Seaweed cultivation remote sensing extraction method and system based on relative darkness and timing stability

Abstract

The application discloses a seaweed cultivation remote sensing extraction method and system based on relative darkness and time sequence stability. In view of the problems that the traditional spectral method is easily disturbed by the water environment and the deep learning method depends on a large number of samples, the application provides an unsupervised extraction scheme, and the core steps of the scheme include the following steps: a morphological structure element suitable for the size of a cultivation raft is constructed, the background of the sea surface is inhibited through morphological bottom-hat transformation, the weak'relative dark' signal of the large seaweed is reversed and enhanced into an 'absolute bright' signal with high signal-to-noise ratio, the position of the cultivation raft is fixed, the time sequence median synthesis is further conducted on the enhanced image sequence, the moving noise is automatically filtered out, and the growth cycle signal is completed, and finally, the extraction is realized through adaptive threshold segmentation. The application is free from the dependence on absolute spectral values, has high robustness and calculation efficiency, and is suitable for large-scale business mapping under complex sea conditions.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing and feature extraction technology, specifically to a method for extracting remote sensing ground feature information by combining morphological and spatiotemporal sequence analysis. This invention is particularly suitable for extracting ground features in remote sensing images that exhibit spatially relatively dark characteristics and temporally stable spatial locations, especially large-scale seaweed farming areas. Background Technology

[0002] Large-scale seaweed farming is an important part of the global marine economy. Utilizing remote sensing information technology to conduct large-scale, periodic spatial distribution monitoring of large-scale seaweed farming areas is an urgent need for achieving scientific and efficient management, sustainable industrial development, and scientific research in the industry.

[0003] Currently, existing remote sensing methods for extracting large-scale seaweed farming areas are mainly divided into two categories: one is the traditional method based on "absolute spectral classification", such as supervised classification methods like spectral index thresholding and random forests; the other is the data-driven deep learning method, such as semantic segmentation networks like U-Net.

[0004] However, all of the above-mentioned existing technologies have significant drawbacks in application:

[0005] 1. The problem of "absolute spectrum" failure: The traditional spectral index method is highly dependent on the clarity of the water. When the sea is turbid or there are solar flares, the weak spectral signal of the aquaculture raft is easily drowned out by background noise, resulting in extraction failure.

[0006] 2. High Cost of "Black Box" Models: While deep learning methods offer acceptable accuracy, they require massive amounts of manually labeled datasets, and model training demands extremely high computational power, making rapid migration and generalization across different sea areas difficult. In summary, existing technologies either suffer from poor robustness due to reliance on unstable "absolute spectra" or are prohibitively expensive and difficult to implement due to their dependence on "black box" models and massive amounts of data. Therefore, there is an urgent need to develop a novel, robust, and efficient remote sensing method for extracting large-scale seaweed cultivation areas that breaks free from traditional spectral dependence, requires no training samples, has low computational cost, and is efficient. Summary of the Invention

[0007] The purpose of this invention is to overcome the problems of poor robustness, low signal-to-noise ratio, and difficulty in model transfer caused by the reliance on "absolute spectrum" in existing technologies (such as spectral index method and supervised classification method), as well as the bottlenecks of deep learning methods that rely on massive labeled samples, high computing power cost, and opaque algorithms. The invention provides a remote sensing extraction method and system for seaweed aquaculture based on relative darkness and time-series stability.

[0008] The specific technical solution adopted in this invention is as follows:

[0009] In a first aspect, the present invention provides a remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability, which includes the following steps:

[0010] S1. Obtain multi-temporal multispectral remote sensing image sequences of the target sea area within a complete large-scale seaweed cultivation cycle, and obtain pre-processed image sequences through image preprocessing.

[0011] S2. Based on the prior typical spatial size of a single large seaweed farming raft in the target sea area, define a morphological structural element with a spatial scale of 0.8 to 1.5 times the prior typical spatial size, so that the morphological structural element can basically or completely cover a single large seaweed farming raft in the image space.

[0012] S3. For each scene in the preprocessed image sequence, apply the morphological structuring element to perform morphological undercap transformation, thereby converting each original pixel value in the image into the corresponding spatial relative darkness value of the highlight, and obtaining a single-scene signal enhancement image.

[0013] S4. Perform time-series median synthesis on all single-scene signal enhancement images to obtain a spatiotemporal aggregated image by utilizing the temporal stability of the spatial location of the large seaweed farming area.

[0014] S5. Perform threshold segmentation on the spatiotemporal aggregated image to extract the spatial range of the large seaweed farming area.

[0015] As a preferred embodiment of the first aspect, the spatial resolution of the multi-temporal multispectral remote sensing image sequence is not less than 10 meters, and it includes at least red, green, blue, and near-infrared spectral bands.

[0016] As a preferred embodiment of the first aspect above, the image preprocessing operation includes one or more combinations of geometric correction, radiometric correction, atmospheric calibration, cloud masking, and land masking.

[0017] As a preferred embodiment of the first aspect above, the a priori typical space size is 190 to 200 meters.

[0018] As a preferred embodiment of the first aspect above, the morphological cap transformation specifically includes the following sub-steps:

[0019] S31. Take each image in the preprocessed image sequence as the original image, and apply the morphological structuring element to perform morphological closing operation to obtain an estimated background image; in the morphological closing operation, the original image is first subjected to morphological dilation processing and then morphological erosion processing.

[0020] S32. Subtract the corresponding original image from the estimated background image to obtain the single-scene signal enhancement image corresponding to each scene image.

[0021] As a preferred embodiment of the first aspect above, the threshold segmentation employs the Otsu method, which automatically determines the threshold.

[0022] Secondly, the present invention provides a remote sensing extraction system for seaweed cultivation based on relative darkness and temporal stability, comprising:

[0023] The image preprocessing module is used to acquire multi-temporal multispectral remote sensing image sequences of the target sea area within a complete large-scale seaweed cultivation cycle, and to obtain preprocessed image sequences through image preprocessing.

[0024] The structuring element definition module is used to define a morphological structuring element with a spatial scale of 0.8 to 1.5 times the prior typical spatial size of a single large seaweed farming raft in the target sea area, so that the morphological structuring element can basically or completely cover a single large seaweed farming raft in the image space.

[0025] The signal enhancement module is used to apply the morphological structuring element to perform morphological undercap transformation on each image in the preprocessed image sequence, thereby converting each original pixel value in the image into the corresponding spatial relative darkness value of the highlight, and obtaining a single-scene signal-enhanced image.

[0026] The spatiotemporal aggregation module is used to perform time-series median synthesis of all single-scene signal enhancement images, thereby utilizing the temporal stability of the spatial location of large seaweed farming areas to obtain a spatiotemporal aggregated image.

[0027] The threshold segmentation module is used to perform threshold segmentation on the spatiotemporal aggregated image to extract the spatial range of the large seaweed farming area.

[0028] Thirdly, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can realize the remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability as described in any of the first aspects above.

[0029] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the seaweed aquaculture remote sensing extraction method based on relative darkness and temporal stability as described in any of the first aspects above.

[0030] Fifthly, the present invention provides a computer electronic device, which includes a memory and a processor;

[0031] The memory is used to store computer programs;

[0032] The processor is configured to, when executing the computer program, implement the remote sensing extraction method for seaweed aquaculture based on relative darkness and time-stability as described in any of the first aspects above.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] 1. This invention innovatively introduces the feature of "relative spatial darkness" to achieve complete immunity to water background.

[0035] Existing technologies are mostly based on the "absolute spectral values" of pixels, while this invention utilizes morphological cap transformation to extract the "spatial relative darkness" of pixels. Regardless of whether the seawater background is clear, turbid, or eutrophic, as long as the aquaculture raft exhibits a dark characteristic relative to the surrounding seawater, this relative difference can be accurately captured. This "differential approach" eliminates spectral interference from complex water color environments in principle, and its robustness far exceeds that of traditional spectral indices.

[0036] 2. This invention achieves a cascaded enhancement of the signal-to-noise ratio from a "weak dark signal" to a "strong bright signal".

[0037] Large seaweed typically appears as a weak dark signal in raw images, making it extremely difficult to segment. This invention uses an algorithm to normalize the background to zero and forcibly invert the target signal into a bright signal. This signal inversion and enhancement mechanism increases the contrast between the target and the background by several orders of magnitude, allowing for high-precision segmentation with only a simple Otsu threshold, eliminating the need for a complex classifier.

[0038] 3. This invention integrates a spatiotemporal complementarity mechanism, which balances the integrity of the extraction with noise resistance.

[0039] To address the issue of signal loss in single-scene images due to growth cycles (such as the seedling stage) or instantaneous noise from sea waves, this invention utilizes temporal median synthesis. This step cleverly leverages the physical difference between the stationary spatial location of aquaculture facilities and the instantaneous movement of noise. While eliminating moving noise, it completes the spatial morphology of the target from a temporal dimension, achieving "removing falsehoods and preserving truth."

[0040] 4. As a pure "white-box" algorithm, this invention has extremely high engineering application value.

[0041] Unlike deep learning "black boxes" that rely on a large number of samples for training, this invention is based on clear physical and geometric morphological principles, requiring no manually labeled samples and not relying on GPU computing power. The algorithm has low computational complexity and can be directly deployed on cloud platforms such as Google Earth Engine (GEE) to achieve minute-level automated monitoring of large-scale sea areas. Attached Figure Description

[0042] Figure 1This is a flowchart of the steps in the remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability in this invention.

[0043] Figure 2 This is a schematic diagram of the process for enhancing single-scene signals through morphological bottom-hat transformation in this invention.

[0044] Figure 3 This is a diagram showing the module composition of the seaweed aquaculture remote sensing extraction system based on relative darkness and temporal stability in this invention.

[0045] Figure 4 This is a schematic diagram of the structure of a computer electronic device.

[0046] Figure 5 This is a schematic diagram of the result of single-scene signal enhancement through morphological cap transformation in this invention; wherein, (a) is a partial schematic diagram of an original remote sensing image, showing a large seaweed raft that is “relatively dark” and a seawater background that is “relatively bright”; (b) is a schematic diagram of the “estimated background” image obtained after performing morphological closing operation on (a); (c) is a schematic diagram of the “signal enhancement” image obtained by subtracting (a) from (b) through cap transformation, showing the signal of the large seaweed raft that is “absolutely bright” and the background that is suppressed to zero.

[0047] Figure 6 This is a schematic diagram of the spatiotemporal aggregation denoising result based on median synthesis of time series in this invention; where (a) is a "signal enhancement image" at a certain moment, in which large seaweed is in the vigorous growth period and the signal is relatively strong; (b) is a "signal enhancement image" at another moment, in which the signal of large seaweed rafts is missing or weak due to different growth stages, and there is a lot of instantaneous noise such as waves; (c) is a "spatiotemporal aggregation image" obtained after median synthesis of multiple temporal images such as (a) and (b), showing the ideal result of noise removal and signal completion.

[0048] Figure 7 This is a schematic diagram of the target extraction result by threshold segmentation in this invention; where (a) is a spatiotemporal aggregated image; and (b) is the final binarized map of the large seaweed farming area obtained after applying Otsu threshold segmentation to (a).

[0049] Figure 8 Comparison of remote sensing extraction results of large seaweed cultivation areas with different radii for the structural element. Detailed Implementation

[0050] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. Technical features in various embodiments of the present invention can be combined accordingly without mutual conflict.

[0051] This invention provides a remote sensing extraction method for seaweed aquaculture based on relative darkness and temporal stability. This method departs from the traditional approach of extracting large-scale seaweed aquaculture areas using "absolute spectral classification," and achieves highly robust, high signal-to-noise ratio, and highly computationally efficient automated extraction based on a profound understanding of the physical essence of ground features, namely "spatial relative darkness" and "temporal stability of spatial location." The specific implementation process and technical principles of this method are described in detail below.

[0052] It should be noted that the term "seaweed" in this invention refers to "macro-seaweed." Macro-seaweed is a type of embryoless, multicellular spore-bearing plant visible to the naked eye, mainly composed of filamentous, membranous, or thallus structures, lacking true root, stem, and leaf differentiation, and includes species from the phyla Phaeophyta, Rhodophyta, and Chlorophyta. Representative species of "macro-seaweed" include kelp, laver, and Gracilaria.

[0053] like Figure 1 As shown, in a preferred embodiment of the present invention, the above-mentioned remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability includes the following steps S1 to S5:

[0054] S1. Obtain multi-temporal, multispectral remote sensing image sequences of the target sea area within a complete large-scale seaweed cultivation cycle, and obtain preprocessed image sequences through image preprocessing.

[0055] It should be noted that the aforementioned target sea area refers to the sea area where large-scale seaweed aquaculture areas need to be extracted; the specific location is not limited and depends on actual application requirements. In the multi-temporal multispectral remote sensing image sequence of this invention, the specific type of remote sensing image is not limited. To ensure the accuracy of extraction, the spatial resolution of the remote sensing image is generally required to be no less than 10 meters, and it must contain at least red, green, blue, and near-infrared spectral bands to ensure effective identification of the spatial structure of the aquaculture rafts. The temporal resolution of the remote sensing image in the target sea area should also be as high as possible to meet the requirements of subsequent time series median synthesis.

[0056] In embodiments of the present invention, the Sentinel-2 L2A image set can be used to construct the aforementioned multi-temporal multispectral remote sensing image sequence. This image set has a spatial resolution of 10 meters, has undergone atmospheric correction, and includes red, green, blue, and near-infrared spectral bands.

[0057] It should also be noted that the above image preprocessing operations need to be selected based on the specific remote sensing image data used, and may include one or more combinations of geometric correction, radiometric correction, atmospheric calibration, cloud masking, and land masking. In practical applications, the selection can be flexible according to the image quality, with the goal of obtaining high-quality remote sensing images after processing.

[0058] S2. Based on the prior typical spatial size of a single large seaweed farming raft in the target sea area, define a morphological structural element with a spatial scale of 0.8 to 1.5 times the prior typical spatial size, so that the morphological structural element can basically or completely cover the single large seaweed farming raft in the image space.

[0059] It should be noted that the "prior typical spatial size" in this invention specifically refers to the typical spatial size of a single large seaweed farming raft in the target sea area under a top-down view (i.e., perpendicular to the sea surface), determined through long-term observation, literature review, and field measurement. The statistical distribution range of this size can be determined through sample statistics, and the upper limit of the range or a value close to the upper limit is taken as the "prior typical spatial size." In actual sea areas, the top surface of a single large seaweed farming raft is generally square, and its width does not exceed 200 meters. Therefore, the prior typical spatial size can be set to 190-200 meters. However, since this size is a physical spatial size, and morphological structural elements are mainly used for image manipulation of pixel space, the physical spatial size can be mapped to the pixel spatial size in one step. Assuming the prior typical spatial size is denoted as A, A can be converted into the number of pixels covered based on the remote sensing image resolution. In one embodiment, if the prior typical spatial size A = 200 meters and the remote sensing spatial resolution is 10 meters, then the morphological structural element needs to cover at least 20 pixels in the pixel space. If a circular morphological structuring element is used, considering that calculations can only be performed using a single pixel as the smallest unit in pixel space, the diameter of the circular morphological structuring element should be set to 21 pixels. In this case, the surface space covered in the physical world would be approximately 210 meters, which is sufficient to cover large seaweed farming rafts of 200 meters or less. If a square morphological structuring element is used, it can be a square of 20 pixels × 20 pixels. Furthermore, it is permissible for the spatial scale of the morphological structuring element to be slightly smaller than the prior typical spatial size of a single large seaweed farming raft in the target sea area. Generally speaking, the spatial scale of the morphological structuring element should be 0.8 to 1.5 times the prior typical spatial size, calculated using the diameter of the circle or the side length of the square.

[0060] The spatial scale of the morphological structural elements defined in this invention can basically or completely cover a single large seaweed farming raft in the image space. This is the key to the success of background estimation in this invention. Both excessively large and excessively small sizes will result in poor extraction results.

[0061] S3. For each scene in the preprocessed image sequence, apply the morphological structuring element to perform morphological undercap transformation, thereby converting each original pixel value in the image into a corresponding bright spatial relative darkness value, that is, converting the slight grayscale difference between the large seaweed farming raft and the background in the image into a high-contrast foreground signal, and obtaining a single-scene signal enhancement image.

[0062] It should be noted that the morphological undercap transformation described above in this invention is intended to convert the original pixel values ​​into corresponding spatial relative darkness, essentially calculating a normalized "relative darkness index." This index is completely immune to drastic spectral changes in seawater backgrounds (whether clear, turbid, or eutrophic), exhibiting extremely high robustness and can be used to accurately identify large seaweed farming rafts in the ocean. For example... Figure 2 As shown, in an embodiment of the present invention, the above-mentioned morphological bottom-hat transformation can be implemented through the following sub-steps:

[0063] S31. Using each image in the preprocessed image sequence as the original image, perform a morphological closing operation using the morphological structuring element to obtain an estimated background image. The "morphological closing operation" can consist of two consecutive operations: (a) First, apply the morphological dilation processing to the original image using the morphological structuring element; (b) Then, apply the same morphological erosion processing to the dilated image obtained in step (a). The physical meaning of the background image obtained in this step is an estimated smooth, "pure seawater background" image without large seaweed rafts.

[0064] S32. For each image in the preprocessed image sequence, the estimated background image is subtracted from the corresponding original image (i.e., background - original image) to obtain the single-scene signal enhancement image corresponding to that image. The aforementioned "spatial relative darkness" is quantified in this single-scene signal enhancement image, and subsequent time series median synthesis and threshold segmentation can be performed based on this image.

[0065] S4. Perform time-series median synthesis on all single-scene enhanced images to obtain a spatiotemporal aggregated image by utilizing the temporal stability of the spatial location of the large seaweed farming area.

[0066] It should be noted that "time series median synthesis" is an existing data compression and denoising method for processing multi-temporal remote sensing images. This method calculates the median at the pixel level for all available images within a specified time window for the same area (the median value of each pixel is taken as the value of that pixel in the composite image after sorting the observation values ​​at all time points), ultimately generating a new spatiotemporal aggregated image that represents the typical state of land cover during that time period. The purpose of introducing time series median synthesis in this step is to utilize the physical characteristic of "spatial location temporally stable" of aquaculture rafts to achieve two functions: first, to filter out instantaneous moving noise such as ships, thin clouds, solar flares, and ocean waves; and second, to compensate for the loss of single-scene signals caused by different aquaculture stages (such as the timing of seedling placement and the speed of growth in large seaweed aquaculture), thereby aggregating the most complete and cleanest aquaculture area.

[0067] S5. Perform threshold segmentation on the spatiotemporal aggregated image to extract the spatial range of the large seaweed farming area.

[0068] It should be noted that, theoretically, any image segmentation method with good segmentation effect can be used for the above threshold segmentation. In the embodiments of the present invention, the Otsu method, which can be automatically implemented, is preferred. It can automatically find the optimal segmentation threshold based on the histogram of the aggregated image, thereby accurately extracting the spatial range of the large seaweed farming area, that is, the outer boundary of each large seaweed farming raft in the large seaweed farming area.

[0069] It should be noted that the remote sensing extraction method steps for seaweed aquaculture based on relative darkness and temporal stability shown in S1 to S5 above can essentially be implemented in the form of computer programs or software functional modules.

[0070] Therefore, based on the same inventive concept, such as Figure 3 As shown, the present invention also provides a remote sensing extraction system for seaweed cultivation based on relative darkness and time-stability, corresponding to the remote sensing extraction method for seaweed cultivation based on relative darkness and time-stability provided in the above embodiments, which includes:

[0071] The image preprocessing module is used to acquire multi-temporal multispectral remote sensing image sequences of the target sea area within a complete large-scale seaweed cultivation cycle, and to obtain preprocessed image sequences through image preprocessing.

[0072] The structuring element definition module is used to define a morphological structuring element with a spatial scale of 0.8 to 1.5 times the prior typical spatial size of a single large seaweed farming raft in the target sea area, so that the morphological structuring element can basically or completely cover a single large seaweed farming raft in the image space.

[0073] The signal enhancement module is used to apply the morphological structuring element to perform morphological undercap transformation on each image in the preprocessed image sequence, thereby converting each original pixel value in the image into the corresponding spatial relative darkness value of the highlight, and obtaining a single-scene signal-enhanced image.

[0074] The spatiotemporal aggregation module is used to perform time-series median synthesis of all single-scene signal enhancement images, thereby utilizing the temporal stability of the spatial location of large seaweed farming areas to obtain a spatiotemporal aggregated image.

[0075] The threshold segmentation module is used to perform threshold segmentation on the spatiotemporal aggregated image to extract the spatial range of the large seaweed farming area.

[0076] Furthermore, based on the same inventive concept, such as Figure 4 As shown, the present invention also provides a computer electronic device corresponding to the remote sensing extraction method for seaweed cultivation based on relative darkness and time-stability provided in the above embodiments, which includes a memory and a processor;

[0077] The memory is used to store computer programs;

[0078] The processor is configured to, when executing the computer program, implement the remote sensing extraction method for seaweed aquaculture based on relative darkness and time-stability as described above.

[0079] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0080] Therefore, based on the same inventive concept, this invention provides a computer-readable storage medium corresponding to the seaweed aquaculture remote sensing extraction method based on relative darkness and time-stability. The storage medium stores a computer program, which, when executed by a processor, can realize the seaweed aquaculture remote sensing extraction method based on relative darkness and time-stability as described above.

[0081] Therefore, based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can realize the remote sensing extraction method for seaweed cultivation based on relative darkness and time-series stability as described above.

[0082] Specifically, in the computer-readable storage medium of the above three embodiments, the stored computer program is executed by a processor, which can perform the aforementioned steps S1 to S5.

[0083] It is understood that the aforementioned storage media may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Furthermore, the storage media may also be various media capable of storing program code, such as USB flash drives, external hard drives, magnetic disks, or optical discs.

[0084] It is understood that the processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0085] It should also be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. In the embodiments provided in this application, the division of steps or modules in the system and method is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules or steps may be combined or integrated together, and a module or step may also be split.

[0086] The present invention will further demonstrate the detailed implementation process and technical effects of the seaweed aquaculture remote sensing extraction method based on relative darkness and temporal stability shown in steps S1 to S5 on a specific dataset through a specific embodiment, so as to facilitate understanding of the essence of the present invention.

[0087] Example

[0088] This embodiment extracts large-scale seaweed farming areas in the coastal waters of China based on Sentinel-2 imagery. This extraction method is implemented on the Google Earth Engine (GEE) cloud platform. A complete demonstration follows. Figure 1The specific implementation of each step in the technical process shown is as follows: Steps one to five correspond to steps S1 to S5 above, respectively. The specific technical framework and principles will not be completely repeated. The main focus is on the specific data source, some specific parameter settings and the results of each step in this embodiment.

[0089] Step 1: Data Acquisition and Preprocessing (corresponding to S1)

[0090] (1) Data source: Sentinel-2 L2A image set on the GEE platform was selected. The image set has a spatial resolution of 10 meters, has been atmospherically corrected, and includes red, green, blue, and near-infrared spectral bands.

[0091] (2) Timing selection: Based on the cultivation pattern of large seaweed in the target sea area, select all available images within a complete cultivation cycle (in this embodiment, based on the cultivation pattern of laver in Lianyungang, Jiangsu, from October 1, 2024 to May 1, 2025) to form a multi-temporal multispectral remote sensing image sequence.

[0092] (3) Preprocessing: Using the QA band (Quality Assessment) built into the Sentinel-2 image, the multi-temporal multispectral remote sensing image sequence is masked for clouds, cirrus clouds and cloud shadows to obtain a cloudless image sequence (or the cloud-covered area is set to null), i.e., a cloudless image sequence.

[0093] Step 2: Definition of morphological structural elements (corresponding to S2)

[0094] (1) Prior knowledge: Through literature review and field measurement, it was statistically determined that the typical width of a raft aquaculture unit (single raft) in China's coastal waters (the width of the square top surface of the part floating on the sea surface in the top view) is generally no more than 200 meters. Therefore, the a priori typical spatial dimension of a single large seaweed aquaculture raft is set at 200 meters.

[0095] (2) Definition of Morphological Structural Element: For a Sentinel-2 image with a resolution of 10 meters, a width of 200 meters corresponds to 20 pixels. Therefore, to ensure that the morphological structural element can always "cross" the aquaculture raft and "step" on the pure seawater background around it when sliding, a structural element larger than 20 pixels must be defined. In this embodiment, a circular structural element with a radius r of 10 pixels is defined (its diameter is 21 pixels, and the surface coverage width is 210 meters). The selection of this structural element scale can effectively fit the background sea surface features and is a key guarantee for the accuracy of background suppression in the subsequent step three.

[0096] Of course, in some computing platforms of other embodiments, an equivalent 21×21 pixel square structural element may also be used.

[0097] Step 3: Signal enhancement based on morphological transformation (corresponding to S3)

[0098] For each image in the cloudless image sequence obtained after the first preprocessing step, the process is performed one by one. The principle of this step is as follows: Figure 2 As shown. Figure 5 Image (a) shows a pristine scene from a cloudless image sequence, in which large seaweed farming rafts appear “relatively dark” than the seawater.

[0099] In this step, for any original image in the cloudless image sequence, the specific operation is as follows:

[0100] (1) Morphological closing operation:

[0101] (a) Dilation: First, the original image is filtered by the maximum value using the structuring element (radius 10 pixels) defined in step two. Physically, this means that as the window slides across, the dark pixels of the large seaweed raft within the window will be "submerged" and "filled" by the bright pixels of the surrounding seawater.

[0102] (b) Erosion: Next, the structuring element is applied to perform minimum filtering on the "dilation image" obtained in step (a). The physical meaning of this is to smooth out some small bright spot noise (such as small boats) that may have been generated during the previous dilation process, ensuring that the generated background is smooth and continuous.

[0103] The combination of (a) and (b) above constitutes the "morphological closing operation," and the final result is a pure seawater background image. Figure 5 Taking (a) as an example, the result after morphological closing operation is as follows: Figure 5 As shown in (b), a perfectly restored raster of "imaginary pure seawater background" was obtained.

[0104] (2) Background subtraction (bottom hat transformation):

[0105] For any original image in a cloudless image sequence, the corresponding original image is subtracted from the "pure seawater background image" obtained through morphological closing operations. This step calculates the spatial relative darkness of each pixel, thereby generating a single-scene signal-enhanced image. The specific principle is as follows: a sea surface background fitting image is constructed using morphological closing operations, and then the corresponding original image is subtracted from it. For ordinary seawater areas, the background fitting value is basically consistent with the original observation value; after subtraction, the pixel value approaches 0, and the background is suppressed. However, for large seaweed farming rafts exhibiting dark characteristics, their original pixel values ​​are significantly lower than the background fitting value; after subtraction, they are transformed into significant positive values ​​(high-brightness signals). Through this difference operation, the 'absolute grayscale' in the original image is successfully converted into spatial relative darkness reflecting the difference between the target and the background.

[0106] In this embodiment, we continue to use Figure 5 Taking image (a) as an example, Figure 5 Image (b) minus Figure 5 Image (a) in the image ultimately yields the following result: Figure 5 The single-view signal-enhanced image shown in (c) shows how the large seaweed farming rafts, which were originally “relatively dark”, are inverted and enhanced into a high-contrast “absolutely bright” signal.

[0107] Step 4: Spatiotemporal Convergence and Noise Suppression (corresponding to S4)

[0108] All the single-scene signal-enhanced images produced in step three were subjected to time-series median composite analysis on the GEE platform. Some results are shown below. Figure 6 As shown. This step is based on the prior knowledge that the aquaculture rafts are fixed in location and stable in time-series images. It can play a dual role in removing instantaneous noise and compensating for signal loss, resulting in a spatiotemporal aggregated image with a very high signal-to-noise ratio, complete signal, and clean background, as shown. Figure 6 As shown in (c).

[0109] like Figure 6 As shown in (b), occasional bright noise (red) in the image only appears in single-scene images and is automatically removed as an "extreme outlier" when taking the median; while weak signals in some images (such as seedlings in the early stages or those with poor growth) are compensated by strong signals in other images, ultimately resulting in a spatiotemporal aggregated image with complete and clean signals, such as... Figure 6 As shown in (c).

[0110] Step 5: Extraction from large-scale seaweed cultivation areas (corresponding to S5)

[0111] The spatiotemporal aggregated imagery obtained in step four is used to directly generate a high-precision binarized aquaculture area distribution map by applying the Otsu method segmentation algorithm, which can automatically set thresholds. This raster data can be further converted into vector data for subsequent area statistics and spatial analysis. Since the specific implementation of the Otsu algorithm is existing technology, it will not be described in detail here. Figure 7 The image comparison before and after thresholding is shown, where (a) is a spatiotemporal aggregated image. Figure 7 Image (a) in the image corresponds to Figure 6 Image (c) Figure 7 (b) in the image is the final binarized map of the large seaweed farming area obtained after applying Otsu threshold segmentation to the spatiotemporal aggregated image.

[0112] Furthermore, this embodiment further includes an experiment comparing the effects of different sized structuring elements on the extraction results. In the experiment, the radius *r* of the morphological structuring element in step two was adjusted, with comparisons set at *r* = 30, 20, 15, 10, 8, and 5 pixels respectively. The remaining steps were implemented as described in steps one through five. The results show that there is a significant scale effect in the selection of structuring elements.

[0113] When the scale is too large (r > 15 pixels), the local background fitting is too smooth and cannot effectively remove the spectral heterogeneity caused by the complex sea environment, resulting in a large amount of residual background noise in the results.

[0114] When the scale is too small, i.e. smaller than the target feature size (Radius < 8 pixels), although the background is completely suppressed, some small aquaculture patches with weak signals or fragmented patterns (as shown in the red dashed box in the figure) suffer excessive erosion, resulting in the loss of target information.

[0115] Overall, extraction results were better when r was 15, 10, or 8 pixels. In this embodiment, r=10 pixels (20 pixels in diameter for a circular structuring element) was ultimately selected as the optimal parameter to preserve the spatial details of the aquaculture area to the greatest extent while ensuring background purity. To avoid loss of generality, it is recommended that the spatial scale of the morphological structuring element be set to 0.8 to 1.5 times the a priori typical spatial size of a single large seaweed aquaculture raft, which yields better extraction results.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A remote sensing extraction method for seaweed aquaculture based on relative darkness and temporal stability, characterized in that, Includes the following steps: S1. Obtain multi-temporal multispectral remote sensing image sequences of the target sea area within a complete large-scale seaweed cultivation cycle, and obtain pre-processed image sequences through image preprocessing. S2. Determine the distribution range of the spatial size statistics of a single large seaweed farming raft in the target sea area under the top view by sample statistics. Take the upper limit of the range or the spatial size value close to the upper limit of the range as the prior typical spatial size. Based on the prior typical spatial size of a single large seaweed farming raft in the target sea area, define a morphological structural element with a spatial scale of 0.8 to 1.5 times the prior typical spatial size, so that the morphological structural element can basically cover or completely cover a single large seaweed farming raft in the image space. S3. For each scene in the preprocessed image sequence, apply the morphological structuring element to perform morphological undercap transformation, thereby converting each original pixel value in the image into the corresponding spatial relative darkness value of the highlight, and obtaining a single-scene signal enhancement image. S4. Perform time-series median synthesis on all single-scene signal enhancement images to obtain a spatiotemporal aggregated image by utilizing the temporal stability of the spatial location of the large seaweed farming area. S5. Perform threshold segmentation on the spatiotemporal aggregated image to extract the spatial range of the large seaweed farming area.

2. The remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability according to claim 1, characterized in that, The spatial resolution of the multi-temporal multispectral remote sensing image sequence is not less than 10 meters, and it includes at least red, green, blue, and near-infrared spectral bands.

3. The remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability according to claim 1, characterized in that, The image preprocessing includes one or more combinations of geometric correction, radiometric correction, atmospheric calibration, cloud masking, and land masking.

4. The remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability according to claim 1, characterized in that, The a priori typical space size is 190~200 meters.

5. The remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability according to claim 1, characterized in that, The morphological cap transformation specifically includes the following sub-steps: S31. Take each image in the preprocessed image sequence as the original image, and apply the morphological structuring element to perform morphological closing operation to obtain an estimated background image; in the morphological closing operation, the original image is first subjected to morphological dilation processing and then morphological erosion processing. S32. Subtract the corresponding original image from the estimated background image to obtain the single-scene signal enhancement image corresponding to each scene image.

6. The remote sensing extraction method for seaweed cultivation based on relative darkness and temporal stability according to claim 1, characterized in that, The threshold segmentation uses the Otsu method, which automatically determines the threshold.

7. A remote sensing extraction system for seaweed cultivation based on relative darkness and temporal stability, characterized in that, include: The image preprocessing module is used to acquire multi-temporal multispectral remote sensing image sequences of the target sea area within a complete seaweed cultivation cycle, and to obtain preprocessed image sequences through image preprocessing. The structuring element definition module is used to determine the distribution range of the spatial size statistics of a single large seaweed farming raft in the target sea area under the top view through sample statistics. The spatial size value at or close to the upper limit of the range is taken as the prior typical spatial size. Based on the prior typical spatial size of a single large seaweed farming raft in the target sea area, a morphological structuring element with a spatial scale of 0.8 to 1.5 times the prior typical spatial size is defined, so that the morphological structuring element can basically cover or completely cover the single large seaweed farming raft in the image space. The signal enhancement module is used to apply the morphological structuring element to perform morphological undercap transformation on each image in the preprocessed image sequence, thereby converting each original pixel value in the image into the corresponding spatial relative darkness value of the highlight, and obtaining a single-scene signal-enhanced image. The spatiotemporal aggregation module is used to perform time-series median synthesis of all single-scene signal enhancement images, thereby utilizing the temporal stability of the spatial location of large seaweed farming areas to obtain a spatiotemporal aggregated image. The threshold segmentation module is used to perform threshold segmentation on the spatiotemporal aggregated image to extract the spatial range of the large seaweed farming area.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it can realize the remote sensing extraction method for seaweed cultivation based on relative darkness and time-series stability as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the remote sensing extraction method for seaweed aquaculture based on relative darkness and temporal stability as described in any one of claims 1 to 6.

10. A computer electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to, when executing the computer program, implement the remote sensing extraction method for seaweed aquaculture based on relative darkness and temporal stability as described in any one of claims 1 to 6.

Images