion characteristics of slow-moving targets is required; sub-images are set based on echo size; a small echo or too few foreground candidate points in a small area indicates that the target is too small and requires multi-frame accumulation processing; irregular forward and backward motion can be detected by thresholding; and if the target stops at a certain location based on its historical trajectory, the judgment threshold at that location can be lowered, which is beneficial for stationary target detection.

[0048] To further enhance the detection of moving targets, we improved the relevant architectures, models, and algorithms. We primarily introduced the target's motion and shape characteristics, adopted a large-area sub-image convolution statistical thresholding method, and further enhanced the recognition of moving targets based on the target's motion characteristics in the previous and next frames.

[0049] To further improve the detection of stationary and small targets, a model algorithm is needed that combines the target's motion and appearance characteristics, using foreground candidates from the previous frame to further determine the target's maneuverability at the current location. A multi-frame echo accumulation and small-area sub-image convolution statistical thresholding method are used. Furthermore, a dual-channel processing method for moving, stationary, and small targets is employed to achieve effective detection and tracking of moving, stationary, and small targets.

[0050] Segmentation operation, preliminary judgment and correlation of target and clutter: Preliminary judgment: When the radar scan main pulse arrives, the value of a certain (i, j) position represents the i-th azimuth and j-th distance sampling value (echo intensity) h + (i, j) is compared with the h(i, j, k) sample library (k = 1, 2, ... n, equivalent to the clutter depth) array at this position one by one, and the number of array values ​​(of the (i, j) position) whose echo intensity at this position is greater than or equal to that in the sample library is m ij The sample target threshold ratio (probability) is recorded as λ0 (temporarily set to 0.7), and the number of times greater than or equal to can be expressed as:

[0051]

[0052] in, is a step function. x is an arbitrary independent variable.

[0053] If m is satisfied ij If nλ0 is greater than or equal to nλ0, the echo at that position is considered to be a foreground candidate (tentative) target, otherwise the echo is considered to be background clutter. This means that the probability that the echo is judged to be a foreground candidate is 1-λ0.

[0054] Set the foreground candidate position recording area:

[0055]

[0056] Among them, g - (i, j) is the foreground candidate position recording area of ​​the previous frame.

[0057] If the target at that location in the previous frame is also a foreground candidate, the foreground candidate location in this frame will be recorded for use in the next frame, and the relevant program will be triggered to start.

[0058] If m is satisfied ij <nλ0, the above formula is judged as clutter background, and this trigger starts the update program.

[0059] Related procedures: With the position (i, j) of a foreground candidate target g(i, j) as the center, call the cache g of the previous frame -(i, j) area, the total number of cells around the cell (i, j) (including this cell, based on the size of the typical target, the number of cells n1 occupied by the azimuth line, the number of cells n2 occupied by the length of the distance, and adjusted appropriately according to the target type) is N = n1 × n2 squares (appropriately increased to n1, n2 are odd numbers greater than 1; for the sake of convenience, the quantity before the multiplication sign is the azimuth interval, the corresponding azimuth line length should be the azimuth angle multiplied by the distance of the echo point, and the quantity after the multiplication sign is the distance interval point. A fan-shaped area is formed in the polar coordinate plane, and the subsequent similarity will not be explained. Here it is equivalent to setting a candidate area slightly larger than the target, and N is the total number of small squares in the square), with (i, j) as the center, calculate g - The number of "1"s in the N=n1×n2 cells around (i, j). Let the total number of foreground candidates in the previous frame be N ij (less than N), let the threshold ratio (probability) of the foreground candidate target in the grid (n1×n2 cells, the central cell is (i, j)) be recorded as λ1 (a value can be set temporarily and can be adaptively adjusted later).

[0060]

[0061] Among them, l and m are independent variables, representing the displacement centered at (i, j).

[0062] When N is satisfied ij ≥Nλ1, the foreground candidate target at position (i, j) is considered to be the real target, and the system output value h out (i,j)=h + (i,j). Returns to the calling program.

[0063] When N is satisfied ij <Nλ1, the foreground target at this position is considered not to be a real target based on the previous frame, and the system output value is: h out (i, j) = 0. This means that the output of the target is temporarily suspended. Return to the calling program.

[0064] Update procedure: Based on the image update method, if the echo value at a certain position is determined to be clutter background, if the probability p0 is set (such as 0.2), a value in the sample library is randomly replaced with this probability.

[0065] In radar clutter background processing, other aspects are the same, except that the earliest background value in the sample library is replaced sequentially (specifically: h(i,j,k-1)=h(i,j,k), k=2,3,…,n, and h(i,j,n)=h +(i, j). Or another way: suppose we add an old pointer k1, set the record area: z(i, j) = k1; 1≤k1≤n, the pointer specifies the oldest value, and each time it is triggered, the new value replaces the oldest value h(i, j, k1 = h + (i,j), pointer cyclic update When the background is cluttered, the trigger update program is started. After about n updates, the entire sample library at a certain location can be replaced.

[0066] Adaptive radar echo image update method: After reading (updating) all distances at a certain position (i, j), the distance values ​​at the next position become the end line of the current (unupdated) frame. The position and end line of the current frame rotate clockwise. When the distance of the next position arrives, all point data frames at that position (including all buffers, sample libraries, data records, etc.) are recorded in the corresponding data record of the previous (previous) frame for subsequent use.

[0067] Subgraph convolution operations, preliminary identification of moving targets, and correlation: Regions are set based on target characteristics: A large candidate region is set for moving targets based on target speed (an M1×M2 region surrounding the foreground point, where M1 and M2 represent the maximum distance a typical target travels during the scan, approximately the number of cells in the azimuth and distance units; this can be adjusted based on the motion data of specific targets). For example, if the target's speed is relevant, consider a speedboat, for example, which typically has a maximum speed of 40–50 knots and travels approximately 50–60 meters in 2.5 seconds, occupying approximately 24×24 cells. Targets moving within this large candidate region remain within this circle. Because the end line of the current frame (not updated) may intersect with the large candidate region, the subsequent processing of the azimuth requires a certain delay.

[0068] Based on the target's shape, set the target sliding window (a × b) cells. The theoretical size is the longest side of the target, such as the length of the target ship. A near-square shape can be used, slightly larger than the ship itself. The target ship (hull) size is (c × d). Where a and c are the number of bearing cells occupied by the target, respectively. b and d are the number of range cells occupied by the target, respectively.

[0069] Convolution statistics calculation and threshold benchmark: In the M1×M2 large candidate area around the foreground posterior selection point (i, j), a sliding window (a×b) is used to slide through the foreground candidate position record area g of the M1×M2 unit of the large candidate area around the foreground posterior selection point (i, j) - (i, j), each time it slides (up or down or left or right) one grid, the number of "1" in the foreground candidate position record area of ​​the calculation window (centered at (k, l)) is n kl .

[0070] The main reference is the speedboat target value, and other situations can be determined based on the target characteristics. For all foreground points, temporarily set the background probability λ2 = 0.8 to 80%. The probability of judging the foreground is 20%. Its threshold reference value: N ij =cdλ2+M1M2(1-λ2).

[0071] Target determination and discussion of related issues: When n kl ≥N ij , the foreground candidate point (i, j) is considered the true target. Based on the distance between the two center positions, the target's motion speed can be estimated, and the size of the large candidate area can be adjusted in the next frame. Since it is a target, the target's contour circle can be recorded and extracted at the foreground candidate position in the current and previous frames. This is important for extended target tracking. Due to the deep (foreground) sample library (n = 30), the moving target detection effect is statistically analyzed experimentally: 0.1 knot (very slow) unidirectional targets can be detected in clutter areas.

[0072] For non-directional, oscillating, stationary targets (such as buoys and anchored target vessels), the above inequalities may not be satisfied many times. This can lead to loss of tracking or track disappearance due to non-satisfactory tracking conditions. This requires further improvement and processing. This (abnormal) information needs to be input into another parallel module.

[0073] If no sliding window region can be found that satisfies the above inequality, it can be directly determined as clutter and no further processing is performed. For some special areas, the foreground point (i, j) can be determined as a suspected target, which requires multiple frames of observation to confirm.

[0074] Generally speaking, increasing the threshold value increases the target processing miss rate (true targets misidentified as clutter) and decreases the false alarm rate (clutter misidentified as true targets). Lowering the threshold value decreases the miss rate and increases the false alarm rate. Depending on the application and sea area, appropriate adjustments may be required.

[0075] Step S3: Based on statistical analysis of the clutter background and various target characteristics, relevant thresholds and parameters are adaptively adjusted to detect and track various targets, thereby improving the detection and tracking capabilities of various targets. To further enhance the system's adaptability to various environments and targets, necessary parameter statistics are collected for the clutter environment, target motion characteristics, and echo characteristics of stationary targets. Relevant thresholds and system parameters are adjusted accordingly to enhance the method's adaptability to these environments and targets. Adaptive threshold and parameter processing: Parameter statistics are collected by region for the clutter environment, by motion characteristics for moving targets, and by history and echo characteristics for stationary targets. The discrimination thresholds and relevant parameters are adjusted accordingly to detect and track various targets. Specifically, the discrimination thresholds and relevant parameters for typical targets are adjusted accordingly, significantly improving the detection and tracking capabilities of slow and small targets.

[0076] Adaptive threshold and parameter processing: The system adaptively sets and adjusts thresholds based on regional statistics in clutter environments, facilitating the initial separation of foreground candidates from the background. Statistical analysis is performed on small foreground candidate targets, flags are set, and multiple frames of foreground candidate points are accumulated or thresholds are lowered before determining whether a moving or stationary target is detected. The target tracking module provides necessary target track information (position and speed) to facilitate targeted threshold adjustment at a specific location and the setting of large and small regions and sub-images. This enables effective detection of adaptive clutter environments and slow, small targets.

[0077] Several special issues and improvements: Detection and tracking of stationary and small targets: There are two situations for stationary targets. One is the transition from a moving target to a stationary state. In this case, the necessary speed, profile, and position information is fed back by the track tracking module. Lowering the threshold reference value of the target position can increase the probability of detection. The other is a stationary target that may experience small, non-directional swings or rotations in a circle. This requires the use of an accumulation method or an automatic threshold method, parallel implementation of small candidate area convolution, introduction of (multi-frame) amplitude accumulation (very effective for small target detection), and extraction of the first-order moment of the echo center to comprehensively determine the stationary state of the target. This plays an important role in small target detection.

[0078] The input of the static target processing module has two cases. First, it is directly input from the foreground candidate and is parallel to the input of the moving target detection module. Second, the abnormal state output from the moving target detection module is used as the input of the static channel. The sub-graph convolution calculation is similar, but it uses the method of small candidate area (a×b) and sub-graph (c×d). Its threshold reference value is: N ij =cdλ3+ab(1-λ3). λ3 is the background probability of the channel. This parameter needs to be included in the adaptive parameter adjustment range. Other methods are basically the same as those for moving target detection.

[0079] In parallel with the above method, moving target and stationary target point detection methods are respectively input into the multi-hypothesis tracking module for further tracking processing. The tracking processing module can further reduce the false alarm rate and missed detection rate of the track.

[0080] Parameter statistics and threshold adjustment: The target information sent by the target tracking module can be used to calculate the target's motion state and parameters. This information can be used to add reference information to previous target detection, adjust the threshold of relevant positions, and introduce new judgment parameters. This helps stabilize the new sequence points of the tracked target and obtain the target outline. This can maximize the stability and reliability of target detection and tracking.

[0081] The results of the sub-image convolution can be used to further calculate the distribution of sliding windows within the large candidate region that meet the threshold. If there is only one sliding window in the large candidate region, the velocity vector at that point can be counted to appropriately change the shape of the large candidate region in the next frame, significantly reducing the amount of sliding window calculations. If several sliding windows meet the threshold requirement, this is due to a large number of close targets, extended targets, or strong clutter. New decision logic needs to be established, or the mean and variance of the large candidate region can be counted to increase the threshold of the sub-image convolution.

[0082] In a (special) large area where the radar's bearing lines and range resolution are per unit area, the mean and variance of all foreground signals are calculated to determine the clutter threshold for that area. This adaptive processing plays a significant role in improving system performance. For small targets, a lower threshold can be established, and multi-frame accumulation methods can be used to increase the echo amplitude. The threshold can then be further adjusted based on subsequent statistical information or target track information.

[0083] To speed up target detection output, during initialization, the average (smaller value) of the foreground area in the first few frames can be used as multiple foreground inputs. In the tracking module, the starting and following conditions can be reduced to speed up target tracking.

[0084] To reduce computational complexity, we believe that moving target detection can adopt a first-come, first-served approach. When a high level of external information input is used to form predictions, computational complexity can be significantly reduced. Therefore, when providing target feedback, reference information should be provided whenever possible. Threshold calculation and parameter setting must be considered in conjunction with the clutter environment, target characteristics, and the specific characteristics of specific sea areas, rivers, lakes, and other regions. Comprehensive consideration, system optimization, and automatic adaptation are required based on various circumstances.

[0085] The four modules in the detection have different focuses. The first module only recognizes echoes, not targets, and obtains foreground candidate points. The second module inputs foreground candidates, considers the characteristics, shape, motion parameters, etc. of moving targets for target detection, and outputs moving target traces. The third module inputs foreground candidates, considers the characteristics, shape, drift parameters, etc. of stationary targets for target detection, and outputs stationary target traces. Moving and stationary target traces are input into the tracking part. The fourth module performs statistical analysis on the target track parameters, target characteristics, clutter parameters and other information of the tracking part, and adaptively adjusts the clutter threshold, small target threshold, correlation calculation parameters, area and sub-image parameters, etc., to achieve effective detection of slow and small targets. The tracking part further improves the target track processing effect. The tracking part is the output object of the detection part.

[0086] The present invention includes utilizing the optical detection VIBE (Visual Background Extractor) method to achieve preliminary separation of the radar clutter environment and the target, performing regional convolution on large and small areas of the measurement point based on target information and determining target existence through statistical parameters, and effectively achieving separation of the target from the background and detection of moving and stationary targets based on statistical data related to the motion characteristics of the target in the previous and next frames.

[0087] Legend and Notes: Figure 2 As shown in the radar echo map of a certain aquaculture area, the radar strong clutter area (the part circled by the dotted box in the figure is the aquaculture area) presents a large number of clutter areas, where targets exist but cannot be tracked, and small boats cannot be tracked.

[0088] like Figure 3 As shown in the image below, the effect of moving target detection is shown. A large number of water-filled buoys create a patchy radar echo, even blurring the radar echo of a small fishing boat. In the aquaculture area, strong clutter and non-coherent moving target detection of a small wooden boat (left image: radar image, upper right image: photoelectric image, lower right image: infrared image) reveal the presence of a listed target within the area and the tracks of several small boats.

[0089] like Figure 4 The following figure shows a radar echo map near a coastal bridge. The radar is positioned at the lower right corner of the bridge. Three kilometers away, there are virtually no raw echoes from small boats. Ten kilometers away, the (dark) radar echoes completely indistinguish the river channel and its banks. Fishery administration vessels (7-meter-long, fiberglass-clad vessels with numerous metal objects) can only be accurately observed when they approach the radar from a distance. Fishing boats can be followed when they are not docked, but are difficult to track near shore. The dense echo area on the right is not caused by a fixed target, but rather by radial reflections from the vessel. It can be seen that the intensity is higher near the radar, and the tail gradually decreases away from the radar. The artificial circle represents the radar target.

[0090] like Figure 5 As shown, the echoes from a certain river channel and its sides are unable to detect small targets. While the original echoes can stably track large ships in the channel, the echoes from small boats near the bank are obscured by reflected clutter or mixed with other clutter, making them virtually invisible and preventing effective target detection. The land-based shielding zone method can only suppress targets on the river channel; traditional methods are difficult to suppress clutter from the banks and reflections into the river channel.

[0091] like Figure 6The following figure compares the effects of preliminary (background difference) moving target detection with those without it. The left side shows the effect of the most basic (background difference) moving target detection image. All stationary targets are eliminated, moving targets are detected, and a large number of residual echoes remain. The right side shows the image without moving target processing, and the overall echo intensity in the corresponding area is very high. Conventional detection algorithms cannot obtain a stable cross-threshold signal, resulting in frequent tracking interruptions. Further improvement of moving target algorithms is needed.

[0092] like Figure 7 The image below shows the effect of stable tracking of a boat crossing a bridge. Using moving target detection, the boat moves from left to right under the bridge, indicated by a track line (the black line may be a bit faint). Despite strong reflected clutter, stable tracking is achieved, and the track is relatively smooth. The clutter shown in the image is the remaining part after processing.

[0093] like Figure 8 Figure 2 shows the stable tracking of a ship in a 10km range. Using moving target detection, stable tracking of targets can be achieved even at long radar distances. Four markers provide tracking information for four targets.

[0094] like Figure 9 As shown in the figure, the effect of stable tracking of a target that is stationary (with a sign) after movement, the application of stationary target detection can achieve tracking of targets that are always stationary or stationary after movement, and the sign indicates the specific information of the target track.

[0095] like Figure 10 As shown in the figure, the system can track slow and small targets. By combining moving and stationary target detection methods, it can track buoys, slow and small targets, and even some floating objects. Tracking floating objects on the water surface shows that the system can track targets with very low reflections and very slow movements.

[0096] Implementation process and improvement method: The present invention first improves the VIBE method based on the characteristics of radar echoes, selects new parameters, preliminarily establishes a radar clutter background environment, and uses thresholds to achieve preliminary separation of the target (foreground) from the clutter (background), obtaining foreground candidates. In its initial application, it was found that the separation effect was good for large target echoes. However, target motion detection was not possible. The first improvement method: A large candidate area and sub-image convolution method and threshold are used to determine the position of the target in the previous frame, and the moving target is further confirmed according to the motion characteristics. This is very effective for high-speed targets. Due to the depth of the sample library, by gradually adjusting the relevant parameters, very slow, unidirectional moving targets can be detected. However, for stationary targets with small, non-directional swings or rotations (this is the echo characteristic of anchored ships on the sea surface, and the deeper the sea, the larger the target's swing amplitude may be), it cannot pass the motion target detection, and the target disappears during tracking. Experimental analysis shows that stationary targets on the sea surface are relatively stationary, and it is necessary to design a processing channel specifically for detecting stationary targets. The second improved method utilizes small candidate regions and sub-image convolution and thresholding, introducing (multi-frame) amplitude accumulation and echo center first-order moment optimization to improve small target preprocessing. This approach forms a parallel processing relationship with the moving target detection module. The outputs of these two modules are fed into the multi-hypothesis tracking module for further processing, effectively tracking both moving and stationary targets. Because the thresholds and parameters used in the aforementioned processing model primarily focus on typical targets and clutter environments, they are not adaptable to a wider range of sea clutter environments and target characteristics. Therefore, adaptive adjustment of the thresholds and related parameters is required. The third improved method utilizes the intermediate and output data of each module to perform statistical analysis of the characteristics of moving and stationary targets and clutter environments. This method obtains numerical features such as the mean, variance, and covariance of clutter in a specific region, as well as the motion parameters and profile of specific targets. The corresponding thresholds, regional parameters, and sub-image parameters are then adjusted based on the corresponding echo points to provide special attention to specific targets. For small target detection, dual detection thresholds for typical and small targets can be set in the initial foreground and background separation module, and preprocessed on the input echoes to enhance detection effectiveness. Many minor problems were found in the actual test, and the software modules were modified and improved accordingly. From the actual test results, this method is effective.

[0097] The present invention is used for moving target detection in a clutter environment. The implementation principle and improvement method are given, and the implementation process of the system is described in detail and the problems are discussed. After preliminary verification, the method is improved several times. The implemented module is applied in a multi-element joint perception system, achieving good moving target and stationary target detection effects, and greatly improving the detection and tracking capabilities of slow and small targets.

[0098] Radar echo detection based on optical detection methods has certain practical significance. However, the optical detection VIBE method already has a suppressive effect. After years of dedicated research, based on actual sea and river surface environments, a unique algorithm module was designed. This improved VIBE method and subgraph convolution-related methods, combining radar echo characteristics, achieves effective detection of both moving and stationary targets, and enhances the detection of slow and small targets. Continuous technical improvements are made during application. Multiple practical application tests have demonstrated the effectiveness of this proposed method.

[0099] The present invention solves the problem of moving target detection in a strong sea clutter environment. In aquaculture areas, there are a large number of water-carrying floats, and the radar echoes from these floats seriously interfere with radar target detection. Based on the different radar echo characteristics of sea surface targets, clutter, noise, etc., and referring to the optical detection VIBE (Visual Background Extractor) method and adaptive improvements, regional convolution is performed on a large area of ​​the measurement point based on target information and statistical parameters are used to determine the presence of the target, thus realizing a method for detecting and tracking moving targets in a strong clutter environment. VIBE was proposed by Dliver Barnich in 2009. It is a modeling method based on image background. The method mainly consists of (background) initialization, foreground target segmentation operation (determination), and background model. Due to the significant differences between optical and radar target detection, the team employed a foreground extraction (RVIFE) approach. Based on radar target performance, they implemented technical improvements such as automatic parameter extraction, sub-image convolution, and adaptive dual thresholding. This enabled them to create dynamic clutter maps for water targets in strong clutter environments and developed a moving target detection and processing module. Application testing significantly improved the detection and tracking of targets in areas such as aquaculture areas (where there is a lot of buoy clutter) and river bridges (where there is a lot of reflected clutter).

[0100] An embodiment of the present invention also discloses a system for detecting moving targets in a strong sea clutter environment, comprising the following modules: Module M1: Compares the radar echo data to a sample library to perform preliminary separation of clutter and targets. This separation is achieved using optical detection (VIBE). Based on radar data characteristics and combined with optical detection, the radar scans multiple frames. The radar data is then compared with the sample library to perform preliminary target separation or update the sample library.

[0101] like Figure 1As shown, the system includes submodules for database construction, initialization (sample database), first correlation calculation, first threshold determination, and update (Yangben database). Based on radar echo characteristics and video image principles, and utilizing the fundamental principles of optical detection (VIBE), this method achieves a preliminary separation of foreground candidates from background clutter in circular scanning echo images. Differences from optical detection include image shape and scanning method, distinct differences in foreground and background between radar and optical methods, range-dependent resolution of azimuth lines (higher resolution at close range, lower at greater distances), and consideration of clutter characteristics in sample database updates.

[0102] Module M2: Target detection based on preliminary separation of clutter and target. Effective detection of moving targets based on this preliminary separation. Effective detection of stationary targets based on this preliminary separation. By incorporating the target's motion and shape characteristics, large-area sub-image convolution statistical thresholding is used to discriminate moving targets based on the target's motion characteristics in the preceding and following frames. Multi-frame echo accumulation and small regions are selected, and after accumulation, sub-image convolution statistical thresholding is applied to identify stationary targets. Dual-channel processing of moving and stationary targets is used to detect both moving and stationary targets.

[0103] like Figure 1 As shown in the figure, it includes foreground candidates (large candidate area), sub-image convolution, second correlation calculation (sliding window accumulation), second threshold judgment, and moving position. By introducing the target's motion and shape characteristics, a statistical threshold judgment is adopted for large-area sub-image convolution. Moving targets are detected based on the target motion characteristics of the previous and next frames. Since the foreground candidates come from module M1 and have more background values, very slow targets can be detected, while high-speed targets are easily detected.

[0104] like Figure 1 As shown in the figure, the algorithm includes foreground candidates (small candidate areas), sub-image convolution, third correlation calculation (small target accumulation), third threshold judgment, and motion position. By taking into account the shape characteristics and dynamic irregularities of stationary targets, the algorithm uses multi-frame accumulation of small target candidates and statistical threshold judgment for small-area sub-image convolution. The motion judgment of stationary targets is significantly different from that of moving targets.

[0105] Module M3: Based on statistical analysis of the clutter background and various target characteristics, it adaptively adjusts thresholds and parameters to detect and track various targets. Adaptive statistical adjustment of thresholds and parameters enables effective detection of slow and small targets. Adaptive threshold and parameter processing: Parameter statistics are calculated by region for clutter environments, by motion characteristics for moving targets, and by historical and echo characteristics for stationary targets. Target discrimination thresholds and parameters are adjusted accordingly to detect and track various targets.

[0106] like Figure 1As shown, the system includes modules for clutter parameter statistics, target characteristic statistics, target motion parameter statistics, and adaptive (regional) threshold and parameter calculation. Thresholds are adaptively set and adjusted based on regional statistics in clutter environments, facilitating the initial separation of foreground candidates from the background. A lower threshold can be set (or adjusted based on the tracking target's position) for small targets. A dual threshold is formed in module M1 to achieve initial target separation and detection. Alternatively, statistical analysis is performed on small foreground candidate targets, with flags set and sub-image sizes adjusted to accommodate the identification of small, stationary targets. Multiple frames of foreground candidate points are accumulated or thresholds lowered before determining whether a moving or stationary target is being identified. Based on target feature statistics (including for small targets; the line connecting the third correlation calculation to the target characteristic statistics is not shown in the figure due to excessive crossover), the size and shape of large and small regions, as well as sub-images, can be adjusted. This significantly reduces the search range and convolution computation workload for sub-image convolution of tracked targets. The tracking module provides essential target track information (position and speed) to facilitate targeted adjustment of thresholds for a specific location and the setting of large and small regions and sub-maps. Based on comprehensive statistical results and information feedback, it adaptively adjusts various judgment thresholds, correlation, large and small regions, sub-maps, and other parameters to achieve effective detection of slow and small targets in adaptive clutter environments.

[0107] Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, and units provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same functions of the system and its various devices, modules, and units provided by the present invention in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; the devices, modules, and units for implementing various functions can also be considered as both software modules implementing the method and structures within the hardware component.

[0108] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. A method for detecting moving targets in a strong sea clutter environment, characterized in that: The steps include: Step S1: Compare the radar echo data to be tested with the sample library to perform preliminary separation of clutter and targets; Step S2: Detecting the target based on the preliminary separation of clutter and target; In step S2, the motion and shape characteristics of the target are introduced, and a large-area sub-image convolution statistical threshold is used for discrimination, and the moving target is detected based on the motion characteristics of the target in the previous and next frames; multiple frames of echo accumulation and a small area are selected, and after the accumulation, the sub-image convolution statistical threshold is used to judge the stationary target; Adopt dual-channel processing of moving targets and stationary targets to detect moving targets and stationary targets; According to the target's moving speed, a large candidate area of ​​moving targets is set up, M1×M2, where M1 and M2 are the number of azimuth and distance cells occupied by the maximum distance traveled by a typical target during the scanning time. According to the target shape, the target sliding window is set to a×b units, and the size of the target hull is c×d, where a and c are the number of azimuth cells occupied by the target, and b and d are the number of distance cells occupied by the target; In the M1×M2 large candidate area around the foreground candidate point (i, j), a sliding window a×b is used to slide through the foreground candidate position record area g(i, j) of the M1×M2 unit in the large candidate area around the foreground candidate point (i, j). The number of "1"s n in the foreground candidate position record area is calculated every time the window slides one grid. kl ; When n kl ≥N ij , threshold value of moving target: N ij =cdλ2+ M1 M2 (1-λ2), where λ2 is the background probability of the moving target channel. The foreground posterior point (i, j) is considered to be the true target. Based on the distance relationship between the two center positions, the target's moving speed is estimated, and the size of the large candidate area is adjusted in the next frame. For stationary targets, a small candidate area a×b and a sub-image c×d sub-image convolution calculation method are used. The threshold reference value of the stationary target is: N ， ij =cdλ3+ab(1-λ3), where λ3 is the background probability of the stationary target channel.

2. The method for detecting moving targets in a strong sea clutter environment according to claim 1, characterized in that: In step S1, optical detection VIBE is used to separate radar targets from clutter. Based on the characteristics of radar data and combined with optical detection, the radar data to be tested is compared with the sample library through multi-frame scanning to preliminarily separate the target or update the sample library.

3. The method for detecting moving targets in a strong sea clutter environment according to claim 1, wherein: The method further comprises step S3: performing statistical analysis based on the clutter background and characteristics of various targets, adaptively adjusting relevant thresholds and parameters, and detecting and tracking various targets.

4. The method for detecting moving targets in a strong sea clutter environment according to claim 3, characterized in that: In step S3, adaptive threshold and parameter processing are performed: parameter statistics are performed on the clutter environment by area, on the moving targets by motion characteristics, and on the stationary targets by history and echo characteristics. The discrimination thresholds of the relevant targets are adjusted accordingly and relevant parameters are set to detect and track various targets.

5. A moving target detection system in a strong sea clutter environment, characterized by: Includes the following modules: Module M1: compares the radar echo data to be tested with the sample library to perform preliminary separation of clutter and targets; Module M2: Target detection based on preliminary separation of clutter and target; In the module M2, the motion and shape characteristics of the target are introduced, and a large-area sub-image convolution statistical threshold is adopted to detect the moving target based on the target motion characteristics of the previous and next frames; multiple frames of echo accumulation and a small area are selected, and after accumulation, a sub-image convolution statistical threshold is applied to determine the stationary target; Adopt dual-channel processing of moving targets and stationary targets to detect moving targets and stationary targets; According to the target's moving speed, a large candidate area of ​​moving targets is set up, M1×M2, where M1 and M2 are the number of azimuth and distance cells occupied by the maximum distance traveled by a typical target during the scanning time. According to the target shape, the target sliding window is set to a×b units, and the size of the target hull is c×d, where a and c are the number of azimuth cells occupied by the target, and b and d are the number of distance cells occupied by the target; In the M1×M2 large candidate area around the foreground candidate point (i, j), a sliding window a×b is used to slide through the foreground candidate position record area g(i, j) of the M1×M2 unit in the large candidate area around the foreground candidate point (i, j). The number of "1"s n in the foreground candidate position record area is calculated every time the window slides one grid. kl ; When n kl ≥N ij , threshold value of moving target: N ij =cdλ2+ M1 M2 (1-λ2), where λ2 is the background probability of the moving target channel. The foreground posterior point (i, j) is considered to be the true target. Based on the distance relationship between the two center positions, the target's moving speed is estimated, and the size of the large candidate area is adjusted in the next frame. For stationary targets, a small candidate area a×b and a sub-image c×d sub-image convolution calculation method are used. The threshold reference value of the stationary target is: N ， ij =cdλ3+ab(1-λ3), where λ3 is the background probability of the stationary target channel.

6. The moving target detection implementation system in a strong sea clutter environment according to claim 5 is characterized in that: In the module M1, optical detection VIBE is used to separate radar targets from clutter. Based on the characteristics of radar data and combined with optical detection, the radar data to be tested is compared with the sample library through multi-frame scanning to preliminarily separate the target or update the sample library.

7. The system for detecting moving targets in a strong sea clutter environment according to claim 5, characterized in that: The system also includes module M3: based on the clutter background and various target characteristics, statistical analysis is performed, and relevant thresholds and parameters are adaptively adjusted to detect and track various targets.

8. The moving target detection system in a strong sea clutter environment according to claim 7 is characterized in that: In the module M3, adaptive threshold and parameter processing are performed: parameter statistics are performed on the clutter environment by area, on the moving targets by motion characteristics, and on the stationary targets by history and echo characteristics. The discrimination thresholds of the relevant targets are adjusted accordingly and relevant parameters are set to detect and track various targets.

Images