A wind direction retrieval method for marine radar images

Abstract

The present invention discloses a method for wind direction retrieval from a marine radar image. The method removes co-frequency interference from the original radar image and normalizes it. In the polar coordinate system of the image, pixels with equal distances from the center of the radar image are selected as a group according to a set distance step size. Pixels with abnormal intensities are eliminated from each group of pixels. Pixels with the highest intensity are then selected from each group of pixels. One-dimensional ideal distance attenuation data is determined based on the pixels with the highest intensity. The one-dimensional ideal distance attenuation data is then fitted to an ideal distance attenuation model using least squares fitting, and its regression parameters are determined. The ideal distance attenuation model is then fitted and compared with the radar image pixel intensities at each azimuth angle to determine the horizontal attenuation component at each azimuth angle. The method then retrieves the sea surface wind direction based on the azimuth-dependence of the wind direction in the marine radar image under horizontal polarization. The method improves the wind direction retrieval accuracy and effectively reduces errors caused by outlier interference and low wind speed in the radar image.

Description

Technical Field

[0001] The invention belongs to the technical field of marine remote sensing, and relates to a method for retrieving wind direction from a marine radar image, in particular to a method for retrieving wind direction from a marine radar image based on echo intensity distance attenuation. Background Art

[0002] Sea surface wind is a crucial parameter for studying the interaction between the ocean and the air. By regulating heat, water vapor, air-sea fluxes, and particulate matter, sea surface winds regulate the coupling between the atmosphere and ocean, thereby maintaining global and regional climate. Currently, sea surface wind direction is primarily measured using in-situ sensors (such as anemometers) mounted on ship masts. However, anemometers are susceptible to atmospheric turbulence and airflow disturbances caused by the ship's superstructure. Even when anemometers are mounted in an unobstructed location, errors in wind parameter measurements can be significant. In recent decades, X-band marine radars, with their high resolution and immediate feedback, have become an effective method for obtaining sea surface wind information. Furthermore, since most ships are already equipped with X-band marine radars, they offer minimal additional cost compared to other remote sensing instruments. These studies have laid the foundation for using marine radars to extract sea surface wind information.

[0003] At present, the main achievements in the field of wind direction retrieval in marine radar images are as follows. There are two types of methods that can be used to retrieve wind direction from X-band marine radar image sequences. The first type of method uses the relationship between the normalized radar cross section (NRCS) and the windward direction of the radar illumination. However, applying this method to obtain wind direction requires a radar image with no obstructions in the entire 360° azimuth. In 2012, Lund proposed a method to determine wind direction by fitting the backscatter intensity as a function of azimuth angle to a harmonic function using the least squares method based on the dependence of radar backscatter intensity on the windward direction (Lund B, Graber H C, Romeiser R. Wind Retrieval From Shipborne Nautical X-Band Radar Data[J]. IEEE Transactions on Geoscience & Remote Sensing, 2012, 50(10):3800-3811.). In 2013, Vicen-Bueno et al. extracted wind direction from temporally integrated and spatially smoothed radar images (Vicen-Bueno R, Horstmann J, Terril E, et al. Real-time ocean wind vector retrieval from marine radar image sequences acquired at grazing angle [J]. Journal of Atmospheric and Oceanic Technology, 2013, 30(1): 127-139.). In 2015, Chen et al. proposed a method based on probability distribution function to retrieve nearshore sea surface wind parameters from radar image sequences (Chen Z, He Y, Zhang B, et al. Determination of nearshore sea surface wind vector from marine X-band radar images [J]. Ocean Engineering, 2015, 96(mar.1): 79-85.).That same year, Liu improved Lund's method, proposing a hyperbola fitting method and adding a curve fitting method (Liu, Ying, Vicen-Bueno, et al. Comparison of Algorithms for Wind Parameters Extraction From Shipborne X-Band Marine Radar Images [J]. IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 2015). However, the additional curve fitting method performs poorly when the wind direction is obstructed. The ocean is a randomly disturbed process with multiple models. While the above method works well in wind and wave conditions, it may not be effective in swell conditions. The foam caused by breaking waves can have a significant impact on the backscattering of electromagnetic waves, thereby affecting wind measurements. In order to solve the shortcomings of wind direction being blocked and the unstable sea surface wind field model, in 2022, Yu et al. proposed a method of expanding discrete wavelet transform and azimuth data to extract wind direction from a single obscured radar image (Yu H, Wang H, Lu Z. Wind-Direction Estimation from Single X-Band Marine Radar Image Improvement by Utilizing the DWT and Azimuth-Scale Expansion Method [J]. Entropy, 2022, 24 (6): 747.). The second type of method is based on the wind streaks in the radar image, which are low-frequency signals with a spatial scale of about 200 to 500 meters and are parallel to the wind direction. After filtering out the high-frequency wave signals through low-pass filtering, the wind streaks can be obtained, and the direction of the wind streaks is the wind direction. Based on this theory, in 2021, Wang Hui proposed the energy spectrum theory (ESM) to retrieve sea surface wind direction from radar image sequences (Wang H, Qiu H, Lu Z, et al. An Energy Spectrum Algorithm for Wind Direction Retrieval From X-BandMarine Radar Image Sequences[J]. IEEE Journal of Selected Topics in AppliedEarth Observations and Remote Sensing, 2021, 14: 4074-4088.).

[0004] The above methods have been continuously developed to account not only for obstructions in the radar field of view but also for obstructions in the wind direction azimuth, achieving good results. However, within the same azimuth, not only are obstructions present, but also target trails formed by obstructing fixed targets. Even when excluding target pixels, it is difficult to calculate the true average intensity at that azimuth, resulting in localized low or high values. Therefore, while data from obstructed directions is excluded during the curve fitting process, data from multiple azimuths due to complex sea conditions may be retained, reducing the accuracy of the results. Furthermore, radar echo intensity is modulated by distance. The distance dependence of the NRCS from the ocean surface can be seen in radar images, and radar echo intensity attenuates with increasing distance. Echo signals from distant sea surfaces not only decrease with increasing distance but also exhibit a higher number of zero-intensity pixels. This results in a lower average echo intensity at that azimuth, which in turn reduces the accuracy of wind direction retrieval. Studies have shown that any radial range dependence in radar imagery is a result of the grazing angle dependence of the NRCS of the ocean surface, which decays cubically with radial range. Therefore, when retrieving wind information from marine radar imagery, the effects of range decay cannot be ignored. Summary of the Invention

[0005] In view of the above-mentioned prior art, the technical problem to be solved by the present invention is to provide a method for retrieving wind direction from a marine radar image, which utilizes the distance attenuation characteristics of the echo intensity in the radar image to retrieve the sea surface wind direction from the image obtained by the X-band marine radar.

[0006] To solve the above technical problems, the present invention provides a method for wind direction retrieval from a marine radar image, comprising:

[0007] Step 1: remove the co-frequency interference from the original radar image obtained by reading the radar file, and normalize the pixel intensity of the radar image;

[0008] Step 2: In the polar coordinate system of the radar image obtained in step 1, select pixels with equal distance from the center of the radar image as a group according to the set distance step size, remove pixels with abnormal intensity in each group of pixels, and then select the pixel with the largest intensity in each group of pixels, and determine the ideal distance attenuation data X based on the pixel with the largest intensity. ra (r), X ra (r)=max θ {X(r,θ)},θ∈[0,2π), where r represents the radial distance from the radar antenna to the sea surface, θ represents the azimuth of the radar image, and X represents the normalized radar image intensity. Then, the least squares fitting is applied to convert X ra (r) is fitted to the ideal distance decay model D(r) to determine the regression parameters of D(r);

[0009] Step 3: Fit and compare the ideal distance attenuation model D(r) with the radar image pixel intensity at each azimuth angle to determine the attenuation horizontal component at each azimuth angle.

[0010] Step 4: Retrieve the sea surface wind direction based on the wind direction and azimuth dependency of the marine radar image under horizontal polarization.

[0011] Furthermore, the step 2 of removing pixels with abnormal intensities from each group of pixels includes:

[0012] The intensity range of 0-1 is evenly divided into m histogram panes, and then the pixels are stored in the histogram panes of the corresponding intensity. The pixels in the panes with a number of pixels less than a threshold T are considered abnormal and are removed.

[0013] Furthermore, the threshold T satisfies:

[0014] T=k×N

[0015] Where N is the number of azimuth lines in the radar image, that is, the number of pixels at the same distance; k is a constant between 0 and 1.

[0016] Furthermore, the ideal distance attenuation model D(r) is:

[0017]

[0018] Where r is the radial distance from the radar antenna to the sea surface, b0 and b1 are regression parameters.

[0019] Furthermore, in step 3, fitting and comparing the ideal distance attenuation model D(r) with the radar image pixel intensity at each azimuth angle to determine the attenuation horizontal component at each azimuth angle includes:

[0020] Step 3.1. Initialize the number of iterations N = 1. Starting from the azimuth angle 0° and going clockwise to 2π, select the radar image pixel intensity X(r,θ) at each azimuth angle in the radar image in turn, and set the weight ω(r) of the echo intensity of 0 in X(r,θ) to zero. The weight ω(r) is:

[0021]

[0022] Where r is the radial distance from the radar antenna to the sea surface, l is the range resolution of the radar; n is the discrete point position of the radial distance, and p is the maximum discrete point position of the radial distance;

[0023] Then, the attenuation horizontal component C of each azimuth is calculated by nonlinear least square method. θ , C θ The calculation formula is:

[0024]

[0025] Among them, δ is the set threshold, δ≤1.

[0026] Step 3.2: Determine whether the set number of iterations N is reached. If so, output the attenuation level component C. θ Otherwise, let N = N + 1, and use the iterative constraint to obtain C θ The value of is optimized, and the weighting function ω(r) and the threshold δ are updated, specifically: The threshold δ is updated to δ / 2; return to step 3.1.

[0027] Furthermore, the step 4 of retrieving the sea surface wind direction based on the wind direction and azimuth dependency of the marine radar image under horizontal polarization includes:

[0028] Step 4.1: Use the least square method to calculate the attenuation horizontal component C θ Fit a cosine function as a function of azimuth:

[0029] C θ =a0+a1cos 2 (0.5(θ-w))

[0030] Where θ is the azimuth of the radar image, a0 and a1 are regression parameters, and w represents the angle of the peak point of the fitting function;

[0031] Step 4.2: Convert the peak point angle w to relative wind direction:

[0032]

[0033] Among them, θ re Indicates relative wind direction;

[0034] Step 4.3: Convert relative wind direction to absolute wind direction:

[0035] θ wind =Rem(θ re +θ ship ,360)

[0036] Among them, θ wind represents the absolute wind direction, θ ship Indicates the bow direction of the ship.

[0037] Beneficial effects of the present invention: The present invention proposes a method for inverting the wind direction of marine radar images based on the echo intensity distance attenuation characteristics. The ideal attenuation model of the radar image is derived by utilizing the dependence of the echo intensity in the radar image on the radial distance from the sea surface. The image collected by the standard marine horizontal polarization X-band radar working under low grazing incidence has only one echo intensity peak in the windward direction. The attenuation function at each azimuth angle is obtained through the ideal attenuation model and the attenuation horizontal component, that is, the average echo intensity, is calculated, and the wind direction is calculated through the horizontal component. After excluding the blocked area, the wind direction is retrieved by the azimuth dependence of the grayscale level. Experimental verification was carried out using measured data, and the effectiveness and superiority of the algorithm were verified by comparing it with the single curve fitting method that directly averages the echo intensity. Compared with the prior art, the wind direction retrieval method proposed by the present invention has the following advantages:

[0038] The present invention utilizes the distance attenuation of the echo intensity of the navigation radar image to retrieve the wind direction, overcomes the interference of fixed obstacles and abnormal values ​​on the sea surface, and can be applied to sea areas affected by fixed objects.

[0039] The present invention calculates the attenuation horizontal component of each azimuth angle through iterative optimization and the least squares method, thereby avoiding the influence of outliers and excessive zero points when the wind speed is low when directly calculating the echo intensity, and improving the accuracy of wind direction retrieval at low wind speeds. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is the entire radar original image in the polar coordinate system;

[0041] Figure 2 It is the normalized whole radar original image in polar coordinate system;

[0042] Figure 3 It is the histogram of echo intensity at a distance of 7.5m;

[0043] Figure 4 is the one-dimensional ideal attenuation data of the radar original image;

[0044] Figure 5 It is an ideal attenuation model fitted based on the ideal attenuation data;

[0045] Figure 6 is the horizontal component of range attenuation at each azimuth;

[0046] Figure 7 It is the cosine function fitted according to the attenuation horizontal component;

[0047] Figure 8 It is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0049] Combine Figure 8 The present invention is divided into the following steps: the first step is to read the radar image to be detected, remove co-channel interference, and perform normalization processing; the second step is to determine the ideal range attenuation model of the marine radar image; the third step is to calculate the attenuation horizontal component of each azimuth; and the fourth step is to calculate the sea surface wind direction. The present invention includes the following steps:

[0050] Step 1: Read the radar file to obtain the original radar image, remove the co-frequency interference and normalize it, which specifically includes:

[0051] Step 1.1: Use the radar image processing program to load the radar file and record the radar image acquisition time, azimuth, radial distance, echo intensity and other information.

[0052] Step 1.2: Use the selected filtering algorithm to perform co-frequency interference suppression on the radar original image.

[0053] Step 1.3: Normalize the radar image data.

[0054] Step 2: Determine the ideal distance attenuation model for the marine radar image. For pixels with the same distance in the polar coordinates of the radar image, select the pixel with the highest intensity value to derive the ideal distance attenuation model, which specifically includes:

[0055] Step 2.1, calculate the threshold T in the histogram statistics.

[0056] The calculation formula of threshold T is:

[0057] T=k×N (1)

[0058] Where:

[0059] N-----The number of azimuth lines in the radar image, that is, the number of pixels at the same distance

[0060] k-----empirical parameter

[0061] Step 2.2, starting from the center of the radar image, select one-dimensional data with the same distance and store them in m histogram bins ranging from 0 to 1. The bins with a number less than a threshold T are considered outliers and excluded;

[0062] Step 2.3: For the pixels with the same distance r determined in step 2.2, select the pixel with the highest intensity value as the value at this distance. Then, with the center of the image as the origin, repeat steps 2.2 to 2.3 at a larger distance until the farthest distance is reached, thereby determining the one-dimensional ideal distance attenuation data X ra (r).

[0063] X ra The calculation formula for (r) is as follows:

[0064] X ra (r)=max θ {X(r,θ)}, θ∈[0,2π) (2)

[0065] Where:

[0066] r-----Radial distance from radar antenna to sea surface

[0067] θ-----azimuth of radar image

[0068] X-----normalized radar image intensity

[0069] Step 2.4: Select the existing ideal distance decay function model D(r) and apply the least squares fitting to X ra (r) is fitted to the ideal distance decay model to determine the regression parameters of D(r).

[0070] The formula of the ideal distance attenuation model D(r) is as follows:

[0071]

[0072] Where:

[0073] r-----Radial distance from radar antenna to sea surface

[0074] b0-----regression parameter

[0075] b1-----regression parameter.

[0076] Step 3: Determine the horizontal attenuation component of each azimuth angle, i.e., the relative average echo intensity, based on the ideal range attenuation model. The ideal attenuation model is fitted and compared with the one-dimensional data of each azimuth angle to determine the horizontal attenuation component of each azimuth angle. This includes:

[0077] Step 3.1, initialize the weighting function ω and the threshold δ=1.

[0078] The initialization calculation formula of ω(r) is as follows:

[0079]

[0080] r-----Radial distance from radar antenna to sea surface

[0081] l-----Range resolution of the radar

[0082] n-----discrete point position of radial distance

[0083] p-----the maximum discrete point position of the radial distance

[0084] Step 3.2: Select the one-dimensional data X(r,θ) at each azimuth from the radar image starting from azimuth 0° to 2π clockwise, set the weight ω(r) of the echo intensity of 0 in X(r,θ) to zero, and calculate the attenuation horizontal component C of each azimuth by minimizing the following formula using the nonlinear least squares method: θ .

[0085] C θ The calculation formula is as follows:

[0086]

[0087] l-----Range resolution of the radar

[0088] n-----discrete point position of radial distance

[0089] p-----the maximum discrete point position of the radial distance

[0090] Step 3.3: Preliminary estimate of C through step 3.2 θ , and then use iterative constraints on C θ The value of is optimized, and the weighting function ω and the threshold δ are updated during the iteration process. The weight ω is updated by formula (6), and the threshold δ is updated to half of the original value after each iteration. The updated parameters ω and δ are passed to formula (5) for the next iteration. The whole process needs to be iterated 3 times to obtain the final attenuation level component C θ , the attenuation horizontal component C at each azimuth θ It has a cosine function dependency on the radar illumination direction and has a peak in the windward direction.

[0091] The update formula of the weighted function ω is as follows:

[0092]

[0093] Step 4: Retrieve the sea surface wind direction. Retrieve the sea surface wind direction using the wind direction and azimuth dependency of the horizontally polarized marine radar image, specifically including:

[0094] Step 4.1, apply the least square method to reduce the attenuation horizontal component C θ Fit a cosine function as a function of azimuth.

[0095] The cosine function is as follows:

[0096] C θ =a0+a1cos 2(0.5(θ-w)) (7)

[0097] θ-----azimuth of radar image

[0098] a0-----Regression parameter

[0099] a1-----Regression parameter

[0100] w-----the angle of the peak point of the fitting function

[0101] Step 4.2: In step 4.1, w in formula (7) is the angle of the peak point of the fitting function. Convert the peak point angle into relative wind direction.

[0102] The formula for calculating relative wind direction is as follows:

[0103]

[0104] θ re -----Relative wind direction

[0105] w-----the angle of the peak point of the fitting function

[0106] Step 4.3, convert the relative wind direction to absolute wind direction.

[0107] The absolute wind direction calculation formula is as follows:

[0108] θ wind =Rem(θ re +θ ship ,360) (9)

[0109] θ wind -----Absolute wind direction

[0110] θ re -----Relative wind direction

[0111] θ ship -----Bow direction.

[0112] The following is an example with reference to specific parameters:

[0113] The present invention uses an X-band marine radar mounted forward of the main mast of a maritime surveillance vessel. Data is collected using short pulses, covering a radial range of 4.2 km with an angular resolution of approximately 1 degree. Each image is acquired in approximately 2.7 seconds. 32 images are stored as a time series. A single radar image contains 2048 lines, each with 560 points. The range resolution is 7.5 meters, and the directional resolution is approximately 0.18 degrees. The wind reference data used in this invention is derived from an anemometer on a ship at sea, provided by the State Oceanic Administration. Like the marine radar, this anemometer is also mounted on the main mast. Wind parameters are recorded in minutes and used to verify the accuracy of the radar-derived wind direction.

[0114] Combined with attachment Figures 1 to 8 , the specific implementation steps of the present invention are:

[0115] The first step is to read the radar file to obtain the original radar image, remove the co-frequency interference and normalize it.

[0116] Step 1.1: Use the radar image processing program to load the radar file and record the radar image acquisition time, azimuth, radial distance, echo intensity and other information.

[0117] In step 1.2, the radar raw image is processed using the selected filtering algorithm to suppress co-channel interference. This embodiment uses the median filtering method to suppress co-channel interference. Specifically, the echo intensity value of each pixel is replaced by the median of the echo intensities of the other eight pixels within a 3×3 neighborhood window.

[0118] Step 1.3: Normalize the radar image data. In this embodiment, the radar image data is digitized and stored as a 14-bit grayscale depth image sequence. That is, the digitized backscatter intensity ranges from 0 to 8192. Therefore, the intensity of each pixel is divided by 8192 to normalize it to an intensity range of 0 to 1, as shown in the figure. Figure 1 is the unnormalized radar original image, Figure 2 The original radar image after removing the same-frequency interference and normalization.

[0119] The second step is to determine the ideal distance attenuation model of the marine radar image. The determination of the ideal attenuation model includes the following steps:

[0120] Step 2.1, using formula (1) to calculate the threshold T in the histogram statistics. In the present invention, N is 2048, and the empirical parameter k is selected as 0.01, so the threshold T = 20.48.

[0121] The calculation formula of threshold T is:

[0122] T=k×N (1)

[0123] Where:

[0124] N-----The number of azimuth lines in the radar image, that is, the number of pixels at the same distance

[0125] k-----empirical parameter

[0126] Step 2.2, starting from the center of the radar image, select one-dimensional data with the same distance and store them into m histogram bins with an intensity range of 0 to 1. In this example, m is selected as 256. The bins with a number less than the threshold T are considered as outliers and excluded. Figure 3 The figure shows the histogram intensity value statistics at a distance of 7.5m. The pixel points where the statistical number of panes is less than the threshold value of 20.48 are considered as outliers and are excluded.

[0127] Step 2.3: For the pixels with the same distance r determined in step 2.2, select the pixel with the highest intensity value as the value at this distance. Figure 3 In the example, we selected an intensity of 0.5186 as the intensity value at 7.5m. Then, with the center of the image as the origin, we repeat steps 2.2 to 2.3 at a larger distance until the farthest distance is reached, thus determining the ideal distance attenuation data X. ra (r), such as Figure 4 Shown is the final extracted X ra (r).

[0128] X ra The calculation formula for (r) is as follows:

[0129] X ra (r)=max θ {X(r,θ)}, θ∈[0,2π) (2)

[0130] Where:

[0131] r-----Radial distance from radar antenna to sea surface

[0132] θ-----azimuth of radar image

[0133] X-----normalized radar image intensity

[0134] Step 2.4: Select the existing ideal distance decay function model D(r) according to relevant literature, and apply the least squares fitting to X ra (r) is fitted to the ideal distance attenuation model to determine the regression parameters of D(r). Figure 4 X in ra (r), we get the regression parameters b0 = 0.632, b1 = 0.875, and draw a curve of the ideal attenuation model, as shown in Figure 5 shown.

[0135] The formula of the ideal distance attenuation model D(r) is as follows:

[0136]

[0137] Where:

[0138] r-----Radial distance from radar antenna to sea surface

[0139] b0-----regression parameter

[0140] b1-----regression parameter

[0141] The third step is to calculate the horizontal attenuation component of each azimuth angle. The horizontal attenuation component of the azimuth angle represents the difference between the attenuation situation at this azimuth angle and the ideal attenuation model, and is used to measure the relative average echo intensity at this azimuth angle. The following steps are included:

[0142] Step 3.1, initialize the weighting function ω and the threshold δ=1.

[0143] The initialization calculation formula of ω(r) is as follows:

[0144]

[0145] r-----Radial distance from radar antenna to sea surface

[0146] l-----Radar range resolution, in this example the range resolution is 7.5m

[0147] n-----discrete point position of radial distance

[0148] p-----The maximum discrete point position of the radial distance, in this example it is 560

[0149] Step 3.2: Select the one-dimensional data X(r,θ) at each azimuth from the radar image starting from azimuth 0° to 2π clockwise, set the weight ω(r) of the echo intensity of 0 in X(r,θ) to zero, and calculate the attenuation horizontal component C of each azimuth by minimizing the following formula using the nonlinear least squares method: θ .

[0150] C θ The calculation formula is as follows:

[0151]

[0152] l-----Radar range resolution, in this example the range resolution is 7.5m

[0153] n-----discrete point position of radial distance

[0154] p-----The maximum discrete point position of the radial distance, in this example it is 560

[0155] Step 3.3: Preliminary estimate of C through step 3.2 θ , and then use iterative constraints on C θ The value of is optimized, and the weighting function ω and the threshold δ are updated during the iteration process. The weight ω is updated by formula (6), and the threshold δ is updated to half of the original value after each iteration. The updated parameters ω and δ are passed to formula (5) for the next iteration. The whole process needs to be iterated 3 times to obtain the final attenuation level component C θ There are 2048 azimuth data lines in the present invention. For the sake of computational efficiency, one line is selected every four lines for computation. That is to say, 512 attenuation horizontal components C are obtained in the end. θ Value, the attenuation horizontal component C of each azimuth θ It has a cosine function dependency on the radar illumination direction and has a peak in the windward direction.

[0156] like Figure 6 As shown, the dotted line represents the fixed shielding area of ​​the ship mast.

[0157] The update formula of the weighted function ω is as follows:

[0158]

[0159] The fourth step is to calculate the sea surface wind direction. This is done by using the dependence of wind direction and azimuth angle in horizontally polarized ocean radar images. This means that X-band marine radars with horizontal polarization have only one peak echo intensity in the upwind direction. This involves the following steps:

[0160] Step 4.1, apply the least square method to reduce the attenuation horizontal component C θ As a function of azimuth, a cosine function is fitted. First, the fixed occlusion area is removed. The data used in this invention will produce an occlusion area of ​​about 60 degrees between 140° and 200°, so the data in this area is excluded. And the attenuated horizontal component C in the occlusion area is removed. θ Fitting to formula (7), the parameters obtained are a0=0.446, a1=0.08, and w=-31.95.

[0161] The cosine function is as follows:

[0162] C θ =a0+a1cos 2 (0.5(θ-w)) (7)

[0163] θ-----azimuth of radar image

[0164] a0-----Regression parameter

[0165] a1-----Regression parameter

[0166] w-----the position of the peak point

[0167] Step 4.2: In step 4.1, w in formula (7) is the peak point position of the fitting curve. Convert the peak point position into the relative wind direction. According to formula (8), the relative wind direction θ in this example is re =360-31.95=328.05°

[0168] The formula for calculating relative wind direction is as follows:

[0169]

[0170] θ re -----Relative wind direction

[0171] w-----the position of the peak point

[0172] Step 4.3: In this example, the reference wind direction is 24° and the bow direction is 40.6°. According to formula (9), the absolute wind direction θ is wind =Rem(328.05+40.6,360)=8.65°. This is 15.35° different from the reference wind direction of 24°.

[0173] The absolute wind direction calculation formula is as follows:

[0174] θ wind =Rem(θ re +θ ship ,360) (9)

[0175] θ wind -----Absolute wind direction

[0176] θ re -----Relative wind direction

[0177] θ ship -----Bow direction

[0178] The algorithm performance verification of the present invention is based on measurement data from a State Oceanic Administration marine surveillance vessel. During experimental testing, the radar's monitoring range is within 4.2 km, and the acquisition time for each image is approximately 2.7 seconds. It is stipulated that 32 images are stored as a time series. The total number of radar images is 2048, with 560 points on each line, a distance resolution of 7.5 meters, and a directional resolution of 0.18 degrees. The wind reference data used in the invention is derived from anemometer data from the State Oceanic Administration. The anemometer is the same as that used for marine radars and is also installed on the main mast of the ship. Wind parameters are recorded in minutes and used to verify the accuracy of the radar-derived wind field.

[0179] For better analysis, we selected 400 radar images from four different wind speed time periods, for a total of 1,600 radar images for testing. We compared this with a single curve fitting method that directly stacks and averages the echo intensity at each azimuth angle. The first phase consisted of radar data with only medium wind speeds and fixed obstructions. The second phase consisted of radar data with only high wind speeds and fixed obstructions. The third phase consisted of radar data with only low wind speeds and fixed obstructions. The fourth phase also consisted of radar data with high wind speeds, but not only was the fixed obstruction caused by masts present, but the radar images also contained interference from targets such as ships. The overall results are shown in Table 1.

[0180] Table 1 Overall search results of wind direction

[0181]

[0182] As can be seen from the table above, the wind direction retrieval method based on the distance attenuation method is more effective and far better than the direct average single curve fitting method. It can well retrieve the wind direction of the radar image, especially in low wind speeds and in the presence of fixed object interference.

[0183] The method proposed in the present invention for performing wind direction retrieval on marine radar images based on the horizontal component of radar image range attenuation improves the accuracy of wind direction retrieval. This method not only overcomes the influence of sea conditions on radar echoes, but also makes full use of the range attenuation characteristics of radar images, thereby significantly improving the accuracy of wind direction retrieval.

Claims

1. A method for wind direction retrieval from marine radar images, characterized in that: include: Step 1: remove the co-frequency interference from the original radar image obtained by reading the radar file, and normalize the pixel intensity of the radar image; Step 2: In the polar coordinate system of the radar image obtained in step 1, select pixels with equal distance from the center of the radar image as a group according to the set distance step size, remove pixels with abnormal intensity in each group of pixels, and then select the pixel with the largest intensity in each group of pixels, and determine the ideal distance attenuation data X based on the pixel with the largest intensity. ra (r), Among them, r represents the radial distance from the radar antenna to the sea surface, θ represents the azimuth of the radar image, and X represents the normalized radar image intensity. Then, the least squares fitting is applied to convert X ra (r) is fitted to the ideal distance decay model D(r) to determine the regression parameters of D(r); Step 3: Fit and compare the ideal distance attenuation model D(r) with the radar image pixel intensity at each azimuth angle to determine the attenuation horizontal component at each azimuth angle. The ideal distance attenuation model D(r) is: Where: r is the radial distance from the radar antenna to the sea surface, b0 and b1 are regression parameters; Step 4: Retrieve the sea surface wind direction based on the wind direction and azimuth dependency of the marine radar image under horizontal polarization.

2. The method for wind direction retrieval from marine radar images according to claim 1, characterized in that: Eliminating pixels with abnormal intensities in each group of pixels in step 2 includes: The intensity range of 0-1 is evenly divided into m histogram panes, and then the pixels are stored in the histogram panes of the corresponding intensity. The pixels in the panes with a number of pixels less than a threshold T are considered abnormal and are removed.

3. The method for wind direction retrieval from marine radar images according to claim 2, characterized in that: The threshold T satisfies: T=k×N Where N is the number of azimuth lines in the radar image, that is, the number of pixels at the same distance; k is a constant between 0 and 1.

4. The method for wind direction retrieval from marine radar images according to claim 1, characterized in that: In step 3, the ideal distance attenuation model D(r) is fitted and compared with the radar image pixel intensity at each azimuth angle to determine the attenuation horizontal component at each azimuth angle, including: Step 3.

1. Initialize the number of iterations N = 1. Starting from the azimuth angle 0° and moving clockwise to 2π, select the radar image pixel intensity X(r,θ) at each azimuth angle in the radar image in turn, and set the weight ω(r) of the echo intensity of 0 in X(r,θ) to zero. The weight ω(r) is: Where r is the radial distance from the radar antenna to the sea surface, l is the range resolution of the radar; n is the discrete point position of the radial distance, and p is the maximum discrete point position of the radial distance; Then, the attenuation horizontal component C of each azimuth is calculated by nonlinear least square method. θ , C θ The calculation formula is: Among them, δ is the set threshold, δ≤1; Step 3.2: Determine whether the set number of iterations N is reached. If so, output the attenuation level component C. θ Otherwise, let N = N + 1, and use the iterative constraint to obtain C θ The value of is optimized, and the weighting function ω(r) and the threshold δ are updated, specifically: The threshold δ is updated to δ / 2; return to step 3.

1.

5. The method for wind direction retrieval from marine radar images according to claim 1, characterized in that: The step 4 of retrieving the sea surface wind direction based on the wind direction and azimuth dependency of the marine radar image under horizontal polarization includes: Step 4.1: Use the least square method to calculate the attenuation horizontal component C θ Fit a cosine function as a function of azimuth: C θ =a0+a1cos 2 (0.5(θ-w)) Where θ is the azimuth of the radar image, a0 and a1 are regression parameters, and w represents the angle of the peak point of the fitting function; Step 4.2: Convert the peak point angle w to relative wind direction: Among them, θ re Indicates relative wind direction; Step 4.3: Convert relative wind direction to absolute wind direction: i wind =Rem(θ re +θ ship ,360) Among them, θ wind represents the absolute wind direction, θ ship Indicates the bow direction of the ship.

Images