A method for identifying and classifying copper algae and green tide of enteromorpha based on characteristic spectrum

Abstract

The present application belongs to the field of marine remote sensing information, and specifically relates to a copper algae golden tide and green tide remote sensing classification and identification method based on characteristic spectrum. The method comprises the following steps: 1) obtaining a Sentinel-2 remote sensing image of a set sea area, and pre-processing the remote sensing image; 2) calculating the normalized vegetation index NDVI of the pre-processed remote sensing image, and extracting a floating macroalgae image according to the calculation result; 3) using a triple mask method to eliminate non-water body interference factors of the floating macroalgae image, and obtaining a macroalgae bloom area; and 4) classifying and identifying the copper algae golden tide and green tide based on the macroalgae bloom area. The present application proposes an improved spectral ratio gradient SRG-1 algorithm to enhance the spectral differentiation ability of the two kinds of algae, and aims to establish a macroalgae bloom classification and extraction method considering the accuracy and efficiency, and to provide technical support for algae bloom disaster monitoring and prevention.

Description

Technical Field

[0001] This invention belongs to the field of marine remote sensing information, specifically a remote sensing classification and identification method for copper algae golden tides and seaweed green tides based on characteristic spectra. Background Technology

[0002] Existing single algal bloom identification methods cannot meet the needs of accurate monitoring. Due to the spectral characteristics of Ulva prolifera and Coleus thunbergii (e.g., Coleus thunbergii has a characteristic absorption valley at 645 nm, while Ulva prolifera has a more significant reflection peak at 705 nm), it is expected that an exponential model integrating spectral characteristics can be used to achieve automated differentiation between the two types of algal blooms, providing technical support for the monitoring and early warning of large algal blooms.

[0003] To address the issue of distinguishing between the green tides of Ulva prolifera and the golden tides of Algae, existing methods based on the green and red light reflectance slope index (SRG, as shown in Equation 1) can classify and identify the two types of large algal blooms.

[0004]

[0005] Among them, R rc_green and R rc_red These represent the Rayleigh-corrected reflectance in the green (560nm) and red (665nm) light bands, respectively. To reduce the influence of seawater background, the Rrc value used in the formula needs to be subtracted from the Rrc value of the adjacent seawater. When the SRG value is positive, it indicates that the target is *Ulva prolifera*; when the SRG value is negative, it indicates that the target is *Ulva prolifera*.

[0006] However, the SRG method still has several limitations in practical applications. First, the spatial heterogeneity of seawater pixel reflectance is low, and the global mean calculation of background values ​​in the SRG index weakens the regional difference characteristics. Second, while dynamically calculating local seawater background values ​​can improve accuracy, it increases computational complexity, and the background values ​​in the formula are easily partially canceled out during subtraction. More importantly, the commonly used Sentinel-2 Level-2A (L2A) atmospheric correction product uses the Sen2Cor atmospheric correction algorithm, which is based on the atmospheric radiative transfer model developed by the LibRadtran laboratory. This algorithm is mainly designed for land surface reflectance assumptions (such as dark vegetation or targets with known reflectance) and does not consider the atmospheric correction needs of water areas, resulting in systemic defects in its application in aquatic environments. Summary of the Invention

[0007] Based on measured spectral data of *Ulva prolifera* and *Gnaphalium affine*, this invention combines Sentinel-2 MSI band characteristics with L1C apparent reflectance data to propose an improved spectral ratio gradient index-1 (SRG-1) algorithm to enhance the ability to distinguish the spectra of the two algae. The aim is to establish a method for classifying and extracting large algal blooms that balances accuracy and efficiency, providing technical support for the monitoring and control of algal bloom disasters.

[0008] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0009] A remote sensing classification and identification method for copper algae golden tides and seaweed green tides based on characteristic spectra includes the following steps:

[0010] 1) Acquire Sentinel-2 remote sensing images of the designated sea area and preprocess them;

[0011] 2) Calculate the Normalized Difference Vegetation Index (NDVI) of the preprocessed remote sensing images, and extract images of floating large algae based on the calculation results;

[0012] 3) Use a triple masking method to eliminate non-water body interference factors in the image of floating large algae to obtain the algal bloom region;

[0013] 4) Classify and identify the golden tide of copper algae and the green tide of seaweed based on large algal bloom regions.

[0014] Step 1) involves preprocessing the remote sensing image, specifically as follows:

[0015] Cloud and cirrus pixels on an image are removed using a publicly available cloud removal dataset mask.

[0016] Step 2) specifically refers to:

[0017] Calculate the NDVI of the remote sensing image, and consider the image composed of pixels with NDVI ≥ 0.2 as floating large algae, wherein the NDVI is specifically:

[0018]

[0019] Among them, R NIR For near-infrared reflectivity, Sentinel-2 selects infrared bands 7 or 8, R RED For red band reflectivity, the corresponding Sentinel-2 selects band 4.

[0020] Step 3) specifically refers to:

[0021] The Normalized Differential Water Index (NDWI) was calculated based on annual Sentinel-2 images, and dynamic water area masking was performed using an NDWI greater than or equal to 0.2 as a threshold.

[0022] Remove land boundaries based on known coastline vector data;

[0023] Based on a global seabed topography database, seabed topography masking is performed using water depth thresholds.

[0024] The Normalized Differential Water Index (NDWI) is specifically:

[0025] NDWI = (R GREEN -R NIR ) / (R GREEN +R NIR )

[0026] Among them, R GREEN For green band reflectivity, Sentinel-2 selects band 3, R NIR For near-infrared reflectivity, select 8 bands for Sentinel 2.

[0027] Step 4) specifically involves:

[0028] Based on large algal bloom regions, the minimum SRG-1 index for each time period of each month is calculated. Algal bloom regions with SRG-1 ≥ -0.28 are designated as copper algae gold tides, and the rest are designated as green algae tides.

[0029] The SRG-1 index is specifically:

[0030]

[0031] Among them, R RED The red band is the reflectivity of the top layer of the atmosphere, corresponding to the 4th band of Sentinel-2, R. GREEN The reflectance of the top layer of the atmosphere in the green band corresponds to the third band of Sentinel-2, λ. RED λ is the center wavelength of the red band. GREEN This is the center wavelength of the green band.

[0032] A remote sensing classification and identification system for copper algae golden tides and seaweed green tides based on characteristic spectra includes:

[0033] The image preprocessing module is used to acquire Sentinel-2 remote sensing images of a designated sea area and preprocess them.

[0034] The floating macroalgae image extraction module is used to calculate the Normalized Difference Vegetation Index (NDVI) of the preprocessed remote sensing image and extract images of floating macroalgae based on the calculation results.

[0035] The triple masking module is used to remove non-water body interference factors from images of floating large algae using the triple masking method, so as to obtain the large algae bloom area.

[0036] The classification and identification module is used to classify and identify the golden tide of copper algae and the green tide of seaweed based on large algal bloom regions.

[0037] A remote sensing classification and identification device for copper algae golden tides and seaweed green tides based on characteristic spectra includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the remote sensing classification and identification method for copper algae golden tides and seaweed green tides based on characteristic spectra when the computer program is executed.

[0038] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements a remote sensing classification and identification method for copper algae golden tides and seaweed green tides based on characteristic spectra.

[0039] The present invention has the following beneficial effects and advantages:

[0040] 1. High accuracy in classifying and identifying green tides and golden tides: This invention randomly generates detection points in different regions across six time phases in the South Yellow Sea and uses a confusion matrix to quantify the accuracy of identifying green tides, golden tides, and non-algal bloom areas using the established method and index. The overall classification and identification performance of this method for the two types of large algal blooms is good, with high consistency (overall accuracy: 83%, Kappa coefficient: 68%). There is some false negative in the identification of green tides (recall rate: 51%), mainly due to the NDVI > 0.2 threshold excluding some low-density green tide feature pixels; however, its identification accuracy remains high (precision rate: 79%), indicating a low false positive rate. In contrast, the recall rate (>80%) and precision rate (>80%) for golden tide identification are both excellent, indicating that this method has better overall performance in extracting golden tides.

[0041] 2. High-precision classification and identification of green and golden algal blooms: The SRG-1 index (spectral ratio gradient index-1) proposed in this invention significantly improves the classification accuracy of golden algal blooms and green algal blooms by optimizing the reflectance gradient calculation of the red band (Sentinel 2 band 4, 665nm) and the green band (Sentinel 2 band 3, 560nm). Confusion matrix analysis based on 902 validation points shows that the method achieves an overall accuracy of 83% and a Kappa coefficient of 68% for classifying the two types of algal blooms, which is significantly superior to traditional methods. Particularly noteworthy is the outstanding performance of this method in identifying golden algal blooms, with both recall and precision exceeding 80%, fully demonstrating the algorithm's superiority in identifying specific algal bloom types.

[0042] 3. SRG-1 Algorithm Optimization Significantly Improves Computational Efficiency: Existing SRG indices rely on Rayleigh-corrected reflectance, but the atmospheric correction models used in commonly used atmospheric reflectance products (e.g., Sentinel-2L2A data) are not optimized for water bodies, leading to systematic errors. This invention improves algorithm stability and significantly simplifies data processing by directly utilizing atmospheric surface reflectance for gradient calculation. This improvement eliminates the need for complex calculations of local seawater background values, greatly reducing computational complexity while maintaining accuracy, resulting in a significant efficiency improvement.

[0043] 4. Highly efficient operational processing capabilities: This invention efficiently removes background noise through triple masking, and combines normalized difference water index (NDWI≥0.2), coastline vector data, and seabed topographic depth threshold (depth<-6m) to effectively eliminate interference from clouds, land, shallow sea vegetation, and ships, avoiding interference from coastal zone and island surrounding vegetation; the entire process has been automated based on a large-scale remote sensing cloud computing platform, and can complete the preprocessing and classification of all data in the Yellow and East China Seas within 1 hour, meeting the needs of large-scale near real-time monitoring. Attached Figure Description

[0044] Figure 1 Flowchart of the method of this invention;

[0045] Figure 2 Comparison of spectral measurement curves of *Ulva prolifera* and *Codonopsis pilosula*;

[0046] Figure 3 Comparative analysis of SRG-1 and NDVI features of green tides and gold tides in the Yellow and East China Seas in 2021 and 2022;

[0047] Figure 4 Image showing the extraction results of green tides of *Ulva prolifera* and golden tides of *Alternaria alternifolia* in the Yellow and East China Seas from May 26 to 30, 2020. Detailed Implementation

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

[0049] like Figure 1 As shown, a remote sensing classification and identification method for copper algae golden tides and seaweed green tides based on characteristic spectra includes the following steps:

[0050] (1) Extraction of floating large algae. Remote sensing images of the Yellow Sea East China Sea were obtained from the Sentinel-2 Multispectral Imager Level-1C (L1C), and cloud and cirrus cloud pixels were removed from the images by combining the cloud removal dataset (Cloud Score+S2_HARMONIZED V1) provided by the Google Earth Engine platform.

[0051] (2) Floating large algae are extracted by calculating NDVI and setting a threshold range of NDVI≥0.2.

[0052] To calculate the spatial distribution dataset of large algal blooms over a long period of time, the NDVI index was constructed by calculating the NDVI index (as shown in Equation 2). The near-infrared (NIR) band used Sentinel-2 band 7 (782.5 nm) and the red (RED) band used Sentinel-2 band 4 (664.5 nm).

[0053] The Normalized Difference Vegetation Index (NDVI) was used to extract large algal blooms, and the calculation formula is shown in Equation (2).

[0054]

[0055] NDVI is related to the photosynthetic capacity and energy absorption capacity of vegetation canopies, and this algorithm has been widely used in studies on terrestrial vegetation cover and land use. Many scholars have used NDVI to extract harmful algal blooms with good results.

[0056] (3) Elimination of non-aquatic interference factors. A triple masking method is adopted. First, the normalized difference water index (NDWI) is calculated based on the annual Sentinel-2 image, and the calculation formula is shown in Equation (3). The NDWI is calculated using data from band 3 (560nm) and band 8A (864nm), and dynamic water area masking is performed with an NDWI greater than or equal to 0.2 as the threshold. Second, high-precision coastline vector data is integrated to remove land boundaries. Third, based on the GEBCO global seabed topography database, seabed topography masking is performed through a water depth threshold (depth < -6m) to remove interference from coastal vegetation near islands.

[0057] NDWI = (R B3 -R B8A ) / (R B3 +R B8A (3)

[0058] (4) Classification and identification of copper algae golden tide and seaweed green tide. Based on the large algal bloom region (pixel set) extracted in the previous step, calculate the minimum SRG-1 value for each time item in each month. Those that meet the threshold of SRG-1≥-0.28 are judged as copper algae golden tide, and those that meet the threshold of SRG-1<-0.28 are judged as seaweed green tide.

[0059] Based on the SRG index proposed by Min et al. (2019), this study proposes the SRG-1 index for distinguishing between green tides and golden tides (as shown in Equation 4). This method is based on the Sentinel-2MSI Level-1C dataset without Rayleigh correction, and the calculation method is as follows:

[0060]

[0061] In the above formula, R B4 and R B3 These are the TOA reflectivities of Sentinel-2 in the 665nm and 560nm bands, respectively.

[0062] Finally, by using multiple masks, a spatial distribution dataset of copper algae golden tides was obtained, after removing noise information such as land, islands, artificial coastlines, and ships.

[0063] Figure 2 To compare the spectral measurement curves of *Ulva prolifera* and *Ulva thunbergii* (orange: *Ulva thunbergii*, green: *Ulva prolifera*), on-site spectral measurements were performed on *Ulva thunbergii* and *Ulva prolifera* at different fresh weights, and the resulting spectral spectra are shown below. Figure 2 As shown in the figure, the horizontal axis represents spectral band information, and the vertical axis represents spectral reflectance. By comparing the spectral characteristics of the two algal blooms, it can be seen that the average reflectance of *Ulva prolifera* at the 560nm green band is 0.016, which is higher than that of *Alternaria alternata* (0.012); while at the 660nm red band, the average reflectance of *Ulva prolifera* (0.007) is significantly lower than that of *Alternaria alternata* (0.011). The inverse difference in reflectance between the red and green bands leads to a significant difference in the slope of their spectral curves, providing a basis for distinguishing *Ulva prolifera* from *Alternaria alternata* based on multispectral remote sensing data.

[0064] Figure 3 This paper presents the comparative analysis results of SRG-1 and NDVI features of green tides and golden tides in the Yellow and East China Seas in 2021 and 2022. To evaluate the adaptability of the SRG-1 algorithm and its threshold, this study selected green tides and golden tides in the Yellow and East China Seas in 2021 and 2022 as validation objects. Based on Sentinel-2L2A data, four validation scenarios were constructed (Golden tide area in the southern Yellow Sea on May 1, 2021; Green tide area in the southern Yellow Sea on June 1, 2021; Golden tide area in the East China Sea on April 16, 2022; and Green tide area in the southern Yellow Sea on June 21, 2022). After NDVI threshold segmentation (0.2 as the critical value), the target pixels were classified and identified, and the frequency distribution characteristics of SRG-1 indices for different categories were statistically analyzed. The results are shown in Figure 3. It can be seen that in the high-density algal bloom region with NDVI≥0.2, the SRG-1 index discrimination threshold of -0.28 can effectively distinguish between the green tide of Ulva prolifera (light green) and the golden tide of Copper algae (red); however, in the low-value range of NDVI<0.2, the SRG-1 indices of the two types of algal blooms show a significant overlap, indicating that the algorithm has limited ability to distinguish low-density algal patches.

[0065] Figure 4 This paper presents the extraction results of *Ulva prolifera* green tide and *Alternaria alternata* golden tide in the Yellow and East China Seas from May 26 to 30, 2020, based on the construction method of this invention. Satellite remote sensing data acquired in parts of the Yellow and East China Seas from May 26 to 30, 2020, were analyzed. Given that the Sentinel-2 satellite has a revisit period of 5 days, this study achieved complete coverage of the Yellow and East China Seas by integrating image data from four adjacent orbits (acquired from May 26 to 30). Figure 4 In the full-field true-color composite image shown in image a, the left-hand experimental area in the South Yellow Sea includes patches of two types of algal blooms: a green tide of *Ulva prolifera* (green) and a golden tide of *Cotinus coggygria* (red), while the East China Sea experimental area shows only a single golden tide of *Cotinus coggygria* (red). A local magnification of a typical mixed algal bloom area in the South Yellow Sea is also shown. Figure 4 (b) In the true-color image of this sea area, two types of floating algae, appearing both green and red, are present simultaneously. Further differentiation using the SRG-1 algorithm successfully distinguished the green tide of *Ulva prolifera* (*Ulva urticata*). Figure 4 c Green area) and copper algae golden tide ( Figure 4 (c yellow area) The separation boundary between the two types of algal blooms is clear, and the spatial distribution characteristics are highly consistent with the visual interpretation results.

Claims

1. A remote sensing classification and identification method for golden tides of copper algae and green tides of Ulva prolifera based on characteristic spectra, characterized in that, Includes the following steps: 1) Acquire Sentinel-2 remote sensing images of the designated sea area and preprocess them. The Sentinel-2 remote sensing images are Level-1C remote sensing images from a multispectral imager. 2) Calculate the Normalized Difference Vegetation Index (NDVI) of the preprocessed remote sensing image, and extract images of floating large algae based on the calculation results; 3) Use a triple masking method to eliminate non-water body interference factors in the image of floating large algae to obtain the large algal bloom area; 4) Classification and identification of copper algae golden tides and green tides of Ulva lactuca based on large algal bloom regions; Step 4) specifically involves: Based on large algal bloom regions, the minimum SRG-1 index for each time period of each month is calculated. Algal bloom regions with SRG-1 ≥ -0.28 are designated as copper algae gold tides, and the rest are designated as green tides of Ulva prolifera. The SRG-1 index is specifically: in, This refers to the red band reflectivity of the upper atmosphere, corresponding to the 4th band of Sentinel-2. This refers to the reflectivity of the top layer of the atmosphere in the green band, corresponding to the third band of Sentinel-2. The center wavelength of the red band. This is the center wavelength of the green band.

2. The remote sensing classification and identification method for copper algae golden tide and Ulva prolifera green tide based on characteristic spectra as described in claim 1, characterized in that, Step 1) involves preprocessing the remote sensing image, specifically as follows: Cloud and cirrus pixels on an image are removed using a publicly available cloud removal dataset mask.

3. The remote sensing classification and identification method for copper algae golden tide and Ulva prolifera green tide based on characteristic spectra as described in claim 1, characterized in that, Step 2) specifically refers to: Calculate the NDVI of the remote sensing image, and consider the image composed of pixels with NDVI ≥ 0.2 as floating large algae, wherein the NDVI is specifically: in, For near-infrared reflectivity, Sentinel-2 selects infrared bands 7 or 8. For red band reflectivity, Sentinel-2 selects band 4.

4. The remote sensing classification and identification method for golden tides of copper algae and green tides of Ulva prolifera based on characteristic spectra according to claim 1, characterized in that, Step 3) specifically refers to: The Normalized Differential Water Index (NDWI) was calculated based on annual Sentinel-2 images, and dynamic water area masking was performed using an NDWI greater than or equal to 0.2 as a threshold. Remove land boundaries based on known coastline vector data; Based on a global seabed topography database, seabed topography masking is performed using water depth thresholds.

5. The remote sensing classification and identification method for copper algae golden tide and Ulva prolifera green tide based on characteristic spectra as described in claim 4, characterized in that, The Normalized Differential Water Index (NDWI) is specifically: in, For green band reflectivity, the third band is selected for Sentinel-2. For near-infrared reflectivity, select 8 bands for Sentinel 2.

6. A remote sensing classification and identification system for copper algae golden tides and Ulva prolifera green tides based on characteristic spectra, used to implement the remote sensing classification and identification method for copper algae golden tides and Ulva prolifera green tides based on characteristic spectra as described in claim 1, characterized in that, include: The image preprocessing module is used to acquire Sentinel-2 remote sensing images of a designated sea area and preprocess them. The floating macroalgae image extraction module is used to calculate the Normalized Difference Vegetation Index (NDVI) of the preprocessed remote sensing image and extract images of floating macroalgae based on the calculation results. The triple masking module is used to remove non-water body interference factors from images of floating large algae using the triple masking method, so as to obtain the large algae bloom area. The classification and identification module is used to classify and identify the golden tide of copper algae and the green tide of seaweed based on large algal bloom regions.

7. A remote sensing classification and identification device for golden tides of copper algae and green tides of Ulva prolifera based on characteristic spectra, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a remote sensing classification and identification method for copper algae golden tide and seaweed green tide based on characteristic spectra, as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements a remote sensing classification and identification method for copper algae golden tides and seaweed green tides based on characteristic spectra, as described in any one of claims 1-5.

Images