A polarimetric SAR ship detection method based on the fusion of statistical quantities of Wasserstein distance

Abstract

The application discloses a polarimetric SAR ship detection method based on fusion of statistics of Wasserstein distance, and relates to the field of polarimetric synthetic aperture radar ship target detection and identification. The method comprises the following steps: S1, acquiring polarimetric synthetic aperture radar data of an original ship and performing filtering processing; S2, performing feature extraction on the polarimetric synthetic aperture radar data after filtering processing to obtain three polarization features; S3, calculating a statistics map of Wasserstein distance of each pixel of a to-be-detected image by using the three polarization features; S4, screening a sea surface background on the statistics map by using a feature prior method, estimating a sea surface background distribution by using a kernel density method combined with a constant false alarm rate detection method, obtaining a threshold according to a preset false alarm rate, and completing pixel-by-pixel judgment on the sea surface background. The application significantly enhances the target / background contrast, and can still achieve high recall, low missed detection and good robustness in complex scenes such as strong sidelobes, strip artifacts and dense targets.

Description

Technical Field

[0001] This invention relates to the field of polarimetric synthetic aperture radar (SAR) ship target detection and identification, and in particular to a polarimetric SAR ship detection method based on Wasserstein distance fusion statistics. Background Technology

[0002] my country possesses vast maritime areas and abundant marine resources. Its coastal regions have developed economies and shipping industries, and maritime transportation plays a vital supporting role in the national economy. In recent years, the situation in the surrounding waters has become increasingly complex, and illegal activities have placed higher demands on maritime order and the protection of rights and interests. To safeguard maritime security and protect sea lanes and marine resources, continuous and accurate target detection of naval vessels is urgently needed. Ship target detection can achieve automated reconnaissance and rapid analysis in military early warning and situational awareness, and in the civilian sector, it can be used for port access control and improved navigation efficiency, thus possessing significant engineering application value.

[0003] Synthetic aperture radar (SAR) is an active microwave remote sensing technology that uses antennas to actively transmit and receive microwaves for ground object detection. It features all-weather, cloud-penetrating, fog-penetrating capabilities and wide-area observation, and has been widely applied in marine monitoring. Traditional single-channel SAR systems can only acquire information on the backscatter intensity of one object, limiting the information dimension. In contrast, fully polarimetric SAR systems can acquire all backscatter information of a target, thus providing a complete interpretation of the scattering mechanism. Therefore, it has become one of the important means of maritime target monitoring in recent years.

[0004] Constant False Alarm Rate (CFAR) is the most widely used target detection algorithm in SAR image processing. CFAR ship detection statistically models the background clutter region, calculates the detection statistics corresponding to the raw data using a pixel-by-pixel sliding window approach, and obtains a detection threshold for the detection statistics based on a given distribution model and a set false alarm rate. This threshold segmentation then binarizes the image to obtain the detection result. When the statistical model matches the actual clutter distribution in a specific scenario, the clutter characteristics are well-fitted, resulting in good CFAR detection performance. However, when the statistical model mismatches with the actual distribution, the performance of CFAR detection will be significantly affected, especially in complex sea clutter backgrounds, where fixed distribution assumptions often fail to provide a sufficient fit. Kernel density estimation is a non-parametric probability density function estimation method, typically using a parzen window to interpolate and fit the probability density using a kernel function centered on the observation sample. When the data sample is large enough, this method can accurately estimate clutter samples. However, traditional kernel density windows often select a fixed local area or the entire image as clutter samples. The former has the problem of a small number of samples, which cannot fully describe the clutter distribution information of the entire image. The latter will mix target pixels into the samples, which will raise the detection threshold and easily lead to more missed detections.

[0005] In traditional CFAR methods, the difference between ship targets and background targets is mainly characterized by radar backscattering intensity. In polarimetric SAR data, the difference between polarimetric scattering mechanisms can further improve the contrast between targets and clutter. In practical applications, the dimensions and value ranges of different polarimetric features are inconsistent, and their sensitivity to sea state and imaging varies. Simple stitching or fixed weights can easily be dominated by a single channel, leading to unstable target-background distinction in different scenarios. Target leakage and sidelobe diffusion in dense scenes can also damage background samples, making weak targets easily obscured in the fusion result. Therefore, how to achieve scale fusion and robust fusion in a multi-dimensional polarimetric feature space, and combine it with adaptive and sample-clean background estimation to obtain a controllable false alarm rate threshold, has become a technical problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to overcome the aforementioned technical deficiencies and propose a ship target detection and recognition method based on Waserstein distance fusion statistics and feature prior kernel density window. This method significantly improves the ship detection performance in dense targets and complex backgrounds, achieving excellent detection accuracy and robustness in experimental data.

[0007] This invention proposes a polarimetric SAR ship detection method based on Wassstein distance fusion statistics. The specific steps are as follows: S1, acquire the original polarimetric synthetic aperture radar data of the ship and perform filtering processing; S2, extract features from the filtered polarimetric synthetic aperture radar data to obtain three polarimetric features; S3, use the three polarimetric features to calculate the statistical map of Wassstein distance for each pixel of the image to be detected; S4, use feature prior to screen the sea surface background on the statistical map, use the kernel density method combined with the constant false alarm rate detection method to estimate the sea surface background distribution, calculate the threshold according to the preset false alarm rate, and perform pixel-by-pixel decision on the sea surface background.

[0008] In some embodiments, step S2 involves extracting features from the filtered polarimetric synthetic aperture radar data to obtain three polarimetric features, wherein the three polarimetric features obtained by feature extraction are... Span and ;in, Span The characteristic is the total scattered power measured by the polarimetric synthetic aperture radar system; It is the first polarization characteristic, used to represent the randomness of target scattering; It is the second polarization characteristic, used to represent the polarization scattering angle.

[0009] In some embodiments, step S3, calculating the Wasserstein distance statistics for each pixel of the image to be detected using the three polarization features, includes the following steps: S31, performing scale normalization on the three polarization features; S32, constructing the detection unit, guard band, and background ring based on local sliding window and feature prior, intersecting the background ring with the feature prior mask to obtain an effective background set; then calculating the Wasserstein distance and forming the statistics map; wherein, step S32 includes constructing a three-dimensional feature vector F through the three polarization features, and calculating the Wasserstein distance between the detection unit in the target region and the effective background set based on the three-dimensional feature vector F of each pixel, that is, calculating the one-dimensional Wasserstein distance between the detection unit and the sea surface background by projection, and then averaging the one-dimensional Wasserstein distance to approximate the three-dimensional Wasserstein distance to obtain the statistics map of the entire image.

[0010] In some embodiments, step S31, performing scale normalization on the three polarization features, includes: processing the polarization features... Span Maximum values ​​are pruned by percentage, and the pruned data is then standardized using quantiles; Using a one-sided nonlinear transformation, for The portion is logarithmically standardized; Dead zone mapping is used to suppress overlapping regions, so that the Span and After the scale standardization process, the three polarization features are ultimately positioned between 0 and 1.

[0011] In some embodiments, the three-dimensional feature vector F is: for each pixel (i,j) in the image to be detected, the three polarization features after standardization are used to construct a three-dimensional feature vector F:

[0012]

[0013] in, Span _n, , Standardized Span and feature.

[0014] In some embodiments, in step S32, the unit to be detected is located at the central pixel position of the image to be detected, and the side length of the window is w; the protective band is located around the window, and the width of the protective band is g, which can prevent data leakage in the unit to be detected; the background ring is located around the protective band, and the width of the background ring is b, which is used as a background candidate area.

[0015] In some embodiments, in step S32, when the effective background set is lower than a preset threshold, the width of the background ring is increased until the effective background set reaches the preset threshold; wherein, the feature prior mask is based on the Obtained by threshold segmentation.

[0016] In some embodiments, in step S32, the Wasserstein distance is projected into a one-dimensional form using the following formula.

[0017]

[0018] Among them, the unit direction is The projection operator is , and Represents the three-dimensional feature vector F Joint distribution P and Q In direction v Projected distribution on, For a uniform measure on a unit sphere, where, in actual calculations, it is used... K The Monte Carlo average in each direction approximates the three-dimensional Wasserstein distance:

[0019]

[0020] In some embodiments, the kernel density method includes selecting a kernel density window and utilizing the... The kernel density window is segmented by a threshold to obtain a region for fitting the clutter probability density distribution, and the clutter statistical distribution is fitted using this region; the step S4, which uses the kernel density method combined with the constant false alarm rate detection method to estimate the sea surface background distribution and calculate the threshold according to the preset false alarm rate, includes: obtaining the threshold based on the clutter statistical distribution fitted by the kernel density window and the preset false alarm rate, and then comparing the pixels of the sea surface background with the threshold to obtain a binary image of the ship detection result.

[0021] In some embodiments, a terminal device is also included, comprising a memory and a processor, the memory storing a computer program running on the processor, wherein the processor executes the computer program to implement the steps of the polarimetric SAR ship detection method based on Wasserstein distance fusion statistics as described in any of the above embodiments.

[0022] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0023] In this invention, firstly, using Span , and Three polarization features are fused using Wasserstein distance to construct a detection statistic, which fully utilizes the information of the full polarization and improves the separability between the target and the background. In addition, when calculating the statistic, samples are drawn from the background ring after prior screening. When the samples are insufficient, the background ring is adaptively increased to maintain sensitivity to weak targets in dense scenes and ensure the reliability of the estimation. Finally, the background is nonparametrically modeled using a kernel density window after feature prior, which incorporates global background clutter information to ensure detection stability in complex backgrounds. Attached Figure Description

[0024] 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.

[0025] Figure 1 This is a flowchart of a polarimetric SAR ship detection method based on Wassstein distance fusion statistics provided in an embodiment of the present invention;

[0026] Figure 2 These are two Pauli pseudo-color images of fully polarimetric SAR experimental data provided in this embodiment of the invention;

[0027] Figure 3 These are two truth maps of fully polarimetric SAR experimental data provided in this embodiment of the invention;

[0028] Figure 4 This is the probability density distribution of the three polarization features provided in the embodiments of the present invention in the ship region and the background region;

[0029] Figure 5 This is a schematic diagram of the statistical calculation area window structure provided in an embodiment of the present invention;

[0030] Figure 6 These are the sea clutter probability density function and Parzen Window fitting results provided in this embodiment of the invention;

[0031] Figure 7 These are two binary images of the detection results of fully polarimetric SAR experimental data provided in this embodiment of the invention;

[0032] Figure 8 This is a block diagram of a ship detection device based on Wassstein distance fusion statistics provided in an embodiment of the present invention;

[0033] Figure 9 This is a schematic diagram of the structure of a ship detection device based on Wassstein distance fusion statistics provided in an embodiment of the present invention. Detailed Implementation

[0034] 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.

[0035] To address the problems existing in the prior art, this invention provides a polarimetric SAR ship detection method based on Wasserstein distance fusion statistics. The invention will be described in detail below with reference to the accompanying drawings.

[0036] like Figure 1 As shown in the figure, the polarimetric SAR ship detection method based on Wassstein distance fusion statistics provided in this embodiment of the invention includes the following steps:

[0037] S1. Acquire raw SAR ship data and perform preprocessing.

[0038] This embodiment uses two fully polarimetric SAR images in the L-band of the Japanese ALOS-2 satellite, with a range resolution of 8.7m and an azimuth resolution of 5.3m. The Pauli pseudo-color image and the corresponding ground truth image are shown below. Figure 2 and Figure 3As shown, the coherence matrix T of the original data is subjected to fine Lee filtering to suppress the influence of coherence speckle noise and to some extent enhance the contrast between ship targets and clutter.

[0039] S2. Feature extraction is performed on the preprocessed polarimetric SAR data to obtain... Span , and Three polarization characteristics.

[0040] S3. Calculate the statistical map of Wasserstein distance for each pixel in the image to be detected using the three polarization features;

[0041] S4. Use feature priors to screen the sea surface background on the statistical map, use kernel density method combined with constant false alarm rate detection method to estimate the distribution of the sea surface background, calculate the threshold according to the preset false alarm rate, and make a pixel-by-pixel decision on the sea surface background.

[0042] S2 involves extracting features from the preprocessed polarimetric SAR data to obtain... Span , and Three polarization characteristics, among which, total polarization power Span The total scattered power measured by the polarimetric SAR system is the sum of the target backscattering intensities, and is defined as:

[0043] (1)

[0044] Where Tr() represents the trace of the matrix, and T is the polarization coherence matrix.

[0045] polarization characteristics and This was proposed by Yin et al. in 2016. Two parameters in the polarimetric decomposition method. (J. Yin, WM Moon and J. Yang, "Novel Model-Based Method for Identification of Scattering Mechanisms in Polarimetric SAR Data," in IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 1, pp. 520-532, Jan. 2016, doi: 10.1109 / TGRS.2015.2461431.)

[0046] in, It is a rotation invariant, namely:

[0047] (2)

[0048] in,

[0049] (3)

[0050] (4)

[0051] in, and These are the estimated average common-polarization amplitude ratio and phase difference. The scattering coefficient is represented when the incident polarization is q and the receiving polarization is p, where H is the horizontal polarization and V is the vertical polarization. The range of values ​​for is [ The amplitude of co-polarized coherence The impact is significant. When When the amplitude ratio and phase difference variance are relatively small, the scattering process exhibits non-uniformity in backscattering behavior. For In larger regions of high coherence, all scattering may originate from a single scattering mechanism. The value is mainly determined by Decision. If so This is equivalent to not considering the randomness of scattering. Pick as follows:

[0052] (5)

[0053] In practical applications, due to incoherent averaging It is less than 1, therefore a new parameter is defined. Used for calculation right Impact:

[0054] (6)

[0055] It can be used to describe the randomness of target scattering. Its sign is determined by the phase difference of the co-polarization channels. If, within a resolving cell, almost all scatterers have the same scattering mechanism, and the orientation angle is consistent with the dielectric constant, then the co-polarization correlation coefficient... The value is relatively large. close to .

[0056] S3, calculating the statistical map of the Wasserstein distance for each pixel in the image to be detected using the three polarization features, is to fully utilize the polarization information of different features to enhance the contrast between the ship target and background clutter for subsequent detection processes, including:

[0057] S31. First, the three polarization features are scaled according to the feature distribution law.

[0058] Figure 4 The probability density distributions of the three polarization feature ship regions and the background region are shown respectively.

[0059] Wherein, the scale standardization is:

[0060] For the total polarization power Span It reflects the overall backscattered energy within a pixel. Ship targets have high intensity, while background areas have low intensity, but are often dominated by a small number of strong pixels, resulting in a large dynamic range. To suppress the influence of extreme values ​​on the feature distribution, the maxima are cropped by a percentage, and the cropped data is then standardized using quantiles.

[0061] The characteristics represent the transition of scattering properties from surface scattering to secondary scattering. Sea surfaces are mostly scattered at smaller angles, while ship targets tend to be scattered at larger angles. In the intermediate segment (such as volume scattering / mixed scattering), the ship target and sea clutter distribution overlap. Linear scaling in this case would amplify the noise in the overlapping area by equal weight. Therefore, a three-segment dead-zone mapping is used. This dead-zone mapping refers to: flattening the intermediate overlapping area (keeping a constant value within this area), and using linear mapping outside the overlapping area. This effectively separates the more discriminative ends, achieving feature standardization. For information on selecting the overlapping area, please refer to [reference needed]. The boundary line of a two-dimensional plane.

[0062] The feature representation indicates the influence of randomness and coherence, directly characterizing the offset caused by random scattering. When surface scattering or volume scattering dominates, Positive values ​​are typically greater than 0, while negative values ​​are typically less than 0 when secondary scattering is dominant. Therefore, positive values ​​generally correspond to sea clutter regions, while negative values ​​primarily indicate ship regions. However, due to interference from sidelobes and artifacts, the central ship region exhibits a more complex scattering mechanism. It approaches 0 (the negative side); however, in scenarios with high scattering complexity due to interference, it shifts significantly towards the negative side, moving away from 0. Therefore, for The feature employs a one-sided nonlinear transformation, only for The target is weighted and then logarithmically standardized to improve the distinction between ship targets and interference signals such as side lobes and artifacts.

[0063] For each pixel (i,j) in the data graph, construct a three-dimensional feature vector from the three standardized polarization features:

[0064] (7)

[0065] in, Span _n, , Standardized Span , , feature.

[0066] S32. Construct the target unit, guard band and background ring based on local sliding window and feature prior, which are used to select the region for calculating the statistics of each pixel.

[0067] The cell under test (CUT) is a square window with an odd side length w centered on each pixel in the image. The set of pixels within the cell under test is denoted as the target region. X A square annular protective zone of width g is set around the unit to be detected. This area is not included in any statistics to avoid the target echo leakage affecting the background estimation. A square annular area of ​​width b is then extended outside the protective zone, which is denoted as the candidate background area. R Specific forms are as follows: Figure 5 As shown.

[0068] Step S32 further includes: intersecting the background ring with the feature prior mask to obtain an effective background set. Specifically, to avoid inaccurate clutter estimation caused by a large number of ship targets mixed into the background of dense target areas, candidate background regions for each pixel are selected. R Remove For pixels with features less than 0, i.e., only sea clutter regions are retained, resulting in a feature prior mask. M The effective background set for each pixel is obtained by intersecting the candidate background ring with the feature prior mask. :

[0069] (8)

[0070] Set background sample number threshold ,like:

[0071] (9)

[0072] When the number of valid background samples is lower than the preset threshold, the width of the background ring b is gradually increased until the number of samples meets the requirements.

[0073] S32 further includes calculating the Wasserstein distance and forming the statistical graph, including projecting the Wasserstein distance into a one-dimensional form.

[0074] Wasserstein distance describes the distance between data distributions. P Transform into distribution Q The minimum cost required at that time. LetP , Q for If two probability distributions on the given surface have finite p-th order moments ( Then the p-th order Wasserstein distance is defined as:

[0075] (10)

[0076] in, For all and P , Q A set of joint distributions with consistent margins. Indicates from position x Transport to location y The quality. In this invention, "quality" refers to the probability weight of a sample in the polarization feature space.

[0077] when Then, the above distance is transformed into a first-order form, namely the first-order Wasserstein distance:

[0078] (11)

[0079] Also known as Earth Mover's Distance (EMD), it describes the area difference between two distribution quantile curves.

[0080] In real sea conditions, the polarization characteristics of targets and sea clutter often exhibit complex patterns such as unilateral shifts and regional overlaps. The Wasserstein distance can simultaneously capture the quantile differences caused by overall displacement and shape changes, and is not easily biased by a small number of strong scattering points or extreme samples. Therefore, this invention constructs three-dimensional features. Through the three-dimensional feature vector of each pixel F Calculate the target region detection unit and the effective background set. The Wasserstein distance is used to obtain the detection statistics for the entire image. Because the three features are scaled uniformly, the statistics measure the overall distribution shape difference, avoiding the situation where a single strong channel dominates the decision.

[0081] Direct computation in high-dimensional space Need to The search for the optimal solution is costly. This invention uses the sliced ​​Wasserstein distance to project the high-dimensional distribution into one dimension along several directions, and then projects the one-dimensional distribution... The average is calculated, balancing efficiency and discriminative power. Let the unit direction be... The projection operator is Then the Sliced ​​Wasserstein distance is expressed as:

[0082] (12)

[0083] in and Representing three-dimensional feature vectors F Joint distribution P and Q In direction v Projected distribution on, It is a uniform measure on a unit sphere. In actual calculations, it is used... K The Monte Carlo average from each direction is approximately approximated as follows:

[0084] (13)

[0085] At this point, each one-dimensional The quantile grid method can be used for rapid calculation, which involves sorting the distributions of the target and background data, performing linear interpolation, and then approximating the integral using the quantile difference:

[0086] (14)

[0087] in, and Let quantile functions be the corresponding values ​​for the two sets of samples. Let be the quantile network selected on (0,1).

[0088] S4. Based on feature priors, screen the sea surface background samples on the statistical map, use the kernel density method combined with the constant false alarm rate detection method to estimate the background distribution, and calculate the threshold according to the preset false alarm rate to complete the pixel-by-pixel decision.

[0089] Among them, the kernel density method is a non-parametric probability density function estimation method. It uses a kernel function centered on the observed samples to interpolate and fit the probability density, which can better estimate complex sea clutter scenes. However, the selection of the kernel density window region is closely related to the final result and should cover a sufficient background area while excluding target pixels. Since the secondary scattering-dominant region usually conforms to polarization characteristics... The characteristics of this feature are used here. Thresholding is performed on the entire image to remove regions dominated by secondary scattering. Specifically, this involves selecting... The region is used as the Parzen Window, and finally, the two sides are lightly clipped to suppress the anomalous tail, resulting in the region used to fit the clutter distribution.

[0090] Assume the background sample set is The probability density function is estimated using a kernel density window, i.e.:

[0091] (15)

[0092] in, This is the kernel function for the kernel density window. Since the obtained detection statistics exhibit a unimodal and approximately symmetrical distribution, a Gaussian kernel function is chosen to fit the clutter.

[0093] (16)

[0094] at this time X The probability density estimation results are as follows:

[0095] (17)

[0096] in, N Indicates the number of background samples. h This is the width of the kernel function, calculated as follows:

[0097] (18)

[0098] in, The standard deviation of the background sample is given.

[0099] Figure 6 The result of fitting the sea clutter probability density function and the Parzen Window is used to adjust the fitted distribution according to the set false alarm rate. The detection threshold is calculated using the following formula:

[0100] (19)

[0101] in, t The threshold value is set to the Wasserstein fusion statistic. Each pixel in the entire image is compared to this threshold for a decision, resulting in a binary image representing the initial ship detection results.

[0102] In an embodiment of the present invention, given a false alarm rate... .

[0103] The preliminary results are then subjected to a closed-loop operation involving expansion followed by corrosion to fill the holes, resulting in the final ship inspection results.

[0104] right Figure 2 Two polarimetric SAR data sets were used for ship detection, and the final results are as follows: Figure 7 As shown, this invention demonstrates excellent detection performance in complex scenes and with dense targets. Tables 1 and 2 present the object-level detection metrics for the two images and compare them with four classic methods.

[0105] Table 1. Evaluation index results of the comparison methods in Experiment Scenario 1

[0106]

[0107] Table 2 Evaluation index results of the comparison methods in Experiment Scenario 2

[0108]

[0109] F1 is an indicator combining recall and false alarm rate; a higher F1 value indicates better actual detection performance. Tables 1 and 2 show that the method of this invention significantly improves recall. Compared to other classic methods, it maintains detection performance even in scenarios with weak or dense targets, indicating that the constructed statistics are more sensitive to the joint distribution differences between the target and the background. Simultaneously, the false alarm rate is more controllable, thanks to the feature prior constraint-based kernel density estimation-constant false alarm rate detection method (Parzen-CFAR), which achieves good thresholds under different background complexities.

[0110] In this invention, firstly, using Span , and Three polarization features are fused using Wasserstein distance to construct a detection statistic, which fully utilizes the information of the full polarization and improves the separability between the target and the background. In addition, when calculating the statistic, samples are drawn from the background ring after prior screening. When the samples are insufficient, the background ring is adaptively increased to maintain sensitivity to weak targets in dense scenes and ensure the reliability of the estimation. Finally, the background is nonparametrically modeled using a kernel density window after feature prior, which incorporates global background clutter information to ensure detection stability in complex backgrounds.

[0111] Figure 8 This is a block diagram of a ship detection device based on Wassstein distance fusion statistics provided in an embodiment of the present invention. This device is used in a polarimetric SAR ship detection method based on Wassstein distance fusion statistics. (Refer to...) Figure 8 The device includes a processing unit 810, a feature extraction unit 820, a detection unit 830, and a pixel decision unit 840. Wherein:

[0112] The processing unit 810 is used to acquire the original polarimetric synthetic aperture radar data of the ship and perform filtering processing.

[0113] The feature extraction unit 820 is used to extract features from the filtered polarimetric synthetic aperture radar data to obtain three polarimetric features.

[0114] The detection unit 830 is used to calculate a statistical map of the Wasserstein distance of each pixel in the image to be detected using the three polarization features.

[0115] The pixel decision unit 840 uses feature priors to screen the sea surface background on the statistical map, uses kernel density method combined with constant false alarm rate detection method to estimate the distribution of the sea surface background, calculates the threshold according to the preset false alarm rate, and performs pixel-by-pixel decision on the sea surface background.

[0116] In this invention, firstly, using Span , and Three polarization features are fused using Wasserstein distance to construct a detection statistic, which fully utilizes the information of the full polarization and improves the separability between the target and the background. In addition, when calculating the statistic, samples are drawn from the background ring after prior screening. When the samples are insufficient, the background ring is adaptively increased to maintain sensitivity to weak targets in dense scenes and ensure the reliability of the estimation. Finally, the background is nonparametrically modeled using a kernel density window after feature prior, which incorporates global background clutter information to ensure detection stability in complex backgrounds.

[0117] Figure 9 This is a schematic diagram of the structure of a ship detection device based on Wassstein distance fusion statistics provided in an embodiment of the present invention, as shown below. Figure 9 As shown, ship detection equipment based on Wasserstein distance fusion statistics can include the above-mentioned... Figure 8 The ship detection device shown is based on Wasserstein distance fusion statistics. Optionally, the ship detection device 910 based on Wasserstein distance fusion statistics may include a first processor 2001.

[0118] Optionally, the ship detection device 910 based on Wasserstein distance fusion statistics may also include a memory 2002 and a transceiver 2003.

[0119] The first processor 2001, memory 2002, and transceiver 2003 can be connected via a communication bus.

[0120] The following is combined Figure 9 A detailed introduction to each component of the Wasserstein distance fusion statistics-based ship detection equipment 910 is provided below:

[0121] The first processor 2001 is the control center of the ship detection device 910 based on Wasserstein distance fusion statistics. It can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0122] Optionally, the first processor 2001 can perform various functions of the ship detection device 910 based on Wassstein distance fusion statistics by running or executing software programs stored in memory 2002 and calling data stored in memory 2002.

[0123] In a specific implementation, as one example, the first processor 2001 may include one or more CPUs, for example... Figure 9 CPU0 and CPU1 are shown in the diagram.

[0124] In a specific implementation, as one example, the ship detection device 910 based on Wassstein distance fusion statistics can also include multiple processors, for example... Figure 9 The first processor 2001 and the second processor 2004 are shown in the diagram. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). Here, a processor can refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0125] The memory 2002 is used to store the software program that executes the present invention, and is controlled by the first processor 2001 to execute it. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0126] Optionally, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or may exist independently, and may be connected via the interface circuit of the ship detection device 910 based on Wasserstein distance fusion statistics. Figure 9 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0127] The transceiver 2003 is used to communicate with network devices or with terminal devices.

[0128] Alternatively, transceiver 2003 may include a receiver and a transmitter. Figure 9 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0129] Optionally, the transceiver 2003 can be integrated with the first processor 2001, or it can exist independently, and can be connected to the interface circuit of the ship detection device 910 based on Wasserstein distance fusion statistics. Figure 9 (Not shown in the image) is coupled to the first processor 2001, and this embodiment of the invention does not specifically limit this.

[0130] It should be noted that, Figure 9 The structure of the ship detection device 910 based on Wasserstein distance fusion statistics shown in the figure does not constitute a limitation on the router. Actual ship detection devices based on Wasserstein distance fusion statistics may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0131] Furthermore, the technical effect of the ship detection equipment 910 based on Wassstein distance fusion statistics can be referred to the technical effect of the polarization SAR ship detection method based on Wassstein distance fusion statistics described in the above method embodiments, and will not be repeated here.

[0132] It should be understood that the first processor 2001 in the embodiments of the present invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or it may be any conventional processor, etc.

[0133] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0134] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. 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., 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 includes one or more sets of 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. A semiconductor medium can be a solid-state drive.

[0135] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0136] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0137] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0138] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0139] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0143] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] 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 variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A polarimetric SAR ship detection method based on Wassstein distance fusion statistics, the specific steps of which are as follows: S1. Acquire the original polarimetric synthetic aperture radar data of the ship and perform filtering processing; S2. Feature extraction is performed on the filtered polarimetric synthetic aperture radar data to obtain three polarimetric features; among them... The three polarization features obtained from the feature extraction are Span, and Among them, the Span feature is the total scattered power measured by the polarimetric synthetic aperture radar system; It is the first polarization characteristic, used to represent the randomness of target scattering; It is the second polarization characteristic, used to represent the polarization scattering angle; S3. Calculate the statistical map of Wasserstein distance for each pixel in the image to be detected using the three polarization features; S4. Use feature priors to screen the sea surface background on the statistical map, use kernel density method combined with constant false alarm rate detection method to estimate the distribution of sea surface background and calculate the threshold according to the preset false alarm rate, and make a pixel-by-pixel decision on the sea surface background. Specifically, S3, calculating the statistical map of the Wasserstein distance for each pixel in the image to be detected using the three polarization features, includes the following steps: S31. Perform scale standardization on the three polarization features; S32. Construct the target unit, guard band, and background ring based on local sliding window and feature prior. Intersect the background ring with the feature prior mask to obtain an effective background set. Then, construct a three-dimensional feature vector F through the three polarization features. Calculate the Wasserstein distance between the target unit and the effective background set based on the three-dimensional feature vector F of each pixel. That is, calculate the one-dimensional Wasserstein distance between the target unit and the sea surface background by projection. Then, average the one-dimensional Wasserstein distance to approximate the three-dimensional Wasserstein distance to obtain the statistical map of the entire image. In S32, The Wasserstein distance can be projected into a one-dimensional form using the following formula. ； Among them, the unit direction is The projection operator is , and This represents the projection distribution of the joint distribution P and Q formed by the three-dimensional feature vectors F onto the direction v. For uniform measurement on a unit sphere, The first-order Wasserstein distance; In actual calculations, the three-dimensional Wasserstein distance is approximated using the Monte Carlo average of K directions: 。 2. The polarimetric SAR ship detection method based on Wassstein distance fusion statistics according to claim 1, wherein, S31, the scale normalization process for the three polarization features includes: The maximum values ​​of the polarization feature Span are truncated by a percentage, and the truncated data is then standardized by quantiles. The Using a one-sided nonlinear transformation, for Log-standardize the portion; The Dead zone mapping is used to suppress overlapping regions. Make the Span, and After the scale standardization process, the three polarization features are ultimately positioned between 0 and 1.

3. The polarimetric SAR ship detection method based on Wassstein distance fusion statistics according to claim 2, wherein, The three-dimensional feature vector F is constructed by using the standardized three polarization features to construct a three-dimensional feature vector F for each pixel (i,j) in the image to be detected. ； Wherein, Span_n, , These are the standardized Span, and feature.

4. The polarimetric SAR ship detection method based on Wassstein distance fusion statistics according to claim 3, wherein, In step S32, The detection unit is located at the central pixel position of the image to be detected, and the window side length of the detection unit is w; The protective strip is disposed around the window, and the width of the protective strip is g. The protective strip can prevent data leakage in the unit to be detected. The background ring is disposed around the protective strip, and the width of the background ring is b. The background ring is used as a background candidate area.

5. A polarimetric SAR ship detection method based on Wassstein distance fusion statistics according to claim 4, wherein, In step S32, when the effective background set is lower than a preset threshold, the width of the background ring is increased until the effective background set reaches the preset threshold. Wherein, the feature prior mask is based on the Obtained by threshold segmentation.

6. The polarimetric SAR ship detection method based on Wassstein distance fusion statistics according to claim 5, wherein the kernel density method includes selecting a kernel density window and utilizing the... The kernel density window is segmented by a threshold to obtain a region for fitting the clutter probability density distribution, and the clutter statistical distribution is obtained by fitting this region. The step S4, which involves estimating the sea surface background distribution using the kernel density method combined with the constant false alarm rate detection method and calculating the threshold according to the preset false alarm rate, includes: A threshold is obtained based on the clutter statistical distribution fitted by the kernel density window and the preset false alarm rate. Then, the pixels of the sea surface background are compared with the threshold to obtain a binary image of the ship detection result.

7. A terminal device, comprising a memory and a processor, wherein the memory stores a computer program running on the processor, wherein, When the processor executes the computer program, it implements the steps of the polarization SAR ship detection method based on Wasserstein distance fusion statistics as described in any one of claims 1 to 6.

Images