Multispectral remote sensing accurate extraction method for offshore wind turbine point location

Abstract

The invention discloses a multispectral remote sensing accurate extraction method for an offshore wind turbine point location, and the method comprises the steps: 1, collecting a multi-temporal satellite remote sensing image covering a coastal zone region, carrying out the water body information enhancement and threshold segmentation processing, generating a sea-land binary image, and separating an offshore research region; 2, collecting and dividing a multi-stage remote sensing data set of the offshore wind plant; constructing an offshore wind turbine enhancement index capable of effectively distinguishing the offshore wind turbine and the background ground feature, and generating an offshore wind turbine enhancement image; constructing a visible light superposition index capable of retaining shadow information of the offshore wind turbine, and generating an offshore wind turbine shadow enhanced image; 3, multi-directional gray-level co-occurrence matrixes are calculated for the multi-stage offshore wind turbine enhanced images respectively, and a low-noise wind turbine binary image is obtained through synthesis; and 4, performing mathematical morphology expansion processing and vectorization on the low-noise fan binary image, and performing calculation to obtain accurate position data of the offshore fan.

Description

Technical Field

[0001] The present invention relates to the technical field of ocean engineering monitoring, and particularly relates to a method for accurately extracting multi-spectral remote sensing of offshore wind turbine positions. Background Art

[0002] With the rapid development of the global economy, the extensive use of fossil fuels poses a serious threat to world energy security. As a clean and renewable energy source, wind energy has received extensive attention. The offshore wind energy industry is developing steadily, and the construction of offshore wind farms worldwide continues to expand, with the scale constantly increasing. Mastering the situation of offshore wind power projects using sea areas is of great significance for standardizing the sea use activities of the offshore wind power industry and promoting the scientific and reasonable development and utilization of sea area space resources.

[0003] Currently, the identification of offshore wind turbines mainly uses microwave SAR images, and the differences in SAR backscattering coefficients are used to distinguish high backscattering objects (wind turbines) from low backscattering objects (seawater background). Since SAR remote sensing images lack color information and do not perform well in showing details and textures, it is difficult to distinguish wind turbines from similar targets such as offshore substations and offshore converter stations; when the quality of short-term synthetic SAR images is not high, the synthetic period needs to be extended to ensure the quality of the synthetic images, and the advantages compared with multi-spectral images are not obvious.

[0004] From the existing research, there are few identification methods for offshore wind power that involve multi-spectral remote sensing data. Therefore, researching methods for identifying and positioning offshore wind turbines based on multi-spectral remote sensing images can provide important technical support for monitoring the construction and operation of offshore wind farms and can enrich and improve the methods for identifying offshore wind turbines based on satellite remote sensing.

[0005] When using multi-spectral images to identify offshore wind turbines, the following technical problems mainly exist: 1) In turbid sea areas or intertidal zones, the spectral characteristics of offshore wind turbines are not obvious, and it is easy to be confused with the surrounding environmental background, making it difficult to accurately extract offshore wind turbines; 2) Offshore wind farms often contain facilities such as offshore substations. Such targets have similar spectral characteristics to offshore wind turbines, and existing identification methods are likely to misjudge such targets as offshore wind turbines, thus increasing the error of the extraction results. Summary of the Invention

[0006] Object of the Invention. In view of the above problems, the present invention proposes a method for accurately extracting multi-spectral remote sensing of offshore wind turbine positions.

[0007] Technical Solution. To achieve the object of the present invention, the present invention proposes a method for accurately extracting multi-spectral remote sensing of offshore wind turbine positions, and the method includes the following steps:

[0008] S10. Collect multi-temporal satellite remote sensing images covering the coastal zone area, perform water body information enhancement and threshold segmentation processing to generate a land-sea binary image, isolate the offshore study area, and delimit the area range for extracting offshore wind turbines;

[0009] S20. Under the area range delimited in step S10, collect and divide multi-phase remote sensing datasets of offshore wind farms; construct an offshore wind turbine enhancement index that can effectively distinguish offshore wind turbines from background ground objects to generate an offshore wind turbine enhancement image; construct a visible light superposition index that can retain the shadow information of offshore wind turbines to generate an offshore wind turbine shadow enhancement image;

[0010] S30. Calculate the gray-level co-occurrence matrices in multiple directions for the multi-phase offshore wind turbine enhancement images respectively, extract the contrast, variance, and cluster prominence of the images, and merge them to generate multi-phase composite images; use the random forest algorithm for classification to obtain multi-phase noisy wind turbine binary images; use logical "AND" operation to stack the multi-phase classification results for noise reduction processing, and synthesize to obtain the wind turbine binary image;

[0011] S40. Perform mathematical morphological dilation processing and vectorization on the wind turbine binary image obtained in S30, and sequentially perform mathematical morphological area screening, boundary distance judgment, combined with Roberts operator edge detection and support vector machine SVM discrimination, and point density data screening to obtain the position data of offshore wind turbines.

[0012] Further, in S10, collecting multi-temporal satellite remote sensing images covering the coastal zone area, performing water body information enhancement and threshold segmentation processing to generate a land-sea binary image, isolating the offshore study area, and delimiting the area range for extracting offshore wind turbines includes:

[0013] Step S11. Collect multi-temporal satellite remote sensing images covering the coastal zone area with a resolution requirement of not more than 30 meters, and complete the de-clouding and median synthesis processing of the images;

[0014] Step S12. Adopt the maximum spectral index synthesis algorithm to calculate the modified normalized difference water index MNDWI index for each pixel at the same position in the multi-temporal images one by one, select the maximum value of the pixels as the pixel value at the corresponding position in the synthesized image, and create a water body information enhancement image. The calculation formula of MNDWI is as follows:

[0015]

[0016] In the formula, R g is the reflectance of the green light band; R swir is the reflectance of the short-wave infrared band;

[0017] Step S13: Calculate the optimal threshold for land-water separation of the water body information enhanced image using the Otsu algorithm, and generate a land-sea binary image through threshold segmentation processing; extract the sea area part from the binary image and perform vectorization operation to obtain the vector of the sea area of the offshore research area, and separate the offshore research area.

[0018] Further, in step S20, collect and divide multi-phase remote sensing datasets of offshore wind farms; construct an offshore wind turbine enhancement index that can effectively distinguish offshore wind turbines from background ground objects to generate an offshore wind turbine enhanced image; construct a visible light superposition index that can retain the shadow information of offshore wind turbines to generate an offshore wind turbine shadow enhanced image, including:

[0019] Step S21: Under the limited area of the vector result in S13, collect multi-spectral satellite remote sensing images of multiple different time phases, with the ground resolution required to be no greater than 15 meters and the cloud cover less than 10%; on the premise of ensuring that each time-phase image can cover the offshore wind farm, divide the multi-time-phase image data into at least two image sets;

[0020] Step S22: Construct an offshore wind turbine enhancement index OWTEI using the bands with spectral differences between offshore wind turbines and background ground objects; calculate the offshore wind turbine enhancement index for all images in each image set to obtain the corresponding offshore wind turbine enhanced image. The calculation formula of OWTEI is as follows:

[0021]

[0022] In the formula, R r is the reflectance of the red light band; R g is the reflectance of the green light band; R b is the reflectance of the blue light band; R nir is the reflectance of the near-infrared band;

[0023] Step S23: Construct a visible light superposition index SUM rgb that can retain the shadow information of offshore wind turbines; calculate the visible light superposition index for all images in each image set to obtain the corresponding offshore wind turbine shadow enhanced image. The calculation formula of SUM rgb is as follows:

[0024] SUM rgb = R r + R g + R b .

[0025] Further, in S30, for multi-phase enhanced images of offshore wind turbines, calculate the gray-level co-occurrence matrices in multiple directions respectively, extract the contrast, variance, and cluster prominence of the images, and merge them to generate multi-phase synthetic images; classify using the random forest algorithm to obtain multi-phase binary images of noisy wind turbines; use the logical "AND" operation to stack the multi-phase classification results for noise reduction processing, and the synthetic binary image of the wind turbine includes:

[0026] Step S31: Perform median synthesis processing within each set of images in each phase to obtain the median synthetic image of each phase; extract the enhanced images of offshore wind turbines in the median synthetic images of each phase and calculate the gray-level co-occurrence matrices in four directions of 0°, 90°, 45°, and 135° respectively to generate contrast, variance, and cluster prominence images;

[0027] Step S32: Synthesize the blue, green, red, and near-infrared bands of the median synthetic images of each phase, combine with the enhanced OWTEI images of offshore wind turbines and the contrast, variance, and cluster prominence images to generate multi-phase synthetic images;

[0028] Step S33: For the synthetic images of each phase, use the random forest classification method to distinguish offshore wind turbines from background ground objects respectively, and fill the statistical values into the neighboring pixels to obtain multi-phase binary images of noisy wind turbines;

[0029] Step S34: Use the logical "AND" operation to stack the multi-phase binary images of noisy wind turbines, perform noise reduction processing, and synthesize to obtain the binary image of the wind turbine.

[0030] Further, in step S40, perform mathematical morphological dilation processing and vectorization on the binary image of the low-noise wind turbine, and then perform mathematical morphological area screening, boundary distance judgment, combined with Roberts operator edge detection and support vector machine SVM discrimination, and point density data screening in sequence. Finally, obtain the accurate position data of the offshore wind turbine including:

[0031] Step S41: Perform mathematical morphological dilation calculation on the binary image of the low-noise wind turbine to obtain the dilated binary image. In the mathematical morphological processing, the structure element SE selects a square kernel, and the size S calculation formula is as follows:

[0032]

[0033] In the formula, GSD 1 is the ground resolution of the enhanced image of the offshore wind turbine in step S22, and Round represents rounding calculation; the size of the square kernel is S rows × S columns;

[0034] Step S42: After performing vectorized dilation processing on the binary image, calculate the area of each vector patch; by setting the area threshold range from 1000m 2 to 20000m 2Extract offshore wind turbine spots; further calculate the center of gravity of the extracted offshore wind turbine spots to obtain M original offshore wind turbine point location data.

[0035] Step S43: traverse the original offshore wind turbine point data to generate M N×N feature matrix windows. The feature matrix window size calculation formula is:

[0036]

[0037] In the formula, GSD 2 is the spatial resolution of the offshore wind turbine shadow enhancement image in step S23, and Round represents rounding calculation;

[0038] Step S44: determine whether the M feature windows intersect with the surface vector boundary of the offshore study area generated in step S13. If so, the offshore wind turbine point corresponding to the feature window is determined as a coastal interference point and is removed; the offshore wind turbine shadow enhancement image in step S23 is cropped using the remaining feature window position and window size, and its value is linearly stretched to 0-255 to obtain a standardized spot image of the offshore wind turbine shadow;

[0039] Step S45: Use the Roberts operator to perform edge detection on the standardized patch image of the offshore wind turbine shadow; divide and label the offshore wind turbine and offshore booster station sample data set based on the detection results; input the sample data set based on edge detection and its classification label through support vector machine (SVM) model training, and save the trained classification model; use the classification model to discriminate the edge detection results, distinguish the offshore wind turbine from the offshore booster station, and obtain the preliminary processed wind turbine point data;

[0040] Step S46: Based on the preliminarily processed wind turbine point data, the number of wind turbines in the wind turbine point grid neighborhood is divided by the area of ​​the neighborhood to obtain a point density grid image; wherein, the pixel size of the point density grid is set to 1000 meters, the neighborhood analysis is circular, and the neighborhood radius is 8 pixels; according to the geographic coordinate information in the wind turbine point data, the point density value corresponding to each wind turbine point in the point density grid image is extracted, and those with a point density less than 0.06 are determined as cloud interference points and removed, so as to finally obtain accurate wind turbine point data.

[0041] Beneficial technical effects: Compared with the prior art, the technical solution of the above invention has the following beneficial technical effects:

[0042] (1) The present invention utilizes the bands with significant spectral differences between offshore wind turbines and background features, as well as the spectral change information between bands, to originally propose an offshore wind turbine enhancement index. The enhanced image of the offshore wind turbine calculated using this index can reduce the high reflectivity of the background and the significant differences between background features, while retaining the high reflectivity characteristics of the wind turbine, significantly enhancing the difference between the wind turbine and the environmental background, and having good effects in clear waters, turbid waters, and intertidal zones;

[0043] (2) The present invention proposes to calculate the multi-directional gray-level co-occurrence matrix and extract contrast, variance, and saliency as one of the feature parameters. Synthesize the blue, green, red, and near-infrared images of the median composite images in each period and the enhanced image of the offshore wind turbine, generate a multi-period 8-band composite image, use random forest classification and stack multi-period data to retain the patch image that meets the wind turbine morphology and has position invariance; among them, calculating the multi-directional gray-level co-occurrence matrix as one of the feature parameters, and the preliminary classification method of using random forest classification and stacking short-term multi-temporal data is also applicable to other microwave SAR satellite image data, and this preliminary classification method itself has universality.

[0044] (3) The post-processing method proposed by the present invention is as follows: utilize the enhanced image of the offshore wind turbine shadow, combine Roberts operator edge detection and support vector machine SVM discrimination, and point density threshold screening; this process greatly improves the accuracy of the wind turbine point results;

[0045] (4) The present invention proposes an offshore wind turbine point extraction process including calculating the enhancement index, preliminary classification of the wind turbine, and post-processing of the wind turbine points. It can not only identify and extract the offshore wind turbine points, but also distinguish the booster station and the wind turbine in the wind farm, making it possible to monitor offshore wind turbines using multi-spectral satellite remote sensing technology, and also providing important technical support for identifying offshore wind turbines using multi-spectral satellite remote sensing technology;

[0046] (5) Achieve precise multi-spectral remote sensing extraction of the position of offshore wind turbines. This method provides a feasible solution for identifying and extracting offshore wind turbine points using multi-spectral remote sensing technology, and can also provide technical support for analyzing the extraction of small targets with fixed positions and morphological characteristics in the sea and intertidal zones. Description of the Drawings

[0047] Figure 1 Schematic diagram of the steps of the multi-spectral remote sensing extraction method for offshore wind turbine points in an embodiment;

[0048] Figure 2 Flowchart of the multi-spectral remote sensing extraction method for offshore wind turbine points in an embodiment;

[0049] Figure 3 Process diagram of the water-land boundary extraction in an embodiment;

[0050] Figure 4 Schematic diagram of the enhanced image of an offshore wind turbine and the enhanced image of the shadow of the offshore wind turbine in an embodiment;

[0051] Figure 5 Schematic diagram of the contrast, variance and saliency features of the gray-level co-occurrence matrix in an embodiment;

[0052] Figure 6 Schematic diagram of the noise reduction effect of the different-phase stack in an embodiment;

[0053] Figure 7 Schematic diagram of the gray-scale image of the feature window and the edge detection by the Roberts operator in an embodiment;

[0054] Figure 8 Schematic diagram of the extraction results of an offshore wind turbine and an offshore substation in an embodiment;

[0055] Figure 9 Schematic diagram of the extracted positions of the offshore wind turbines and the point density distribution in an embodiment. Detailed implementation manners

[0056] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] Referring to "embodiment" herein means that the specific features, structures or characteristics described in connection with the embodiment may be included in at least one embodiment of the present application. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0058] In one embodiment, an offshore and intertidal wind farm located in the coastal area of a certain coastal province (such as Jiangsu Province) is selected as the research area to carry out the research on the extraction of offshore wind turbine positions by multi-spectral remote sensing. Sentinel-2 MSI remote sensing images (spatial resolution: 10 m) covering the research area and Landsat-8 surface reflectance images (spatial resolution: 30 m) are collected as data sources. The time span of the Sentinel-2 MSI remote sensing images is from November 1, 2023 to February 29, 2024, and the time span of the Landsat-8 surface reflectance images is from January 1, 2023 to December 31, 2023. Google Earth Engine cloud platform (hereinafter referred to as GEE), Envi 5.6 is used as the remote sensing image processing tool, ArcGIS 10.8 is used as the vector data processing tool, and Matlab is used as the post-processing discrimination tool.

[0059] The above method for identifying satellite multi-spectral remote sensing of offshore wind turbines based on the analysis of multi-spectral satellite remote sensing image data refers to Figure 2 As shown, the specific steps are as follows:

[0060] S1: Collect multi-temporal satellite remote sensing images covering the coastal zone, perform water body information enhancement and threshold segmentation processing to generate a land-sea binary image, separate the offshore research area, and limit the area range for offshore wind turbine extraction. The specific operations are as follows:

[0061] S11: Collect multi-temporal satellite remote sensing images covering the coastal zone with a resolution requirement of not more than 30 m, and complete the de-clouding and median synthesis processing of the images;

[0062] S12: Collect 64 multi-temporal remote sensing images covering the coastal zone of the research area with a cloud cover of less than 10% from Landsat-8 through GEE, and the time span is from January 1, 2023 to December 31, 2023; The image databases of GEE have all completed radiometric calibration, atmospheric correction, and geometric correction; Use the improved normalized difference water index MNDWI to enhance the water body, such as Figure 3 a, where the calculation formula of MNDWI is as follows:

[0063]

[0064] In the formula, R g is the reflectance of the green light band; R swir is the reflectance of the short-wave infrared band.

[0065] Use the maximum spectral index synthesis algorithm to create an enhanced image of water body information at higher tide levels, such as Figure 3 b, that is, calculate the MNDWI index for each pixel at the same position in the multi-temporal images one by one, and finally select the maximum value pixel as the pixel at the corresponding position in the synthesized image.

[0066] S13: Calculate the optimal threshold of the water body information enhanced image using the Otsu algorithm in GEE, and obtain the binary image of water and land separation through threshold segmentation, as Figure 3 c. Generate the vector of the sea study area surface in ArcGIS 10.8, and make local adjustments to the vector boundary to obtain the final vector of the sea study area surface, as Figure 3 d.

[0067] S2: Under the regional scope defined in step S10, collect and divide multi-phase remote sensing datasets of offshore wind farms; construct an offshore wind turbine enhancement index that can effectively distinguish offshore wind turbines from background ground objects, and generate an offshore wind turbine enhancement image; construct a visible light superposition index that can retain the shadow information of offshore wind turbines, and generate an offshore wind turbine shadow enhancement image; the specific operations are as follows:

[0068] S21: Collect 146 multi-temporal remote sensing images covering the offshore and intertidal wind farm areas with cloud cover less than 10% from Sentinel-2 MSI through GEE; the Sentinel-2 MSI database in GEE has completed radiometric calibration, atmospheric correction, and geometric correction; divide the 146 images into two image sets according to the time periods from November 1, 2023 to December 31, 2023 and from January 1, 2024 to February 29, 2024 (divide the time periods as much as possible under the condition of meeting the imaging quality), and each of the two sets has 73 images.

[0069] S22: Calculate the offshore wind turbine enhancement index in GEE, and add the turbine feature bands to each original image, as shown in Figure 4 a. Among them, based on the bands with significant spectral differences between offshore wind turbines and background ground objects and the variation information between bands, establish the offshore wind turbine enhancement index OWTEI. The calculation formula of OWTEI is as follows:

[0070]

[0071] In the formula, R r is the reflectance of the red light band; R g is the reflectance of the green light band; R b is the reflectance of the blue light band; R nir is the reflectance of the near-infrared band;

[0072] S23: Calculate the visible light superposition index in GEE, and add the turbine shadow feature bands to each original image, as shown in Figure 4 b. Among them, the formula of the visible light superposition index is as follows:

[0073] SUM rgb =R r +R g+R b

[0074] In the formula, R r is the reflectance in the red light band; R g is the reflectance in the green light band; R b is the reflectance in the blue light band.

[0075] S3: Calculate the gray-level co-occurrence matrices in multiple directions for multi-temporal enhanced images of offshore wind turbines respectively, extract the contrast, variance, and cluster prominence of the images, and combine them to generate multi-temporal composite images; classify using the random forest algorithm to obtain the noisy wind turbine binary images in multiple periods; use the logical "AND" operation to stack the classification results in multiple periods for noise reduction processing, and synthesize to obtain the wind turbine binary image. The specific operations are as follows:

[0076] S31: In GEE, perform median composite processing within the collection on the two sets of image data for extraction and classification (both have added two enhanced bands) to obtain two median composite images at different time periods; calculate the gray-level co-occurrence matrices in four directions of 0°, 90°, 45°, and 135° for the enhanced images of offshore wind turbines in each period, and extract the contrast, variance, and salience features. Refer to Figure 5 shown, (a) is the enhanced image of the offshore wind turbine, (b) is the schematic diagram of the contrast parameter of the gray-level co-occurrence matrix, (c) is the schematic diagram of the variance parameter of the gray-level co-occurrence matrix, and (d) is the schematic diagram of the salience parameter of the gray-level co-occurrence matrix;

[0077] S32: Synthesize the blue, green, red, and near-infrared bands of the median composite images in each period, combine with the enhanced OWTEI image of the offshore wind turbine and the contrast, variance, and cluster prominence images to generate multi-temporal composite images;

[0078] S33: For the composite images in each period in GEE, use the random forest classification method to distinguish the offshore wind turbines from the background ground objects respectively, and statistically fill the numerical values into the neighboring pixels. When exporting in GEE, it is default to fill according to the image resolution to obtain the noisy wind turbine binary images in each period;

[0079] S34: In GEE, use the logical "AND" operation to stack the noisy wind turbine binary images in multiple periods for noise reduction processing, and synthesize to obtain the low-noise wind turbine binary image. Refer to Figure 6 shown, this figure is the schematic diagram of the noise reduction effect of different time-phase stacking in this embodiment.

[0080] S4: Perform mathematical morphological dilation processing and vectorization on the low-noise wind turbine binary image, and then sequentially perform mathematical morphological area screening, boundary distance judgment, combined with Roberts operator edge detection and support vector machine SVM discrimination, and point density data screening to finally obtain the accurate position data of the offshore wind turbine. The specific operations are as follows:

[0081] S41: Perform mathematical morphological dilation calculation on the low-noise binary image of the wind turbine and the environmental background in Envi 5.6 to obtain the dilated binary image. In the mathematical morphological processing, the structuring element (SE) is selected as a square kernel, and the size calculation formula is as follows:

[0082]

[0083] In the formula, GSD 1 is the ground resolution of the enhanced image of the offshore wind turbine in step S22, and Round represents rounding calculation; the size of the square kernel is S rows × S columns. The ground resolution of the Sentinel 2 MSI satellite image is 10 meters, and the size of the square kernel can be obtained as 11 grid pixel units.

[0084] S42: Vectorize the dilated binary image, calculate the area of each patch, and extract the offshore wind turbine patches according to the patch area threshold discrimination. Further extract the centroid position of the patch vector to obtain the original offshore wind turbine point data (the number is 2,961). The area range of the above-mentioned offshore wind turbine surface vector is generally between 1,000 m 2 and 20,000 m 2 , but the actual range needs to be determined specifically for the dilated binary image;

[0085] S43: Traverse the preliminarily processed wind turbine point data in matlab to generate 2,961 N×N feature matrix windows. The feature matrix window size calculation formula is:

[0086]

[0087] In the formula, GSD 2 is the spatial resolution of the enhanced image of the offshore wind turbine shadow in step S23, and Round represents the rounding calculation formula. The ground resolution of the Sentinel 2 MSI satellite image is 10 meters, and the window size can be obtained as 43 grid pixel units.

[0088] S44: First, in matlab, determine whether the 2,961 feature windows intersect with the boundary of the offshore research area surface vector generated in step S13. If they intersect, the offshore wind turbine point corresponding to the feature window is determined as a coastal interference point and removed; then, use the remaining feature window positions and window sizes to crop the enhanced image of the offshore wind turbine shadow in step S23 and linearly stretch its values to 0 - 255 to obtain the standardized patch image of the offshore wind turbine shadow.

[0089] Reference Figure 7As shown in the figure, this is the grayscale image of the feature window and the schematic diagram of edge detection by the Roberts operator in this embodiment. Among them, (a) is the image of the offshore wind turbine window patch, (b) is the image of the offshore substation window patch, (c) is the schematic diagram of offshore wind turbine edge detection, and (d) is the schematic diagram of offshore substation edge detection;

[0090] S45: Use the Roberts operator in Matlab to calculate and extract the edges of the normalized patch image of the offshore wind turbine shadow, manually select samples of offshore wind turbines and offshore substations from it, and train the support vector machine SVM; use the trained model to judge and classify the positions of offshore wind turbines and offshore substations to obtain the preliminarily processed wind turbine position data. The discrimination results are as Figure 8 shown in the figure, this is the schematic diagram of offshore wind turbines and offshore substations.

[0091] S46: In ArcGIS 10.8, according to the preliminarily processed wind turbine position data, divide the number of wind turbines in the raster neighborhood of the wind turbine position by the area of the neighborhood. The pixel size of the set point density raster is 1000 meters, the neighborhood analysis is circular, and the neighborhood radius is 8 pixels to calculate the point density raster image; the positions distributed in the area with a point density less than 0.06 are cloud spot interference points; use the point density threshold to screen out the cloud spot interference points to obtain the final accurate wind turbine positions. Refer to Figure 9 shown in the figure, this is the schematic diagram of the distribution of offshore wind turbine positions and point density. This result can be used to monitor and identify the distribution of offshore and intertidal wind turbines, providing a reference basis for the construction and evaluation of offshore and intertidal wind farms.

[0092] Therefore, this patent intends to form a multi-spectral remote sensing precise extraction technology system for offshore wind turbine positions starting from technologies such as offshore wind turbine feature enhancement, morphological enhancement based on gray-level co-occurrence matrix, multi-temporal image stack denoising, support vector machine SVM discrimination, and point density threshold screening, to solve the problems of low extraction accuracy of offshore wind turbines in turbid sea areas or intertidal zones and being easily confused with similar offshore targets. The purpose is to provide a feasible solution for identifying and extracting offshore wind turbine positions using multi-spectral remote sensing technology, and at the same time, it can also provide technical support for analyzing the extraction of small targets with fixed positions and morphological characteristics in offshore and intertidal zones.

Claims

1. A multi-spectral remote sensing accurate extraction method for offshore wind turbine locations, characterized in that: The method comprises the following steps: S10, collect multi-temporal satellite remote sensing images covering the coastal area, perform water body information enhancement and threshold segmentation processing, generate sea-land binary images, separate the offshore study area, and limit the regional scope of offshore wind turbine extraction; S20, collecting and dividing multiple periods of offshore wind farm remote sensing data sets within the area defined in step S10; Construct an offshore wind turbine enhancement index that can effectively distinguish offshore wind turbines from background objects and generate an offshore wind turbine enhanced image; Construct a visible light superposition index that can preserve the shadow information of offshore wind turbines and generate an enhanced image of offshore wind turbine shadows; S30, respectively calculate the grayscale co-occurrence matrix in multiple directions for the enhanced images of offshore wind turbines in multiple periods, extract the contrast, variance, and cluster prominence of the images, and merge them to generate a multi-period synthetic image; use the random forest algorithm to classify and obtain the noisy binary images of wind turbines in multiple periods; use the logical "and" operation to stack the classification results of multiple periods for noise reduction, and synthesize to obtain the binary image of the wind turbine; S40, performing mathematical morphological expansion processing and vectorization on the binary image of the wind turbine obtained in S30, and sequentially performing mathematical morphological area screening, boundary distance judgment, combining Roberts operator edge detection and support vector machine (SVM) discrimination and point density data screening to obtain the location data of the offshore wind turbine.

2. The multi-spectral remote sensing accurate extraction method of offshore wind turbine points according to claim 1 is characterized in that: In S10, multi-temporal satellite remote sensing images covering the coastal area are collected, and water information is enhanced and threshold segmented to generate sea-land binary images, separate the offshore study area, and limit the area for offshore wind turbine extraction to include: Step S11, collecting multi-temporal satellite remote sensing images covering the coastal area, with a resolution requirement of no more than 30 meters, and completing cloud removal and median synthesis processing of the images; Step S12, using the maximum spectral index synthesis algorithm, calculate the improved normalized difference water index MNDWI index for all pixels at the same position in the multi-temporal image one by one, select the maximum value of the pixel as the pixel value of the corresponding position in the synthetic image, and create a water information enhanced image. The calculation formula of MNDWI is as follows: In the formula, R g is the reflectivity of green light band; R swir is the reflectivity in the short-wave infrared band; Step S13, using the Otsu algorithm to calculate the optimal threshold for water-land separation of the water body information enhanced image, generating a sea-land binary image through threshold segmentation processing; and extracting the sea area part of the binary image and performing vectorization operation to obtain the surface vector of the offshore study area, thereby separating the offshore study area.

3. The multi-spectral remote sensing accurate extraction method of offshore wind turbine points according to claim 2 is characterized in that: In step S20, multiple periods of offshore wind farm remote sensing data sets are collected and divided; an offshore wind turbine enhancement index that can effectively distinguish offshore wind turbines from background objects is constructed to generate an offshore wind turbine enhanced image; Constructing a visible light superposition index that can preserve the shadow information of offshore wind turbines and generating an enhanced image of offshore wind turbine shadows includes: Step S21: In the limited area of ​​the surface vector result of S13, a plurality of multispectral satellite remote sensing images of different time phases are collected, with the ground resolution requirement being no greater than 15 meters and the cloud cover being less than 10%; on the premise of ensuring that the images of each time phase can cover the offshore wind farm, the multi-temporal image data are divided into at least two image sets; Step S22: construct an offshore wind turbine enhancement index OWTEI using the bands with spectral differences between the offshore wind turbine and the background objects; calculate the offshore wind turbine enhancement index for all images in each period of the image set to obtain the corresponding offshore wind turbine enhanced image. The calculation formula of OWTEI is as follows: In the formula, R r is the reflectivity of red light band; R g is the reflectivity of green light band; R b is the reflectivity of blue light band; R nir is the reflectivity in the near-infrared band; Step S23: Constructing a visible light superposition index SUM that can preserve the shadow information of offshore wind turbines rgb ; Calculate the visible light superposition index for all images in each image set to obtain the corresponding offshore wind turbine shadow enhancement image, SUM rgb The calculation formula is as follows: SUM rgb =R r +R g +R b 。 4. The multi-spectral remote sensing accurate extraction method of offshore wind turbine points according to claim 3 is characterized in that: In S30, the gray-level co-occurrence matrix of multiple directions is calculated for multiple periods of offshore wind turbine enhanced images, and the contrast, variance, and cluster prominence of the images are extracted, and the multi-period synthetic images are generated by merging. The random forest algorithm is used for classification to obtain multiple periods of noisy wind turbine binary images. The multi-period classification results of the stack are stacked using logical "AND" operations to perform noise reduction, and the resulting binary images of the wind turbine include: Step S31: performing median synthesis processing on the image set of each period to obtain the median synthetic image of each period; The enhanced images of offshore wind turbines in the median synthetic images of each period were extracted and the gray-level co-occurrence matrices at four directions of 0°, 90°, 45°, and 135° were calculated to generate contrast, variance, and cluster prominence images; Step S32: synthesizing the blue, green, red and near-infrared bands of the median synthetic images of each period, combining the OWTEI enhanced image of the offshore wind turbine and the contrast, variance and cluster prominence images, to generate a multi-period synthetic image; Step S33: for each period of synthetic images, the random forest classification method is used to distinguish the offshore wind turbines from the background objects, and the statistical values ​​are filled into the neighboring pixels to obtain the noisy wind turbine binary images of multiple periods; Step S34: using a logical “AND” operation, stacking multiple phases of noisy fan binary images, performing noise reduction processing, and synthesizing the fan binary image.

5. The multi-spectral remote sensing accurate extraction method of offshore wind turbine points according to claim 4 is characterized in that: In step S40, the low-noise wind turbine binary image is subjected to mathematical morphological dilation processing and vectorization, and then mathematical morphological area screening, boundary distance judgment, Roberts operator edge detection and support vector machine SVM discrimination and point density data screening are performed in sequence, and finally the precise location data of the offshore wind turbine is obtained, including: Step S41: Perform mathematical morphological expansion calculation on the low-noise fan binary image to obtain an expanded binary image. In the mathematical morphological processing, the structural element SE selects a square kernel, and the size S is calculated as follows: Where GSD1 is the ground resolution of the offshore wind turbine enhanced image in step S22, Round means rounding calculation; the size of the square kernel is S rows × S columns; Step S42: After vector dilation processing of the binary image, the area of ​​each vector spot is calculated; by setting the area threshold range to 1000m 2 Up to 20000m 2 Extract offshore wind turbine spots; further calculate the center of gravity of the extracted offshore wind turbine spots to obtain M original offshore wind turbine point location data. Step S43: traverse the original offshore wind turbine point data to generate M N×N feature matrix windows. The feature matrix window size calculation formula is: Wherein, GSD2 is the spatial resolution of the offshore wind turbine shadow enhancement image in step S23, and Round represents rounding calculation; Step S44: determine whether the M feature windows intersect with the surface vector boundary of the offshore study area generated in step S13. If so, the offshore wind turbine point corresponding to the feature window is determined as a coastal interference point and is removed; the offshore wind turbine shadow enhancement image in step S23 is cropped using the remaining feature window position and window size, and its value is linearly stretched to 0-255 to obtain a standardized spot image of the offshore wind turbine shadow; Step S45: Use the Roberts operator to perform edge detection on the standardized patch image of the offshore wind turbine shadow; divide and label the offshore wind turbine and offshore booster station sample data set based on the detection results; input the sample data set based on edge detection and its classification label through support vector machine (SVM) model training, and save the trained classification model; use the classification model to discriminate the edge detection results, distinguish the offshore wind turbine from the offshore booster station, and obtain the preliminary processed wind turbine point data; Step S46: Based on the preliminarily processed wind turbine point data, the number of wind turbines in the wind turbine point grid neighborhood is divided by the area of ​​the neighborhood to obtain a point density grid image; wherein, the pixel size of the point density grid is set to 1000 meters, the neighborhood analysis is circular, and the neighborhood radius is 8 pixels; according to the geographic coordinate information in the wind turbine point data, the point density value corresponding to each wind turbine point in the point density grid image is extracted, and those with a point density less than 0.06 are determined as cloud interference points and removed, so as to finally obtain accurate wind turbine point data.

Images