Method for detecting wake direction of surface ship from synthetic aperture radar image
By using convolutional neural networks and Curvelet transform, the direction of ship wakes in synthetic aperture radar images is automatically detected, solving the problem of time-consuming and laborious manual interpretation in existing technologies, and achieving efficient and accurate detection of ship wake directions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NATIONAL SATELLITE OCEAN APPLICATION SERVICE
- Filing Date
- 2026-01-30
- Publication Date
- 2026-05-12
AI Technical Summary
Existing synthetic aperture radar image technology cannot automatically obtain the direction of a ship's wake, requiring manual interpretation, which is time-consuming, labor-intensive, and dependent on the technical level of the interpreters.
A convolutional neural network model combined with Curvelet transform is used to automatically detect the direction of the ship's wake. By obtaining the coordinates of the four vertices of the rectangle, performing Curvelet transform, calculating the average value and angle of the Curvelet coefficient matrix, eliminating 180-degree directional ambiguity, and determining the direction of the ship's wake.
It achieves fully automated detection of the wake direction of surface ships in synthetic aperture radar images, improving detection efficiency and accuracy, with strong anti-interference capabilities, and the method principle is clear and easy to implement in engineering.
Smart Images
Figure CN122017834A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of marine remote sensing, marine target detection and harmonic analysis technology, and specifically relates to a method for detecting the direction of the wake of a ship on the water surface from a synthetic aperture radar image. Background Technology
[0002] Satellite remote sensing technology can acquire observational images of large sea areas. Visible light images are easily affected by cloud cover and lighting conditions, while synthetic aperture radar (SAR) is less affected by cloud cover and lighting conditions, allowing for observation of the sea surface even in cloud cover or at night. Most existing technologies detect ship targets from SAR images. For example, patent publication number CN106170819B discloses a rapid ship target detection method from SAR images. This method performs preliminary target screening of SAR images, dividing the image into several sub-image blocks; finally, it uses CUDA technology to perform constant false alarm rate (CFAR) detection on three different image distributions to detect valid ship targets. However, it cannot automatically obtain the direction of ship wakes. Obtaining the direction of ship wakes requires further manual interpretation, which is time-consuming, labor-intensive, and requires a certain level of technical expertise from the interpreters. The direction of a surface ship's wake can determine its direction of travel. Obtaining the direction of travel of surface ships is of great significance for maritime traffic control, combating smuggling, and maintaining maritime security, and is also important for the work of the coast guard and navy. Summary of the Invention
[0003] The purpose of this invention is to propose a method for detecting the direction of the wake of a surface ship from a synthetic aperture radar image.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0005] A method for detecting the direction of a surface ship's wake from a synthetic aperture radar image, the method being as follows:
[0006] S1. Construct and train a convolutional neural network model for detecting surface ship targets;
[0007] S2. Input the synthetic aperture radar image to be detected into the convolutional neural network model in S1. If a surface ship target is detected, obtain the coordinates of the four vertices of the rectangle surrounding the ship and its wake, and determine the coordinates of the center point of the rectangle.
[0008] S3. Extract the synthetic radar sub-image corresponding to the rectangular area from the original synthetic aperture radar image to be detected based on the coordinates of the four vertices of the rectangular frame;
[0009] S4. Perform Curvelet transform on the radar sub-images captured in S3 to obtain Curvelet coefficient matrices at different scales and in different directions;
[0010] S5. Based on the spatial resolution of synthetic aperture radar images, the width of ship wakes, and the width of Curvelet functions at different scales, determine the scale number of the Curvelet coefficients used to detect the direction of ship wakes on the water surface. It is required that the width of the Curvelet function corresponding to the selected scale number is closest to the number of pixels of the ship wake width.
[0011] S6. For the selected scale used to detect the orientation of surface ships, calculate the average value of the Curvelet coefficient matrix in different directions at that scale, and sort the average values.
[0012] S7. For the selected scale used to detect the direction of the ship's wake, determine the index of the direction with the maximum average value of the Curvelet coefficient matrix;
[0013] S8. For the selected scale used to detect the direction of the ship's wake, calculate the angle corresponding to the direction with the maximum average value of the Curvelet coefficient matrix based on the total number of directions, the angle corresponding to the first direction Curvelet, and the index of the direction with the maximum average value of the coefficient matrix.
[0014] S9. Determine the normal direction of the ship's wake based on the angle corresponding to the direction of the maximum average value of the Curvelet coefficient matrix;
[0015] S10. Determine the direction of the ship's wake based on the perpendicular relationship between the ship's wake line and its normal, and obtain two angles. At this time, the direction of the ship's wake has a 180-degree directional ambiguity.
[0016] S11. Surface ship targets have a strong backscattering coefficient, which appears as a bright spot in synthetic aperture radar images. Determine the coordinates of the center point of the bright spot.
[0017] S12. Starting from the center point of the rectangle surrounding the ship and its wake, and ending at the center point of the bright spot corresponding to the ship, determine a vector, with counterclockwise as the positive direction, and calculate the angle between this vector and the geographic north.
[0018] S13. Among the two possible angles for the direction of the ship's wake obtained in S10, select the angle that is closest to the angle obtained in S12 as the direction of the ship's wake. By eliminating the 180-degree directional ambiguity, the direction of the ship's wake on the water surface can be determined.
[0019] Furthermore, in step S1, the convolutional neural network model includes, but is not limited to, the YOLO series of object detection models.
[0020] Furthermore, in step S4, the Curvelet transform is a post-wavelet transform that can decompose the image at different scales and directions to obtain Curvelet coefficients at different scales, directions and positions. For a certain scale and direction, the Curvelet coefficients at different positions form a Curvelet coefficient matrix.
[0021] Furthermore, based on the resolution of the synthetic aperture radar image and the number of rows and columns of the selected rectangle, the number of scales and directions at each scale are set. The larger the number of rows and columns of the radar sub-image corresponding to the rectangle, the larger the number of scales and directions. The number of scales can be set to 5 to 7, and the number of directions can be set to 64, 256, 512 or 1024.
[0022] Furthermore, based on the spatial resolution of the radar image and the width of the ship's wake, the number of pixels corresponding to the width of the ship's wake is determined. The scale index of the Curvelet coefficients used to detect the direction of the ship's wake is then determined by combining the widths of Curvelet functions at different scales. The width of the target ship's wake can be obtained empirically; assuming the width of the target ship's wake is denoted as... The spatial resolution of radar images is denoted as The number of pixels corresponding to the width of the ship's wake is ,function Represents the real number Rounding; Once the number of scales for the Curvelet transform is determined, the spatial width of the Curvelet function at each scale can be determined; Select the scale whose spatial width of the Curvelet function is closest to the number of pixels of the ship's wake width as the Curvelet coefficient scale used to detect the direction of the ship's wake on the water surface, and record the scale number.
[0023] Furthermore, in step S6, for the selected scale used to detect the direction of the ship's wake, the index of the direction with the maximum average value of the Curvelet coefficient matrix is determined. The method for calculating the average value of the coefficient matrix is as follows: for the coefficient matrix of each direction, the absolute value of the Curvelet coefficient at each position in the coefficient matrix is calculated, and the average value of the absolute values of all elements in the matrix is calculated.
[0024] Furthermore, the specific method in S8 is as follows, assuming the first... There is a scale, and the total number of directions at that scale is denoted as... The angle corresponding to the first direction Curvelet is denoted as The index of the direction of maximum average value of the coefficient matrix is denoted as . The angular step size of the Curvelet rotation. The angle of the direction with the maximum average value of the coefficient matrix is The unit is angle.
[0025] Furthermore, in S10, the normal direction of the ship's wake is perpendicular to the ship's line, and the angle between the two is 90 degrees, in degrees.
[0026] Furthermore, the specific method in S12 is as follows: the center point of the rectangle surrounding the ship and its wake is... , The center point of the bright spot corresponding to the ship is denoted as , The vector originating from the center point of the rectangle and ending at the center point of the bright spot is denoted as... , Along the geographical northward direction, take any point above the center point of the rectangle, and denote it as... , , Let be any positive integer, with the center point of the rectangle as the starting point. The vector with endpoint is denoted as , Then the angle between the two vectors is ,in for The inverse cosine function, "" represents the dot product of two vectors.
[0027] Furthermore, the specific calculation method in S13 is to calculate the two angles obtained in S10 and the included angle obtained in S12. The angle with the smallest absolute value among the two differences determines the direction of the ship's wake.
[0028] The above technical solution can achieve the following beneficial effects:
[0029] The method achieves fully automated detection of the wake direction of surface ships in synthetic aperture radar images, eliminating the need for manual interpretation and greatly improving efficiency. By combining the target detection capabilities of convolutional neural networks with the multi-scale and multi-directional analysis capabilities of Curvelet transform, the method achieves high direction detection accuracy and strong anti-interference ability. The method has a clear principle, well-defined steps, and is easy to implement in engineering, making it suitable for real-time maritime monitoring systems. Attached Figure Description
[0030] Figure 1 It is a method flowchart.
[0031] Figure 2 In (a), the white rectangle is the bounding box labeled by the deep learning model. (b) Sub-images are extracted from the original radar image based on the coordinates of the four vertices of the bounding box.
[0032] Figure 3This is a schematic diagram of the structure of the Curvelet coefficients obtained after performing Curvelet transform on a synthetic aperture radar sub-image.
[0033] Figure 4 It is the average value of the Curvelet coefficient matrix in different directions (this figure is only an example, and the actual number of directions may not be 128).
[0034] Figure 5 It is a diagram showing the normal direction of the ship's wake.
[0035] Figure 6 It is a tail trajectory diagram to be determined.
[0036] Figure 7 This is a schematic diagram of two vectors. One vector starts at the center of the rectangle surrounding the ship and its wake and ends at the center of the corresponding white bright spot on the ship. The other vector starts at the center of the rectangle surrounding the ship and its wake and ends at any point due north of the center of the rectangle.
[0037] Figure 8 This is a schematic diagram showing the direction of the stern of a ship. Detailed Implementation
[0038] The following is an appendix Figure 1-8 The present invention will be further described in conjunction with the embodiments:
[0039] Example: A method for detecting the direction of a surface ship's wake from synthetic aperture radar images. Figure 1 As shown, the method includes a surface ship identification step, a step of obtaining a rectangular frame surrounding the ship and its wake, a step of obtaining the Curvelet coefficients of the radar image corresponding to the rectangular frame, a step of determining the scale number of the Curvelet coefficients used for detection, a step of calculating the average value of the Curvelet coefficient matrix in each direction at the selected scale, a step of determining the direction corresponding to the largest average value of the coefficient matrix at the selected scale, a step of determining the normal direction of the ship's wake, a step of eliminating 180-degree directional ambiguity, and a step of determining the direction of the ship's wake, which is used to automatically detect the direction of the surface ship's wake from the synthetic aperture radar image.
[0040] The specific method is as follows:
[0041] Step 1: Build and train a convolutional neural network model for detecting surface ship targets;
[0042] The second step involves inputting the synthetic aperture radar image to be detected into the convolutional neural network model. If a surface ship target is detected, the coordinates of the four vertices of the rectangle surrounding the ship and its wake are obtained, and the coordinates of the center point of the rectangle are determined. Taking the YOLO11 model as an example, YOLO11 can automatically complete this task, and the size of the rectangle is adjustable. Figure 2 In (a), the white box is the rectangular box obtained by the convolutional neural network model, the striped signal in the figure is the ship's wake, and the white bright spot is the ship on the water.
[0043] Step 3: The white box is a rectangular box marked by the convolutional neural network model. Based on the coordinates of the four vertices of the rectangle, a sub-image is extracted from the original radar image.
[0044] Step 4: Perform Curvelet transform on the captured radar sub-image to obtain Curvelet coefficient matrices at different scales and in different directions.
[0045] The Curvelet transform is a post-wavelet transform that can decompose an image at different scales and directions, obtaining Curvelet coefficients at different scales, directions, and locations. For a given scale and direction, the Curvelet coefficients at different locations form a Curvelet coefficient matrix. The number of scales and directions at each scale of the Curvelet transform are set according to the resolution of the synthetic aperture radar image and the number of rows and columns of the selected bounding box. The larger the number of rows and columns of the radar sub-image corresponding to the bounding box, the larger the number of scales and directions; the number of scales can be set to 5 to 7. The number of directions can be set to 64, 256, 512, or 1024. For example, performing a Curvelet transform on a radar image with a scale of 5, where scales 1 and 5 are not directional, scale 2 has 64 directions, scale 3 has 128 directions, and scale 4 has 128 directions. This yields Curvelet coefficient matrices at different scales and directions, such as... Figure 3 As shown.
[0046] Step 5: Based on the spatial resolution of the synthetic aperture radar image, the width of the ship's wake, and the width of the Curvelet function at different scales, determine the scale number of the Curvelet coefficient used to detect the direction of the ship's wake on the water surface. It is required that the width of the Curvelet function corresponding to the selected scale number is closest to the number of pixels of the ship's wake width.
[0047] The width of the target ship's wake can be obtained empirically; let's assume the width of the target ship's wake is denoted as... The spatial resolution of radar images is denoted as The number of pixels corresponding to the width of the ship's wake is ,function Represents the real number Rounding. Once the number of scales for the Curvelet transform is determined, the spatial width of the Curvelet function at each scale can be determined. Select the scale whose spatial width of the Curvelet function is closest to the number of pixels of the ship's wake width as the Curvelet coefficient scale used to detect the direction of the ship's wake on the water surface, and record the scale number.
[0048] For example, based on the spatial resolution of the radar image and the width of the ship's wake, the number of pixels corresponding to the width of the ship's wake can be determined. For instance, if the spatial resolution of the radar image is 2 meters and the width of the ship's wake is 20 meters, then it corresponds to 10 pixels in the radar image. The index of the Curvelet coefficient scale used to detect the ship's direction can be determined based on the width of the input radar image.
[0049] Step 6: For the selected scale used to detect the orientation of surface ships, calculate the average value of the Curvelet coefficient matrix in different directions at that scale, and sort the average values. The method for calculating the average value of the coefficient matrix is as follows: for the coefficient matrix in each direction, calculate the absolute value of the Curvelet coefficient at each position in the coefficient matrix, and calculate the average of the absolute values of all elements in the matrix. Figure 4 (As shown).
[0050] Step 7: For the selected scale used to detect the direction of the ship's wake, determine the index of the direction with the maximum average value of the Curvelet coefficient matrix;
[0051] Step 8: For the selected scale used to detect the direction of the ship's wake, calculate the angle corresponding to the direction with the maximum average value of the Curvelet coefficient matrix based on the total number of directions, the angle corresponding to the first direction Curvelet, and the index of the direction with the maximum average value of the coefficient matrix.
[0052] Assuming we choose the first There is a scale, and the total number of directions at that scale is denoted as... The angle corresponding to the first direction Curvelet is denoted as The index of the direction of maximum average value of the coefficient matrix is denoted as . The angular step size of the Curvelet rotation. The angle of the direction with the maximum average value of the coefficient matrix is The unit is angle.
[0053] Step 9: Determine the normal direction of the ship's wake based on the angle corresponding to the direction of the maximum average value of the Curvelet coefficient matrix; obtain the normal direction of the ship's wake, and predict the wake direction based on the normal direction. Figure 6 The yellow arrow indicates the direction of the wake normal, the white arrow indicates geographic north, and the red and blue arrows represent the wake directions to be determined.
[0054] Step 10: Determine the direction of the ship's wake based on the perpendicular relationship between the ship's wake line and its normal. Two angles can be obtained. At this time, there is a 180-degree directional ambiguity in the direction of the ship's wake.
[0055] Step 11: Surface ships have a strong backscattering coefficient, which appears as a bright spot in synthetic aperture radar images. Determine the coordinates of the center point of the bright spot. Large and medium-sized surface ships usually have a strong radar backscattering coefficient, and therefore appear as bright spots in radar images.
[0056] Step 12: Using the center point of the rectangle surrounding the ship and its wake as the starting point and the center point of the corresponding bright spot on the ship as the ending point, determine a vector. Taking the counterclockwise direction as the positive direction, calculate the angle between this vector and the geographic north direction. Figure 7 (As shown).
[0057] Specifically, the center point of the rectangle surrounding the ship and its wake is , The center point of the bright spot corresponding to the ship is denoted as , The vector originating from the center point of the rectangle and ending at the center point of the bright spot is denoted as... , Along the geographical northward direction, take any point above the center point of the rectangle, and denote it as... , , Let be any positive integer, with the center point of the rectangle as the starting point. The vector with endpoint is denoted as , Then the angle between the two vectors is ,in for The inverse cosine function, "" represents the dot product of two vectors.
[0058] Step 13: Among the two possible angles for the ship's wake direction obtained in S10, select the angle closest to the angle obtained in S12 as the ship's wake direction, eliminating 180-degree directional ambiguity, and determine the direction of the ship's wake on the water surface. Figure 8 (As shown).
[0059] The calculation method is as follows: In S13, the specific calculation method is to calculate the two angles obtained in S10 and the included angle obtained in S12. The angle with the smallest absolute value among the two differences determines the direction of the ship's wake.
[0060] The above descriptions are all preferred embodiments of the present invention. For those skilled in the art, any modifications to the present invention in various equivalent forms without departing from the principle of the present invention shall fall within the protection scope of the appended claims.
Claims
1. A method for detecting the direction of a surface ship's wake from a synthetic aperture radar image, characterized in that: The method is as follows: S1. Construct and train a convolutional neural network model for detecting surface ship targets; S2. Input the synthetic aperture radar image to be detected into the convolutional neural network model in S1. If a surface ship target is detected, obtain the coordinates of the four vertices of the rectangle surrounding the ship and its wake, and determine the coordinates of the center point of the rectangle. S3. Extract the synthetic radar sub-image corresponding to the rectangular area from the original synthetic aperture radar image to be detected based on the coordinates of the four vertices of the rectangular frame; S4. Perform Curvelet transform on the radar sub-images captured in S3 to obtain Curvelet coefficient matrices at different scales and in different directions; S5. Based on the spatial resolution of synthetic aperture radar images, the width of ship wakes, and the width of Curvelet functions at different scales, determine the scale number of the Curvelet coefficients used to detect the direction of ship wakes on the water surface. It is required that the width of the Curvelet function corresponding to the selected scale number is closest to the number of pixels of the ship wake width. S6. For the selected scale used to detect the orientation of surface ships, calculate the average value of the Curvelet coefficient matrix in different directions at that scale, and sort the average values. S7. For the selected scale used to detect the direction of the ship's wake, determine the index of the direction with the maximum average value of the Curvelet coefficient matrix; S8. For the selected scale used to detect the direction of the ship's wake, calculate the angle corresponding to the direction with the maximum average value of the Curvelet coefficient matrix based on the total number of directions, the angle corresponding to the first direction Curvelet, and the index of the direction with the maximum average value of the coefficient matrix. S9. Determine the normal direction of the ship's wake based on the angle corresponding to the direction of the maximum average value of the Curvelet coefficient matrix; S10. Determine the direction of the ship's wake based on the perpendicular relationship between the ship's wake line and its normal, and obtain two angles. At this time, the direction of the ship's wake has a 180-degree directional ambiguity. S11. Surface ship targets have a strong backscattering coefficient, which appears as a bright spot in synthetic aperture radar images. Determine the coordinates of the center point of the bright spot. S12. Taking the center point of the rectangle surrounding the ship and its wake as the starting point and the center point of the corresponding bright spot of the ship as the ending point, determine a vector, with counterclockwise as the positive direction, and calculate the angle between this vector and the geographic north direction. ; S13. Among the two possible angles for the direction of the ship's wake obtained in S10, select the angle that is closest to the angle obtained in S12 as the direction of the ship's wake. By eliminating the 180-degree directional ambiguity, the direction of the ship's wake on the water surface can be determined.
2. The method according to claim 1, characterized in that: In step S1, the convolutional neural network model includes, but is not limited to, the YOLO series of object detection models.
3. The method according to claim 1, characterized in that: In step S4, the Curvelet transform is a post-wavelet transform that can decompose an image at different scales and directions to obtain Curvelet coefficients at different scales, directions and positions. For a certain scale and direction, the Curvelet coefficients at different positions form a Curvelet coefficient matrix.
4. The method according to claim 1, characterized in that: The number of scales and directions at each scale are set according to the resolution of the synthetic aperture radar image and the number of rows and columns of the selected rectangle. The larger the number of rows and columns of the radar sub-image corresponding to the rectangle, the larger the number of scales and directions. The number of scales can be set to 5 to 7, and the number of directions can be set to 64, 256, 512 or 1024.
5. The method according to claim 1, characterized in that: Based on the spatial resolution of the radar image and the width of the ship's wake, the number of pixels corresponding to the width of the ship's wake is determined. The scale index of the Curvelet coefficients used to detect the direction of the ship's wake on the water surface is then determined by combining the widths of Curvelet functions at different scales. The width of the target ship's wake can be obtained empirically; assuming the target ship's wake width is denoted as... The spatial resolution of radar images is denoted as The number of pixels corresponding to the width of the ship's wake is ,function Represents the real number Rounding; Once the number of scales for the Curvelet transform is determined, the spatial width of the Curvelet function at each scale can be determined; Select the scale whose spatial width of the Curvelet function is closest to the number of pixels of the ship's wake width as the Curvelet coefficient scale used to detect the direction of the ship's wake on the water surface, and record the scale number.
6. The method according to claim 1, characterized in that: In step S6, for the selected scale used to detect the direction of the ship's wake, the index of the direction with the maximum average value of the Curvelet coefficient matrix is determined. The method for calculating the average value of the coefficient matrix is as follows: for the coefficient matrix of each direction, the absolute value of the Curvelet coefficient at each position in the coefficient matrix is calculated, and the average value of the absolute values of all elements in the matrix is calculated.
7. The method according to claim 1, characterized in that: The specific method in S8 is as follows, assuming the first... There is a scale, and the total number of directions at that scale is denoted as... The angle corresponding to the first direction Curvelet is denoted as The index of the direction of maximum average value of the coefficient matrix is denoted as... The angular step size of the Curvelet rotation. The angle of the direction with the maximum average value of the coefficient matrix is The unit is angle.
8. The method according to claim 1, characterized in that: In S10, the normal direction of the ship's wake is perpendicular to the ship's wake, and the angle between the two is 90 degrees, with the unit being angle.
9. The method according to claim 1, characterized in that: The specific method in S12 is as follows: the center point of the rectangle surrounding the ship and its wake is... , The center point of the bright spot corresponding to the ship is denoted as , The vector originating from the center point of the rectangle and ending at the center point of the bright spot is denoted as... , ; Along the geographical northward direction, take any point above the center point of the rectangle, and denote it as... , , Let be any positive integer, with the center point of the rectangle as the starting point. The vector ending at point is denoted as , Then the angle between the two vectors is ,in for The inverse cosine function, "" represents the dot product of two vectors.
10. The method according to claim 1, characterized in that: The specific calculation method in S13 is to calculate the two angles obtained in S10 and the included angle obtained in S12. The angle with the smallest absolute value among the two differences determines the direction of the ship's wake.