Rapid laver culture area extraction method based on sentinel No.1 radar image

Abstract

The invention belongs to the field of ocean remote sensing information, and particularly relates to a laver culture area rapid extraction method based on sentinel No.1 radar images. Comprising the following steps: 1) acquiring a sentry No.1 radar image, and analyzing and processing pixels in the image to obtain a nori raft frame initial product; 2) pretreating the initial product of the laver raft frame to obtain a continuous raft frame product; 3) extracting three masks by using the auxiliary optical image and confused pixels around the high-reflection area; and 4) performing mask processing on the continuous raft frame product by using an extracted result to obtain a spatial distribution grid data set of the laver culture raft frame. According to the invention, the optical image is prevented from being limited by a cloud layer and a turbid water body, the time sequence is integrated through a maximum value synthesis algorithm, and the complete spatial distribution of the raft frame submerged by the high tide level is captured.

Description

Technical Field

[0001] The present invention belongs to the field of marine remote sensing information, and specifically relates to a method for rapidly extracting laver aquaculture areas based on Sentinel-1 radar images. Background Art

[0002] In recent years, the rapid development of marine pastures has provided a new way to address the global food crisis. Dozens of alternating long sand ridges and grooves are distributed in the western part of the southern Yellow Sea and the northern Jiangsu Shoal. The vast intertidal zone and the sand ridges of the northern Jiangsu sandbars have been used for laver (Pyropia) cultivation, making it an important aquaculture base in China. However, the unregulated expansion of coastal aquaculture has caused a series of environmental problems. Rapid and accurate detection of aquaculture rafts will contribute to the scientific planning and precise management of coastal aquaculture.

[0003] Inherent defects of optical images: Traditional methods are mostly based on Landsat or Sentinel-2 optical images, which are vulnerable to cloud cover, seawater turbidity, and suspended sediment interference. For example, some studies have pointed out that insufficient fusion of optical features will lead to low classification accuracy. Although the visible light images of unmanned aerial vehicles have high resolution, their coverage is limited and they are dependent on weather conditions.

[0004] Inefficient processing of interfering pixels: The reflection characteristics of interfering targets such as wind farms and ships are similar to those of aquaculture rafts, and it is difficult for traditional extraction methods to completely eliminate isolated noises. The literature uses a deep learning algorithm based on an improved U-Net model for aquaculture area extraction, but it requires a large amount of training data and has a high calculation cost.

[0005] Lack of single-temporal information: Existing methods are mostly based on single-temporal images, without considering the influence of the submerged state of rafts (such as high tide) on the reflection signal during the aquaculture cycle, as well as the spectral changes of laver itself during the growth process, resulting in incomplete area extraction.

[0006] Sentinel-1 has the following advantages:

[0007] (1) All-weather and all-time imaging ability

[0008] Equipped with a C-band synthetic aperture radar (SAR), it actively emits microwaves and receives echoes, and is not restricted by clouds, haze, or day-night light.

[0009] Compared with other SAR satellites such as Radarsat and ALOS-2, Sentinel-1 data is freely available and has a higher coverage frequency.

[0010] (2) High revisit cycle and stable observation

[0011] Under the networking of two satellites (Sentinel-1A / 1B), the revisit cycle in the equatorial region is 6 days, and it is shortened to 1-3 days in the polar region, which is suitable for monitoring dynamic changes. Summary of the Invention

[0012] The object of the present invention is to provide a rapid extraction method for laver cultivation areas based on Sentinel-1 radar images, so as to overcome the inherent defects of optical images, the inefficient processing of interfering pixels, and the lack of single-temporal information in existing methods.

[0013] The technical solution adopted by the present invention to achieve the above object is as follows:

[0014] A rapid extraction method for laver cultivation areas based on Sentinel-1 radar images, comprising the following steps:

[0015] 1) Obtain Sentinel-1 radar images, analyze and process the pixels in the images to obtain the initial products of laver rafts;

[0016] 2) Pretreat the initial products of laver rafts to obtain continuous raft products;

[0017] 3) Use auxiliary optical images and confused pixels around high-reflection areas to extract three masks;

[0018] 4) Use the extracted results to perform masking processing on the continuous raft products to obtain a spatial distribution grid dataset of laver cultivation rafts.

[0019] The step 1) includes the following steps:

[0020] 1.1) Use the VH polarization mode data of Sentinel-1 radar and screen out the SAR images in the study area;

[0021] 1.2) Extract all valid observation values from December of the previous year to February of the current year for each pixel in the study area, and generate a single image by selecting the maximum VH value in this time period through pixel-level time series analysis;

[0022] 1.3) Select the area where -20 < VH < -10 in the image as the initial products of cultivation rafts.

[0023] The step 2) is specifically:

[0024] Perform median filtering on the initial products of cultivation rafts to eliminate small-area noises and retain the continuous raft structure to form continuous raft products.

[0025] The step 3) includes the following steps:

[0026] 3.1) For the pixels where VH ≥ -10 in the image, that is, the high-reflection confusion targets, use maximum filtering to extract the surrounding confused pixels;

[0027] 3.2) Obtain Sentinel-2 optical images in the same time period as Sentinel-1, perform cloud removal processing on them, and calculate the normalized difference water index NDWI;

[0028] 3.3) Take the area where NDWI < 0 as the non-water body index and extract the non-water body area;

[0029] 3.4) Extract the artificial shoreline based on the existing land vector boundary layer.

[0030] The calculation of the normalized difference water index NDWI is specifically as follows:

[0031] NDWI = (R B3 - R B8A ) / (R B3 + R B8A )

[0032] where R B3 represents the Sentinel-2 band 3 data, and R B8A represents the Sentinel-2 band 8A data.

[0033] The specific steps of step 4) are as follows:

[0034] Perform three-layer masking filtering on the continuous raft product in sequence. Among them, the first layer of mask is the confused pixels around the high-scattering area, the second layer of mask is the non-water body area, and the third layer of mask is the artificial shoreline.

[0035] A rapid extraction system for laver cultivation areas based on Sentinel-1 radar images, comprising:

[0036] An image acquisition module, configured to acquire Sentinel-1 radar images, analyze and process the pixels in the images, and obtain the initial laver raft product;

[0037] A data smoothing module, configured to smooth the extracted initial raft product, remove the salt-and-pepper noise inside the raft structure, and generate a continuous raft product;

[0038] A mask acquisition module, configured to extract three masks by using auxiliary optical images and confused pixels around the high-reflection area;

[0039] A mask processing module, configured to perform mask processing on the continuous raft product by using the extracted results to obtain the raster dataset of the spatial distribution of the laver cultivation raft.

[0040] The present invention has the following beneficial effects and advantages:

[0041] 1. All-weather anti-interference ability: Based on the VH band data of Sentinel-1 SAR, it avoids the limitations of optical images by clouds and turbid water bodies, and integrates the time series through the maximum value synthesis algorithm (maximizing the VH value from the growth period to the harvest period) to capture the complete spatial distribution of the raft submerged at high tide.

[0042] 2. Triple mask layer denoising:

[0043] The first layer: Eliminate the ship instantaneous signal in the area where the VH value > -10 through maximum value filtering;

[0044] The second layer: Non-water area mask where Sentinel-2 NDWI < 0;

[0045] The third layer: Eliminate the artificial shoreline with high-precision administrative boundaries.

[0046] This combined strategy is more efficient than the single deep network feature fusion in the literature and does not require complex model training.

[0047] 3. Process automation and scalability: Rely on the Google Earth Engine platform to achieve rapid processing of SAR images, which is more suitable for large-scale monitoring than the drone visible light method in the literature, and at the same time avoids the computational burden of multi-source optical feature fusion. Description of the Drawings

[0048] Figure 1 Technical roadmap for extracting aquaculture raft frames in the Yellow Sea;

[0049] Figure 2 Dual-polarization imaging map of Sentinel-1, where (a) is the VV image and (b) is the VH image;

[0050] Figure 3 Comparison analysis chart before and after filtering;

[0051] Among them, the red pixels represent the seaweed aquaculture areas identified during the intermediate processing. The white pixels correspond to the areas where the VH value is higher than -10. The white areas in Figures c and d are wind farms and their shadows. The black background pixels are the areas where the VH value is lower than -20. The black backgrounds in Figures a - d are all sea water areas. The red part in Figure a is the seaweed aquaculture area identified before smoothing, and the red part in Figure b is the result after smoothing. The red pixels in Figure c are the interference areas around the wind farm that cannot be directly removed by threshold filtering. After maximum value filtering in Figure d, such confused pixels are completely eliminated. Figure e is the high-resolution remote sensing image of the wind farm;

[0052] Figure 4 Schematic diagram of the spatial distribution and annual change characteristics of the purple laver aquaculture raft frames in the northern Jiangsu Shoal from 2017 to 2023;

[0053] Among them, Figures a - e are the spatial distribution of the aquaculture raft frames, and Figure f is the annual change characteristics of the aquaculture area. Detailed Implementation Modes

[0054] The following further elaborates on the present invention in conjunction with the drawings and embodiments.

[0055] The main variety of seaweed cultured in the sea area of the northern Jiangsu Shoal is Pyropia yezoensis, and the semi-floating raft culture mode is mainly adopted. The culture raft frame consists of 4 bamboo poles perpendicular to the tidal flat, 2 horizontal bamboo poles, an intermediate net curtain that provides an attachment substrate for the growth of Pyropia yezoensis, and two mooring ropes that connect the bamboo poles and the net curtain. This raft frame structure enables the net curtain to have an appropriate drying time during ebb tide, resulting in a significant difference in its reflection signal from the surrounding sea water. Therefore, the seaweed culture area can be identified through Sentinel-1 radar (SAR) images. The specific extraction process of the present invention is as Figure 1 shown, and specifically includes the following steps:

[0056] 1. Select Sentinel-1 radar images to extract the Pyropia yezoensis culture boundary. The VH polarization mode is selected, and mainly the period from December of the previous year to February of the current year is selected as the screening period for Sentinel-1 SAR images. During this period, the distribution of the culture raft frames is relatively complete, and interference pixels generated by the activities of harvesting boats from March to May can be avoided.

[0057] 2. Synthesize the most complete VH image of the current year by selecting the maximum value of the time series to further extract the Pyropia yezoensis raft frames.

[0058] 3. Perform morphological smoothing processing on the culture raft frames. For the suspected raft frame areas, the median filtering algorithm is used to eliminate small-area noises and retain the continuous raft frame structure ( Figure 2 comparisons of the extraction results before and after filtering in a and b), as the continuous raft frame product.

[0059] 4. Select the area of -20 < VH < -10 as the initial product of the culture raft frames. The pixels with VH ≥ -10 belong to the ships and offshore wind farms in the high-scattering area. There are also some pixels around them with VH values between -20 and -10. Therefore, the extracted pixels with VH ≥ -10 are used to extract the surrounding confused pixels by maximum value filtering (such as Figure 3 the pixels around c)

[0060] 5. Assist the optical images to optimize the extraction results. Combine the Sentinel-2 optical images during the research period (December of the previous year to February of the current year). Through cloud removal processing, use the data of band 3 (560 nm) and band 8A (864 nm) to calculate the normalized difference water index (NDWI, calculation formula attached later), and use the area with NDWI < 0 as the non-water index for subsequent masking processing.

[0061] NDWI = (R B3 - R B8A ) / (R B3 + R B8A )

[0062] 6. Construct an accurate administrative vector boundary layer, mainly for the removal of some artificial shorelines that will be submerged seasonally.

[0063] 7. Finally, the continuous raft product obtained by threshold extraction is subjected to three-layer masking to delete confusing pixels: the first layer is the pixels around the high scattering area, the second layer is the non-water area, and the third layer is the artificial coastline, and finally the spatial distribution raster dataset of laver farming rafts is obtained.

[0064] The growing season of Porphyra yezoensis is from the end of September to April of the following year, and the harvesting and storage period of mature Porphyra yezoensis is from the end of March to May. This study selected December of the previous year to February of the current year as the screening period for Sentinel-1 SAR images. During this period, the distribution of aquaculture rafts was relatively complete, and the interference pixels generated by harvesting boat activities from March to May could be avoided. The application of the maximum value algorithm can effectively incorporate the pixel information of partially submerged rafts at high tide.

[0065] like Figure 2 As shown in Figure 1, the Sentinel-1 Ground Range Detection (GRD) data includes four polarization modes: HH (horizontal transmission / horizontal reception), HV (horizontal transmission / vertical reception), VV (vertical transmission / vertical reception) and VH (vertical transmission / horizontal reception), but the GEE platform only provides VV and VH bands. By comparing VV ( Figure 2 a) and VH( Figure 2 b) The maximum value calculation results of the band show that although both polarization bands can reflect the laver farming area, the VH band can more effectively eliminate the background mudflat interference ( Figure 2 The mudflat features in the VV polarization mode in a are obvious and difficult to identify and remove), so the present invention uses the VH band to construct the data set.

[0066] like Figure 3 As shown in the figure, morphological algorithms are used in image processing to smooth the image and remove interfering pixels: a circular median filter algorithm with a radius of 10 meters is used to smooth the breeding raft area (the smoothing effect is shown in Figure 3 ab), a circular maximum filtering algorithm with a radius of 30 meters is used to eliminate isolated interference pixels around ships and wind power generation facilities that are difficult to filter through the VH threshold (see the elimination effect Figure 3 cd).

[0067] Pixels with VH values ​​between -20 and -10 were initially identified as seaweed farming areas through threshold filtering. To mask interference targets such as ships, wind power generation facilities, land, and submerged artificial shorelines, three mask layers were superimposed: the first layer was composed of pixels with VH values ​​greater than -10 after maximum filtering; the second layer was non-water areas with NDWI values ​​less than 0 calculated based on the Sentinel-2 optical images of the same period; and the third layer was submerged artificial shorelines extracted based on high-precision administrative boundary data. Finally, a raster dataset of the spatial distribution of seaweed farming was generated.

[0068] Figure 4Spatial distribution of Porphyra cultivation rafts in the northern Jiangsu Shoal from 2017 to 2023 obtained using the method of the present invention ( Figure 4 a- Figure 4 e) and interannual variation characteristics of the cultivation area ( Figure 4 f).

Claims

1. A method for rapid extraction of laver cultivation areas based on Sentinel-1 radar images, characterized in that: It includes the following steps: 1) Obtain Sentinel-1 radar images, analyze and process the pixels in the images to obtain the initial products of laver raft frames; 2) Preprocess the initial products of laver raft frames to obtain continuous raft frame products; 3) Extract three masks using auxiliary optical images and confused pixels around high-reflection areas; 4) Use the extracted results to perform mask processing on the continuous raft frame products to obtain the raster dataset of the spatial distribution of laver cultivation raft frames.

2. The method for rapid extraction of laver cultivation areas based on Sentinel-1 radar images according to claim 1, characterized in that: The said step 1) includes the following steps: 1.1) Use the VH polarization mode data of Sentinel-1 radar and screen out the SAR images in the study area; 1.2) Extract all valid observations from December of the previous year to February of the current year for each pixel in the study area, and select the maximum VH value in this period through pixel-level time series analysis to generate a single image; 1.3) Select the area where -20 < VH < -10 in the image as the initial products of cultivation raft frames.

3. The method for rapid extraction of laver cultivation areas based on Sentinel-1 radar images according to claim 1, characterized in that: The said step 2) is specifically: Perform median filtering on the initial products of cultivation raft frames to eliminate small-area noise and retain the continuous raft frame structure to form continuous raft frame products.

4. The method for rapid extraction of laver cultivation areas based on Sentinel-1 radar images according to claim 1, characterized in that: The said step 3) includes the following steps: 3.1) For the pixels where VH ≥ -10 in the image, that is, the high-reflection confusion targets, use maximum filtering to extract the surrounding confused pixels; 3.2) Obtain the Sentinel-2 optical image in the same time period as Sentinel-1, perform cloud removal on it, and calculate the Normalized Difference Water Index (NDWI); 3.3) Take the area where NDWI < 0 as the non-water body index and extract the non-water body area; 3.4) Extract the artificial shoreline based on the existing land vector boundary layer.

5. The method for rapid extraction of laver cultivation areas based on Sentinel-1 radar images according to claim 4, characterized in that: The calculation of the Normalized Difference Water Index (NDWI) is specifically: NDWI(R B3 -R B8A ) / (R B3 +R B8A ) Among them, R B3 Indicates Sentinel-2 Band 3 data, R B8A Indicates Sentinel-2 Band 8A data.

6. The method for rapid extraction of laver cultivation areas based on Sentinel-1 radar images according to claim 1, characterized in that: The said step 4) is specifically: Perform three-layer mask filtering processing on the continuous raft frame products in sequence. Among them, the first layer of mask is the confused pixels around the high-scattering area, the second layer of mask is the non-water body area, and the third layer of mask is the artificial shoreline.

7. A rapid extraction system for laver cultivation areas based on Sentinel-1 radar images, characterized in that: It includes: An image acquisition module, used to obtain Sentinel-1 radar images, analyze and process the pixels in the images to obtain the initial products of laver raft frames; A data smoothing module, used to smooth the extracted initial products of raft frames, remove the salt-and-pepper noise inside the raft frame structure, and generate continuous raft frame products; A mask acquisition module, used to extract three masks using auxiliary optical images and confused pixels around high-reflection areas; A mask processing module, used to perform mask processing on the continuous raft frame products using the extracted results to obtain the raster dataset of the spatial distribution of laver cultivation raft frames.

Images