Parameter space multi-channel target search method for weak target detection in sea clutter

Abstract

This invention discloses a parameter-space multi-channel target search technology for weak target detection in sea clutter. For weak target detection in sea clutter, the radar echo signal is first processed, and a low target detection threshold is used for preliminary target detection to obtain raw point information. Utilizing the continuity of target motion in space and the temporal correlation of target echo data from several consecutive frames, the raw points obtained from multiple scanning cycles are accumulated. Considering the diversity of target motion trends, a multi-channel parallel search method is employed to find possible tracks. For each search channel, considering information from multiple dimensions such as range, azimuth, and time, the raw points are batch-processed with parameter transformation. In the parameter space, possible target tracks, i.e., potential tracks, are obtained, and the probability of their existence is calculated. This probability is then used to accurately detect weak targets in the clutter.

Description

Technical Field

[0001] This invention belongs to the field of radar technology, specifically relating to a multi-channel parallel target search technology based on multi-scale parameter transformation. Background Technology

[0002] The traditional task of maritime radar detection is mainly to detect and track ships at sea, primarily targeting targets under conditions of high signal-to-noise ratio. However, in recent years, with changes in the marine environment and the development of marine science and stealth technology, the demand for detecting weak targets on the sea surface, such as small ships, maritime search and rescue vessels, divers, periscopes, unmanned surface vessels, and small drones, has been increasing, but the detection performance of radar is not satisfactory.

[0003] Traditional radar target detection involves processing single-frame echo data with constant false alarm rate (CFAR) before performing threshold detection, extracting data points, and then using data points from multiple frames for target tracking. During threshold detection, a higher threshold increases the false alarm rate, potentially missing weak targets; a lower threshold, while improving detection probability, also increases the false alarm rate and complicates the data association algorithms. Therefore, traditional target detection techniques struggle to reliably detect weak targets on the sea surface.

[0004] Research on detection technologies for weak targets on the sea surface has thus become a current research hotspot. Improving radar's ability to detect weak targets on the sea surface is one of the challenges of modern radar. Track-before-detection (TBD) technology, as an emerging weak target detection technique, accumulates energy from multiple frames of radar echo information to detect weak targets on the sea surface with low signal-to-clutter ratios, and has become a research hotspot for scholars both domestically and internationally. In recent years, many TBD algorithms have been researched by scholars at home and abroad, such as Hough Transform TBD, Dynamic Programming TBD, Particle Filter TBD, Maximum Likelihood Probability TBD, and Random Set TBD. Because radar not only needs to complete target detection in complex environments such as noise, clutter, and interference, but also needs to deal with complex environments such as dense targets and diverse movement trends, TBD technology is still mainly in the theoretical research stage.

[0005] To address the shortcomings of existing radar technologies in detecting weak targets in sea clutter, there is an urgent need to propose a new technology for detecting weak targets in sea clutter. This new technology should achieve higher accuracy than traditional methods and be easy to implement in engineering. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing radar detection technologies in reliably detecting weak targets on the sea surface, and to propose a novel weak target detection technology in sea clutter. This technology utilizes multiple frames of radar echoes for information accumulation, improving the target signal-to-clutter-to-noise ratio. It performs parallel searches in multiple search channels, taking into account information from multiple dimensions such as range, azimuth, and time. It employs parameter transformation batch processing technology to cluster target echo signals and detects weak targets in clutter by calculating the target's presence probability. This achieves higher detection accuracy than traditional methods and is easily implemented in engineering.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] A parameter-space multi-channel target search technique for detecting weak targets in sea clutter is characterized by the following steps: First, the radar echo signal is processed, and a low target detection threshold is used for preliminary target detection to obtain raw point information. Utilizing the continuity of target motion in space and the temporal correlation of target echo data across several consecutive frames, the raw point information obtained from multiple scan cycles is accumulated. Considering the diversity of target motion trends, a multi-channel parallel search method is employed to find potential tracks. For each search channel, considering information from multiple dimensions including range, azimuth, and time, the raw point information undergoes parameter transformation and batch processing. In the parameter space, potential target tracks are obtained, and their existence probability is calculated. This probability is then used to accurately detect weak targets in the clutter.

[0009] This method performs joint processing on multiple frames of data, predicts the target trajectory, and accumulates energy across the multiple frames to preserve target information to the maximum extent, avoiding the signal-to-noise ratio loss caused by signal processing or CFAR in single-frame detection. Simultaneously, this method considers both distance-time and azimuth-time information of the track, thus effectively preventing the generation of false tracks and improving the probability of detecting weak targets in sea clutter while reducing the false alarm rate.

[0010] The preferred scheme for weak target detection in sea clutter employs a sliding window algorithm. First, it accumulates raw data points obtained from multiple scan cycles, sorts them chronologically, selects data points within a certain angular range, and performs a Doppler transformation to convert the distance-time information of the data points into Doppler velocity and intercept information. The transformed intercepts are then sorted, and clustering is used to select potential tracks. For the selected tracks, another parameter transformation is performed to convert the azimuth-time information of the data points into turning rate and intercept information. The same process is applied to data points within other angular ranges. After these two parameter transformations and multi-channel search, tracks that meet the criteria are identified as potential tracks. Simultaneously, a motion trend fitting method is used to obtain the prior probability of each potential track. After searching all channels, all potential tracks are sorted according to their existence probability, and those with higher prior probabilities are identified as confirmed tracks, i.e., target trajectories. Finally, based on the existence probability of potential tracks and the data point sharing relationships among them, weak targets in the clutter are detected.

[0011] The preferred scheme is a weak target detection algorithm in sea clutter that employs parameter transformation batch processing and multi-channel search. This algorithm uses two parameter transformation batch processing steps. First, the observation data in the Cartesian coordinate system is processed... Coordinates transformed into parameter space ,Right now:

[0012] (1)

[0013] In equation (1), For searching a channel. For a point on a straight line. There must be two unique parameters. and satisfy:

[0014] (2)

[0015] A straight line in a Cartesian coordinate system can be obtained through the slope of this line. and intercept To define;

[0016] Measurements with the same slope and intercept can be considered a straight line, i.e., a track. The weak target detection algorithm in sea clutter uses the above method to perform parameter space transformations in both the RT two-dimensional plane and the Az-T two-dimensional plane, and conducts multi-channel searches to find possible tracks. Finally, based on the searched tracks, weak targets in the clutter are detected.

[0017] This algorithm is suitable for detecting targets in cluttered environments. , In parameter space , The optimal intercept size, the performance of this algorithm depends on the accumulation time of the measurements and the parameters. , In two aspects, the longer the accumulation time of the measured values, the higher the quality of the initial track; parameters , A smaller value results in a higher quality initial track, but it also increases the risk of missed alarms. , The selection should be based on the actual measurement error of the radar.

[0018] In a preferred embodiment, the calculation of the existence probability of potential tracks is based on the Probability Hypothesis Density (PHD) to calculate the likelihood of each potential track being a real target, thereby obtaining the prior probability of the potential track, assuming that the potential track has been obtained. Given a measurement value, the prior probability of the potential trajectory is... It can be represented as:

[0019] (3)

[0020] in, It is a PhD with potential flight paths. Let be the probability density of false targets in the radar observation area. When false targets exist in the radar observation area, they are uniformly distributed within the area. If the volume of the radar observation area is... Then the probability density of false targets in the radar observation area In practical radar applications, the radar observation area refers to the range covered by the radar's main beam. It can be represented as:

[0021] (4)

[0022] To obtain the prior probability of a potential trajectory, a PHD filter is needed, which is obtained from the potential trajectory. The PHD filter estimates a number of measurements. The PHD filter has two main computational steps: prediction and update.

[0023] The main operational steps of the PHD filter are as follows:

[0024] (5)

[0025] (6)

[0026] in,

[0027] (1) It is a state belonging to the target state space, which contains information such as position and velocity. It is the first The second detection yields a set of measurement values, which contains the values ​​from the first detection. Measurement information obtained from secondary target detection;

[0028] (2) For the first The updated PhD. For the first The updated PhD. To utilize right The predictions made;

[0029] (3) For from the first Time to the The state transition function at time t, In order to be in The state of the target at any given time can typically be represented by a state transition matrix;

[0030] (4) For the first The measurement function for the next detection, which represents the target state. The obtained target measurement value is This function can typically be represented by a measurement matrix;

[0031] (5) for The probability density of generating new targets in the target state space at any given time. This represents the probability density distribution of clutter. For detection probability, From Time's up The state at time is The probability of the target continuing to survive.

[0032] In the recursive calculation of the PHD filter, it is assumed that the PHD can be represented by a mixture Gaussian distribution. The PhD at time t can be written as:

[0033] (7)

[0034] in, The first in the mixture Gaussian distribution The weights of a Gaussian model, The number of Gaussian models. For the first The expectation of a Gaussian model, For the first The covariance of a Gaussian model. The PHD's predictions and updates can be rewritten as:

[0035] (8)

[0036] (9)

[0037] in, It is a predicted PhD. The number of Gaussian models predicted. For the predicted mixture Gaussian distribution, the th The weights of a Gaussian model, For the predicted first The expectation of a Gaussian model, For the predicted first Covariance of a Gaussian model;

[0038] In the PHD filter update, the following calculations are required:

[0039] (10)

[0040] (11)

[0041] (12)

[0042] (13)

[0043] (14)

[0044] in, Time observation matrix, To measure the noise covariance.

[0045] The beneficial effects of this invention are:

[0046] 1. Use a lower target detection threshold for preliminary target detection to obtain original point information and effectively prevent missed detections;

[0047] 2. Utilize multi-frame radar echoes to accumulate information and improve the target signal-to-clutter-to-noise ratio;

[0048] 3. For each search channel, taking into account multiple dimensions of information such as distance, orientation, and time, perform parameter transformation batch processing on the original points;

[0049] 4. In the preferred parameter space , intercept , To ensure the high quality of the initial flight path and prevent missed alarms;

[0050] 5. Calculate the probability of existence of potential tracks and use the probability of existence to improve the accuracy of detecting weak targets in clutter. Attached Figure Description

[0051] Figure 1 This is a flowchart of an algorithm for detecting weak targets in sea clutter;

[0052] Figure 2 This is a schematic diagram illustrating one application scenario of the present invention;

[0053] Figure 3 This is a schematic diagram illustrating the target detection effect in sea clutter targeted by this invention. Detailed Implementation

[0054] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings: These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0055] Example: A parameter-space multi-channel target search technique for weak target detection in sea clutter. For weak target detection in sea clutter, the radar echo signal is first processed to remove co-frequency asynchronous interference, noise, and some sea clutter interference. A low target detection threshold is used for preliminary target detection to obtain raw point information. Utilizing the continuity of target motion in space and the temporal correlation of target echo data from several consecutive frames, the raw points obtained over multiple scan cycles are accumulated. Considering the diversity of target motion trends, a multi-channel parallel search method is employed to find possible tracks. For each search channel, considering information from multiple dimensions including range, azimuth, and time, the raw points undergo parameter transformation batch processing. In the parameter space, possible target tracks, i.e., potential tracks, are obtained, and the probability of their existence is calculated. This probability is then used to accurately detect weak targets in the clutter.

[0056] This method performs joint processing on multiple frames of data, predicts the target trajectory, and accumulates energy across the multiple frames to preserve target information to the maximum extent, avoiding the signal-to-noise ratio loss caused by signal processing or CFAR in single-frame detection. Simultaneously, this method considers both distance-time and azimuth-time information of the track, thus effectively preventing the generation of false tracks and improving the probability of detecting weak targets in sea clutter while reducing the false alarm rate.

[0057] The algorithm for detecting weak targets in sea clutter is as follows: Figure 1As shown. This algorithm employs a sliding window approach. First, it accumulates raw point traces obtained from multiple scan cycles, sorts them chronologically, selects points within a certain angular range, and performs a parameter space transformation in Doppler dimension, converting the distance-time information of the points into Doppler velocity and intercept 1 information. The transformed intercepts are then sorted, and clustering is used to select potential tracks. For the selected tracks, another parameter space transformation is performed, converting the azimuth-time information of the points into turning rate and intercept 1 information. The same processing method is then applied to points within the remaining angular ranges. See [link to relevant documentation]. Figure 2 .

[0058] After the two parameter space transformations and multi-channel searches described above, tracks that meet the conditions are identified as potential tracks. Simultaneously, a motion trend fitting method is used to obtain the prior probability of each potential track's existence. After completing the search of all channels, all potential tracks are sorted according to their existence probabilities, and those with higher prior probabilities are identified as confirmed tracks, i.e., target trajectories. Finally, based on the existence probabilities of potential tracks and the point-sharing relationships between them, weak targets in clutter are detected.

[0059] The algorithm for detecting weak targets in sea clutter employs a parameter transformation batch processing and multi-channel search method. This algorithm uses two parameter transformation batch processing steps. First, the observation data in the Cartesian coordinate system is processed... Coordinates transformed into parameter space ,Right now:

[0060] (1)

[0061] In equation (1), For searching a channel. For a point on a straight line. There must be two unique parameters. and satisfy:

[0062] (2)

[0063] A straight line in a Cartesian coordinate system can be obtained through the slope of this line. and intercept To define.

[0064] Measurements with the same slope and intercept of 1 can be considered a straight line, i.e., a track. The weak target detection algorithm in sea clutter uses the above method to perform parameter space transformations in both the RT two-dimensional plane and the Az-T two-dimensional plane, and conducts multi-channel searches to find possible tracks. Finally, based on the searched tracks, weak targets in the clutter are detected.

[0065] This algorithm is suitable for detecting targets in cluttered environments. , In parameter space , The intercept size is 1. The performance of this algorithm depends on the accumulation time of the measurements and the parameters. , Two aspects. The longer the accumulation time of the measured values, the higher the quality of the initial track; parameters , A smaller value results in a higher quality initial track, but it also increases the likelihood of missed detections. Parameter , The selection should be based on the actual measurement error of the radar.

[0066] The probability calculation of potential tracks is based on the Probability Hypothesis Density (PHD) to calculate the likelihood of each potential track being a real target, thus obtaining the prior probability of the potential track. It is assumed that the potential tracks have already been obtained. Given a measurement value, the prior probability of the potential trajectory is... It can be represented as:

[0067] (3)

[0068] in, It is a PhD with potential flight paths. Let be the probability density of false targets in the radar observation area. We assume that false targets exist in the radar observation area, and that these false targets follow a uniform distribution within the radar observation area. If the volume of the radar observation area is... Then the probability density of false targets in the radar observation area In practical radar applications, the radar observation area refers to the range covered by the radar's main beam. It can be represented as:

[0069] (4)

[0070] To obtain the prior probability of a potential trajectory, a PHD filter is needed, which is obtained from the potential trajectory. The PHD is estimated using a set of measurements. The PHD filter has two main computational steps: prediction and update. The main computational steps of the PHD filter are shown below:

[0071] (5)

[0072] (6)

[0073] in,

[0074] (1) It is a state belonging to the target state space, which contains information such as position and velocity. It is the first The second detection yields a set of measurement values, which contains the values ​​from the first detection. Measurement information obtained from secondary target detection;

[0075] (2) For the first The updated PhD. For the first The updated PhD. To utilize right The predictions made;

[0076] (3) For from the first Time to the The state transition function at time t, In order to be in The state of the target at any given time can typically be represented by a state transition matrix;

[0077] (4) For the first The measurement function for the next detection, which represents the target state. The obtained target measurement value is This function can typically be represented by a measurement matrix;

[0078] (5) for The probability density of generating new targets in the target state space at any given time. This represents the probability density distribution of clutter. For detection probability, From Time's up The state at time is The probability of the target continuing to survive.

[0079] In the recursive calculation of the PHD filter, it is assumed that the PHD can be represented by a mixture Gaussian distribution. The PhD at time t can be written as:

[0080] (7)

[0081] in, The first in the mixture Gaussian distribution The weights of a Gaussian model, The number of Gaussian models. For the first The expectation of a Gaussian model, For the first The covariance of a Gaussian model. The PHD's predictions and updates can be rewritten as:

[0082] (8)

[0083] (9)

[0084] in, It is a predicted PhD. The number of Gaussian models predicted. For the predicted mixture Gaussian distribution, the th The weights of a Gaussian model, For the predicted first The expectation of a Gaussian model, For the predicted first The covariance of a Gaussian model.

[0085] In the PHD filter update, the following calculations are required:

[0086] (10)

[0087] (11)

[0088] (12)

[0089] (13)

[0090] (14)

[0091] in, Time observation matrix, To measure the noise covariance.

[0092] The embodiments of the present invention have been verified through actual testing that this novel technology can achieve higher accuracy than traditional methods and is easy to implement in engineering.

[0093] The implementation steps of this invention can be summarized as follows:

[0094] Step 1: Perform signal processing on the radar echo data to remove co-frequency asynchronous interference, noise, and some sea clutter interference. Then, use a lower target detection threshold to perform preliminary target detection and obtain the original point information.

[0095] Step 2: Accumulate the original points obtained from multiple scanning cycles and sort them by time.

[0096] Step 3: Perform parameter space transformations in the distance-time and azimuth-time two-dimensional planes, respectively. Then, in the transformed parameter space, considering the diversity of target motion trends, employ multi-channel parallel search and clustering methods to find possible tracks, i.e., potential tracks. Furthermore, a motion trend fitting method is used to obtain the prior probability of each potential track.

[0097] Step 4: Detect weak targets in clutter based on the existence probability of potential tracks and the point sharing relationship between potential tracks.

[0098] Figure 2 The image shows one of the scenarios in which the parameter space multi-channel target search technology for weak target detection in sea clutter is applied according to the present invention.

[0099] Figure 3 The sea clutter is largely suppressed, and most targets are detected, resulting in high accuracy in target detection. Therefore, the generation of false tracks can be effectively avoided, increasing the probability of detecting weak targets in sea clutter while reducing the false alarm rate. The targets marked by the boxes in the figure are those detected by this invention in this application scenario.

[0100] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A parameter-space multi-channel target search method for weak target detection in sea clutter, characterized in that, For weak target detection in sea clutter, the radar echo signal is first processed, and a low target detection threshold is used for preliminary target detection to obtain raw point information. Utilizing the continuity of target motion in space and the temporal correlation of target echo data across several consecutive frames, the raw point information obtained from multiple scan cycles is accumulated. Considering the diversity of target motion trends, a multi-channel parallel search method is employed to find potential tracks. For each search channel, considering information from multiple dimensions such as range, azimuth, and time, the raw point information undergoes parameter transformation and batch processing. In the parameter space, potential target tracks are obtained, and their existence probability is calculated. This probability is then used to accurately detect weak targets in the clutter. The algorithm for weak target detection in sea clutter employs a sliding window approach. First, it accumulates raw data points obtained from multiple scan cycles, sorts them chronologically, selects data points within a certain angular range, and performs a Doppler transformation to convert the distance-time information of the data points into Doppler velocity and intercept information. The transformed intercepts are then sorted, and clustering is used to select potential tracks. For the selected tracks, a second parameter transformation is performed to convert the azimuth-time information of the data points into turning rate and intercept information. This same process is applied to data points within other angular ranges. After these two parameter transformations and multi-channel search, tracks that meet the criteria are identified as potential tracks. Simultaneously, a motion trend fitting method is used to obtain the prior probability of each potential track's existence. After searching all channels, all potential tracks are sorted according to their existence probabilities, and those with higher prior probabilities are identified as confirmed tracks, i.e., target trajectories. Finally, based on the existence probabilities of potential tracks and the data point sharing relationships among them, weak targets in the clutter are detected. The algorithm for detecting weak targets in sea clutter employs a parameter transformation batch processing and multi-channel search method. This algorithm uses two parameter transformation batch processing steps: first, it processes the observation data in Cartesian coordinates... Coordinates transformed into parameter space ,Right now: （1） In equation (1), For searching channels; for points on a straight line There must be two unique parameters. and satisfy: （2） A straight line in a Cartesian coordinate system can be obtained through the slope of this line. and intercept To define; For measurements that satisfy the same slope and intercept, they can be considered as a straight line, i.e. a track. The weak target detection algorithm in sea clutter uses the above method to perform parameter space transformation in the RT two-dimensional plane and the Az-T two-dimensional plane respectively, and performs multi-channel search to search for possible tracks. Finally, based on the searched tracks, weak targets in clutter are detected. This algorithm is suitable for detecting targets in cluttered environments. , In parameter space , The intercept size, the performance of this algorithm depends on the accumulation time of the measurements and the parameters. , In two aspects, the longer the accumulation time of the measured values, the higher the quality of the initial track; parameters , A smaller value results in a higher quality initial track, but it also increases the risk of missed alarms. , The selection should be based on the actual measurement error of the radar.

2. The parameter space multi-channel target search method for weak target detection in sea clutter according to claim 1, characterized in that, The probability of existence of potential tracks is calculated based on the probability hypothesis density (PHD). This yields the prior probability of each potential track as a real target. Represented as: （3） in, The number of potential tracks for which measurements have been obtained. It is a PhD with potential flight paths. Let be the probability density of false targets in the radar observation area. When false targets exist in the radar observation area, they are uniformly distributed within the area. If the volume of the radar observation area is... Then the probability density of false targets in the radar observation area In practical radar applications, the radar observation area refers to the range covered by the radar's main beam. It can be represented as: （4） To obtain the prior probability of a potential trajectory, a PHD filter is needed, which is obtained from the potential trajectory. The PHD is estimated using a set of measurements; the PHD filter has two main computational steps: prediction and update. The main operational steps of the PHD filter are as follows: （5） （6） in, (1) It is a state belonging to the target state space, which contains information such as position and velocity. It is the first The second detection yields a set of measurement values, which contains the values ​​from the first detection. Measurement information obtained from secondary target detection; (2) For the first The updated PhD. For the first The updated PhD. To utilize right The predictions made; (3) For from the first Time to the The state transition function at time t, In order to be in The state of the target at any given time can typically be represented by a state transition matrix; (4) For the first The measurement function for the next detection, which represents the target state. The obtained target measurement value is This function can typically be represented by a measurement matrix; (5) for The probability density of generating new targets in the target state space at any given time. This represents the probability density distribution of clutter. For detection probability, From Time's up The state at time is The probability of the target continuing to survive; In the recursive calculation of the PHD filter, it is assumed that the PHD can be represented by a mixture Gaussian distribution. The PhD at time t can be written as: （7） in, The first in the mixture Gaussian distribution The weights of a Gaussian model, The number of Gaussian models. For the first The expectation of a Gaussian model, For the first The covariance of the Gaussian model; the prediction and update of the PHD can be re-expressed as: （8） （9） in, It is a predicted PhD. The number of Gaussian models predicted. For the predicted mixture Gaussian distribution, the th The weights of a Gaussian model; In the PHD filter update, the following calculations are required: （10） （11） （12） （13） （14） in, Time observation matrix, To measure the noise covariance.

3. The parameter space multi-channel target search method for weak target detection in sea clutter according to claim 1, characterized in that, The specific steps include the following: Step 1: Perform signal processing on the radar echo data to remove co-frequency asynchronous interference, noise, and some sea clutter interference. Then, use a lower target detection threshold to perform preliminary target detection and obtain the original point information. Step 2: Accumulate the raw points obtained from multiple scanning cycles and sort them according to time; The third step is to perform parameter space transformation in the distance-time two-dimensional plane and the azimuth-time two-dimensional plane respectively. Then, in the transformed parameter space, in view of the diversity of target motion trends, a multi-channel parallel search and clustering method is used to find possible tracks, i.e. potential tracks. The motion trend fitting method is used to obtain the prior probability of each potential track. Step 4: Detect weak targets in clutter based on the existence probability of potential tracks and the point sharing relationship between potential tracks.

Images