A method for identifying a wind turbine in a SAR image of a complex tidal flat subsoil

Abstract

The present application relates to a kind of tidal flat complex underlying surface wind turbine SAR identification method, comprising: generating annual SAR time series composite image;The mean value of each pixel in the preset backscattering coefficient interval of the composite image is calculated, and high backscattering value image is generated;High backscattering value image is binarized and segmented;The morphological processing is carried out to the binarized image to enhance high backscattering target;Multi-scale screening mechanism based on geometric features is established;Difference spatial resolution processing is implemented, and wind turbine point vector data is output.The beneficial effects of the present application are: the present application effectively improves the identification precision of tidal flat and open sea wind turbine, and provides reliable technical support for fine evaluation and management of offshore wind power resources.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, and in particular relates to a SAR identification method for wind turbines on complex tidal flat surfaces. Background Technology

[0002] Currently, marine wind turbine identification technology based on synthetic aperture radar (SAR) image features mainly extracts physical characteristics such as the target's backscattering coefficient and polarization properties, and combines them with texture statistical features for analysis. This technology has shown significant effectiveness in open sea applications. However, when applied to tidal flat areas, it faces numerous technical challenges. The main reason is the unique complexity of the tidal flat environment; its underlying surface differs significantly from open sea areas, and target morphology exhibits diverse characteristics. Furthermore, it is affected by multiple factors such as tidal dynamics, sandbar migration, and vegetation cover, leading to a high false positive and false negative rate for traditional identification methods in this region. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a SAR identification method for wind turbines on complex tidal flat surfaces.

[0004] Firstly, a SAR identification method for wind turbines on complex tidal flat surfaces is provided, including:

[0005] Step 1: Obtain SAR image data for the entire year in time series, and perform preprocessing to generate SAR backscattering coefficients; and select VV polarization images for the target year from the SAR image set to generate annual SAR time series composite images.

[0006] Step 2: Calculate the mean value of each pixel in the composite image within a preset backscattering coefficient range, and generate a high backscattering value image to eliminate temporary moving targets and drifting targets;

[0007] Step 3: Perform binarization segmentation on the high backscattering value image based on the threshold T to generate a binarized image;

[0008] Step 4: Perform morphological processing on the binarized image to enhance high backscattered targets;

[0009] Step 5: Establish a multi-scale screening mechanism based on geometric features to eliminate interfering targets and noise points;

[0010] Step 6: Implement differentiated spatial resolution processing and output wind turbine point vector data.

[0011] As a preferred option, it also includes:

[0012] Step 7: Evaluate the performance of SAR in identifying wind turbines using metrics such as missed classification, misclassification, correct identification, accuracy, recall, and F1 score.

[0013] Preferably, in step 1, the preprocessing includes noise removal and radiometric calibration.

[0014] Preferably, in step 3, a grid-adaptive backscattering filter is used to compare the mean backscattering coefficient within a preset backscattering coefficient range with a threshold T to generate a binarized image; the expression for the threshold T is:

[0015]

[0016] Where, σ max σ represents the maximum value of the backscattering coefficient within the grid. min This represents the minimum backscattering coefficient within the grid.

[0017] Preferably, in step 4, the morphological treatment includes expansion and etching operations.

[0018] Preferably, in step 5, pixel statistics of connected regions are performed on the binarized image after morphological processing to determine the size of each connected region.

[0019] In a second aspect, a SAR identification system for wind turbines on complex tidal flat surfaces is provided, for performing any of the methods described in the first aspect, including:

[0020] The acquisition module is used to acquire SAR image data for the entire year in time series, perform preprocessing to generate SAR backscattering coefficients, and filter VV polarization images of the target year from the SAR image set to generate annual SAR time series composite images.

[0021] The calculation module is used to calculate the mean value of each pixel in the synthetic image within a preset backscattering coefficient range, and generate a high backscattering value image to eliminate temporary moving targets and drifting targets.

[0022] The segmentation module is used to perform binarization segmentation on high backscattering value images based on a threshold T, generating a binary image;

[0023] The processing module is used to perform morphological processing on the binarized image to enhance high backscattered targets;

[0024] A module is established to create a multi-scale screening mechanism based on geometric features to remove interfering targets and noise points;

[0025] The implementation module is used to perform differential spatial resolution processing and output wind turbine point vector data.

[0026] Thirdly, a computer storage medium is provided, wherein a computer program is stored therein; when the computer program is run on a computer, the computer causes the computer to perform any of the methods described in the first aspect.

[0027] Fourthly, an electronic device is provided, comprising:

[0028] Memory, used to store computer programs;

[0029] A processor for executing the computer program to implement the method as described in any of the first aspects.

[0030] The beneficial effects of this invention are as follows: This invention leverages the temporal stability advantage of time-series Sentinel-1 SAR imagery to effectively suppress transient interference noise by constructing annual time-series images. In the SAR wind turbine identification layer, this invention designs a processing framework. First, it removes temporary moving and drifting targets by generating high backscattering values ​​in the imagery. Second, it uses a grid-adaptive backscattering filter to distinguish high-scattering targets from the low-scattering sea surface background. Furthermore, it combines morphological opening and closing operations to enhance the backscattering characteristics of turbine targets. Finally, this invention innovatively adopts a differentiated resolution processing scheme for marine and tidal flat areas, effectively improving the identification accuracy of wind turbines in tidal flats and open sea areas, providing reliable technical support for the refined assessment and management of offshore wind power resources. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the overall technical process of the present invention;

[0032] Figure 2 This is a schematic diagram of the SAR backscattering coefficients obtained by preprocessing Sentinel-1 SAR on January 10, 2018.

[0033] Figure 3 It is an annual SAR time series image generated from 142 Sentinel-1 SAR images from January 1 to December 31, 2018;

[0034] Figure 4 This is a schematic diagram of the results after removing temporary interfering targets from annual SAR time-series images;

[0035] Figure 5 This is a schematic diagram of high backscattering target extraction after adaptive threshold segmentation of annual SAR time series images;

[0036] Figure 6 This is a schematic diagram of a high backscattering target obtained by morphological operations on annual SAR time-series images.

[0037] Figure 7This is a schematic diagram of potential wind turbines obtained after area screening of annual SAR time series images;

[0038] Figure 8 This is a schematic diagram showing the identification results of wind turbines in a local area of ​​Jiangsu.

[0039] Figure 9 This is a schematic diagram of the wind turbine identification results in Jiangsu Province from 2015 to 2024.

[0040] Figure 10 This is a schematic diagram showing the identification accuracy and recall rate of wind turbines in Jiangsu Province from 2015 to 2024. Detailed Implementation

[0041] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0042] Example 1:

[0043] To address the problems of existing technologies, Embodiment 1 of this application provides a SAR identification method for wind turbines on complex tidal flat surfaces, such as... Figure 1 As shown, it includes:

[0044] Step 1: Acquire synthetic aperture radar (SAR) image data for the entire year and preprocess it to generate SAR backscattering coefficients; then select VV polarization images for the target year from the SAR image set to generate annual SAR time series synthetic images.

[0045] Specifically, such as Figures 2-3 As shown, the remote sensing image data source is the 2018 Sentinel-1 SAR remote sensing image of the coastal area of ​​Jiangsu Province. The preprocessing includes noise removal and radiometric calibration.

[0046] Furthermore, the SAR backscattering coefficient is an important parameter describing the Earth's surface's ability to reflect incident electromagnetic waves, and its unit is decibel (dB). Statistical analysis shows that the backscattering coefficient of calm sea surfaces ranges from -30dB to -20dB, while that of tidal flats ranges from -15dB to -5dB; and that of metal structures (wind turbines) ranges from -5dB to +5dB.

[0047] Step 2: Calculate the mean value σ(x,y) of each pixel in the composite image within the preset backscattering coefficient range, and generate a high backscattering value image to eliminate temporary moving targets and drifting targets.

[0048] For example, the percentile of the backscattering coefficient in the 80-100% range of annual time-series SAR images is statistically analyzed, and the mean backscattering coefficient of each pixel within this range is calculated to generate images with high backscattering values, thereby achieving the removal of temporary moving targets and drifting targets.

[0049] Specifically, using the upper quantile average method, the annual VV values ​​for each pixel are first sorted in descending order. Then, the mean backscattering coefficient of the top 20% of pixels (i.e., the 80%-100% quantile range) is calculated, and this mean value is assigned to that pixel, thus generating a value like... Figure 4 The image shown has high backscattering values. Moving targets (such as ships) are typically transient, appearing only at a few points in the year-round imagery, while stationary targets (such as wind turbines) consistently generate high backscattering for most of the time. This method can remove transient moving targets and drifting targets, such as ships, and focus on persistently high reflectivity areas such as the specular reflection from wind turbine blades.

[0050] Step 3: Perform binarization segmentation on the high backscatter value image based on the threshold T to generate a binarized image B(x,y).

[0051] For example, a grid-adaptive backscattering filter is used to distinguish high-scattering targets from low-scattering sea surface backgrounds by using a "semi-minimum-maximum threshold T". The filter compares the mean backscattering coefficient within the 80-100% range in step 2 with the threshold T to generate a filter as shown below. Figure 5 The binarized image shown makes the target clearer, revealing the spatial distribution of targets with high backscattering. Pixels greater than the threshold T are assigned a value of 1; pixels less than or equal to the threshold are assigned a value of 0.

[0052] Specifically, a grid-adaptive backscattering filter is used to compare the mean backscattering coefficient within a preset backscattering coefficient range with a threshold T to generate a binarized image B(x,y).

[0053] The mesh adaptive backscattering filter is a binarization method based on image processing techniques. Its core idea is to dynamically adjust the threshold to classify pixels in an image into two categories: targets (wind turbines) and background (sea surface). Specifically, a "semi-minimum-maximum threshold T" is used to distinguish between high-scattering targets and low-scattering sea surface background, expressed as follows:

[0054]

[0055] The threshold T is set to the maximum value of the backscattering coefficient within the grid (σ). max ) and minimum value (σ) min The backscattering coefficient of each pixel is half of the mean. By comparing the backscattering coefficient of each pixel with the threshold T, a binary image B(x,y) is generated, which can effectively distinguish the wind farm target from the seawater background.

[0056] The binarized image B(x,y) is defined as:

[0057]

[0058] Where σ(x,y) is the backscattering coefficient at position (x,y) (unit: dB); T is the threshold (unit: dB);

[0059] B(x,y)=1: represents a man-made target (offshore wind turbine); B(x,y)=0: represents a natural background (calm sea surface).

[0060] Step 4: Perform morphological processing on the binarized image to enhance high backscattered targets.

[0061] Step 5: Establish a multi-scale screening mechanism based on geometric features to eliminate interfering targets and noise points.

[0062] Step 6: Implement differentiated spatial resolution processing and output wind turbine point vector data.

[0063] Example 2:

[0064] Based on Example 1, Example 2 of this application provides a more specific SAR identification method for wind turbines on complex underlying surfaces of tidal flats, including:

[0065] Step 1: Obtain SAR image data for the entire year in time series, and perform preprocessing to generate SAR backscattering coefficients; then select VV polarization images for the target year from the SAR image set to generate annual SAR time series composite images.

[0066] Step 2: Calculate the mean value of each pixel in the composite image within a preset backscattering coefficient range to generate a high backscattering value image to eliminate temporary moving targets and drifting targets.

[0067] Step 3: Perform binarization segmentation on the high backscatter value image based on the threshold T to generate a binarized image.

[0068] Step 4: Perform morphological processing on the binarized image to enhance high backscattered targets.

[0069] In step 4, such as Figure 6As shown, considering the noise and texture interference in the generated binarized image, morphological analysis methods are used to remove large and small interference objects and enhance high backscattering targets. The morphological processing includes dilation and erosion operations. Dilation outputs the maximum pixel value in the neighborhood, enlarging the target and filling small holes, thus enhancing target connectivity. Erosion outputs the minimum pixel value in the neighborhood, eliminating isolated noise points and retaining only the main target. The number of erosion and dilation operations is 1-2 times, depending on the situation.

[0070] Specifically, the corrosion calculation formula is expressed as follows:

[0071]

[0072] The expansion formula is expressed as follows:

[0073]

[0074] A(k,l) is the pixel value (0 or 1) at position (k,l) of the input image.

[0075] The neighborhood N(i,j) is a circular region centered at (i,j) with radius r, satisfying:

[0076]

[0077] Step 5: Establish a multi-scale screening mechanism based on geometric features to eliminate interfering targets and noise points (such as exposed tidal flats, islands, oil platforms, and other targets, as well as small noise points).

[0078] In step 5, pixel statistics of connected regions are performed on the binarized image after morphological processing to determine the size of each connected region. A maximum pixel count limit of 256 is set for a single connected region; if the number of pixels in a region exceeds this value, it is truncated to 256. Furthermore, an 8-neighborhood connection method is used, meaning that a pixel is considered to be connected to its neighbors in the four cardinal directions (up, down, left, right, and four diagonals). This processing method can effectively identify and count the connected regions of different targets (such as offshore wind turbines) in the image and calculate their area.

[0079] For example, such as Figure 7 As shown, target selection is performed based on area size range (1600-20000 square meters) to generate the final binarized image. The first step is area selection, setting the maximum pixel count limit for a single connected region to 256 and using an 8-neighborhood connection method. Next, logical operations are used to filter regions that meet the area criteria, resulting in the spatial distribution of potential wind turbines. Here, 1 represents a connected region with an area between 1600-20000 square meters (potential offshore wind farms); 0 represents regions that are too small or too large (noise, ships, etc.).

[0080] The 8-neighborhood expression is as follows:

[0081] N8(i,j)={(i-1,j-1),(i-1,j),(i-1,j+1),(i,j-1),(i,j+1),(i+1,j-1),(i+1,j),(i

[0082] +1,j+1)}

[0083] For pixel coordinates (i,j), its 8-neighborhood includes pixels in the surrounding 8 directions, namely all adjacent pixels in the horizontal, vertical, and diagonal directions.

[0084]

[0085] After area screening, the spatial distribution of potential wind turbines was obtained. Next, the underlying surface in this result needs to be divided into marine areas and tidal flats. To achieve this, the composite image from step 1 can be used for observation.

[0086] Step 6: Implement differentiated spatial resolution processing and output wind turbine point vector data.

[0087] Specifically, the spatial resolution of images from different sea areas and tidal flats is differentiated to achieve high-precision wind turbine identification in the tidal flat area. The raster images are converted into point vector data, with each point corresponding to a wind turbine.

[0088] For example, a high spatial resolution of 30 meters is used to ensure the accuracy of small target recognition for the complex intertidal topography (tidal flat area) in the coastal waters of Jiangsu, while a spatial resolution of 300 meters is used to balance computational efficiency for open sea areas. Then, the previously processed binarized raster images are converted into vector data formats region by region, and finally exported in shapefile (shp) format. Here, the vector spatial resolution for the tidal flat output is tentatively set to 8 meters, and the vector spatial resolution for the non-tidal flat output is set to 10 meters. Each point corresponds to an offshore wind farm area, such as... Figure 8 As shown, these can be displayed on the map as red and blue dots respectively, making them easy to view and analyze.

[0089] Step 7: Evaluate the performance of SAR in identifying wind turbines using the metrics of Failure to classify (FN), Misclassification (FP), Correct Identification (TP), Precision (P), Recall (R), and F1 score.

[0090] In remote sensing image recognition and classification tasks, final comparative validation is a crucial step in evaluating model performance and recognition accuracy. By comparing the performance of different methods or models on the same dataset, their strengths and weaknesses in terms of recognition accuracy, misclassification, and missed detection can be comprehensively analyzed. Commonly used evaluation metrics include missed detection (FN), misclassification (FP), and correct detection (TP), which reflect the model's recognition ability and robustness from different dimensions. Accuracy (P) measures the proportion of real targets among all identified targets, reflecting the accuracy of the recognition results. Recall (R) measures the proportion of correctly identified targets among all real targets, reflecting the comprehensiveness of the recognition. Finally, to more comprehensively evaluate the model's recognition performance, the F1 score is usually used as a comprehensive metric, calculated using the following formula:

[0091]

[0092] This achieves a balance between accuracy and comprehensiveness. A systematic analysis of these indicators allows for a more scientific evaluation of the target recognition model's performance, providing a theoretical basis for subsequent optimization and application.

[0093] For example, such as Figure 9 and Figure 10 As shown in the figure, in this application, various indicators were calculated for remote sensing image data of Jiangsu Province from 2015 to 2024. The results show that the accuracy and recall rate are both greater than 95%, and the number of identification errors is relatively small, indicating that the method has high accuracy and stability in target recognition tasks and has good application prospects.

[0094] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be referred to each other, and will not be repeated in this application.

[0095] Example 3:

[0096] Based on Example 1, Example 3 of this application provides a SAR identification system for wind turbines on complex tidal flat surfaces, including:

[0097] The acquisition module is used to acquire SAR image data for the entire year in time series, perform preprocessing to generate SAR backscattering coefficients, and filter VV polarization images of the target year from the SAR image set to generate annual SAR time series composite images.

[0098] The calculation module is used to calculate the mean value of each pixel in the synthetic image within a preset backscattering coefficient range, and generate a high backscattering value image to eliminate temporary moving targets and drifting targets.

[0099] The segmentation module is used to perform binarization segmentation on high backscattering value images based on a threshold T, generating a binary image;

[0100] The processing module is used to perform morphological processing on the binarized image to enhance high backscattered targets;

[0101] A module is established to create a multi-scale screening mechanism based on geometric features to remove interfering targets and noise points;

[0102] The implementation module is used to perform differential spatial resolution processing and output wind turbine point vector data.

[0103] It should be noted that the system provided in this embodiment is the corresponding system of the method provided in embodiment 1. Therefore, the parts that are the same as or similar to those in embodiment 1 in this embodiment can be referred to each other, and will not be described again in this application.

Claims

1. A SAR identification method for wind turbines on complex tidal flat surfaces, characterized in that, include: Step 1: Acquire SAR image data for the entire year in time series and preprocess it to generate SAR backscattering coefficients; Then, VV polarization images for the target year are selected from the SAR image set to generate annual SAR time series composite images; Step 2: Calculate the mean value of each pixel in the composite image within a preset backscattering coefficient range, and generate a high backscattering value image to eliminate temporary moving targets and drifting targets; Step 3: Perform binarization segmentation on the high backscattering value image based on the threshold T to generate a binarized image; Step 4: Perform morphological processing on the binarized image to enhance high backscattered targets; Step 5: Establish a multi-scale screening mechanism based on geometric features to eliminate interfering targets and noise points; In step 5, pixel statistics of connected regions are performed on the binarized image after morphological processing to determine the size of each connected region; The maximum pixel count limit for a single connected region is set to 256. If the number of pixels in a region exceeds this value, it will be truncated to 256. In addition, an 8-neighborhood connection method is adopted, that is, a pixel is considered to be a connected region with its adjacent pixels in the directions of its top, bottom, left, right and four diagonals. The 8-neighborhood expression is as follows: For pixel coordinates Its 8-neighborhood includes pixels in the surrounding 8 directions, that is, all adjacent pixels in the horizontal, vertical, and diagonal directions, represented as: Step 6: Implement differentiated spatial resolution processing and output wind turbine point vector data.

2. The SAR identification method for wind turbines on complex tidal flat surfaces according to claim 1, characterized in that, Also includes: Step 7: Evaluate the performance of SAR in identifying wind turbines using metrics such as missed classification, misclassification, correct identification, accuracy, recall, and F1 score.

3. The SAR identification method for wind turbines on complex tidal flat surfaces according to claim 2, characterized in that, In step 1, the preprocessing includes noise removal and radiometric calibration.

4. The SAR identification method for wind turbines on complex tidal flat surfaces according to claim 3, characterized in that, In step 3, a grid-adaptive backscattering filter is used to compare the mean backscattering coefficient within a preset backscattering coefficient range with a threshold T to generate a binarized image; the expression for the threshold T is: in, This represents the maximum value of the backscattering coefficient within the grid. This represents the minimum backscattering coefficient within the grid.

5. The SAR identification method for wind turbines on complex tidal flat surfaces according to claim 4, characterized in that, In step 4, the morphological treatment includes expansion and etching operations.

6. A SAR identification system for wind turbines on complex tidal flat surfaces, characterized in that, For performing the method according to any one of claims 1 to 5, comprising: The acquisition module is used to acquire SAR image data for the entire year in time series, perform preprocessing to generate SAR backscattering coefficients, and filter VV polarization images of the target year from the SAR image set to generate annual SAR time series composite images. The calculation module is used to calculate the mean value of each pixel in the synthetic image within a preset backscattering coefficient range, and generate a high backscattering value image to eliminate temporary moving targets and drifting targets; The segmentation module is used to perform binarization segmentation on high backscattering value images based on a threshold T, generating a binary image; The processing module is used to perform morphological processing on the binarized image to enhance high backscattered targets; A module is established to create a multi-scale screening mechanism based on geometric features to remove interfering targets and noise points; The implementation module is used to perform differential spatial resolution processing and output wind turbine point vector data.

7. A computer storage medium, characterized in that, The computer storage medium stores a computer program; when the computer program is run on the computer, it causes the computer to perform the method described in any one of claims 1 to 5.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 5.

Images