A heterogeneous radar data fusion method and system for maritime radar systems

Abstract

The application belongs to the technical field of maritime radar and navigation and positioning, and particularly discloses a heterogeneous radar data fusion method and system for a maritime radar system. The method firstly takes the target ship trajectory obtained by GPS / Beidou as a space-time reference datum, and performs collection and extraction, data cleaning and preprocessing, joint matching based on time and trajectory, coordinate calibration and conversion and other processing on the original data of two heterogeneous radars, so as to convert the original data of the two heterogeneous radars to a unified GPS / Beidou coordinate system. Then, an adaptive weighted trajectory fusion method based on an improved local information entropy and a multi-scale sliding window is proposed. The method performs probability modeling and local information entropy calculation on observation residuals in the sliding window of different spatial scales, and combines the prior weight of radar performance to adaptively adjust the fusion weight of the trajectory point, so as to realize high-precision trajectory fusion of the heterogeneous radar data.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of maritime radars and navigation and positioning, and particularly relates to a heterogeneous radar data fusion method and system for a maritime radar system, and is particularly suitable for high-precision trajectory fusion between two heterogeneous radar data. BACKGROUND

[0002] With the development of global maritime traffic intelligence, the demand for dynamic perception of complex sea conditions (such as foggy days and strong clutter environments) by ships has significantly increased. Radar systems have become the core sensors for ship navigation, maritime situation monitoring, and collision avoidance decision-making. In current mainstream maritime radar equipment, solid-state pulse compression radars are widely used in shore-based monitoring and long-range target early warning due to their long detection range and strong electromagnetic penetration capability. Electronically scanned array radars (ESA) have become the preferred equipment for intelligent ship near-field high-precision tracking with higher spatial resolution and faster target refresh frequency. The above two types of radars have essential differences in working mechanism. Specifically, pulse compression technology is used in pulse compression radars to achieve long-range detection, and the data format is mainly polar coordinates (range, azimuth). ESA radars achieve electronic scanning through phased array technology and output point cloud data in Cartesian coordinates (X, Y). The heterogeneous characteristics of the two types of radars pose natural challenges to data fusion.

[0003] In maritime dynamic environments, the data from a single radar often cannot fully and accurately reflect the true state of a target ship. To improve the accuracy of target detection and trajectory tracking and enhance the robustness of the system in complex working conditions, multi-radar information fusion technology has become a key research direction in recent years. Fusion technology not only requires spatial coordinate conversion and time alignment of data from different radars, but also requires deep data fusion at the target association and trajectory reconstruction level, so as to fully utilize the complementary advantages of multi-source perception and achieve collaborative perception and joint estimation of targets. Therefore, there is an urgent need for a multi-radar fusion method that is suitable for maritime environments and adapts to the dynamic characteristics of ship targets. This method should be able to fuse target observation data from different types of radars in a unified reference coordinate system, construct a unified multi-source target trajectory representation, and improve the positioning accuracy and tracking continuity of ship targets in complex environments, thereby providing stable and reliable perception support for intelligent ship navigation systems and maritime monitoring platforms.

[0004] Most of the existing multi-radar fusion methods are borrowed from the multi-sensor fusion technology in the land transportation field. These methods are usually designed for sensors such as laser radar and camera, rely on relatively stable environmental features and fixed motion patterns, and are usually based on high-precision GPS / IMU assisted positioning. If these fusion strategies are directly migrated to the maritime field, significant challenges will be faced, such as large uncertainty of target motion state, lack of environmental structure, strong radar measurement noise, etc. In addition, maritime radar data often contains strong dynamic interference and background clutter, further increasing the difficulty of multi-source data association and fusion. In summary, if the existing multi-sensor fusion strategies for land transportation are directly migrated to the maritime environment, the following technical problems will be faced: 1. Lack of multi-radar data fusion methods suitable for maritime environment: Most of the existing multi-sensor fusion methods serve land transportation systems and are difficult to be directly applied to the data fusion of marine pulse pressure radar and electronic scanning radar. The target in the maritime scenario is highly dynamic and the background interference is complex, and there is still a lack of special fusion technology that can efficiently adapt to the characteristics of marine sensors. 2. Difficulty in time and space alignment between multi-radar data, weak fusion foundation: The sampling frequency, resolution, coordinate system, etc. of different types of radars are different, which makes it difficult to achieve accurate time synchronization and spatial mapping before target registration and data fusion, seriously affecting the fusion accuracy and stability. 3. Large target information matching error, insufficient fusion precision and reliability: The feature expression methods of multi-radar observed targets in the maritime environment are different, and target matching is prone to mismatch or omission, resulting in discontinuous fusion trajectories and serious jumps, affecting the overall performance and safety protection capability of the ship navigation system. SUMMARY

[0005] The purpose of the present application is to propose a heterogeneous radar data fusion method for maritime radar systems. The method converts two different types of radar data into a unified GPS / Beidou coordinate system through data calibration, and proposes an adaptive weighted trajectory fusion method based on improved local information entropy and multi-scale sliding window to realize adaptive weighted trajectory fusion between two heterogeneous radar data, thereby improving the trajectory fusion accuracy between heterogeneous radars in dynamic maritime environments.

[0006] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0007] A heterogeneous radar data fusion method for maritime radar systems includes the following steps:

[0008] Step 1. The target ship trajectory obtained by GPS / Beidou is taken as a space-time reference datum, and the original data of two heterogeneous radars are converted to a unified GPS / Beidou coordinate system through the steps of original data collection and extraction, data cleaning and preprocessing, joint matching based on time and trajectory, coordinate calibration and conversion, so that data calibration before fusion is completed.

[0009] Step 2. An adaptive weighted trajectory fusion method based on improved local information entropy and multi-scale sliding window is proposed, the observation residual is probabilistically modeled and the local information entropy is calculated in the sliding window of different spatial scales, and the fusion weight of the trajectory point is adaptively adjusted in combination with the prior weight of the heterogeneous radar performance, so as to realize the fusion of the heterogeneous radar trajectory.

[0010] In addition, on the basis of the above-mentioned heterogeneous radar data fusion method for maritime radar system, the application further proposes a corresponding heterogeneous radar data fusion system for maritime radar system, which adopts the following technical scheme:

[0011] A heterogeneous radar data fusion system for maritime radar system comprises the following modules:

[0012] The data calibration module is used for taking the target ship trajectory obtained by GPS / Beidou as a space-time reference datum, collecting and extracting the original data of two heterogeneous radars, data cleaning and preprocessing, joint matching based on time and trajectory, coordinate calibration and conversion, converting the original data of two heterogeneous radars to a unified GPS / Beidou coordinate system, and completing data calibration before fusion.

[0013] The data fusion module is used for proposing an adaptive weighted trajectory fusion method based on improved local information entropy and multi-scale sliding window, probabilistically modeling the observation residual in the sliding window of different spatial scales, and calculating the local information entropy, and adaptively adjusting the fusion weight of the trajectory point in combination with the prior weight of the heterogeneous radar performance, so as to realize the fusion of the heterogeneous radar trajectory.

[0014] The application has the following advantages:

[0015] As described above, the present application relates to a heterogeneous radar data fusion method for a maritime radar system, which includes two parts of data calibration before fusion and data fusion. In the data calibration before fusion stage, the present application modularizes the steps of data preprocessing, track matching, coordinate conversion, etc., unifies the data format, reduces the intermediate conversion links, reduces the computational complexity, and improves the overall processing efficiency. In addition, by introducing a unified GPS / Beidou reference track as the fusion reference, combined with the spatial rigid transformation and time synchronization mechanism, the high consistency of different types of radar data (such as pulse compression radar and electronic scanning radar) in space and time is ensured, which lays a unified data foundation for subsequent fusion weight calculation and residual modeling, effectively reduces data drift and error accumulation, and significantly improves the target positioning and tracking accuracy after fusion. Through the close coupling of the data calibration link and the data fusion link, the present application ensures the comparability of the residual and the accuracy of the weight estimation, making the fusion process have higher consistency and reliability. In the data fusion part, the present application innovatively proposes an adaptive weighted track fusion method based on improved local information entropy and multi-scale sliding window, which models the observation residual and calculates the local information entropy in the sliding window of different spatial scales, and combines the prior weight of the heterogeneous radar performance to adaptively adjust the fusion weight of the track point, so as to realize the fusion of heterogeneous radar tracks. Through the comprehensive action of multi-scale residual information entropy and prior performance weight, this method can significantly improve the contribution of low-noise observation data, while dynamically suppressing the influence of high-noise or uncertain observation points, thereby effectively improving the continuity, smoothness and accuracy of the fused track. The present application is particularly suitable for high-speed maritime environment, and adopts an efficient data association and matching algorithm based on time stamp nearest neighbor matching and spatial nearest neighbor matching, which can quickly establish the corresponding relationship of different radar observation data in a large-scale data stream, ensure the real-time and correctness of the track point pair, meet the timeliness requirement of online fusion, have good real-time response ability, accurately fuse multi-source radar information in the process of ship rapid movement, and improve the dynamic perception ability and navigation robustness of the system in complex scenes. The present application is suitable for various types of maritime radar systems, has good universality, realizes the data fusion of heterogeneous multi-source perception system, and provides a basic support for building a high-precision maritime intelligent perception platform. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Flow chart of the heterogeneous radar data fusion method for a maritime radar system in embodiment 1 of the present application;

[0017] Figure 2 Pulse compression radar, electronic scanning radar and GPS track diagram after data preprocessing in embodiment 1 of the present application;

[0018] Figure 3is a converted electronic scanning radar trajectory comparison chart with GPS trajectory in embodiment 1 of the present application;

[0019] Figure 4 is a converted solid-state pulse pressure radar trajectory comparison chart with GPS trajectory in embodiment 1 of the present application;

[0020] Figure 5 is a converted trajectory alignment effect chart in embodiment 1 of the present application;

[0021] Figure 6 is a flow chart of the weighted trajectory fusion algorithm based on local information entropy in embodiment 1 of the present application;

[0022] Figure 7 is a fused trajectory comparison chart in embodiment 1 of the present application;

[0023] Figure 8 is a comparison chart of the fused trajectory and AIS true value in the longitude direction in embodiment 1 of the present application;

[0024] Figure 9 is a comparison chart of the fused trajectory and AIS true value in the latitude direction in embodiment 1 of the present application;

[0025] Figure 10 is a comparison chart of the total error of the fused trajectory and AIS true value (longitude and latitude) in embodiment 1 of the present application;

[0026] Figure 11 is an error distribution chart of the longitude and latitude distance after the local information entropy weighted fusion algorithm in embodiment 1 of the present application;

[0027] Figure 12 is an error distribution chart of the speed after the local information entropy weighted fusion algorithm in embodiment 1 of the present application;

[0028] Figure 13 is an error distribution chart of the heading after the local information entropy weighted fusion algorithm in embodiment 1 of the present application;

[0029] Figure 14 is an error distribution chart of the trajectory after the local information entropy weighted fusion algorithm in embodiment 1 of the present application. DETAILED DESCRIPTION

[0030] The present application will be further described in detail below in combination with the drawings and specific embodiments:

[0031] Embodiment 1

[0032] The embodiment describes a heterogeneous radar data fusion method for a maritime radar system, which takes the high-precision trajectory of a target ship obtained by GPS / Beidou as a space-time reference datum, fuses the heterogeneous observation data of two different types of radars A and B, eliminates the space-time differences between multiple source radars through coordinate system conversion, time synchronization calibration and trajectory feature matching, etc. Under the unified geographic coordinate system, the optimized fusion algorithm is used to jointly process the multi-source target information, and the high-precision joint estimation and dynamic reconstruction of the ship trajectory are realized. The method can significantly improve the space-time consistency and robustness of multi-radar perception data in complex maritime environment, and provide high-reliable fusion perception technology support for real-time detection, continuous tracking and intelligent navigation of ship targets.

[0033] In the embodiment 1, the radars A and B are, for example, solid-state pulse compression radars and electronic scanning array radars respectively. The embodiment 1 realizes the data fusion of solid-state pulse compression radars and electronic scanning array radars through the steps, aiming to significantly improve the spatial consistency, time synchronization and anti-interference robustness of trajectory data.

[0034] The method of the application is generally divided into two steps, namely the data calibration step before fusion and the data fusion step. The data calibration step before fusion includes raw data acquisition and extraction, data cleaning and preprocessing, joint matching based on time and trajectory, coordinate calibration and conversion, etc. The data fusion step is based on improved local information entropy (Local Information Entropy, LIE) and multi-scale sliding window for adaptive weighted trajectory fusion. The data calibration step before fusion and the data fusion step complement each other, and finally form a unified and high-precision multi-source fusion ship trajectory.

[0035] As shown in Figure 1 The heterogeneous radar data fusion method for a maritime radar system in the embodiment includes the following steps:

[0036] Step 1. Taking the trajectory of a target ship obtained by GPS / Beidou as a space-time reference datum, the raw data of two heterogeneous radars are collected and extracted, the data is cleaned and preprocessed, the joint matching based on time and trajectory is performed, and the coordinate calibration and conversion are performed. The raw data of the two heterogeneous radars is converted to a unified GPS / Beidou coordinate system, and the data calibration before fusion is completed.

[0037] Step 1.1. Raw data acquisition and extraction.

[0038] Firstly, the trajectory data of the target ship is collected from multiple heterogeneous perception devices, mainly including solid-state pulse radar, electronic scanning radar and Beidou / GPS system. Among them, the solid-state pulse radar and electronic scanning radar provide observation data containing longitude, latitude, track number (hjid), timestamp, speed over ground (SOG) and course over ground (COG) and other fields.

[0039] And the Beidou / GPS system provides unified reference high-precision latitude and longitude position and timestamp information. This step 1.1 constructs the basic data interface between the multi-source perception system, providing basic data support for subsequent registration and fusion.

[0040] Next, taking the conversion of the original data of two heterogeneous radars to the unified GPS coordinate system as an example, the conversion is described.

[0041] Step 1.2. Data cleaning and preprocessing.

[0042] Because the data structures of different types of radar devices are inconsistent, the observation accuracy and sampling frequency differ, the original data needs to be uniformly cleaned and standardized to ensure the comparability and uniformity of the data. First, extract the key fields (such as track ID, longitude, latitude, timestamp, speed, heading, etc.) from the source data; delete missing values, duplicate records and static records with speed of 0; convert the timestamp to UTC format and normalize it to seconds; all longitude and latitude data are expressed in angular units (°); arrange the trajectory in ascending order of timestamp and use linear interpolation to complete the trajectory gaps to enhance the time alignment accuracy; coarsely align the radar data and GPS trajectory to lay the foundation for subsequent fine matching.

[0043] Step 1.3. Joint matching based on time and trajectory.

[0044] To ensure the consistency of the same target in different types of radar data sources in space and time, a matching method combining time window and spatial constraint is adopted. First, group the pulse radar and electronic scanning radar data by track number hjid, considering them as the same target; within the set time window range (±2s), use the conditions of spatial distance less than 50 meters, speed difference less than 1 knot and heading difference within 10 degrees for screening; use the nearest neighbor matching strategy to screen the trajectory point pairs that meet the above constraints; output the high-quality matching point pair set for calibration and fusion.

[0045] Step 1.4. Coordinate calibration and conversion based on improved ICP algorithm.

[0046] In order to accurately map all radar observation data to a unified geographic coordinate system (GPS), an improved Weighted Iterative Closest Point (WICP) coordinate calibration algorithm is proposed.

[0047] Compared with the traditional ICP algorithm, the improved ICP calibration algorithm in this embodiment introduces a weighted centroid mechanism to improve the robustness and matching accuracy in the presence of observation noise and incomplete point cloud overlap.

[0048] The detailed process of the improved ICP calibration algorithm is as follows:

[0049] Step 1.4.1. Initial setting and point cloud definition. Let the track points in the radar data constitute the source point cloud , and the GPS track data constitute the target point cloud .

[0050] wherein , ∈R 2 represents a two-dimensional coordinate point, and m and n are the number of source points and target points, respectively. Initialize the rigid transformation matrix T=[R∣t]. Wherein R∈R 2×2 is a rotation matrix, and t∈R 2 is a translation vector. Initially set R=I and t=0.

[0051] Step 1.4.2. Closest point matching. For each source point ∈P, find its closest Euclidean distance matching point ∈Q in the target point cloud Q, and construct the matching point pair set .

[0052] Step 1.4.3. Weighted centroid calculation. To enhance the influence of key point pairs in calibration, introduce a confidence weight (typically related to factors such as point pair distance, speed consistency, etc.), and calculate the weighted centroid of the matching point pair:

[0053] , .

[0054] wherein , are the weighted centroids of the source point cloud and the target point cloud, respectively.

[0055] Step 1.4.4. Point cloud decentralization.

[0056] Center all matching points to obtain a zero-mean point set: , .

[0057] Step 1.4.5. Covariance matrix and rotation matrix solving.

[0058] Construct the weighted covariance matrix:

[0059] .

[0060] Singular Value Decomposition (SVD) of H:

[0061] .

[0062] Get the rotation matrix:

[0063] .

[0064] If R is required to be a positive definite rotation matrix (i.e. det(R)=1), modify V if det(R)=-1.

[0065] Step 1.4.6. Translation vector solving.

[0066] Solve the translation vector according to the weighted centroid difference:

[0067] .

[0068] Step 1.4.7. Rigid transformation update and iteration.

[0069] Combine the current rotation and translation into a new rigid transformation matrix T=[R∣t], and update the coordinate of the source point cloud:

[0070] .

[0071] Repeat steps 1.4.2 to 1.4.7 until the convergence condition is met (here, the convergence condition, for example, refers to the target function error change below the threshold ε, or reaches the maximum iteration number N).

[0072] Step 1.4.8. Optimal transformation solving target function.

[0073] Finally, minimize the following target function to get the optimal rigid transformation:

[0074] .

[0075] where, represents the Euclidean norm, w i is the confidence weight of the i-th matched point.

[0076] Through the above calibration steps, the optimal rigid transformation matrix of the pulse compression radar and the electronically scanned radar relative to the GPS coordinate system is obtained, and then all radar data is mapped to the high-precision geographic coordinate system to realize the consistency alignment of the coordinate system.

[0077] The application fuses the observation data of the pulse compression radar and the electronically scanned array radar by constructing a unified fusion framework based on a reference trajectory, establishes a target-level correspondence relationship, and improves the accuracy of target matching. With the aid of a spatial transformation model and a time synchronization mechanism, consistent expression of data of different radar systems in space and time dimensions is realized, a unified spatial framework is provided for weighted fusion of heterogeneous radars in subsequent steps, and the performance of the fused trajectory in terms of position accuracy and time synchronization is significantly improved.

[0078] Step 2. An adaptive weighted trajectory fusion method based on improved local information entropy and multi-scale sliding window is proposed. The observation residual is probabilistically modeled and the local information entropy is calculated in the sliding window of different spatial scales, and the fusion weight of the trajectory point is adaptively adjusted in combination with the prior weight of the performance of the heterogeneous radar, so as to realize the fusion of the heterogeneous radar trajectory.

[0079] After completing the coordinate calibration of the multi-source radar data and unifying to the GPS geographic coordinate system, in order to further improve the spatial accuracy and continuity of the fused trajectory and solve the residual deviation problem caused by the observation errors of different radars, the application proposes an adaptive weighted trajectory fusion method based on improved local information entropy LIE and multi-scale sliding window. The method dynamically gives the fusion weight in combination with the difference in radar performance, realizes high-precision fusion of the heterogeneous radar trajectory. The core idea is to probabilistically model the observation residual in the sliding window of different spatial scales, calculate the local information entropy, combine the prior weight of the radar performance, adaptively adjust the fusion weight of the trajectory point, realize the robust estimation and weighted fusion of the data uncertainty.

[0080] As shown in Figure 6 , this step 2 is specifically:

[0081] Step 2.1. Construct a matched trajectory point pair set of the solid-state pulse compression radar and the electronically scanned radar, and calculate the Euclidean distance of each matched trajectory point pair in the matched trajectory point pair set in space, i.e. the observation residual.

[0082] Each matched trajectory point pair is composed of observation coordinates of the solid-state pulse compression radar and the electronically scanned radar at the same time.

[0083] It is defined that in the coordinate calibration stage of step 1, the matched trajectory point pair set with high matching accuracy obtained is .

[0084] Wherein is the observation coordinate of radar A at time observed coordinates , observed coordinates of radar B at the same time corresponding matching observed coordinates , the trajectory data of radar A (such as pulse radar) and radar B (such as electronic scanning radar) are unified to the same GPS coordinate system. The Euclidean distance in space, i.e. the observation residual, of each pair of matching trajectory points is calculated, and the formula is as follows:

[0085] ;

[0086] all distance vectors ; where N is the number of matching point pairs, .

[0087] Step 2.2. For each pair of matching trajectory points, the distribution probability of the observation residual in the neighborhood is counted in the sliding window of different spatial scales, and the local information entropy under the corresponding spatial scale is calculated.

[0088] For each pair of matching trajectory points Pi, the distribution probability of the observation residual in the neighborhood is counted in the sliding window of different spatial scales, and the corresponding local information entropy is calculated. Different spatial scales here refer to sliding windows of different radii, such as three different radii of sliding windows, i.e. r1=10m, r2=20m, r3=50m, so that the residual distribution can be counted at different scales.

[0089] For each spatial scale r m sliding window, define the following neighborhood set:

[0090] .

[0091] where is the coordinate of the trajectory point ; is the trajectory point coordinate at the same time or adjacent time as the trajectory point ; is the Euclidean distance between the two points; is the neighborhood trajectory point set centered on the trajectory point under the scale .

[0092] The observation residual of each pair of matching trajectory points in the neighborhood, i.e. the Euclidean distance of the observed coordinates of the same target by the two radars, is denoted as , and is divided into B intervals, the frequency of each interval is counted, and the residual distribution probability is calculated:

[0093] .

[0094] where denotes the probability that the observation residual in the neighborhood falls into the th interval; denotes the frequency of the th interval, i.e., the number of trajectory points; denotes the total number of trajectory points in the neighborhood.

[0095] The local information entropy of the current spatial scale r m is calculated as follows:

[0096] .

[0097] Finally, the local information entropy of different spatial scales is weighted and fused to obtain the multi-scale local information entropy of the trajectory point :

[0098] .

[0099] where is a small positive number to prevent the logarithm from being zero, is the weight of different scales.

[0100] The present application introduces a multi-scale sliding window mechanism in the analysis of multi-radar observation residuals. By modeling the probability of observation residual distribution in multiple spatial scales (i.e., different radius windows, such as 10 meters, 20 meters, 50 meters), the local information entropy index is calculated respectively. Multi-scale analysis can comprehensively depict the local consistency and uncertainty characteristics of observation residuals, avoiding noise sensitivity due to too small neighborhood or smoothing distortion caused by too large neighborhood in a single scale. This method enables the fusion algorithm to simultaneously perceive fine-grained local abnormal points and large-scale trends, significantly improving the spatial accuracy, abnormal detection ability and robustness of trajectory fusion.

[0101] Step 2.3. Prior weight design based on the performance of two heterogeneous radars.

[0102] In order to reasonably reflect the observation reliability of different radars in different ranging ranges, target characteristics and observation environments, the prior weight , is comprehensively valued according to the historical measurement accuracy, noise level, stability index and ranging ability of each radar.

[0103] Define X∈{A,B} represents radar A or radar B, then the prior weight , is calculated according to the following formula:

[0104] .

[0105] where​ RMS error of radar X historical observations, corresponding to historical measurement accuracy, the smaller the value, the lower the long-term observation error of the radar, the prior weight increases accordingly. The sliding average of observation residuals calculated within a fixed length sliding time window (e.g. ±5 seconds) centered on the current track point timestamp, corresponding to the short-term noise level and stability indicator, the larger the value, the stronger the recent observation volatility, the prior weight decreases accordingly. is a constant coefficient used to adjust the relative initial weight of the radar; is a constant to prevent division by zero; is an observation residual sensitivity adjustment parameter; is a distance correction factor, corresponding to the ranging ability, used to dynamically correct the weight according to the current measured target distance R, usually monotonically decreasing as the distance increases, to reflect the impact of the decline in observation accuracy at long distances.

[0106] The current measured target distance refers to the ranging distance between the target ship observed by the current radar and the radar itself.

[0107] Step 2.4. For each pair of matched track points, design a corresponding adaptive fusion weight function based on observation residuals, multi-scale local information entropy, and prior weight parameters, and obtain their adaptive fusion weights under radar A and radar B respectively.

[0108] In order to fully integrate the spatial residuals, uncertainty characteristics of track observations and performance differences of radar systems, an adaptive fusion weight function is designed for each matched track point pair, which is used to dynamically determine the contribution of each observation data in weighted fusion. By measuring the confidence and credibility of a track point observation, the larger the weight value, the higher the proportion of the observation in the fused track. The specific calculation formula of the fusion weight of each matched track point pair under radar X is as follows:

[0109] ;

[0110] where, is the fusion weight of track point i under radar X; is the observation residual of track point i under radar X; is the multi-scale local information entropy of track point i under radar X; is the prior performance weight of radar X; is a Sigmoid function, which maps the linear combination result to the interval (0, 1) to avoid infinite amplification or tending to zero of the weight; , are adjustment parameters used to control the sensitivity of observation residuals and multi-scale local information entropy to the weight respectively. ​

[0111] When the residual distance of the trajectory point Smaller, information entropy A lower value indicates that the observation is more stable and reliable. When the value of the sigmoid value tends to be negative, the output of the sigmoid function approaches 1, and the weight... Larger values ​​indicate a decrease in observation quality or an increase in uncertainty. The Sigmoid output tends towards 0, and the weights automatically decrease, thus weakening their impact on the fusion result. Prior weights... It is used to combine the historical performance differences of different radars (such as the ranging capability's ability to distinguish small targets, and the accuracy performance of electronic scanning / pulse compression in different ranges) for global correction, so that different data sources can be reasonably divided in the fusion process.

[0112] To address the significant performance differences between electronically scanned radar (ESR) and pulse compression radar (PCR) in terms of detection range, resolution, response time, and noise level, this invention dynamically designs a priori weighting function based on historical measurement mean square error (RMSE), residual moving average, and ranging capability. This priori weighting not only reflects the average accuracy of each type of radar under typical conditions but also incorporates performance degradation trends within the target ranging interval (e.g., decreased accuracy of ESR at long ranges) for correction. This allows the weight allocation to be adjusted in real-time according to observation conditions, thereby rationally determining the contribution ratio of different radar data in the fusion process. This mechanism fully leverages the advantages of high-performance radars within suitable ranges, effectively suppressing the negative impact of low-confidence observations on the fusion results and improving the overall trajectory reliability. This fusion weighting can dynamically identify and enhance the contribution of reliable observation points while suppressing noisy points and high-uncertainty observations, effectively improving the overall accuracy and robustness of the fused trajectory and meeting the accuracy requirements of multi-source radar fusion in complex maritime environments.

[0113] Step 2.5. After obtaining the adaptive fusion weights of each pair of matched trajectory points under the two heterogeneous radars, Radar A and Radar B, the final fused trajectory coordinates are calculated by combining the observation coordinates of the two heterogeneous radars and using a weighted average strategy.

[0114] After step 2.4, the adaptive fusion weights of each pair of matched trajectory points i under radar A and radar B are obtained. , Then, a weighted average strategy is used to calculate the final fused trajectory coordinates. The specific process is as follows:

[0115] Let radar A and radar B be at time... The observed coordinates are respectively , .

[0116] The coordinates of the merged trajectory points The calculation formula is as follows:

[0117] ;

[0118] wherein, 、 are the fusion weights of the i th track point in radar A and radar B respectively.

[0119] The weighted average formula is essentially a confidence weighted interpolation, so that the data source with high weight contributes more to the fusion result, ensuring that the fused track has higher stability and continuity while preserving the observation accuracy. Through this method, all track points can be fused point by point to generate a complete high-precision multi-source radar track sequence.

[0120] The entropy-driven weighting strategy proposed in the present application quantitatively reflects the uncertainty of each pair of matched track points observation data based on local information entropy: the more concentrated the residual distribution, the lower the entropy value, indicating strong observation consistency and high reliability, and giving higher weight when fusing; when the entropy value is high, it means that the observation residual distribution around the track point is dispersed and fluctuates greatly, and the uncertainty is significant, thereby automatically reducing its influence on the fusion result. This mechanism is derived from the meaning of entropy in information theory (the lower the entropy, the more ordered and reliable the information), without relying on fixed thresholds or empirical parameters. Through the Sigmoid function, the residual and entropy are mapped to the weight interval (0~1), which can not only enhance the contribution of reliable observations, but also dynamically suppress abnormal points, ensuring the continuity and accuracy of the fused track. This method has low computational complexity, strong robustness, and is easy to embed into real-time multi-radar perception systems, significantly improving fusion quality and system stability.

[0121] Step 2.6. Repeat steps 2.2 to 2.5 above until the weighted fusion calculation of each time track point is completed, and connect all the fused track coordinates in time sequence to build the final fused track sequence.

[0122] After completing the weighted fusion calculation of each time track point, all the fused points are connected in time sequence to build the final fused track sequence ; N is the number of matched point pairs, i∈[1,N].

[0123] The fused track not only retains the spatial resolution of the two types of radar observation data, but also significantly reduces the influence of single radar noise and residual through the multi-scale entropy weight and dynamic weighting mechanism, thus having higher smoothness and higher accuracy.

[0124] The multi-radar data fusion method based on local information entropy weighting in this embodiment can efficiently convert the data of solid-state pulse pressure radar and electronic scanning radar, thereby realizing the fusion processing of different types of radar data.

[0125] Of course, the radars A and B in the embodiment are not limited to the data fusion between the solid-state pulse compression radar and the electronically scanned array radar. For example, the fusion can also be between radars of other types, such as between radars of different working frequency bands, for example, between X-band and S-band radars, or between a pulse Doppler radar and a continuous wave radar.

[0126] In addition, in order to comprehensively evaluate the fusion result quality of the fusion method proposed in the application, the AIS trajectory is selected as the true value reference, and the following multiple indexes are used for error analysis, and the calculation formulas of the indexes are as follows:

[0127] RMSE (root mean square error): quantifying the overall deviation of the fusion trajectory from the true value, and the calculation formula is:

[0128] .

[0129] wherein, is the coordinate of the fusion trajectory point, is the AIS true value coordinate at the corresponding moment, and N is the total number of samples.

[0130] Maximum error: .

[0131] Minimum error: .

[0132] Mean error: .

[0133] Standard deviation: .

[0134] Figure 1 is a fusion method flowchart of the solid-state pulse compression radar, the electronically scanned radar and the GPS. The figure shows the overall process of data fusion of multiple radars (solid-state pulse compression radar, electronically scanned radar) and the GPS system.

[0135] The process is divided into original data acquisition, data preprocessing, trajectory matching, coordinate system conversion, weighted trajectory fusion based on local information entropy, error evaluation and true value comparison, etc. The calibration and fusion of radar data are gradually realized. The input and output and operation steps of each stage are clearly presented, and the whole process from the original data to the final fusion result is shown.

[0136] Figure 2 is a pulse compression radar, electronically scanned radar and GPS trajectory diagram after data preprocessing. The figure shows the original trajectory data of the pulse compression radar, the electronically scanned radar and the GPS before coordinate conversion.

[0137] Wherein the blue trajectory line is the trajectory of the target ship collected by the shipborne GPS system, the yellow trajectory is the trajectory of the target ship scanned by the pulse compression radar, and the red trajectory line is the trajectory of the target ship obtained by the electronic scanning radar.

[0138] Figure 3 is the comparison chart of the converted electronic scanning radar and GPS trajectory. This chart shows the comparison of the electronic scanning radar trajectory before and after conversion. The blue trajectory line represents the trajectory of the shipborne GPS system as a reference trajectory. The red trajectory line is the electronic scanning radar data before conversion; and the yellow trajectory line is the trajectory of the electronic scanning radar after coordinate conversion.

[0139] Figure 4 is the comparison chart of the converted solid-state pulse compression radar and GPS trajectory. This chart shows the comparison of the solid-state pulse compression radar data before and after conversion with the GPS trajectory. The blue trajectory line is the trajectory of the shipborne GPS system as a reference. The yellow is the solid-state pulse compression radar trajectory before conversion, and the dark blue trajectory line is the solid-state pulse compression radar trajectory after coordinate conversion.

[0140] Figure 5 shows the alignment effect of all radar data and GPS trajectory after ICP algorithm. The pulse compression radar, electronic scanning radar and GPS show consistent trajectories in the unified coordinate system. The blue trajectory line is the GPS trajectory as a reference. The red is the converted trajectory of the electronic scanning radar, and the dark blue is the converted trajectory of the solid-state pulse compression radar.

[0141] Figure 7 is the trajectory comparison chart after fusion by the method of the present application. This chart shows the comparison effect of all radar data and GPS trajectory after the local information entropy weighted fusion algorithm. The vertical axis represents latitude, and the horizontal axis represents longitude (for ease of display, each scale value of the horizontal axis needs to be superimposed by 120.197 to be the actual longitude, and the actual longitude of the leftmost side is 120.197), the black trajectory line is the GPS trajectory as a reference after fusion; and the red trajectory line is the fused trajectory of the solid-state pulse compression radar and the electronic scanning radar. It can be seen from Figure 7 that the fused trajectory after the method of the present application has high consistency with the AIS true value, which shows that the fused trajectory has high precision.

[0142] Figure 8 , Figure 9 and Figure 10 respectively show the error comparison of the fused trajectory and the AIS true value in the longitude and latitude directions and the time sequence variation of the total error in longitude and latitude. Specifically, Figure 8 and Figure 9 intuitively reflect the positioning accuracy of the fused trajectory in the east-west direction and the north-south direction, and through comparison with the AIS true value, the performance of the fusion algorithm in different spatial dimensions can be evaluated.Figure 10 The overall spatial error of the fusion trajectory relative to the true value is shown, and the dynamic change trend of the error over time is reflected, which reflects the timing stability and accuracy of the fusion process. Through the calculation of the statistical indicators of the error, the root mean square error (RMSE) of the fusion trajectory is 27.485 meters, the maximum error is 81.725 meters, the minimum error is 0.188 meters, and the average error is 24.515 meters, which indicates that the fusion method can effectively reduce the target positioning error and enhance the spatial precision and consistency of the trajectory.

[0143] Overall, Figure 8 to Figure 10 The error analysis of the fusion method in the multi-source radar data fusion has high accuracy and reliability, and can meet the demand of the intelligent perception system for high-precision target tracking. The vertical axis represents the error, and the horizontal axis represents the timestamp (the timestamp is a digital sequence that identifies the time when an event occurs. In order to facilitate the display, each scale value of the horizontal axis needs to be added by 1721800000 to obtain the actual timestamp, and the leftmost actual timestamp is 1721800000).

[0144] As Figure 11 to Figure 14 The error distribution graphs of the latitude and longitude distance, speed, heading, and trajectory after the local information entropy weighted fusion algorithm are shown respectively. Through Figure 12 、 Figure 13 The speed and heading error graphs also reflect the error control effect of the invention on different motion parameters, which can not only improve the positioning accuracy, but also effectively suppress the measurement deviation of the speed and heading.

[0145] The fusion method of the invention not only enhances the target recognition, positioning and tracking ability of the multi-source radar system, but also significantly improves the perception robustness and navigation safety in complex maritime scenarios, and has good engineering practical value and promotion prospect.

[0146] Embodiment 2

[0147] This embodiment 2 describes a heterogeneous radar data fusion system for a maritime radar system, which is based on the same inventive concept as the heterogeneous radar data fusion method for a maritime radar system in embodiment 1.

[0148] The heterogeneous radar data fusion system for a maritime radar system in this embodiment 2 includes the following modules:

[0149] The data calibration module is used to obtain the target ship trajectory based on GPS / Beidou as the spatio-temporal reference datum, to collect and extract the original data of two heterogeneous radars, to clean and preprocess the data, to perform joint matching based on time and trajectory, to calibrate and convert coordinates, to convert the original data of two heterogeneous radars to a unified GPS / Beidou coordinate system, and to complete data calibration before fusion;

[0150] The data fusion module is used to propose an adaptive weighted trajectory fusion method based on improved local information entropy and a multi-scale sliding window, probability modeling of observation residuals and local information entropy calculation are performed in the sliding window at different spatial scales, and the fusion weight of the trajectory point is adaptively adjusted in combination with the prior weight of the heterogeneous radar performance, so as to realize the fusion of the heterogeneous radar trajectory.

[0151] It should be noted that the functions and effects of each functional module in the heterogeneous radar data fusion system for the marine radar system in this embodiment 2 are realized in the implementation process of the corresponding steps of the method in the above embodiment 1, which will not be repeated here.

[0152] Embodiment 3

[0153] This embodiment 3 describes a computer device. The computer device includes a memory and one or more processors. The executable code is stored in the memory. When the processor executes the executable code, the steps of the above-mentioned embodiment 1 for the marine radar system are realized.

[0154] The computer device in this embodiment is any device or apparatus with data processing capability, which will not be repeated here.

[0155] Embodiment 4

[0156] This embodiment 4 describes a computer readable storage medium, which stores a program. When the program is executed by a processor, the steps of the above-mentioned embodiment 1 for the marine radar system are realized.

[0157] The computer readable storage medium can be an internal storage unit of any device or apparatus with data processing capability, such as a hard disk or memory, or an external storage device of any device or apparatus with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc.

[0158] Of course, the above description is only for the preferred embodiments of the present application, and the present application is not limited to the above-mentioned embodiments. It should be noted that any skilled person in the art can make all equivalent replacements, obvious modifications and forms under the teaching of the present application, which should be protected by the present application.

Claims

1. A method for heterogeneous radar data fusion in marine radar systems, characterized in that, Includes the following steps: Step 1. Using the target ship trajectory obtained by GPS / BeiDou as the spatiotemporal reference benchmark, the raw data of the two heterogeneous radars are collected and extracted, cleaned and preprocessed, jointly matched based on time and trajectory, and calibrated and transformed to convert the raw data of the two heterogeneous radars into a unified GPS / BeiDou coordinate system, thus completing the data calibration before fusion. Step 2. An adaptive weighted trajectory fusion method based on improved local information entropy and multi-scale sliding window is proposed. The observation residuals are probabilistically modeled and local information entropy is calculated within sliding windows of different spatial scales. Combined with the prior weights of heterogeneous radar performance, the fusion weights of trajectory points are adaptively adjusted to achieve the fusion of heterogeneous radar trajectories. Step 2 specifically involves: Step 2.

1. Construct a set of matching trajectory point pairs for two heterogeneous radars, Radar A and Radar B, and calculate the Euclidean distance in space for each pair of matching trajectory points in the set, i.e., the observation residual. Each pair of matching trajectory points consists of the observation coordinates of radar A and radar B at the same time. Step 2.

2. For each pair of matched trajectory points, the distribution probability of the observation residuals in the neighborhood is statistically analyzed within a sliding window at different spatial scales, and the local information entropy at the corresponding spatial scale is calculated. Then, by combining the local information entropy at various spatial scales, a weighted fusion is performed to obtain the multi-scale local information entropy of the trajectory points. Step 2.

3. Design prior weights based on the performance of the two heterogeneous radars; Define X∈{A,B} to represent radar A or radar B, then the prior weights , Calculate using the following formula: ； in The root mean square error of radar X historical observations; It represents the moving average of the observation residuals over a fixed length, centered on the current trajectory point timestamp; These are constant coefficients used to adjust the relative initial weights of the radar; To prevent division by zero of constants; Adjust parameters to observe residual sensitivity; This is a range correction factor used to dynamically adjust radar weights based on the ranging distance, correcting the decrease in observation accuracy caused by increasing distance. Step 2.

4. For each pair of matched trajectory points, design the corresponding adaptive fusion weight function based on the observation residual, multi-scale local information entropy, and prior weight parameters, and obtain its adaptive fusion weight under radar A and radar B respectively. Step 2.

5. After obtaining the adaptive fusion weights of each pair of matched trajectory points under the two heterogeneous radars, Radar A and Radar B, the final fused trajectory coordinates are calculated by combining the observation coordinates of the two heterogeneous radars and using a weighted average strategy. Step 2.

6. Repeat steps 2.2 to 2.5 above until the weighted fusion calculation of all matching trajectory point pairs at each time point is completed, and connect all fused trajectory coordinates according to the time series to construct the final fused trajectory sequence.

2. The heterogeneous radar data fusion method for marine radar systems according to claim 1, characterized in that, In step 1, the heterogeneous radar is a solid-state pulse compression radar and an electronically scanned array radar, or radars operating at different frequencies in the X-band and S-band, or a pulse Doppler radar and a continuous wave radar.

3. The heterogeneous radar data fusion method for marine radar systems according to claim 1, characterized in that, Step 2.1 specifically involves: Define the set of matching trajectory point pairs obtained in the coordinate calibration stage of step 1 as follows: ; in For radar A at time Observation coordinates , For radar B at the same time Corresponding matching observation coordinates The trajectory data of radar A and radar B have been unified into the same GPS coordinate system; The Euclidean distance between each pair of matching trajectory points in space is calculated using the following formula: ； All Construct distance vector Where N is the number of matching point pairs, .

4. The heterogeneous radar data fusion method for marine radar systems according to claim 1, characterized in that, Step 2.2 specifically involves: For each spatial scale r m The sliding window is defined with the following neighborhood set: ； in For trajectory points The coordinates; For the trajectory point Coordinates of trajectory points at the same or adjacent moments; The Euclidean distance between the two points; For scale Below are the trajectory points The set of trajectory points in the neighborhood of the center; The observation residuals of each pair of matching trajectory points in the neighborhood, i.e., the Euclidean distance between the observation coordinates of the two radars for the same target, are denoted as... Divide the data into B intervals and count the frequency in each interval. Calculate the residual probability distribution: ； in This indicates that the observation residual within the neighborhood falls into the first... The probability of each interval; Indicates the first The frequency of each interval, i.e. the number of trajectory points; This represents the total number of trajectory points within the neighborhood; Calculate the current spatial scale r m Local information entropy The formula is as follows: ； Finally, the local information entropy at different spatial scales is weighted and fused to obtain the trajectory points. Multi-scale local information entropy: ； in, To prevent tiny positive numbers from being the zero of the logarithm, Weights for different scales.

5. The heterogeneous radar data fusion method for marine radar systems according to claim 1, characterized in that, Step 2.4 specifically involves: The fusion weight of each pair of matched trajectory points under radar X The specific calculation formula is as follows: ； in, Let i be the fusion weight of trajectory point i under radar X; Let i be the observation residual of trajectory point i under radar X; Let i be the multi-scale local information entropy of trajectory point i under radar X; Let X be the prior performance weight of radar X; For the Sigmoid function; , The parameters are adjusted to control the sensitivity of the influence of observation residuals and multi-scale local information entropy on the weights.

6. The heterogeneous radar data fusion method for marine radar systems according to claim 1, characterized in that, Step 2.5 specifically involves: After step 2.4, the adaptive fusion weights of each pair of matched trajectory points i under radar A and radar B are obtained. , Then, a weighted average strategy is used to calculate the final fused trajectory coordinates. The specific process is as follows: Let radar A and radar B be at time... The observed coordinates are respectively , ; The coordinates of the merged trajectory points The calculation formula is as follows: ； in, , Let be the fusion weights of the i-th trajectory point in radar A and radar B respectively.

7. The heterogeneous radar data fusion method for marine radar systems according to claim 1, characterized in that, In step 2.6, after completing the weighted fusion calculation of trajectory points at each time point, all fused points are... By connecting the time series data, the final fused trajectory sequence is constructed. N represents the number of matching point pairs, where i ∈ [1, N].

8. A heterogeneous radar data fusion system for a maritime radar system for implementing the heterogeneous radar data fusion method for a maritime radar system as described in claim 1, characterized in that, The heterogeneous radar data fusion system for marine radar systems includes the following modules: The data calibration module is used to collect and extract raw data from two heterogeneous radars, clean and preprocess the data, perform joint matching based on time and trajectory, and perform coordinate calibration and transformation, using the target ship trajectory obtained by GPS / BeiDou as a spatiotemporal reference benchmark. It converts the raw data of the two heterogeneous radars into a unified GPS / BeiDou coordinate system, thus completing the data calibration before fusion. The data fusion module proposes an adaptive weighted trajectory fusion method based on improved local information entropy and multi-scale sliding windows. It performs probabilistic modeling and local information entropy calculation on the observation residuals within sliding windows of different spatial scales, and adaptively adjusts the fusion weights of trajectory points by combining the prior weights of heterogeneous radar performance, so as to achieve the fusion of heterogeneous radar trajectories.

Images