A multi-temporal SAR ship target tracking method, system, equipment, and medium

Abstract

This invention belongs to the field of ship target tracking technology and discloses a multi-temporal SAR ship target tracking method, system, device, and medium. It includes: using a global feature matching algorithm to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometric and texture features, narrowing the matching range between multiple targets, and reducing the error range for subsequent local feature matching; using a local feature matching algorithm to address the problem of partial loss of ship internal information caused by radar operating parameters and ship motion; and employing a fast nearest neighbor search function library (FLANN) to measure the similarity between ships with the same name, thereby achieving precise matching of ships with the same name. This invention overcomes the influence of motion on ship matching by comprehensively utilizing global and local features, taking into account multi-level ship feature information. It not only effectively solves the problem of limited feature information for small targets like ships but also overcomes the influence of geometric distortion caused by ship motion, improving ship tracking performance.

Description

Technical Field

[0001] This invention belongs to the field of ship target tracking technology, and particularly relates to a multi-temporal SAR ship target tracking method, system, device, and medium. Background Technology

[0002] Currently, Synthetic Aperture Radar (SAR) offers advantages over other optical and infrared sensors, including immunity to lighting and fog, and the ability to operate 24 / 7. This makes it better suited to complex maritime environments and a crucial tool for monitoring maritime vessels. SAR is widely used for vessel detection and classification. Beyond detection and identification, vessel tracking is another vital aspect of maritime vessel monitoring, significantly contributing to maintaining maritime traffic safety and enhancing maritime early warning capabilities.

[0003] With the rapid development of spaceborne SAR in recent years, systems such as TerraSAR-X, COSMOS-SkyMed, RADARSAT-2, Sentinel-1, and GF-3 have been able to provide high-resolution, multi-temporal SAR data, gradually making ship tracking at sea possible. Furthermore, many countries are shortening revisit times by building satellite constellations and developing multi-source satellite collaborative networking technologies. For example, the Italian COSMO-SkyMed system, consisting of a constellation of four SAR satellites, has a very short revisit time and wide coverage, further improving the ability to track ship targets.

[0004] Ship target tracking based on SAR imagery typically involves acquiring the location information of ship targets from multi-temporal SAR image sequences, then extracting and analyzing the features of the ship targets, and finally matching ships with the same name by measuring the feature similarity between targets, thereby achieving target tracking. The key to this is the extraction and matching of target features. Current SAR image feature matching methods mainly fall into two categories: global feature matching and local feature matching. Global feature matching extracts the overall features of the target features, such as texture and geometry, and matches targets with the same name by measuring similarity. Li Qixue et al. used Hu invariant moment region features, combined with minimum Euclidean distance measure and cosine similarity, to achieve heterogeneous image matching; Li Yanan et al. proposed an improved Hu invariant moment multi-source SAR image matching method. Local feature matching methods mainly include SIFT (Scale-invariant feature transform), SURF (Speeded-up robust features), and ORB (OrientedFAST and Rotated BRIEF), which mainly utilize the feature points within the target to construct feature quantities for matching. These algorithms have good adaptability to target rotation, scale scaling, and noise variations. Lei Yu et al. proposed a new SAR image target matching algorithm that combines Constant False-Alarm Rate (CFAR) and SURF. Li Yi et al. achieved target feature point matching based on SIFT by utilizing the spatial information between multi-source SAR images. In addition, Yin Junkai et al. combined optical and SAR images and used Harris and SIFT algorithms to extract the images to be matched, and achieved target matching by calculating the feature vector of the images through neural networks.

[0005] The aforementioned SAR target matching studies all focused on large, fixed targets such as buildings. Ship targets differ from land-based buildings in that they are smaller and often in motion, making matching them more challenging than for land targets. Applications of feature matching methods to SAR ship tracking include: Lei Lin et al. proposed a ship contour matching method based on partial Hausdorff distance measures; Li et al. proposed a ship image feature point matching algorithm based on principal component analysis and scale-invariant feature transform (PCA-SIFT) features, extracting feature vectors from the ship and reducing the dimensionality of the feature vectors using principal component analysis, then using the nearest neighbor algorithm for feature point matching; Gu Dandan et al. back-projected SAR ship images into a three-dimensional target space, extracted the back-projected scattering map of the target space to characterize the three-dimensional distribution of strong scattering sources of the ship target, used physical optics to predict the 3D hotspot scattering map of candidate ships, and then achieved ship matching through the back-projected scattering map and the hotspot scattering map; Chen Jianhong et al. proposed a heterogeneous high-resolution SAR ship target matching method, using multi-scale SAR-Harris operators to extract effective key points and using normalized correlation coefficients to measure ship target similarity to complete the matching.

[0006] While SAR satellites can effectively shorten revisit cycles and improve target tracking capabilities, current ship target tracking relies on a combination of multi-source SAR satellite data. Due to the different operating parameters of these SAR sensors, applying these heterogeneous SAR images to ship target matching and tracking presents several challenges: First, current heterogeneous SAR satellite target matching is mostly geared towards large targets such as land and buildings. Ship targets possess only a few stable feature points and information, and different types of ships often have similar outlines; relying solely on single feature information such as outlines can lead to mismatches. Second, the different operating parameters of heterogeneous SAR satellites, such as incident angle and frequency bands, result in differences in the image of the same ship target across different SAR images. Third, geometric distortions and positional shifts caused by the ship's own motion also contribute to tracking difficulties.

[0007] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0008] (1) Existing SAR target matching studies are all conducted on large fixed targets such as buildings. Ship targets are different from land buildings. In fact, they are smaller in size and are often in motion, so the matching difficulty is greater than that of land targets.

[0009] (2) Currently, ship target tracking relies on the combination of multi-source SAR satellite data. When applying heterogeneous SAR images to ship target matching and tracking, the target matching of heterogeneous SAR satellites is mostly aimed at large targets such as land and buildings. Ship targets only have a few stable feature points and feature information, and different types of ships have similar outlines. Using only single feature information such as outlines will cause mismatch.

[0010] (3) Different SAR satellites have different operating parameters such as incident angle and band frequency, resulting in different images of the same ship target in different SAR images; at the same time, the geometric distortion and position shift caused by the ship's own motion will also lead to tracking difficulties.

[0011] The difficulty in solving the above problems and defects lies in the fact that, in multi-source SAR ship tracking, ships are small targets and contain relatively little feature information; the loss and change of ship feature information caused by the combined effect of radar operating parameters and ship motion all contribute to the limited accuracy of traditional matching tracking algorithms.

[0012] The significance of solving the above problems and defects is that developing an algorithm suitable for multi-temporal SAR ship target tracking will not only improve the theoretical model of SAR ship tracking and enhance the tracking capability of ships with the same name, but also have important application value in future maritime ship target monitoring. Summary of the Invention

[0013] To address the problems existing in the prior art, this invention provides a multi-temporal SAR ship target tracking method, system, device, and medium, and particularly relates to a multi-temporal SAR ship target tracking method, system, device, and medium based on multi-level feature matching.

[0014] This invention is implemented as follows: a multi-temporal SAR ship target tracking method, the multi-temporal SAR ship target tracking method comprising:

[0015] First, a global feature matching algorithm is used to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometry and texture features, thus narrowing the matching range between multiple targets and reducing the error range for subsequent local feature matching. Second, a local feature matching algorithm is used to address the problem of partial loss of internal ship information caused by radar operating parameters and ship motion. Finally, a fast nearest neighbor search function library, FLANN, is used to measure the similarity between ships with the same name, thereby achieving precise matching of ships with the same name.

[0016] Furthermore, the multi-temporal SAR ship target tracking method includes the following steps:

[0017] Step 1: Extract the target vessel using the CFAR algorithm. The CFAR algorithm is well-established in SAR maritime vessel target detection. It can not only significantly improve the accuracy of SAR vessel detection, but also filter out interference factors such as the vessel background.

[0018] Step two involves using morphological filtering to reduce the impact of SAR background noise and enhance the detection and description capabilities of local feature points. The morphological filtering algorithm can further overcome the impact of target background noise on ship feature extraction and enhance the accuracy of ship feature extraction.

[0019] Step 3: Extract the outline and internal texture features of the ships and construct an overall feature vector. Use Euclidean distance to filter out some similar ships to achieve coarse screening. This not only makes reasonable use of the multi-level feature information of the ships, but also increases the fault tolerance of point-to-point matching in local features.

[0020] Step four involves employing the SURF algorithm based on FLANN feature point matching to achieve accurate tracking of the final ship target. The FLANN algorithm reduces algorithm complexity, improves matching efficiency and speed, while also ensuring matching accuracy.

[0021] Furthermore, in step three, the global and local feature extraction includes:

[0022] Texture and contour features can describe the global features of a ship, and the ship's shape and texture feature information can be extracted by combining HU invariant moments and gray-level co-occurrence matrix.

[0023] The HU invariant moment, as one of the methods for calculating shape-invariant moments, has a good ability to describe the contour information of ship targets. The HU method utilizes second- and third-order normalized center distances to construct seven invariant moments. For a ship image f(x, y) of size M×N, the p+q order matrix and the center distance μ... pq They are defined as follows:

[0024]

[0025]

[0026] in, y pq Represents the normalized central distance. p+q=2,3,...; (x0,y0) are the centroid coordinates, where x0 and y0 represent the gray-level centroids in the horizontal and vertical directions of the image, respectively; a set of seven-dimensional feature vectors are extracted using the normalized center distance.

[0027] The gray-level co-occurrence matrix (GLCM) not only reflects the gray-level distribution characteristics of different types of ships, but also the positional distribution characteristics among pixels with the same or similar gray levels. The original ship image's gray levels are compressed to 16 levels. Feature parameters of the GLCM in the four directions (0°, 45°, 90°, and 135°) are calculated, and the second-order moment A is extracted. sm Entropy E nt The four characteristic parameters are: the second-order statistic moment of inertia Con and the correlation Corr. The mean and variance of the characteristic parameters of the four direction matrices are calculated respectively, forming an eight-dimensional feature vector to represent the texture features of the ship. A total of 15-dimensional feature vectors are formed to represent the global features.

[0028] Furthermore, in step three, the global and local feature extraction also includes:

[0029] An improved SURF algorithm is used to extract local features of ships. In the local feature point extraction stage, the ship image is filtered and a Hessian matrix is ​​constructed to generate stable abrupt change points in the two-dimensional ship image. The Hessian matrix is ​​a square matrix composed of the second-order partial derivatives of a multivariate function, describing the local curvature of the function. For a ship image f(x, y), the Hessian matrix is ​​as follows:

[0030]

[0031] After constructing the scale space of the ship image using a box filter, each pixel processed by the Hessian matrix is ​​compared with points in the two-dimensional image space and the scale space neighborhood to initially locate points of interest. After filtering out weak feature points and erroneously located feature points, the final stable feature points are selected. Then, the Haar wavelet features in the circular neighborhood of the statistical feature points are used to determine the main direction of the feature points and generate feature vectors of the feature points, which constitute the local feature vectors of the ship.

[0032] Furthermore, the multi-temporal SAR ship target tracking method also includes:

[0033] Euclidean distance is used to measure the similarity of overall features between ships; the two images are standardized, and contour and texture feature vectors are extracted using HU invariant moments and GLCM, respectively, with S... o ={S ok |k=1,2,...,n} represents the feature vector of the image of the ship to be tracked, denoted by S. i ={S ik |k=1,2,...,n} represents the feature vector of the i-th ship image.

[0034]

[0035] In this context, the Euclidean distance satisfies dist(i) > 0. The smaller the Euclidean distance, the higher the similarity between ship samples. When the Euclidean distance is too large, it will expand the matching range between multiple targets and increase the error of point-to-point matching in local features. However, when the Euclidean distance is too small, factors such as geometric distortion will affect the global feature extraction and may incorrectly remove correct ship targets with the same name. Based on experimental analysis, setting the threshold of the Euclidean distance to 1 results in the optimal ship matching performance.

[0036] The screened ships undergo further matching. The FLANN algorithm is introduced to calculate the similarity between feature points, and ships with the same name are identified based on the number of matching point pairs. The feature space of the FLANN model is an n-dimensional real vector space, named R. n Its core is to find neighboring points based on Euclidean distance; the sub-vectors of feature points m and n are respectively represented by S. m and S n Let D(m,n) be an expression, then the Euclidean distance is as follows:

[0037] D(m, n) = (S m -S n ·S n -S m );

[0038] Among them, R n All D(m,n) are stored in several structures based on the KD number part; the minimum Euclidean distance to the query point is searched in the entire KD tree, and then the nearest point of the reference point is searched, and the initial matching set of feature point pairs is obtained. The ship with the same name is determined based on the maximum number of feature point pair matching sets.

[0039] Furthermore, the multi-temporal SAR ship target tracking method also includes:

[0040] The performance of the proposed tracking method is verified using two metrics: the number of correct ship target matches (NCM) and the matching accuracy (MP). The MP is calculated as follows:

[0041]

[0042] Where NTM is the total number of matches.

[0043] Another object of the present invention is to provide a multi-temporal SAR ship target tracking system applying the aforementioned multi-temporal SAR ship target tracking method, the multi-temporal SAR ship target tracking system comprising:

[0044] The target vessel extraction module is used to extract target vessels using the CFAR algorithm.

[0045] The local feature point enhancement module is used to reduce the impact of SAR background noise and enhance the detection and description capabilities of local feature points through morphological filtering.

[0046] The feature extraction module is used to extract the outline and internal texture features of the ship and form an overall feature vector. It then uses Euclidean distance to filter out some similar ships, thus achieving a coarse screening.

[0047] The ship target tracking module is used to achieve the final correct tracking of ship targets using the SURF algorithm based on FLANN feature point matching.

[0048] Another object of the present invention is to provide a computer device comprising a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the following steps:

[0049] First, a global feature matching algorithm is used to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometry and texture features, thus narrowing the matching range between multiple targets and reducing the error range for subsequent local feature matching. Second, a local feature matching algorithm is used to address the problem of partial loss of internal ship information caused by radar operating parameters and ship motion. Finally, a fast nearest neighbor search function library, FLANN, is used to measure the similarity between ships with the same name, thereby achieving precise matching of ships with the same name.

[0050] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the following steps:

[0051] First, a global feature matching algorithm is used to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometry and texture features, thus narrowing the matching range between multiple targets and reducing the error range for subsequent local feature matching. Second, a local feature matching algorithm is used to address the problem of partial loss of internal ship information caused by radar operating parameters and ship motion. Finally, a fast nearest neighbor search function library, FLANN, is used to measure the similarity between ships with the same name, thereby achieving precise matching of ships with the same name.

[0052] Another objective of the present invention is to provide an information data processing terminal for implementing the multi-temporal SAR ship target tracking system.

[0053] Combining all the above technical solutions, the advantages and positive effects of this invention are as follows: The multi-temporal SAR ship target tracking method provided by this invention overcomes the influence of motion on ship matching by comprehensively utilizing global and local features, thereby improving ship tracking performance. First, this invention extracts the target ship using the CFAR algorithm, and then uses morphological filtering to reduce the influence of SAR background noise and enhance the detection and description capabilities of local feature points. Second, it extracts the ship's outline and internal texture features to form an overall feature vector, and uses Euclidean distance to filter out some similar ships, achieving a coarse screening purpose and increasing the fault tolerance rate of point-to-point matching in local features. Finally, it uses the SURF algorithm based on FLANN feature point matching to achieve correct tracking.

[0054] This invention addresses the problem of tracking ships with the same name in heterogeneous SAR (SMS) targets. It proposes a multi-level feature matching method for ship tracking in heterogeneous SAR. This method takes into account the multi-level feature information of the ship, effectively solving the problem of limited feature information for small targets like ships, and overcoming the influence of geometric distortion caused by ship motion. Experiments using two sets of data demonstrate the robustness and applicability of the proposed method. The following conclusions are drawn from the experiments:

[0055] (1) In heterogeneous SAR ship target tracking, the ship's motion is the main factor affecting tracking. The speed of the ship's heading change will directly affect the change of aspect ratio characteristic parameters. The geometric distortion caused by the ship's motion will cause changes in the ship's internal geometric features, thus affecting the tracking accuracy.

[0056] (2) Due to the stable geometric constraints of the internal structure of the ship, although the geometric distortion causes the loss of ship features, the ship target still has similar contour information and scattering point distribution characteristics. Reasonable use of such characteristics will help improve the matching accuracy.

[0057] (3) Experimental results show that the algorithm of this invention is more robust and applicable than traditional feature matching algorithms such as SIFT and SURF, with a total matching accuracy of 85.71%, of which the correct matching rate for stationary ships is 90% and the correct matching rate for moving ships is 82.98%. Compared with traditional methods such as SIFT and SURF, the matching accuracy is improved by more than 40%. The algorithm of this invention can maintain good tracking ability for different types of ships and ships with different rotation angles. This invention provides new technical support for heterogeneous SAR maritime ship tracking and has certain practical application value. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of the multi-temporal SAR ship target tracking method provided in the embodiments of the present invention.

[0060] Figure 2 This is a schematic diagram of the multi-temporal SAR ship target tracking method provided in an embodiment of the present invention.

[0061] Figure 3 This is a block diagram of the multi-temporal SAR ship target tracking system provided in an embodiment of the present invention;

[0062] In the diagram: 1. Target vessel extraction module; 2. Local feature point enhancement module; 3. Feature extraction module; 4. Vessel target tracking module.

[0063] Figure 4 This is a flowchart of the extraction process for geometric features and scattering point data of SAR ship targets provided in an embodiment of the present invention.

[0064] Figure 5 is a graph showing the changes in ship geometry, scattering point feature parameters, Euclidean distance for evaluating the similarity of ships with the same name, and ship motion state provided in the embodiments of the present invention.

[0065] Figure 5(a) is a graph showing the relationship between the Euclidean distance of the ship's circumference and the ship's motion state, provided in an embodiment of the present invention.

[0066] Figure 5(b) is a graph showing the change between the Euclidean distance of the ship area and the ship's motion state provided in the embodiment of the present invention.

[0067] Figure 5(c) is a graph showing the change of the Euclidean distance of the ship's length-to-width ratio and the ship's motion state according to an embodiment of the present invention.

[0068] Figure 5(d) is a graph showing the change in the Euclidean distance of the ship's scattering point and the ship's motion state according to an embodiment of the present invention.

[0069] Figure 5(e) is a graph showing the change of Euclidean distance and ship motion state in the ship similarity assessment provided by the embodiment of the present invention.

[0070] Figure 6 is a schematic diagram of the scattering point distribution of different types of stationary ships in the C / X band provided by an embodiment of the present invention.

[0071] Figure 6(a) is a schematic diagram of the scattering point distribution of a stationary oil tanker in the C-band provided by an embodiment of the present invention.

[0072] Figure 6(b) is a schematic diagram of the scattering point distribution of a stationary oil tanker in the X-band provided by an embodiment of the present invention.

[0073] Figure 6(c) is a schematic diagram of the scattering point distribution of a stationary container ship in the C-band provided by an embodiment of the present invention.

[0074] Figure 6(d) is a schematic diagram of the scattering point distribution of a stationary container ship in the X-band provided by an embodiment of the present invention.

[0075] Figure 6(e) is a schematic diagram of the scattering point distribution of a stationary fishing boat in the C-band provided by an embodiment of the present invention.

[0076] Figure 6(f) is a schematic diagram of the scattering point distribution of a stationary fishing boat in the X-band provided by an embodiment of the present invention.

[0077] Figure 7 is a schematic diagram of the scattering point distribution of different types of moving ships in the C / X band provided by an embodiment of the present invention.

[0078] Figure 7(a) is a schematic diagram of the scattering point distribution of a moving container ship in the C-band provided by an embodiment of the present invention.

[0079] Figure 7(b) is a schematic diagram of the scattering point distribution of a moving container ship in the X-band provided by an embodiment of the present invention.

[0080] Figure 7(c) is a schematic diagram of the scattering point distribution of a moving oil tanker in the C-band provided by an embodiment of the present invention.

[0081] Figure 7(d) is a schematic diagram of the scattering point distribution of a moving oil tanker in the X-band provided by an embodiment of the present invention.

[0082] Figure 7(e) is a schematic diagram of the scattering point distribution of a moving container ship in the C-band provided by an embodiment of the present invention.

[0083] Figure 7(f) is a schematic diagram of the scattering point distribution of a moving container ship in the X-band provided by an embodiment of the present invention.

[0084] Figure 7(g) is a schematic diagram of the scattering point distribution of a moving cargo ship in the C-band provided by an embodiment of the present invention.

[0085] Figure 7(h) is a schematic diagram of the scattering point distribution of a moving cargo ship in the X-band provided by an embodiment of the present invention.

[0086] Figure 8This is a schematic diagram of the spatial distribution of ship density in SAR images provided in an embodiment of the present invention; (a) 1#; (b) 2#.

[0087] Figure 9 This is a schematic diagram of the initial screening results after global multi-feature fusion provided in an embodiment of the present invention.

[0088] Figure 10 This is a schematic diagram of a SAR vessel sample to be tracked provided in an embodiment of the present invention.

[0089] Figure 11 This is a schematic diagram of the initial screening results of the ships to be tracked after global multi-feature fusion provided in an embodiment of the present invention.

[0090] Figure 12 This is a schematic diagram of the tracking results of ships with the same name based on local feature matching provided in an embodiment of the present invention. Detailed Implementation

[0091] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0092] To address the problems existing in the prior art, the present invention provides a multi-temporal SAR ship target tracking method, system, device, and medium. The present invention will be described in detail below with reference to the accompanying drawings.

[0093] like Figure 1 As shown, the multi-temporal SAR ship target tracking method provided in this embodiment of the invention includes the following steps:

[0094] S101, the target vessel is extracted using the CFAR algorithm;

[0095] S102 reduces the impact of SAR background noise and enhances the detection and description capabilities of local feature points through morphological filtering.

[0096] S103: Extract the outline and internal texture features of the ship and construct the overall feature vector. Use Euclidean distance to filter out some similar ships to achieve coarse screening.

[0097] S104 employs the SURF algorithm based on FLANN feature point matching to achieve correct tracking of the final ship target.

[0098] The principle diagram of the multi-temporal SAR ship target tracking method provided in this embodiment of the invention is as follows: Figure 2 As shown.

[0099] like Figure 3As shown, the multi-temporal SAR ship target tracking system provided in this embodiment of the invention includes:

[0100] Target vessel extraction module 1 is used to extract target vessels using the CFAR algorithm;

[0101] The local feature point enhancement module 2 is used to reduce the influence of SAR background noise and enhance the detection and description capabilities of local feature points through morphological filtering.

[0102] Feature extraction module 3 is used to extract the outline and internal texture features of the ship and form an overall feature vector. It then uses Euclidean distance to filter out some similar ships, thus achieving coarse filtering.

[0103] Ship target tracking module 4 is used to achieve the final correct tracking of ship targets using the SURF algorithm based on FLANN feature point matching.

[0104] The technical solution of the present invention will be further described below with reference to the embodiments.

[0105] This invention proposes a multi-ship target tracking algorithm among heterogeneous SAR satellites. First, a global feature matching algorithm is used to overcome the difficulty of distinguishing similar ship outlines by leveraging ship geometric and texture features, thus narrowing the matching range among multiple targets and reducing the error range for subsequent local feature matching. Second, a local feature matching algorithm is used to address the loss of internal ship information caused by radar operating parameters and ship motion. Finally, a Fast Approximate Nearest Neighbor Search Library (FLANN) is used to measure the similarity between ships with the same name, thereby achieving precise matching of ships with the same name.

[0106] The main contents of this invention are arranged as follows: Section 1 introduces the experimental data and working parameters used in this invention; Section 2 analyzes the main influencing factors of heterogeneous SAR ship matching; Section 3 introduces the proposed heterogeneous SAR ship matching and tracking method and principle that combines global and local features; Section 4 presents experimental comparisons and accuracy analysis; and Section 5 provides conclusions and discussion.

[0107] 1. Experimental data and its operating parameters

[0108] This invention utilizes data from two pairs of TerraSAR-X and RADARSAT-2 satellites to conduct a matching study of ships with the same name. Detailed data parameters are shown in Table 1. AIS data were acquired simultaneously 30 minutes before and after the SAR imaging time. Ships with the same name in different SAR images were determined by AIS interpolation and visual interpretation. Information such as the number and type of ships with the same name is shown in Table 2.

[0109] The first pair of data was located in the waters near the Strait of Malacca (hereinafter referred to as 1#). The time difference between the two images was approximately 3 minutes. The area was under low wind conditions at the time of acquisition. Both images are VV polarized. A total of 33 pairs of moving vessels with the same name and 210 pairs of stationary vessels with the same name were identified in this area. The moving vessels were mainly large ships such as container ships and cargo ships. TerraSAR-X imagery was used as the base map. Figure 8 (a) shows the spatial distribution of ships, with the highlighted areas being the locations where stationary ships are clustered.

[0110] The second pair of data is located in the waters near the Zhoushan Islands (hereinafter referred to as 2#). The two images were acquired approximately 40 minutes apart under low wind conditions. Both images are HH polarized. A total of 14 pairs of moving vessels with the same name and 78 pairs of stationary vessels with the same name were identified in this area. The moving vessels were mainly container ships. Using TerraSAR-X imagery as the base map, the spatial distribution of the vessels is as follows: Figure 8 As shown in (b).

[0111] Table 1. Introduction to Experimental SAR Data Parameters

[0112]

[0113] Table 2. Number of Ships of Different Types

[0114]

[0115] 2. Analysis of the impact of heterogeneous SAR ship target imaging differences on ship matching

[0116] Variations in the operating parameters of heterogeneous SAR satellites and the motion of ships can affect the imaging results of ships in SAR. This section analyzes in detail the relationships between ship geometry, scattering characteristics, and stationary and moving ships in different radar bands, exploring the main factors affecting ship tracking.

[0117] 2.1 Calculation of Characteristic Parameters of Ships under Different Source SAR

[0118] Geometric features mainly include area, perimeter, and aspect ratio. Figure 4 A flowchart for calculating ship geometry and scattering point features is provided. The ship's perimeter is calculated using Canny edge detection; the area inside the ship is obtained by counting pixels; the aspect ratio is obtained by calculating the ship's minimum circumcircle moment; and the location and number of strong scattering points inside the ship are determined using K-means clustering.

[0119] The similarity of ships with the same name is measured by calculating the feature change rate of ships with the same name in two imaging sessions and the Euclidean distance between the texture and contour features of ships with the same name. That is, the larger the feature change rate and the Euclidean distance, the lower the similarity between the ship targets, and the more difficult it is to match ships with the same name. The feature change rate is obtained by the ratio of the difference between the ship feature values ​​in the C and X bands to the ship feature value in the C band. The formula for calculating the Euclidean distance will be given in Section 3 of this invention.

[0120] Figure 5 shows the curves of the variation of various feature parameters of ships, the Euclidean distance between features of ships with the same name and the motion state in the two sets of data. The horizontal axis represents the ship number, 1-33 and 34-47 are moving ships with the same name in data 1# and 2# respectively, and 48-62 are stationary ships with the same name in the two images.

[0121] As shown in Figure 5, the geometric feature parameters of stationary ships in different SAR systems exhibit relatively stable changes. Figures 5(a) to 5(c) show that the rate of change for the area and perimeter of stationary ships is generally below 20%, especially the aspect ratio, which remains below 10%. However, the geometric features of moving ships fluctuate significantly. The maximum rate of change for perimeter and area exceeds 60%, while the minimum is below 10%. The aspect ratio rate is generally around 20%, with a maximum exceeding 40%. Regarding scattering point features, in Figure 5(d), the rate of change for scattering point features shows no discernible correlation with the ship's motion; regardless of whether the ship is stationary or moving, the rate of change for scattering points is generally below 40%. Furthermore, in Figure 5(e), the Euclidean distances between features of the same name from stationary ships are generally concentrated below 0.6, while the Euclidean distances between features of the same name from moving ships are mainly distributed between 0.8 and 0.9. In conclusion, the similarity between moving ships with the same name in heterogeneous SAR is lower than that between stationary ships with the same name, making matching more difficult.

[0122] Ship motion primarily involves changes in speed and heading. The combined effects of heading and speed cause significant fluctuations in the ship's geometric characteristic parameters, impacting ship matching. As shown in Figures 5(a), 5(b), and 5(d), there is no clear correlation between the perimeter, area, and scattering point characteristic parameters of a moving ship in heterogeneous SAR and changes in heading. However, in Figure 5(c), changes in heading generally show a strong correlation with the aspect ratio characteristic parameter; the faster the heading changes, the greater the change in the aspect ratio characteristic parameter, leading to more severe geometric distortion. Therefore, changes in heading during ship motion are a major factor affecting ship matching.

[0123] 2.2 Analysis of the impact of heterogeneous SAR on the characteristics of stationary ship targets

[0124] As shown in Figure 5, the geometric characteristic parameters of stationary ships are relatively stable across different wavebands, but the rate of change of scattering points shows no clear correlation with the ship's motion. To further explore the influence of wavebands on the characteristic parameters of stationary ships, from... Figure 4 Select oil tankers, container ships, and fishing vessels (such as...) Figure 4 The black dashed boxes represent three types of stationary ships (ship numbers 48, 50, and 52) for case analysis. Figure 5 shows the distribution of their scattering points in the C and X bands. The gradient color legend indicates the intensity change of the ship's scattering points, with darker colors indicating greater scattering point intensity.

[0125] As shown in Figure 5, the number of strong scattering points for the three types of ships in the X-band increased by 18.3%, 13.8%, and 42.6% respectively compared to the C-band. However, as shown in Figure 6, although the number of strong scattering points changed in the SAR images of the X-band and C-band, the distribution of strong scattering points within ships of the same name was roughly similar, and both could reflect information about the various structural features inside the ship. For the three different types of stationary vessels mentioned above, the perimeter change rates for tankers, container ships, and fishing boats are 16.3%, 11.9%, and 19.5%, respectively, and the area change rates are 19.9%, 11%, and 13.7%, respectively. The change rates of their area and perimeter characteristic parameters are all below 20%, indicating relatively small changes. In particular, the length-to-width ratio characteristic parameter shows that the change rates for the length-to-width ratio of the three types of vessels are 7%, 9.5%, and 4.2%, respectively, all below 10%. Therefore, different radar bands have little impact on the geometric characteristic parameters of stationary vessels and almost no impact on the length-to-width ratio. In summary, heterogeneous SAR has little impact on matching stationary vessels of the same name.

[0126] 2.3 Analysis of the impact of heterogeneous SAR on the characteristics of moving vessel targets

[0127] The main reason for the geometric distortion of ships in SAR images is that the changes in the speed and heading of moving ships cause uneven sampling intervals in the azimuth direction, resulting in the ship's imaging position deviating from its actual position and causing geometric distortion.

[0128] To further analyze the influence of different motion states of ships on characteristic parameters in heterogeneous SAR, ships with different motion states were selected from Figure 5 (black solid box in Figure 5, ship numbers 2, 11, 12, and 25). The ship heading change and average speed represent the changes in the ship's motion direction and average speed in two consecutive SAR images, respectively. Figures 7(a) and 7(b) show ships with small heading changes (ship numbers 2 and 11, heading changes of 5° and 1° respectively, and average speeds of 8.5 knots and 15 knots respectively), while Figures 7(c) and 7(d) show ships with large heading changes (ship numbers 12 and 25, heading changes of 25° and 28° respectively, and average speeds of 16.5 knots and 3.8 knots respectively). The gradient color legend represents the change in the intensity of the ship's scattering points; the darker the color, the greater the intensity of the scattering points. As shown in Figure 7, all ships exhibit varying degrees of geometric distortion in the SAR images.

[0129] The motion of a ship is mainly manifested in two aspects: speed and heading. The average speed of a ship is not significantly related to several characteristic parameters or the rate of change of the number of scattering points. Referring to Figure 5, for example, ship number 25, although its average speed is only 3.8 knots, shows significant changes in its perimeter, area, aspect ratio, and scattering point characteristic parameters, at 34.4%, 34.9%, 30.1%, and 34% respectively, all exceeding 30%. Ship number 12, with an average speed as high as 16.5 knots, has relatively smaller changes in its perimeter, area, aspect ratio, and scattering point characteristic parameters, at 17.7%, 12.6%, 27.6%, and 10.4% respectively. As shown in Figure 5(d), there is no discernible pattern between the number of scattering points and the heading. However, in terms of spatial distribution, ships with the same name exhibit similar strong scattering points. This is because ships, as rigid targets, possess geometric invariance in their internal three-dimensional structure, thus resulting in a number of stable scattering points with similar spatial distribution information.

[0130] Combining Figures 5 and 7, it can be seen that there is no significant correlation between the ship's heading and its area, perimeter characteristics, and the rate of change of scattering points. For example, ships numbered 2 and 11, with small heading changes, have perimeter change rates of 21.8% and 17.6%, and area change rates of 5.5% and 21.6%, respectively, with scattering point change rates of 1.7% and 37.2%, respectively. In contrast, ship number 12, with a large heading change, has perimeter and area change rates of only 17.7% and 12.6%, respectively, and a scattering point change rate of 10.4%. However, the change in heading is strongly correlated with the aspect ratio. As shown in Figure 5(c), the aspect ratio changes with the speed of ship rotation. For example, ships numbered 2 and 11, with small heading changes, have aspect ratio change rates of 14.2% and 8.2%, respectively, while ships numbered 12 and 15, with large heading changes, have aspect ratio change rates of 27.6% and 29.8%, respectively. This further indicates that the change in ship heading is the main factor affecting ship matching.

[0131] In summary, compared to spectral bands, ship motion is the primary factor influencing changes in ship characteristic parameters in heterogeneous SAR, particularly changes in ship heading. These changes ultimately lead to difficulties in matching ships with the same name. Therefore, in the process of ship matching in heterogeneous SAR, it is crucial to overcome the impact of ship motion, especially changes in heading. Furthermore, although the geometric distortion caused by motion alters the overall shape and contour features of the ship, moving ships still exhibit similar local scattering characteristics, and ships possess stable scattering points. If the global and local feature information of the ship can be fully utilized, matching ships with the same name will yield good results.

[0132] 3. Ship target feature matching method

[0133] 3.1 Method and Flow

[0134] This invention develops a novel heterogeneous SAR ship tracking method based on multi-level feature matching. By comprehensively utilizing global and local features, it overcomes the influence of motion on ship matching and improves ship tracking performance. The specific process is as follows: Figure 2 As shown, firstly, the target vessel is extracted using the CFAR algorithm, and then morphological filtering is used to reduce the influence of SAR background noise and enhance the detection and description capabilities of local feature points. Secondly, the vessel's outline and internal texture features are extracted to form an overall feature vector. Euclidean distance is used to filter out some similar vessels to achieve coarse screening and increase the fault tolerance rate of point-to-point matching in local features. Finally, the SURF algorithm based on FLANN feature point matching is used to achieve the final correct tracking.

[0135] 3.2 Global and Local Feature Extraction

[0136] Texture and contour features can describe the global features of a ship. This section will combine HU invariant moments and gray-level co-occurrence matrix to extract the shape and texture feature information of the ship.

[0137] HU invariant moments, as a method for calculating shape-invariant moments, has a good ability to describe the contour information of ship targets. It constructs seven invariant moments using second- and third-order normalized center distances, exhibiting translation and scale rotation invariance. For a ship image f(x, y) of size M×N, the p+q order matrix and the center distance μ... pq They are defined as follows:

[0138]

[0139]

[0140] in, y pq Represents the normalized central distance. here p+q=2,3,... (x0, y0) are the centroid coordinates, where x0 and y0 represent the gray-level centroids in the horizontal and vertical directions, respectively. A set of seven-dimensional feature vectors is extracted using the normalized center distance.

[0141] The gray-level co-occurrence matrix (GLCM) not only reflects the gray-level distribution characteristics of different types of ships, but also the positional distribution characteristics among pixels with the same or similar gray levels. First, to reduce computational load and improve retrieval speed without affecting texture feature extraction, the original ship image's gray levels are compressed to 16 levels. To weaken the influence of texture direction on feature values, feature parameters of the GLCM in four directions (0°, 45°, 90°, and 135°) are calculated. This invention extracts the second-order moment A. sm Entropy E nt The four feature parameters are: the second-order statistic, the moment of inertia Con, and the correlation Corr. The mean and variance of the feature parameters of the four direction matrices are calculated respectively, forming an eight-dimensional feature vector to characterize the texture features of the ship. A total of 15 feature vectors are formed to represent the global features.

[0142] This invention utilizes an improved SURF algorithm to extract local features of ships. In the local feature point extraction stage, the ship image is filtered to construct a Hessian matrix, generating stable abrupt change points in the two-dimensional ship image. The Hessian matrix is ​​a square matrix composed of the second-order partial derivatives of a multivariate function, describing the local curvature of the function. For a ship image f(x, y), its Hessian matrix is ​​as follows:

[0143]

[0144] After constructing the scale space of the ship image using a box filter, each pixel processed by the Hessian matrix is ​​compared with the points in the two-dimensional image space and the scale space neighborhood to initially locate the points of interest. After filtering out weak feature points and erroneously located feature points, the final stable feature points are selected. Then, the Haar wavelet features in the circular neighborhood of the statistical feature points are used to determine the main direction of the feature points, generating the feature vectors of the feature points, which constitute the local feature vectors of the ship.

[0145] 3.3 Similarity Evaluation Indicators

[0146] Euclidean distance is used to measure the similarity of overall features between ships. The two images are first standardized, and contour and texture feature vectors are extracted using HU invariant moments and GLCM, respectively, with S... o ={S ok |k=1,2,...,n} represents the feature vector of the image of the ship to be tracked, denoted by S. i ={S ik |k=1,2,...,n} represents the feature vector of the i-th ship image.

[0147]

[0148] In the formula, the Euclidean distance satisfies dist(i) > 0. The smaller the value of the Euclidean distance, the higher the similarity between ship samples. When the Euclidean distance is too large, it will expand the matching range between multiple targets and increase the error of point-to-point matching in local features. However, when the Euclidean distance is too small, factors such as geometric distortion will affect the global feature extraction and may incorrectly remove correct ship targets with the same name. According to experimental analysis, setting the threshold of the Euclidean distance to 1 results in the optimal ship matching performance.

[0149] Further matching processing is performed on the screened ships. The traditional SURF algorithm uses a nearest neighbor distance matching algorithm for feature point matching, which has high complexity and computational burden for 64-dimensional feature point descriptors. To improve matching efficiency and speed while ensuring accuracy, this invention introduces the FLANN algorithm to calculate the similarity between feature points. Ships with the same name are determined based on the number of matching point pairs. The feature space of the FLANN model is typically an n-dimensional real vector space, named Rn. n The core idea is to find neighboring points based on Euclidean distance. The sub-vectors of feature points m and n are respectively represented by S... m and S n Let D(m,n) be an expression, then the Euclidean distance is as follows:

[0150] D(m, n) = (S m -Sn ·S n -S m ) (5)

[0152] Among them, R n All D(m,n) are stored in several structures based on the KD (k-dimensional) number part. The minimum Euclidean distance to the query point is searched in the entire KD tree, thus effectively searching for the nearest point to the reference point. Finally, an initial matching set of feature point pairs is obtained, and ships with the same name are determined based on the maximum number of matching feature point pairs in the set.

[0153] 4. Experimental Results and Analysis

[0154] This invention uses two metrics—the number of correct matches (NCM) and the matching accuracy (MP)—to verify the performance of the proposed tracking method. The MP is calculated as follows:

[0155]

[0156] Where NTM represents the total number of matches. 33 and 14 pairs of moving vessels with the same name were selected from data sets 1 and 2, respectively, and 30 pairs of stationary vessels were selected from these, for a total of 77 vessel samples for matching and tracking experiments. The algorithm of this invention was compared with the GLCM method based on global feature matching, HU invariant moments, and the SURF and SIFT methods based on local feature matching; the results are shown in Table 3.

[0157] Table 3 Comparison of matching accuracy between the algorithm of this invention and traditional methods

[0158]

[0159] When the geometric distortion of a ship due to motion is not significant, ships with the same name can be identified using only the ship's global features. The algorithm for global multi-feature fusion demonstrates good performance, such as... Figure 9 As shown, in most cases, ships undergo a certain degree of geometric distortion during movement, and in this case, global features alone cannot effectively achieve the purpose of tracking.

[0160] Therefore, this example selects a typical vessel to be tracked (vessel number: 4), which is small in size and has significant geometric distortion. This is because when a vessel has geometric distortion, its high-resolution SAR characteristics are as follows: Figure 3 As shown, the ships are first preprocessed using morphological filtering, image normalization, and other methods. Then, Hu invariant moments and GLCM are used to extract the shape and texture features of the ships. Euclidean distance is then used to measure the similarity between ships, achieving the purpose of initial screening. The screening results are as follows. Figure 10 As shown.

[0161] Figure 11 As shown, while texture and shape features alone cannot directly identify ships with the same name, they can still effectively detect such targets in the initial screening stage to a certain extent. Finally, based on the stable local feature point information within the ship's interior, the filtered ships are measured using the SURF algorithm based on FLANN associations. Figure 12 As shown, the similarity between each ship after the initial screening and the ship to be tracked is calculated by the number of matching points, and the ships with the same name are finally identified.

[0162] In summary, this example selects a small vessel with limited feature information and significant geometric distortion. When traditional matching and tracking algorithms cannot achieve accurate tracking, the method demonstrated in this example can successfully achieve tracking.

[0163] This invention addresses the problem of tracking ships with the same name in heterogeneous SAR (SMS) targets. It proposes a multi-level feature matching method for ship tracking in heterogeneous SAR. This method takes into account the multi-level feature information of the ship, effectively solving the problem of limited feature information for small targets like ships, and overcoming the influence of geometric distortion caused by ship motion. Experiments using two sets of data demonstrate the robustness and applicability of the proposed method. The following conclusions are drawn from the experiments:

[0164] 1) In heterogeneous SAR ship target tracking, the ship's motion is the main factor affecting tracking. The speed at which the ship's heading changes will directly affect the changes in the aspect ratio characteristic parameters. The geometric distortion caused by the ship's motion will cause changes in the ship's internal geometric features, thus affecting the tracking accuracy.

[0165] 2) Due to the stable geometric constraints of the internal structure of ships, although geometric distortion causes the loss of ship features, ship targets still have similar contour information and scattering point distribution characteristics. Reasonable use of such characteristics will help improve matching accuracy.

[0166] 3) Experimental results show that the algorithm of this invention is more robust and applicable than traditional feature matching algorithms such as SIFT and SURF, achieving a total matching accuracy of 85.71%, with a correct matching rate of 90% for stationary ships and 82.98% for moving ships. Compared with traditional methods such as SIFT and SURF, the matching accuracy is improved by more than 40%. The algorithm of this invention maintains good tracking capabilities for different types of ships and ships with different rotation angles. This invention provides new technical support for heterogeneous SAR maritime ship tracking and has certain practical application value.

[0167] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0168] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-temporal SAR ship target tracking method, characterized in that, The multi-temporal SAR ship target tracking method includes: firstly, using a global feature matching algorithm to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometric and texture features, thus narrowing the matching range between multiple targets and reducing the error range for subsequent local feature matching; secondly, using a local feature matching algorithm to address the problem of partial loss of ship internal information caused by radar operating parameters and ship motion; and finally, using the FLANN fast nearest neighbor search function library to measure the similarity between ships with the same name, achieving precise matching of ships with the same name. The global feature matching algorithm includes: Texture and contour features can describe the global features of a ship, and the ship's shape and texture feature information can be extracted by combining HU invariant moments and gray-level co-occurrence matrix. The HU invariant moment, as one of the methods for calculating shape-invariant moments, has a good ability to describe the contour information of ship targets. The HU method utilizes second- and third-order normalized center distances to construct seven invariant moments. For a ship image f(x, y) of size M×N, the p+q order matrix and the center distance μ... pq They are defined as follows: ； ； in, , y pq Represents the normalized central distance. , The normalization coefficient is... (x0, y0) are the centroid coordinates, where x0 and y0 represent the gray-level centroids of the image in the horizontal and vertical directions, respectively; a set of seven-dimensional feature vectors are extracted using the normalized center distance; The gray-level co-occurrence matrix (GLCM) not only reflects the gray-level distribution characteristics of different types of ships, but also the positional distribution characteristics among pixels with the same or similar gray levels. The original ship image's gray levels are compressed to 16 levels. Feature parameters of the GLCM in the four directions (0°, 45°, 90°, and 135°) are calculated, and the second-order moment A is extracted. sm Entropy E nt The four feature parameters are: the second-order statistic moment of inertia Con and the correlation Corr. The mean and variance of the feature parameters of the four direction matrices are calculated respectively, forming an eight-dimensional feature vector to characterize the texture features of the ship. A total of 15 feature vectors are formed to represent the global features. The local feature matching algorithm includes: extracting local features of the ship using an improved SURF algorithm; in the local feature point extraction stage, the ship image is filtered and a Hessian matrix is ​​constructed to generate stable abrupt change points in the two-dimensional ship image; the Hessian matrix is ​​a square matrix composed of the second-order partial derivatives of a multivariate function, describing the local curvature of the function. For a ship image f(x, y), the Hessian matrix is ​​as follows: ； After constructing the scale space of the ship image using a box filter, each pixel processed by the Hessian matrix is ​​compared with points in the two-dimensional image space and the scale space neighborhood to initially locate points of interest. After filtering out weak feature points and erroneously located feature points, the final stable feature points are selected. Then, the Haar wavelet features in the circular neighborhood of the statistical feature points are used to determine the main direction of the feature points and generate feature vectors of the feature points, which constitute the local feature vectors of the ship.

2. The multi-temporal SAR ship target tracking method as described in claim 1, characterized in that, The multi-temporal SAR ship target tracking method further includes: using Euclidean distance to measure the similarity of overall features between ships; standardizing two images and extracting contour and texture feature vectors using HU invariant moments and GLCM respectively. The feature vector of the image of the ship to be tracked is represented by... This represents the feature vector of the i-th ship image; ； Among them, the Euclidean distance satisfies dist(i) > 0. The smaller the value of the Euclidean distance, the higher the similarity between ship samples. When the Euclidean distance is too large, it will expand the matching range between multiple targets and increase the error of point-to-point matching in local features. However, when the Euclidean distance is too small, factors such as geometric distortion will affect the global feature extraction and will incorrectly remove the correct ship targets with the same name. According to experimental analysis, the threshold of Euclidean distance is set to 1 to make the ship matching performance optimal. The screened ships undergo further matching. The FLANN algorithm is introduced to calculate the similarity between feature points, and ships with the same name are identified based on the number of matching point pairs. The feature space of the FLANN model is an n-dimensional real vector space, named R. n Its core is to find neighboring points based on Euclidean distance; the sub-vectors of feature points m and N are respectively represented by S. m and S N Let the Euclidean distance of D(m,N) be as follows: ; Among them, R n All D(m,N) are stored in several structures based on the KD number part; the minimum Euclidean distance to the query point is searched in the entire KD tree, and then the nearest point of the reference point is searched, and the initial matching set of feature point pairs is obtained. The ship with the same name is determined based on the maximum number of feature point pair matching sets.

3. The multi-temporal SAR ship target tracking method as described in claim 1, characterized in that, The multi-temporal SAR ship target tracking method further includes: using two indicators, the number of correct ship target matches (NCM) and the matching accuracy (MP), to verify the performance of the proposed tracking method; wherein, the calculation method for MP is as follows: ； Where NTM is the total number of matches.

4. A multi-temporal SAR ship target tracking system implementing the multi-temporal SAR ship target tracking method according to any one of claims 1 to 3, characterized in that, The multi-temporal SAR ship target tracking system includes: The target vessel extraction module is used to extract target vessels using the CFAR algorithm. The local feature point enhancement module is used to reduce the impact of SAR background noise and enhance the detection and description capabilities of local feature points through morphological filtering. The feature extraction module is used to extract the outline and internal texture features of the ship and form an overall feature vector. It then uses Euclidean distance to filter out some similar ships, thus achieving a coarse screening. The ship target tracking module is used to achieve the final correct tracking of ship targets using the SURF algorithm based on FLANN feature point matching.

5. A computer device, performing the method of claim 1, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the following steps: First, a global feature matching algorithm is used to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometry and texture features, thus narrowing the matching range between multiple targets and reducing the error range for subsequent local feature matching. Second, a local feature matching algorithm is used to address the problem of partial loss of internal ship information caused by radar operating parameters and ship motion. Finally, the similarity between ships with the same name is measured using the FLANN fast nearest neighbor search function library, thereby achieving precise matching of ships with the same name.

6. A computer-readable storage medium storing a computer program that performs the method of claim 1, characterized in that, When the computer program is executed by the processor, the processor performs the following steps: First, a global feature matching algorithm is used to overcome the problem of difficulty in distinguishing similar ship outlines by utilizing ship geometry and texture features, thus narrowing the matching range between multiple targets and reducing the error range for subsequent local feature matching. Second, a local feature matching algorithm is used to address the problem of partial loss of internal ship information caused by radar operating parameters and ship motion. Finally, the similarity between ships with the same name is measured using the FLANN fast nearest neighbor search function library, thereby achieving precise matching of ships with the same name.

7. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the multi-temporal SAR ship target tracking system as described in claim 4.

Images