A SAR ship detection method and device combining space-time context information

Abstract

This application relates to a SAR ship detection method and apparatus that incorporates spatiotemporal context information. The method includes: performing sub-view processing on a SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times; constructing a spatiotemporal context covariance matrix based on the SAR sub-views; performing a similarity test on the spatiotemporal context covariance matrix to obtain a similar sample count feature; utilizing the similar sample count feature to enhance the SAR ship image to be detected to obtain an enhanced SAR ship image; and performing ship detection on the enhanced SAR ship image using an adaptive threshold processing method and morphological filtering to obtain a ship detection result. This method can improve ship detection performance.

Description

Technical Field

[0001] The present application relates to the field of radar remote sensing application technology, and in particular to a SAR ship detection method, apparatus, computer equipment, and storage medium that combine space-time context information. Background Art

[0002] Ships are a major man-made target at sea, and robust detection of them is of great value and significance for both military and civilian applications. Compared to optical sensors, synthetic aperture radar (SAR), capable of all-weather and all-day operation, offers unique advantages in ship detection applications. SAR can monitor large areas of the ocean and acquire electromagnetic scattering information, such as amplitude and phase, from scene targets, enhancing target detection capabilities. Currently, ship target detection based on SAR images has become a research hotspot and garnered widespread attention.

[0003] However, current SAR image ship detection methods have employed superpixel segmentation to extract two-dimensional spatial contextual information, overcoming the limitations of inaccurate pixel-level clutter modeling and improving nearshore ship detection accuracy. Furthermore, deep learning-based ship detection methods, which have garnered widespread attention in recent years, have also taken note of the two-dimensional spatial contextual information inherent in SAR images. These methods, such as designing convolution kernels or contextual information extraction modules, extract this information to improve detection accuracy. In addition to spatial contextual information, SAR images also contain temporal contextual information. Subview analysis is a representative method for utilizing this temporal contextual information for ship detection. These methods generate sub-images corresponding to different imaging moments through sub-view processing, and then detect ships based on differences in the temporal and frequency stationarity of the ship and sea in these sub-views. Although numerous ship detection methods have been developed that utilize contextual information in SAR images, ship targets exhibit certain spatial distribution characteristics in medium- and high-resolution radar imagery. First, ship targets are not simply independent strong scattering points, but often occupy multiple consecutive pixel locations in spatial distribution. Due to the complex structure of these adjacent pixels, the scattering mechanisms often vary significantly, exhibiting heterogeneity different from that of the sea surface. Second, ship targets are often separated by a certain distance, resulting in a sparse spatial distribution, unlike the continuous sea surface. Furthermore, ship targets exhibit significant anisotropy, and their scattering fluctuation characteristics differ from those of the sea surface throughout the synthetic aperture time. Consequently, existing methods suffer from poor detection performance in medium- and high-resolution radar imagery. Summary of the Invention

[0004] Based on this, it is necessary to provide a SAR ship detection method, device, computer equipment and storage medium that combines space-time context information to improve ship detection performance in response to the above technical problems.

[0005] A SAR ship detection method combining space-time context information, the method comprising:

[0006] Acquire a SAR ship image to be detected; perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times;

[0007] Construct a space-time context covariance matrix based on the SAR subviews, perform a similarity test on the space-time context covariance matrix, and obtain the similar sample number feature;

[0008] The similar sample number feature is used to enhance the SAR ship image to be detected, and the enhanced SAR ship image is obtained;

[0009] The enhanced SAR ship image is used for ship detection based on the adaptive threshold processing method and morphological filtering to obtain the ship detection results.

[0010] In one embodiment, sub-view processing is performed on a SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times, including:

[0011] Perform one-dimensional Fourier transform of SAR ship images in azimuth to obtain full-aperture azimuth spectrum;

[0012] The amplitude of the full-aperture azimuth spectrum is averaged in the range direction to obtain a weight function, and the inverse of the weight function is calculated and normalized to obtain a correction function;

[0013] The full aperture azimuth spectrum is optimized according to the correction function to obtain an optimized full aperture azimuth spectrum;

[0014] The optimized full aperture azimuth spectrum is divided into r% overlapping parts according to the required number of sub-apertures to obtain N sub-aperture spectra;

[0015] The center of each sub-aperture spectrum is moved to zero Doppler, each sub-aperture spectrum is windowed in azimuth and subjected to an azimuth one-dimensional inverse Fourier transform to obtain N frames of SAR sub-views corresponding to different synthetic aperture times.

[0016] In one embodiment, constructing a spatiotemporal context covariance matrix based on the SAR subviews includes:

[0017] Obtain a pixel set from the SAR subview, determine the neighborhood space of the pixel set, and construct a space-time context vector set of the pixel set in the neighborhood space;

[0018] A spatiotemporal context covariance matrix is ​​constructed from a set of spatiotemporal context vectors.

[0019] In one embodiment, constructing a set of spatiotemporal context vectors for a set of pixels in a neighborhood space includes:

[0020] The space-time context vector set of the pixel set constructed in the neighborhood space is

[0021]

[0022] Among them, v i is pixel p (x,y,t) The corresponding i-th context vector, x, y, t are the coordinates of the pixel on the azimuth axis, distance axis, and time axis. x and y are used to represent the coordinates of the pixel in the two-dimensional image, and t is used to represent the number of image frames. Both represent pixel p (x,y,t) Pixels in the neighborhood K, N v is the number of context vectors in the set V, N p is pixel p (x,y,t) The number of pixels in the context vector v.

[0023] In one embodiment, constructing a spatiotemporal context covariance matrix based on a set of spatiotemporal context vectors includes:

[0024] The space-time context covariance matrix is ​​constructed based on the set of space-time context vectors:

[0025]

[0026] Here, the superscript H represents the conjugate transpose.

[0027] In one embodiment, a similarity test is performed on the spatiotemporal context covariance matrix to obtain a similar sample number feature, including:

[0028] Determine the similar sample selection neighborhood of the pixel set, and calculate the similarity index between the original pixel in the pixel set and the similar pixels in the similar sample selection neighborhood that are different from the central pixel based on the spatiotemporal context covariance matrix and the spatiotemporal context covariance matrix corresponding to multiple similar pixels in the similar sample selection neighborhood; the neighborhood is a rectangular window with a size of N1×N2 in the range and azimuth directions, centered on any original pixel in the pixel set;

[0029] If the similarity index is not less than the preset similarity judgment threshold, the similar pixels corresponding to the similarity index are classified as similar samples of the original pixel, and the set of all similar samples of the original pixel in a certain neighborhood range is obtained;

[0030] The number of similar samples of each pixel in a certain neighborhood range is determined based on the set of all similar samples of the original pixel in a certain neighborhood range; the similar sample number feature is calculated using the similar sample number and the rectangular window size.

[0031] In one embodiment, the similar sample number feature is obtained by calculating the similar sample number and the rectangular window size, including:

[0032] The similar sample number feature is calculated using the similar sample number and rectangular window size:

[0033]

[0034] Among them, SPN represents the number of similar samples.

[0035] In one embodiment, the SAR ship image to be detected is enhanced using the similar sample number feature to obtain the enhanced SAR ship image, including:

[0036] The SAR ship image to be detected is enhanced using the similar sample number feature, and the enhanced SAR ship image is obtained as follows:

[0037] I2=(1-k) M ×I1

[0038] Where I1 and I2 represent the SAR ship image to be detected and the enhanced SAR ship image, respectively, (1-k) M is the weight term, k is the number of similar sample features, and M is the construction index term.

[0039] In one embodiment, ship detection is performed on the enhanced SAR ship image according to the adaptive threshold processing method and morphological filtering to obtain a ship detection result, including:

[0040] The probability density functions of ship targets and sea clutter in the samples are estimated based on the artificially selected ship samples and the true values. Then, the intersection of the probability density functions of ship targets and sea clutter is obtained according to the minimum false positive criterion to obtain the detection threshold.

[0041] The enhanced SAR ship image is binarized using the detection threshold to obtain the preliminary detection results.

[0042] The false alarms whose area is less than half of the preset minimum area of ​​the ship to be detected in the preliminary detection results are removed by morphological filtering to obtain the ship detection results.

[0043] A SAR ship detection device combining space-time context information, the device comprising:

[0044] The sub-view processing module is used to obtain the SAR ship image to be detected; perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times;

[0045] A space-time context covariance matrix construction module is used to construct the space-time context covariance matrix according to the SAR subviews, perform similarity test on the space-time context covariance matrix, and obtain similar sample number features;

[0046] An image enhancement module is used to enhance the SAR ship image to be detected by using the similar sample number feature to obtain an enhanced SAR ship image;

[0047] The ship detection module is used to perform ship detection on the enhanced SAR ship image according to the adaptive threshold processing method and morphological filtering to obtain the ship detection result.

[0048] The above-mentioned SAR ship detection method, device, computer equipment and storage medium combined with space-time context information first obtain SAR subviews corresponding to different synthetic aperture times by performing subview processing on the full-aperture single-view complex SAR image, then use the SAR subview to construct the space-time context covariance matrix to extract the target space-time context information, then combine the space-time context covariance matrix similarity test to obtain the similar sample number feature, then use the similar sample number feature to weighted enhance the ship target of the original SAR image, and finally combine adaptive threshold processing and morphological filtering to achieve ship target detection. The present application extracts the space-time context information in the SAR image by subview processing and constructing the space-time context covariance matrix, and then enhances the ship target by weighting the similar sample number feature, and finally combines adaptive threshold processing to achieve ship detection. It can improve the ship detection performance by utilizing the polarization scattering information of the target in different spatial resolution units and different synthetic aperture times contained in the SAR image, can effectively improve the ship-sea contrast, and reduce the false alarm and missed detection of ship detection in low ship-sea contrast conditions such as large incident angle, providing technical support for sea surface monitoring applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 1 is a flow chart of a SAR ship detection method combining space-time context information in one embodiment;

[0050] Figure 2 Schematic diagram of a sub-view processing flow in one embodiment;

[0051] Figure 3 A schematic diagram of a spatiotemporal context vector according to an embodiment;

[0052] Figure 4 : The full aperture image and sub-view of the GF-3 HV channel in another embodiment; (a) is the full aperture image, (b) is the sub-view Figure 1 , (c) Sub-view Figure 2 , (d) sub-view Figure 3 ;

[0053] Figure 5 is a schematic diagram of a similar sample number feature in one embodiment;

[0054] Figure 6 is a schematic diagram of an enhanced SAR ship image in one embodiment;

[0055] Figure 7 A schematic diagram of threshold determination in one embodiment;

[0056] Figure 8 This is a schematic diagram of the ship detection results of Gaofen-3 data in one embodiment;

[0057] Figure 9 Schematic diagram of ship detection results using Radarsat-2 data in one embodiment; (a) is a schematic diagram of detection results using the SO-CFAR algorithm, (b) is a schematic diagram of detection results using the iterative CFAR algorithm, (c) is a schematic diagram of detection results using the superpixel method, and (d) is a schematic diagram of detection results using the present application;

[0058] Figure 10 This is a structural block diagram of a SAR ship detection device that combines space-time context information in one embodiment;

[0059] Figure 11 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0061] In one embodiment, Figure 1 As shown, a SAR ship detection method combining space-time context information is provided, comprising the following steps:

[0062] Step 102: Acquire a SAR ship image to be detected; perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times.

[0063] The SAR ship images to be detected are not limited to HV polarimetric SAR data, and single polarimetric SAR, dual polarimetric SAR, reduced polarimetric SAR, full polarimetric SAR data and their derived feature data can also be processed similarly.

[0064] Step 104 : constructing a space-time context covariance matrix based on the SAR sub-views, performing a similarity test on the space-time context covariance matrix, and obtaining a similar sample number feature.

[0065] The spatial-temporal context covariance matrix is ​​constructed using SAR subviews to extract the target spatial-temporal context information.

[0066] Step 106 : performing enhancement processing on the SAR ship image to be detected using the similar sample number feature to obtain an enhanced SAR ship image.

[0067] The similar sample number feature is used to weightedly enhance the ship target in the original SAR image, thereby improving the detection performance.

[0068] Step 108: Perform ship detection on the enhanced SAR ship image according to the adaptive threshold processing method and morphological filtering to obtain a ship detection result.

[0069] The enhanced SAR ship image is subjected to ship detection based on the adaptive threshold processing method and morphological filtering to filter out false alarms and improve the accuracy of the detection results.

[0070] In the above-mentioned SAR ship detection method combining spatiotemporal context information, first, SAR subviews corresponding to different synthetic aperture times are obtained by performing subview processing on the full-aperture single-view complex SAR image, and then the spatiotemporal context covariance matrix is ​​constructed using the SAR subview to extract the target spatiotemporal context information, and then the similarity test of the spatiotemporal context covariance matrix is ​​combined to obtain the similar sample number feature, and then the similar sample number feature is used to weight the original SAR image to enhance the ship target, and finally, the ship target detection is achieved by combining adaptive threshold processing and morphological filtering. The present application extracts the spatiotemporal context information in the SAR image by subview processing and constructing the spatiotemporal context covariance matrix, and then enhances the ship target by weighting the similar sample number feature, and finally achieves ship detection by combining adaptive threshold processing. It can improve the ship detection performance by utilizing the polarization scattering information of the target in different spatial resolution units and different synthetic aperture times contained in the SAR image, can effectively improve the ship-sea contrast, and reduce the false alarm and missed detection of ship detection in low ship-sea contrast conditions such as large incident angle, providing technical support for sea surface monitoring applications.

[0071] In one embodiment, sub-view processing is performed on a SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times, including:

[0072] Perform one-dimensional Fourier transform of SAR ship images in azimuth to obtain full-aperture azimuth spectrum;

[0073] The amplitude of the full-aperture azimuth spectrum is averaged in the range direction to obtain a weight function, and the inverse of the weight function is calculated and normalized to obtain a correction function;

[0074] The full aperture azimuth spectrum is optimized according to the correction function to obtain an optimized full aperture azimuth spectrum;

[0075] The optimized full aperture azimuth spectrum is divided into r% overlapping parts according to the required number of sub-apertures to obtain N sub-aperture spectra;

[0076] The center of each sub-aperture spectrum is moved to zero Doppler, each sub-aperture spectrum is windowed in azimuth and subjected to an azimuth one-dimensional inverse Fourier transform to obtain N frames of SAR sub-views corresponding to different synthetic aperture times.

[0077] In a specific embodiment, the sub-view processing obtains SAR sub-views corresponding to different synthetic aperture times, such as Figure 2 As shown, it mainly includes:

[0078] Azimuth Fourier transform: First, a full-aperture single-look complex (SLC) SAR image data is transformed into the range-Doppler domain by performing an azimuth one-dimensional Fourier transform to obtain the full-aperture azimuth spectrum.

[0079] Window effect removal: The full-aperture azimuth spectrum amplitude is averaged in the range direction to obtain a weighting function. The inverse of the weighting function is calculated and normalized to obtain a correction function. The correction function is applied to the full-aperture azimuth spectrum to eliminate the effects of windowing in the azimuth antenna pattern and imaging process.

[0080] Subaperture division: the spectrum data is divided into N subaperture spectra with an overlap of r% according to the required number of subapertures;

[0081] Sub-aperture windowing: Move the center of each sub-aperture spectrum to zero Doppler, and then perform azimuth windowing on each sub-aperture spectrum to improve the response to related point targets;

[0082] Azimuth inverse Fourier transform: Finally, each sub-aperture spectrum is transformed back to the time domain by performing an azimuth one-dimensional inverse Fourier transform, obtaining N frames of SAR sub-views corresponding to different synthetic aperture times.

[0083] In one embodiment, constructing a spatiotemporal context covariance matrix based on the SAR subviews includes:

[0084] Obtain a pixel set from the SAR subview, determine the neighborhood space of the pixel set, and construct a space-time context vector set of the pixel set in the neighborhood space;

[0085] A spatiotemporal context covariance matrix is ​​constructed from a set of spatiotemporal context vectors.

[0086] In a specific embodiment, typically, the pixel p (x,y,t) The neighborhood space K is a pixel p (x,y,t) A w1×w2×w3 cubic pixel block is centered, and a typical value is w1=w2=w3=3;

[0087] like Figure 3 As shown, a typical construction method of the space-time context scattering vector is: v1 = [S1 S 14 S 27 ] T ,v2=[S2 S 14 S 26 ] T ,v3=[S3 S 14 S 25 ] T v4=[S4 S 14 S 24 ] T ,v5=[S5 S 14 S 23 ] T ,v6=[S6 S 14 S 22 ] T v7=[S7 S 14 S 21 ] T ,v8=[S8 S 14 S 20 ] T ,v9=[S9 S 14 S 19 ] T v 10 =[S 10 S 14 S 18 ] T ,v 11 =[S 11 S 14 S 17 ] T ,v 12 =[S 12 S 14 S 16 ] T v 13 =[S 13 S 14 S 15 ] T

[0088] V={v1,v2,v3,v4,v5,v6,v7,v8,v9,v 10 ,v 11 ,v 12 ,v 13}

[0089] In one embodiment, constructing a set of spatiotemporal context vectors for a set of pixels in a neighborhood space includes:

[0090] The space-time context vector set of the pixel set constructed in the neighborhood space is

[0091]

[0092] Among them, v i is pixel p (x,y,t) The corresponding i-th context vector, x, y, t are the coordinates of the pixel on the azimuth axis, distance axis, and time axis. x and y are used to represent the coordinates of the pixel in the two-dimensional image, and t is used to represent the number of image frames. Both represent pixel p (x,y,t) Pixels in the neighborhood K, N v is the number of context vectors in the set V, N p is pixel p (x,y,t) The number of pixels in the context vector v.

[0093] In a specific embodiment, the pixel p (x,y,t) The position in the context vector is consistent; all constructed context vectors , except for pixel p (x,y,t) Except for , other neighboring pixels are not reused; N p Consistent, usually N p =w1, at this time p (x,y,t) is the middle element of the context vector, N v Take 13.

[0094] In one embodiment, constructing a spatiotemporal context covariance matrix based on a set of spatiotemporal context vectors includes:

[0095] The space-time context covariance matrix is ​​constructed based on the set of space-time context vectors:

[0096]

[0097] Here, the superscript H represents the conjugate transpose.

[0098] In one embodiment, a similarity test is performed on the spatiotemporal context covariance matrix to obtain a similar sample number feature, including:

[0099] Determine the similar sample selection neighborhood of the pixel set, and calculate the similarity index between the original pixel in the pixel set and the similar pixels in the similar sample selection neighborhood that are different from the central pixel based on the spatiotemporal context covariance matrix and the spatiotemporal context covariance matrix corresponding to multiple similar pixels in the similar sample selection neighborhood; the neighborhood is a rectangular window with a size of N1×N2 in the range and azimuth directions, centered on any original pixel in the pixel set;

[0100] If the similarity index is not less than the preset similarity judgment threshold, the similar pixels corresponding to the similarity index are classified as similar samples of the original pixel, and the set of all similar samples of the original pixel in a certain neighborhood range is obtained;

[0101] The number of similar samples of each pixel in a certain neighborhood range is determined based on the set of all similar samples of the original pixel in a certain neighborhood range; the similar sample number feature is calculated using the similar sample number and the rectangular window size.

[0102] In a specific embodiment,

[0103] In one embodiment, the similar sample number feature is obtained by calculating the similar sample number and the rectangular window size, including:

[0104] The similar sample number feature is calculated using the similar sample number and rectangular window size:

[0105]

[0106] Among them, SPN represents the number of similar samples.

[0107] In a specific embodiment, the pixel p is determined (x,y,t) The neighborhood W is selected from similar samples. The neighborhood W is generally based on pixel p. (x,y,t) A rectangular window with the size of N1×N2 in the distance and azimuth directions is centered on the image.

[0108] Specifically, with pixel p (x,y,t) The corresponding space-time context covariance matrix is ​​M center , other N in the neighborhood window W neighbor =N1×N2 pixels corresponding to the space-time context covariance matrix is Calculate the center pixel point by point according to the following formula and calculate its relationship with other pixels in the neighborhood window W Similarity index

[0109]

[0110] Among them, N p is the length of the context vector v corresponding to the pixel; |·| means taking the absolute value of the matrix determinant; ln means taking the natural logarithm; the subscript i represents the i-th pixel in the corresponding neighborhood window.

[0111] like Greater than or equal to the similarity judgment threshold th simitest , then the pixel Classified as pixel p (x,y,t) Similar samples of pixel p (x,y,t)The set U of all similar samples in a certain neighborhood range can then be used to count the number of similar samples SPN in a certain neighborhood range for each pixel.

[0112] The number of similar samples is related to the similarity threshold and the size of the neighborhood considered. The similarity threshold is determined by the radar image view count and is generally adjusted between -3 and -0.1. For medium- and high-resolution radar images, the neighborhood considered is usually a 15×15 neighborhood window.

[0113] Generally speaking, the larger the neighborhood range considered, the more likely it is to find more similar samples. To facilitate analysis and comparison, we can further calculate the proportion of similar samples to the neighborhood range considered, that is,

[0114]

[0115] In the formula, the ratio k is the number of similar samples, and it can be seen that k∈[0,1]

[0116] In one embodiment, the SAR ship image to be detected is enhanced using the similar sample number feature to obtain the enhanced SAR ship image, including:

[0117] The SAR ship image to be detected is enhanced using the similar sample number feature, and the enhanced SAR ship image is obtained as follows:

[0118] I2=(1-k) M ×I1

[0119] Where I1 and I2 represent the SAR ship image to be detected and the enhanced SAR ship image, respectively, (1-k) M is the weight term, k is the number of similar sample features, and M is the construction index term.

[0120] In a specific embodiment, the exponential term M is constructed to further increase the distance between pixels with weights close to 1 and pixels with weights close to 0.

[0121] In one embodiment, ship detection is performed on the enhanced SAR ship image according to the adaptive threshold processing method and morphological filtering to obtain a ship detection result, including:

[0122] The probability density functions of ship targets and sea clutter in the samples are estimated based on the artificially selected ship samples and the true values. Then, the intersection of the probability density functions of ship targets and sea clutter is obtained according to the minimum false positive criterion to obtain the detection threshold.

[0123] The enhanced SAR ship image is binarized using the detection threshold to obtain the preliminary detection results.

[0124] The false alarms whose area is less than half of the preset minimum area of ​​the ship to be detected in the preliminary detection results are removed by morphological filtering to obtain the ship detection results.

[0125] In a specific embodiment, the GF-3 large grazing angle HV channel polarimetric SAR data is taken as an example. Figure 4 As shown in (a), the data was acquired by Gaofen-3 in a certain sea area on February 28, 2022. The grazing angle is 66.8°~67.4°, the azimuth resolution is 4.7m, the range resolution is 2.3m, and the data size is 1500×1200 (azimuth×range). There are 23 small boats in the image, but under the condition of large grazing angle, the contrast between the ship and the sea is low. The full aperture spectrum is divided into three sub-views, and the azimuth spectrum overlap rate is set to 80%. After the following steps, Figure 2 After the sub-view processing shown in Figure 4 (b) Figure 4 (c) and Figure 4 (d) shows three sub-views. It should be noted that the input data is not limited to HV polarimetric SAR data, and single-polarimetric SAR, dual-polarimetric SAR, reduced polarimetric SAR, and full polarimetric SAR data can also be processed similarly.

[0126] The space-time context covariance matrix corresponding to each pixel in the second frame sub-view is constructed using the three frame sub-views. Among them, the space-time context vector construction neighborhood K is a 3×3×3 cubic pixel block, according to Figure 3 13 space-time context vectors are constructed in the manner shown, and then a 3×3 space-time context covariance matrix is ​​constructed.

[0127] Combined with the spatial-temporal context covariance matrix similarity test, the similarity sample number feature of each pixel in the second frame sub-view is obtained. The similarity sample selection neighborhood size is 15×15, and the similarity threshold th simitest According to the radar image view count, take -2. The similar sample number characteristics of the whole image are as follows Figure 5 shown.

[0128] The original full-aperture SAR image is further weighted using the similar sample number feature to obtain the following Figure 6 The image of the ship target is enhanced, and it can be seen that compared with Figure 4 (a), the ship target is significantly enhanced, which verifies that this application can effectively improve the contrast between ship and sea. Among them, the weight term (1-k) M The index M is set to 4. It should be noted that the raw data is not limited to HV polarimetric SAR data, and single-polarimetric SAR, dual-polarimetric SAR, reduced polarimetric SAR, full polarimetric SAR data and their derived feature data can also be processed similarly.

[0129] Random selection Figure 6The ship and sea in the white box are used to estimate the probability density function of ship and sea clutter. Figure 7 The threshold determination diagram is shown in Figure 1. According to the intersection of the ship-sea probability density curve, the detection threshold TH is determined to be 0.85. After the detection in the logarithmic domain, morphological filtering is performed to filter out small area false alarms. Figure 8 The final detection results are shown in Figure 2. As can be seen from the figure, all the ship targets in the figure are detected.

[0130] To further verify the detection performance of the proposed method, a comparative experiment was conducted using polarimetric SAR data acquired by Radarsat-2 in a certain sea area on December 16, 2008. The data incidence angle was 61.14° to 61.47°, the data resolution was 9.5m in azimuth, 4.8m in range, and the data size was 300×500 (azimuth × range). There were 138 small boats in the image. The SO-CFAR method, iterative CFAR, and a saliency method based on superpixel segmentation were selected as comparison methods. The experiment used the target-level evaluation index Figure of Merit (FoM) to quantitatively evaluate the detection results. Its calculation formula is:

[0131]

[0132] Where N C 、N FA and N M They represent the number of correct detections, the number of false alarms, and the number of missed detections, respectively.

[0133] Figure 9 The following figure shows the ship detection results from the Radarsat-2 comparative experiment. Rectangular boxes indicate correct detections, triangular boxes indicate false alarms, and prismatic boxes indicate missed detections. As can be seen from the figure, the proposed method effectively reduces missed detections and maintains a good outline of the detected ships. Based on the quantitative results shown in Table 1, the proposed method achieves a FoM value of 96.4%, exceeding SO-CFAR's 89.2%, iterative CFAR's 88.5%, and superpixel methods' 94.2%, achieving the best detection results.

[0134] Table 1

[0135]

[0136] In summary, this application mainly uses the spatiotemporal context information contained in SAR images to enhance ship targets and improve ship target detection performance. Comparative experiments verify that this application can effectively improve ship target detection performance.

[0137] It should be understood that although Figure 1The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0138] In one embodiment, Figure 10 As shown, a SAR ship detection device combining spatiotemporal context information is provided, comprising: a sub-view processing module 1002, a spatiotemporal context covariance matrix construction module 1004, an image enhancement module 1006, and a ship detection module 1008, wherein:

[0139] The sub-view processing module 1002 is used to obtain a SAR ship image to be detected; perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times;

[0140] A space-time context covariance matrix construction module 1004 is used to construct a space-time context covariance matrix according to the SAR subviews, perform a similarity test on the space-time context covariance matrix, and obtain a similar sample number feature;

[0141] An image enhancement module 1006 is configured to enhance the SAR ship image to be detected by utilizing the similar sample number feature to obtain an enhanced SAR ship image;

[0142] The ship detection module 1008 is used to perform ship detection on the enhanced SAR ship image according to the adaptive threshold processing method and morphological filtering to obtain a ship detection result.

[0143] In one embodiment, the sub-view processing module 1002 is further configured to perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times, including:

[0144] Perform one-dimensional Fourier transform of SAR ship images in azimuth to obtain full-aperture azimuth spectrum;

[0145] The amplitude of the full-aperture azimuth spectrum is averaged in the range direction to obtain a weight function, and the inverse of the weight function is calculated and normalized to obtain a correction function;

[0146] The full aperture azimuth spectrum is optimized according to the correction function to obtain an optimized full aperture azimuth spectrum;

[0147] The optimized full aperture azimuth spectrum is divided into r% overlapping parts according to the required number of sub-apertures to obtain N sub-aperture spectra;

[0148] The center of each sub-aperture spectrum is moved to zero Doppler, each sub-aperture spectrum is windowed in azimuth and subjected to an azimuth one-dimensional inverse Fourier transform to obtain N frames of SAR sub-views corresponding to different synthetic aperture times.

[0149] In one embodiment, the module 1004 for constructing a space-time context covariance matrix is ​​further configured to construct a space-time context covariance matrix according to the SAR sub-views, including:

[0150] Obtain a pixel set from the SAR subview, determine the neighborhood space of the pixel set, and construct a space-time context vector set of the pixel set in the neighborhood space;

[0151] A spatiotemporal context covariance matrix is ​​constructed from a set of spatiotemporal context vectors.

[0152] In one embodiment, the module 1004 for constructing a spatiotemporal context covariance matrix is ​​further configured to construct a spatiotemporal context vector set of a pixel set in a neighborhood space, including:

[0153] The space-time context vector set of the pixel set constructed in the neighborhood space is

[0154]

[0155] Among them, v i is pixel p (x,y,t) The corresponding i-th context vector, x, y, t are the coordinates of the pixel on the azimuth axis, distance axis, and time axis. x and y are used to represent the coordinates of the pixel in the two-dimensional image, and t is used to represent the number of image frames. Both represent pixel p (x,y,t) Pixels in the neighborhood K, N v is the number of context vectors in the set V, N p is pixel p (x,y,t) The number of pixels in the context vector v.

[0156] In one embodiment, the module 1004 for constructing a spatiotemporal context covariance matrix is ​​further configured to construct a spatiotemporal context covariance matrix according to a set of spatiotemporal context vectors, including:

[0157] The space-time context covariance matrix is ​​constructed based on the set of space-time context vectors:

[0158]

[0159] Here, the superscript H represents the conjugate transpose.

[0160] In one embodiment, the image enhancement module 1006 is further configured to perform a similarity test on the spatiotemporal context covariance matrix to obtain similar sample number features, including:

[0161] Determine the similar sample selection neighborhood of the pixel set, and calculate the similarity index between the original pixel in the pixel set and the similar pixels in the similar sample selection neighborhood that are different from the central pixel based on the spatiotemporal context covariance matrix and the spatiotemporal context covariance matrix corresponding to multiple similar pixels in the similar sample selection neighborhood; the neighborhood is a rectangular window with a size of N1×N2 in the range and azimuth directions, centered on any original pixel in the pixel set;

[0162] If the similarity index is not less than the preset similarity judgment threshold, the similar pixels corresponding to the similarity index are classified as similar samples of the original pixel, and the set of all similar samples of the original pixel in a certain neighborhood range is obtained;

[0163] The number of similar samples of each pixel in a certain neighborhood range is determined based on the set of all similar samples of the original pixel in a certain neighborhood range; the similar sample number feature is calculated using the similar sample number and the rectangular window size.

[0164] In one embodiment, the image enhancement module 1006 is further configured to calculate a similar sample number feature using the similar sample number and the rectangular window size, including:

[0165] The similar sample number feature is calculated using the similar sample number and rectangular window size:

[0166]

[0167] Among them, SPN represents the number of similar samples.

[0168] In one embodiment, the image enhancement module 1006 is further configured to enhance the SAR ship image to be detected using the similar sample number feature to obtain an enhanced SAR ship image, including:

[0169] The SAR ship image to be detected is enhanced using the similar sample number feature, and the enhanced SAR ship image is obtained as follows:

[0170] I2=(1-k) M ×I1

[0171] Where I1 and I2 represent the SAR ship image to be detected and the enhanced SAR ship image, respectively, (1-k) M is the weight term, k is the number of similar sample features, and M is the construction index term.

[0172] In one embodiment, the ship detection module 1008 is further configured to perform ship detection on the enhanced SAR ship image according to an adaptive threshold processing method and morphological filtering, and obtain a ship detection result including:

[0173] The probability density functions of ship targets and sea clutter in the samples are estimated based on the artificially selected ship samples and the true values. Then, the intersection of the probability density functions of ship targets and sea clutter is obtained according to the minimum false positive criterion to obtain the detection threshold.

[0174] The enhanced SAR ship image is binarized using the detection threshold to obtain the preliminary detection results.

[0175] The false alarms whose area is less than half of the preset minimum area of ​​the ship to be detected in the preliminary detection results are removed by morphological filtering to obtain the ship detection results.

[0176] Regarding the specific definition of a SAR ship detection device that combines spatiotemporal context information, please refer to the definition of a SAR ship detection method that combines spatiotemporal context information above, and will not be repeated here. The various modules in the above-mentioned SAR ship detection device that combines spatiotemporal context information can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0177] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 11 As shown. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a SAR ship detection method combining space-time context information is implemented. The display screen of the computer device can be a liquid crystal display or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a key, trackball, or touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse.

[0178] Those skilled in the art will understand that Figure 11The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0179] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0180] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0181] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the concept of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A SAR ship detection method combining spatiotemporal context information, characterized in that: The method comprises: Acquire a SAR ship image to be detected; perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times; constructing a space-time context covariance matrix according to the SAR subviews, performing a similarity test on the space-time context covariance matrix, and obtaining a similar sample number feature; The similar sample number feature is used to enhance the SAR ship image to be detected, and the enhanced SAR ship image is obtained; performing ship detection on the enhanced SAR ship image according to an adaptive threshold processing method and morphological filtering to obtain a ship detection result; Performing a similarity test on the space-time context covariance matrix to obtain similar sample number features, including: Determine a similar sample selection neighborhood of a pixel set, and calculate a similarity index between an original pixel in the pixel set and similar pixels in the similar sample selection neighborhood that are different from the center pixel based on the spatiotemporal context covariance matrix and the spatiotemporal context covariance matrix corresponding to multiple similar pixels in the similar sample selection neighborhood; the neighborhood is a rectangular window centered on any original pixel in the pixel set and having a size of N1×N2 in the range and azimuth directions; If the similarity index is not less than a preset similarity determination threshold, similar pixels corresponding to the similarity index are classified as similar samples of the original pixel, and a set of all similar samples of the original pixel in a certain neighborhood range is obtained; Determine the number of similar samples within a certain neighborhood range of each pixel based on all similar sample sets of the original pixel within a certain neighborhood range; and calculate a similar sample number feature using the similar sample number and the rectangular window size; The similar sample number feature is calculated using the similar sample number and the rectangular window size, including: The similar sample number feature is calculated using the similar sample number and rectangular window size: Among them, SPN represents the number of similar samples; The similar sample number feature is used to enhance the SAR ship image to be detected, and the enhanced SAR ship image is obtained, including: The SAR ship image to be detected is enhanced using the similar sample number feature, and the enhanced SAR ship image is obtained as follows: I2=(1-k) M ×I1 Where I1 and I2 represent the SAR ship image to be detected and the enhanced SAR ship image, respectively, (1-k) M is the weight term, k is the number of similar sample features, and M is the construction index term.

2. The method according to claim 1, characterized in that Sub-view processing is performed on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times, including: Performing a one-dimensional Fourier transform domain on the SAR ship image in azimuth to obtain a full-aperture azimuth spectrum; Performing range averaging on the amplitude of the full-aperture azimuth spectrum to obtain a weight function, calculating the inverse of the weight function and normalizing it to obtain a correction function; Optimizing the full-aperture azimuth spectrum according to the correction function to obtain an optimized full-aperture azimuth spectrum; dividing the optimized full-aperture azimuth spectrum by an overlap of r% according to the required number of sub-apertures to obtain N sub-aperture spectra; The center of each sub-aperture spectrum is moved to zero Doppler, each sub-aperture spectrum is windowed in azimuth and subjected to an azimuth one-dimensional inverse Fourier transform to obtain N frames of SAR sub-views corresponding to different synthetic aperture times.

3. The method according to claim 1, characterized in that Constructing a spatiotemporal context covariance matrix according to the SAR subviews, including: Acquire a pixel set from the SAR subview, determine a neighborhood space of the pixel set, and construct a space-time context vector set of the pixel set in the neighborhood space; A spatiotemporal context covariance matrix is ​​constructed according to the set of spatiotemporal context vectors.

4. The method according to claim 3, characterized in that Constructing a set of space-time context vectors of a pixel set in the neighborhood space, comprising: The space-time context vector set of the pixel set constructed in the neighborhood space is Among them, v i is pixel p (x,y,t) The corresponding i-th context vector, x, y, t are the coordinates of the pixel on the azimuth axis, distance axis, and time axis. x and y are used to represent the coordinates of the pixel in the two-dimensional image, and t is used to represent the number of image frames. Both represent pixel p (x,y,t) Pixels in the neighborhood K, N v is the number of context vectors in the set V, N p is pixel p (x,y,t) The number of pixels in the context vector v.

5. The method according to claim 4, characterized in that Constructing a space-time context covariance matrix according to the set of space-time context vectors, including: The space-time context covariance matrix is ​​constructed according to the set of space-time context vectors: Here, the superscript H represents the conjugate transpose.

6. The method according to claim 1, characterized in that Performing ship detection on the enhanced SAR ship image according to an adaptive threshold processing method and morphological filtering to obtain a ship detection result, including: The probability density functions of ship targets and sea clutter in the samples are estimated based on the artificially selected ship samples and the true values. Then, the intersection of the probability density functions of ship targets and sea clutter is obtained according to the minimum false positive criterion to obtain the detection threshold. The enhanced SAR ship image is binarized using the detection threshold to obtain a preliminary detection result; The false alarms whose areas are smaller than half of the preset minimum area of ​​the ship to be detected in the preliminary detection results are removed according to morphological filtering to obtain the ship detection results.

7. A SAR ship detection device combining space-time context information, characterized in that: The device comprises: a sub-view processing module, configured to obtain a SAR ship image to be detected; and perform sub-view processing on the SAR ship image to obtain SAR sub-views corresponding to different synthetic aperture times; A space-time context covariance matrix construction module is used to construct a space-time context covariance matrix according to the SAR subview, perform a similarity test on the space-time context covariance matrix, and obtain a similar sample number feature, including: Determine a similar sample selection neighborhood of a pixel set, and calculate a similarity index between an original pixel in the pixel set and similar pixels in the similar sample selection neighborhood that are different from the center pixel based on the spatiotemporal context covariance matrix and the spatiotemporal context covariance matrix corresponding to multiple similar pixels in the similar sample selection neighborhood; the neighborhood is a rectangular window centered on any original pixel in the pixel set and having a size of N1×N2 in the range and azimuth directions; If the similarity index is not less than a preset similarity determination threshold, similar pixels corresponding to the similarity index are classified as similar samples of the original pixel, and a set of all similar samples of the original pixel in a certain neighborhood range is obtained; Determine the number of similar samples within a certain neighborhood range of each pixel based on all similar sample sets of the original pixel within a certain neighborhood range; and calculate a similar sample number feature using the similar sample number and the rectangular window size; The similar sample number feature is calculated using the similar sample number and the rectangular window size, including: The similar sample number feature is calculated using the similar sample number and rectangular window size: Among them, SPN represents the number of similar samples; The image enhancement module is used to enhance the SAR ship image to be detected by using the similar sample number feature to obtain the enhanced SAR ship image, including: The SAR ship image to be detected is enhanced using the similar sample number feature, and the enhanced SAR ship image is obtained as follows: I2=(1-k) M ×I1 Where I1 and I2 represent the SAR ship image to be detected and the enhanced SAR ship image, respectively, (1-k) M is the weight term, k is the number of similar samples, and M is the construction index term; The ship detection module is used to perform ship detection on the enhanced SAR ship image according to an adaptive threshold processing method and morphological filtering to obtain a ship detection result.

Images