Ocean enteromorpha fluorescence remote sensing extraction method based on unmanned aerial vehicle hyperspectrum

Abstract

The invention provides a marine enteromorpha fluorescence remote sensing extraction method based on unmanned aerial vehicle hyperspectrum, and belongs to the technical field of remote sensing monitoring of offshore large floating algae. The method comprises the following steps: acquiring sea surface enteromorpha original image data by using an unmanned aerial vehicle-mounted hyperspectral imager; performing correction processing on the sea surface enteromorpha original image data to obtain initial radiation brightness image data and reflectivity image data; performing denoising processing on the initial radiation brightness image data to obtain denoised radiation brightness image data; and inverting sunlight-induced chlorophyll fluorescence of enteromorpha by utilizing the de-noised radiance image data, the initial radiance image data and the reflectivity image data and combining a Fraunhofer dark line filling principle and a principal component analysis method. By means of the method, reliable inversion of the ocean enteromorpha chlorophyll fluorescence based on unmanned aerial vehicle-mounted hyperspectral image data is achieved, and the method has important application value in the aspect of detection of photosynthetic physiological information of ocean large floating algae such as enteromorpha in the future.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing monitoring of large floating algae in the offshore area, and particularly to a method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on unmanned aerial vehicle (UAV) hyperspectral data. Background Art

[0002] Enteromorpha prolifera is a large floating alga in the ocean, and its large-scale reproduction will disrupt the balance of the nearshore marine ecosystem. When it lands and accumulates, harmful gases and organic matter will be released during the decay process, which not only affects the coastal landscape but also causes significant losses to the economic activities in the coastal zone. Therefore, accurate monitoring of the outbreak range, physiological characteristics, and photosynthetic function of Enteromorpha prolifera is of great significance for marine environmental supervision, ecological protection, and carbon cycle research. The spectral characteristics of Enteromorpha prolifera are similar to those of terrestrial vegetation, showing a typical "red edge" phenomenon, which provides a theoretical basis for remote sensing monitoring.

[0003] Currently, remote sensing monitoring of Enteromorpha prolifera mostly relies on multi-spectral data to construct reflectance spectral indices related to the near-infrared band, and the distribution range and density information of Enteromorpha prolifera are extracted through index thresholds. However, these methods can only reflect the "greenness" or biomass of Enteromorpha prolifera and cannot directly characterize its physiological state and photosynthetic function. In contrast, solar-induced chlorophyll fluorescence (SIF), as a direct by-product of photosynthesis, is widely regarded as an "indicator" of plant photosynthetic efficiency. In the study of terrestrial vegetation, SIF can more directly reflect the dynamic changes of photosynthesis compared with traditional vegetation indices and has become a new means for estimating vegetation productivity. The chlorophyll of Enteromorpha prolifera will also release SIF signals during photosynthesis, and theoretically, its physiological and photosynthetic states can be monitored through its fluorescence characteristics.

[0004] Therefore, a method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on UAV hyperspectral data is needed. Summary of the Invention

[0005] In view of this, the present invention provides a method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on UAV hyperspectral data. The method includes collecting image data of Enteromorpha prolifera on the sea surface by a UAV-borne hyperspectral imager, denoising the image data of Enteromorpha prolifera on the sea surface by using a denoising model, and inversely calculating the solar-induced chlorophyll fluorescence of Enteromorpha prolifera by combining the Fraunhofer dark line filling principle and the principal component analysis method, so as to accurately extract the chlorophyll fluorescence of Enteromorpha prolifera in the ocean.

[0006] For this purpose, the present invention provides the following technical solutions:

[0007] A method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on UAV hyperspectral data, comprising:

[0008] Collecting the original image data of Enteromorpha prolifera on the sea surface;

[0009] Correct the original image data of Enteromorpha prolifera on the sea surface to obtain the initial radiance image data and reflectance image data;

[0010] Denoise the initial radiance image data to obtain the denoised radiance image data;

[0011] Use the denoised radiance image data, the initial radiance image data and the reflectance image data, and combine the Fraunhofer dark line filling principle and the principal component analysis method to invert the solar-induced chlorophyll fluorescence of Enteromorpha prolifera.

[0012] Further, the denoising process of the initial radiance image data includes:

[0013] Construct a denoising model based on the representative coefficient total variation regularization method;

[0014] Solve the denoising model using the alternating direction method of multipliers.

[0015] Further, the denoising model:

[0016]

[0017] s.t.Y = UV T +E+S,V T V = I

[0018] Where: U is the representative coefficient matrix, V is the endmember matrix; E represents the system Gaussian noise, S represents the sparse noise; λ, β and τ are the trade-off parameters for balancing the weights of each item; Y represents the cropped radiance image data matrix.

[0019] Further, the process of solving the denoising model using the alternating direction method of multipliers includes:

[0020] Introduce the augmented Lagrangian function to solve the denoising model.

[0021] Further, the correction process of the original image data of Enteromorpha prolifera on the sea surface includes:

[0022] Radiance correction, geometric correction and reflectance correction.

[0023] Further, the radiance correction includes:

[0024] Convert the DN value of the original image data of Enteromorpha prolifera on the sea surface into radiance data to obtain the radiance data of Enteromorpha prolifera on the sea surface;

[0025] The relationship between the DN value and the radiance: DN = L1*G*t exp +DF

[0026] Among them, DN is the original light intensity value recorded by the hyperspectral imager, dimensionless; L1 is the radiance value; G is the camera gain; t exp is the integration time of the photosensitive element; DF is the dark current noise.

[0027] Further, the geometric correction includes:

[0028] Based on the radiance data of Enteromorpha prolifera on the sea surface, by automatically matching each frame of image data with its corresponding attitude information, longitude and latitude, altitude, and time information, and modifying the altitude compensation of each image data, the radiance data of Enteromorpha prolifera on the sea surface with geographical information is obtained.

[0029] Further, the method for retrieving the solar-induced chlorophyll fluorescence of Enteromorpha prolifera by using the denoised radiance image data, the initial radiance image data, and the reflectance image data, in combination with the Fraunhofer dark line filling principle and the principal component analysis method, includes:

[0030] According to the spectral characteristics of Enteromorpha prolifera, the sensitive bands of solar-induced chlorophyll fluorescence of marine Enteromorpha prolifera are preset;

[0031] Based on the principal component analysis method, the smoothed value of the apparent reflectance is calculated;

[0032] The solar incident radiance data is calculated through the reflectance image data and the denoised radiance image data;

[0033] Using the smoothed value of the apparent reflectance, the sensitive bands of solar-induced chlorophyll fluorescence of marine Enteromorpha prolifera, and the solar incident radiance data, in combination with the Fraunhofer dark line filling principle, the solar-induced chlorophyll fluorescence values corresponding to the original image data of Enteromorpha prolifera on the sea surface are obtained.

[0034] Advantages and positive effects of the present invention:

[0035] The method of the present invention based on the denoising strategy of representative coefficient total variation regularization can effectively suppress noise interference and improve the accuracy of the retrieval result of the solar-induced chlorophyll fluorescence (SIF) of marine Enteromorpha prolifera. This strategy not only significantly reduces the stripe phenomenon during the retrieval of SIF, but also improves the correlation between the SIF signal and the reflectance index.

[0036] In the present invention, the denoising process, the selection of sensitive bands, and the Fraunhofer dark line - principal component joint analysis technology are integrated for the retrieval of the SIF of marine Enteromorpha prolifera, realizing the high-precision extraction of the SIF of Enteromorpha prolifera in the UAV hyperspectral data. Through targeted optimization of band selection and noise suppression, the signal-to-noise ratio of the SIF retrieval result is improved, effectively overcoming the error problems caused by water body background interference and low signal-to-noise ratio. Description of the Drawings

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0038] Figure 1 Flowchart of the method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on unmanned aerial vehicle hyperspectral in Embodiment 1 of the present invention;

[0039] Figure 2 Specific technical flowchart of the method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on unmanned aerial vehicle hyperspectral in Embodiment 2 of the present invention;

[0040] Figure 3 Pseudo-color images of the radiance of the unmanned aerial vehicle before and after denoising processing in Embodiment 2 of the present invention; where: a is the pseudo-color image of the radiance of the unmanned aerial vehicle before denoising processing; b is the pseudo-color image of the radiance of the unmanned aerial vehicle after denoising processing;

[0041] Figure 4 Partial enlarged views of the SIF results retrieved before and after the denoising processing of the radiance image in Embodiment 2 of the present invention; where: a is the SIF before denoising the radiance data; b is the SIF after denoising the radiance data;

[0042] Figure 5 Results of the reflectance index of the Enteromorpha prolifera area extracted in Embodiment 2 of the present invention and the SIF results retrieved before and after the denoising processing of the radiance image; where: a is the reflectance index of Enteromorpha prolifera smoothed by a sliding window spectrum; b is the SIF of Enteromorpha prolifera before denoising the radiance image; c is the SIF of Enteromorpha prolifera after denoising the radiance image;

[0043] Figure 6 Scatter plots of the correlation between the reflectance index of Enteromorpha prolifera extracted in Embodiment 2 of the present invention and the SIF pixel points before and after denoising; where: a is the scatter plot of the correlation between NDVI-NIR2 of Enteromorpha prolifera and the SIF before and after denoising; b is the scatter plot of the correlation between FAI-NIR of Enteromorpha prolifera and the SIF before and after denoising; c is the scatter plot of the correlation between VB_FAH-NIR1 of Enteromorpha prolifera and the SIF before and after denoising. Detailed implementation manners

[0044] To enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0045] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned accompanying drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0046] The present invention provides a method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on UAV hyperspectral data. The main idea is as follows: The denoising strategy for UAV hyperspectral radiance data based on representative coefficient total variation regularization can effectively suppress noise interference; by integrating denoising processing, sensitive band selection, and the combined analysis technology of Fraunhofer dark lines and principal components, high-precision extraction of the SIF of Enteromorpha prolifera in UAV hyperspectral data is achieved.

[0047] Combined with Figure 1 as shown, the method of the present invention is further described with a specific embodiment 1:

[0048] S1. Use a UAV-borne hyperspectral imager to collect the original image data of Enteromorpha prolifera on the sea surface.

[0049] In this embodiment, three UAV hyperspectral image data containing Enteromorpha prolifera are collected in the nearshore waters of Lingshan Island, Qingdao, China at 10:20, 11:00, and 11:40 on June 29, 2023 using a UAV-borne hyperspectral imager (Nano-Hyperspec); the weather is clear during data collection, the spectral resolution of the collected data is 6nm@20μm slit, the wavelength range is 400–1000nm, and the fluorescence Fraunhofer dark line inversion window near 760nm can be detected.

[0050] Pre-set the acquisition area and plan the flight path. Place a target in an open and flat area with good lighting in the acquisition area for reflectance correction of the post-acquisition image data. The target has 3 reflectance levels, namely: 56% (white area), 32% (gray area), and 11% (black area). Adjust the instrument exposure time to avoid overexposure of the white area of the target. The exposure time settings are: 19.994 ms, 17.996 ms, and 21.993 ms respectively.

[0051] Input parameters such as the flight altitude, frame rate, and lens focal length for data acquisition, automatically calculate the flight speed, and set the actual flight speed to be slightly less than the automatically calculated speed. The flight altitude for the acquisition of the three sets of image data is set to approximately 290 m. According to the sensor field of view angle, the ground spatial resolution that can be acquired is approximately 0.25 m, which can capture the fine spatial details of Enteromorpha prolifera. The frame rate is calculated by (1000 / dwell time), and the value of the dwell time is slightly greater than or equal to the exposure time. The dwell times for the three sets of image data are set to 20.099 ms, 18.101 ms, and 22.098 ms respectively; the lens focal length is 12 mm for all, and the corresponding field of view angle is 22°.

[0052] The three sets of image data acquired are raw3232_20230629_3, raw5536_20230629_4, and raw5808_20230629_5, and the data dimension is M×N×270; the M×N of the three sets of image data are 1758×5274, 2686×2552, and 2766×2748 respectively.

[0053] S2. Perform correction processing on the original image data of Enteromorpha prolifera on the sea surface to obtain the initial radiance image data and reflectance image data.

[0054] 1) Radiance correction: The original data collected by the spectral imager is 12-bit DN value data. Load the configuration file for radiance calibration at the factory corresponding to the integration time, convert the DN value data into radiance data, and obtain the initial radiance image data, whose unit is mW / cm2 / sr / μm. The relationship between the DN value and the radiance is established through the following formula:

[0055] DN = L1 * G * t exp + DF

[0056] In the formula, DN is the original light intensity value recorded by the hyperspectral imager, dimensionless; L1 is the radiance value (only considering the first-order spectral splitting of the grating); G is the camera gain; texp is the integration time of the photosensitive element; DF is the dark current noise.

[0057] 2) Geometric correction: Load the image data of the three RGB bands collected by the hyperspectral imager. By automatically matching each frame of image data with the corresponding attitude information (roll angle, heading angle, pitch angle, etc.), and using the information such as longitude, latitude, altitude, and time recorded, perform geometric correction on the image data, and modify the altitude compensation of each image data to obtain image data with geographical information. The altitude compensations for the three image data are -18.2 m, -8.2 m, and -9 m respectively.

[0058] 3) Reflectance correction: Load the initial radiance image data L of a total of 270 bands M×N×270 , find the pixel points with good target quality on the image data as the reference radiance values. Based on the known reflectance of the target itself, import the matching header file, calculate and generate the "Scene White Reference" reference file as the solar incident radiance, adjust the correction parameters, and perform reflectance correction on the image data to obtain the reflectance image data R M×N×270 .

[0059]

[0060] In the formula: M and N respectively represent the number of rows and columns of the initial radiance image data matrix; M×N for the three image data are: 1758×5274, 2686×2552, 2766×2748; ρ represents the known reflectance of the target imported from the header file; I represents the solar incident radiance.

[0061] S3. Denoise the initial radiance image data to obtain the denoised radiance image data;

[0062] S31. Perform band clipping and spectral dimension normalization on the initial radiance image data to obtain the clipped radiance image data and the clipped radiance image data matrix.

[0063] According to the principle of filling Fraunhofer dark lines, chlorophyll fluorescence extraction requires radiance data in bands including the absorption lines of the sun and the earth's atmosphere. Use ENVI software to statistically analyze the radiance spectral information of Enteromorpha pixels; according to the spectral characteristics of Enteromorpha, perform band clipping to select 50 bands of radiance data in the range of 676.409–785.864 nm including the oxygen absorption line.

[0064] Perform unit conversion on the initial radiance image data after band clipping, from mW / cm2 / sr / μm to mW / m2 / sr / nm; perform normalization processing in the spectral dimension and save the characteristic parameters of the data before normalization.

[0065]

[0066] where Y ∈ R MN×50 is a three - dimensional tensor matrix representation of hyperspectral radiance data with a spectral dimension of 50.

[0067] S32. Use the preset regularization parameter combination to denoise the normalized hyperspectral radiance data. In this embodiment, preferably, the regularization parameter combination is λ = 1, β = 50, τ = 0.8.

[0068] 1) Since there is a strong correlation between the spectral bands of the hyperspectral image, that is, the low - rank property of the spectrum, use low - rank matrix factorization (LRMF) to model the noise - free cropped radiance image data as:

[0069] X = UV T

[0070] In the formula, the matrix X ∈ R with rank R MN×50 represents the noise - free hyperspectral radiance data matrix; U ∈ R MN ×50 is the representative coefficient matrix, and V ∈ R 50×R is the endmember matrix.

[0071] 2) Input the cropped radiance image data matrix Y ∈ R MN×50 contaminated by mixed noise and the regularization parameters λ = 1, β = 50, τ = 0.8 into the denoising model of representative coefficient total variation regularization:

[0072]

[0073] s.t. Y = UV T + E + S, V T V = I

[0074] In the formula: E represents other system Gaussian noise, and S represents sparse noise.

[0075] Perform SVD operation on Y as UΣV, and initialize U = U R Σ R , V = V R . Use the alternating direction method of multipliers (ADMM) iterative algorithm framework to solve the denoising model. ADMM solves the model by introducing the augmented Lagrangian function: After introducing the auxiliary variables G1 and G2, the augmented Lagrangian function of the model is expressed as:

[0076]

[0077] where μ is the penalty parameter, is the Lagrange multiplier.

[0078] Under the ADMM framework, by fixing other variables and updating one variable at a time, the variables Gi, V, U, and E are alternately updated to approximate the global optimal solution and solve the model, obtaining the denoised image data X ∈ R M×N×50 .

[0079] S33. Normalize and restore the denoised image data.

[0080] Input the normalized characteristic parameters to restore the denoised image data, obtaining the denoised radiance image data; keep the unit of the radiance data as mW / m2 / sr / nm for the next step of SIF inversion.

[0081]

[0082] S4. Use the denoised radiance image data, the initial radiance image data, and the reflectance image data, and combine the Fraunhofer dark line filling principle and the principal component analysis method to invert the solar-induced chlorophyll fluorescence of Enteromorpha prolifera.

[0083] S41. According to the spectral characteristics of the Enteromorpha prolifera pixels in the UAV radiance data, select the sensitive band with the best SIF inversion effect for marine Enteromorpha prolifera.

[0084] Use ENVI software to perform statistical analysis on the radiance spectra of Enteromorpha prolifera pixels in the initial radiance image data and the denoised radiance image data, compare the spectral shapes, the low-value bands of the absorption lines, and the relatively smooth bands on its left and right, and select 6 different combinations of O2-A absorption band window ranges and absorption line inner and outer band value ranges to crop the radiance image data (L 800×500×50 , L”) 800×500×50 ) for SIF inversion experiments. Compare the experimental results to determine the sensitive band with the best SIF extraction effect for Enteromorpha prolifera;

[0085] In this embodiment, set the O2-A absorption band window range to 756.825–779.163 nm, the wavelength outside the absorption line to 756.825 nm, the wavelength inside the absorption line to the band with the minimum radiance value between 759.059–770.228 nm, and set the dynamic selection of the wavelength inside the absorption line.

[0086] λ out = 756.825 nm

[0087]

[0088] Among them, λ out is the wavelength outside the absorption line; λ in is the wavelength inside the absorption line; λ represents the O2-A absorption band.

[0089] S42. Using the initial radiance image data and the denoised radiance image data, combined with the Fraunhofer dark line filling principle and the principal component analysis method, the solar-induced chlorophyll fluorescence of Enteromorpha prolifera is retrieved.

[0090] Among them, the estimation of SIF within the absorption line:

[0091]

[0092] In the formula: The radiance L is the initial radiance image data and the denoised radiance image data of 50 bands in the range of 676.409 - 785.864 nm including the oxygen absorption line; corresponding to the target selected during reflectance correction, using the known reflectance of the target, the radiance data of the target pixels in the initial radiance image data and the denoised radiance image data are corrected to obtain the solar incident radiance data I for the retrieval of Enteromorpha prolifera SIF.

[0093] is the ratio of the smoothed apparent reflectance inside and outside the absorption line, estimated by the following formula:

[0094]

[0095] is the estimated value of the fluorescence ratio inside and outside the absorption line:

[0096]

[0097] In the formula, is the smoothed apparent reflectance in the O2-A absorption band; is the solar radiance smoothed using the interpolation method;

[0098] Excluding the solar radiance data in the O2-A absorption band, interpolate the solar radiance I based on the shape-preserving piecewise cubic interpolation method built into MATLAB to obtain the smoothed

[0099] Simulating the smoothed apparent reflectance based on the method of extracting principal components by principal component analysis

[0100] In this embodiment, 480 groups of training samples are simulated by setting parameter combinations with different values using the SCOPE model, with a total of 2,880 reflectance spectral data. The spectral sampling interval is 1 nm, and there are 211 bands in the range of 640–850 nm. The simulated training samples are interpolated to the wavelengths of 50 bands in the range of 676.409–785.864 nm of the UAV hyperspectral data as the training dataset, and principal component analysis (PCA) is performed to extract the principal components (PCs) of the reflectance spectra. After PCA decomposition, the cumulative contribution rate of the first 7 principal components reaches 99.99%, which can explain 99.99% of the shape feature information of the reflectance spectral changes;

[0101] The apparent reflectance is:

[0102]

[0103] Use the linear combination of the first 7 principal components to simulate the reflectance curve after smoothing the O2-A absorption band:

[0104]

[0105] In the formula: k i is the weight coefficient corresponding to the i-th principal component. Refer to the apparent reflectance information after removing the O2-A absorption band, and estimate the weight coefficient k i .

[0106] Combined with Figure 1 Taking Specific Embodiment 2 as an example, the beneficial effects of the method of the present invention are further illustrated:

[0107] Taking the second image raw5536_20230629_4 in the original image data of Enteromorpha prolifera on the sea surface collected in Embodiment 1 as an example, the specific implementation manner in the present invention is described.

[0108] 1. The second image is a UAV hyperspectral image containing Enteromorpha prolifera collected by a UAV-borne hyperspectral imager (Nano-Hyperspec) near the coastal waters of Lingshan Island, Qingdao, China at about 11:00 on June 29, 2023. The image size is 2686×2552×270, and the band range is 400–1000 nm, with a total of 270 bands.

[0109] 2. Perform radiance correction, geometric correction, and reflectance correction on the collected image.

[0110] 3. Perform band cropping and denoising on the corrected UAV hyperspectral radiance image.

[0111] Using ENVI software to statistically analyze the radiance spectral information of Enteromorpha pixels, according to its spectral characteristics, perform band cropping to select the radiance data L' of 50 bands in the range of 676.409 - 785.864 nm including the oxygen absorption line 2686×2552×50 Convert the unit of the radiance image data after band cropping to mW / m2 / sr / nm (L 2686×2552×50 ), and then perform normalization processing in the spectral dimension and save the characteristic parameters of the data before normalization

[0112] Compare the results of multiple experiments with different parameter combinations, and determine that the input regularization parameters in the embodiment are λ = 1, β = 50, τ = 0.8; perform denoising processing on the normalized data according to the denoising strategy in Embodiment 1 to obtain the denoised image data X ∈ R 2686×2552×50 .

[0113] Input the saved normalized characteristic parameters, restore the denoised image data, obtain the denoised radiance image data, and keep its unit as mW / m2 / sr / nm (L” 2686×2552×50 ) for the next step of SIF inversion

[0114] Appendix Figure 3 are the false color images of radiance before and after denoising. As shown in the figure, the noise of the UAV hyperspectral radiance image is significantly improved

[0115] 4. According to the spectral characteristics of Enteromorpha pixels in the UAV radiance data, select the sensitive bands with the best SIF extraction effect for marine Enteromorpha. Compare the spectral shapes, the low-value bands of the absorption lines, and the relatively smooth bands on its left and right of the radiance spectra of Enteromorpha pixels in the radiance images before and after denoising (L 2686×2552×50 , L” 2686×2552×50 ), select 6 different combinations of O2-A absorption band window ranges and wavelength values inside and outside the absorption line, crop the radiance images (L 800×500×50 , L” 800×500×50 ) of the dense Enteromorpha area for SIF inversion experiments. Compare the experimental results, determine the sensitive bands with the best SIF extraction effect for Enteromorpha based on the radiance data of the present invention, set the O2-A absorption band window range as 756.825 - 779.163 nm, and the wavelength λ out outside the absorption line as 756.825 nm

[0116] Set the wavelength inside the absorption line to dynamically select bands

[0117]

[0118] Then, use the UAV hyperspectral radiance image data before and after denoising (L 2686×2552×50 , L”2686×2552×50 ) By combining the Fraunhofer dark line filling principle and the principal component analysis method, the solar-induced chlorophyll fluorescence (SIF) of Enteromorpha prolifera is retrieved. The present invention uses formula (9) to estimate the SIF within the absorption line:

[0119]

[0120] In the formula: the radiance L is the radiance image data of 50 bands from 676.409 - 785.864 nm including the oxygen absorption line before and after denoising. Corresponding to the target selected during reflectance correction, using the known reflectance of the target, the radiance data of the target pixels in the image before and after denoising are corrected to obtain the solar incident radiance data I for Enteromorpha prolifera SIF retrieval.

[0121] is the ratio of the smoothed apparent reflectance inside and outside the absorption line:

[0122]

[0123] is the estimated value of the fluorescence ratio inside and outside the absorption line:

[0124]

[0125] In the formula: is the smoothed apparent reflectance in the O2-A absorption band; is the solar radiance smoothed using the interpolation method.

[0126] Excluding the solar radiance data in the O2-A absorption band, based on the shape-preserving piecewise cubic interpolation method built into MATLAB, the solar radiance I is interpolated to obtain the smoothed

[0127] Based on the method of extracting principal components by principal component analysis, the smoothed apparent reflectance is simulated Interpolate the 2880 reflectance spectral data simulated using the SCOPE model to the wavelengths of 50 bands from 676.409 - 785.864 nm of the UAV hyperspectral data as the training dataset, and perform principal component analysis (Principal Component Analysis, PCA) to extract the principal components (Principal Components, PC) of the reflectance spectrum. After PCA decomposition, the cumulative contribution rate of the first 7 principal components is 99.99%, which can explain 99.99% of the shape feature information of the reflectance spectrum change.

[0128] Apparent reflectance:

[0129]

[0130] Use the linear combination of the first 7 principal components to simulate the reflectance curve after smoothing in the O2-A absorption band:

[0131]

[0132] where: k i is the weight coefficient corresponding to the i-th principal component. Refer to the apparent reflectance information after removing the O2-A absorption band, and estimate the weight coefficient k by the least squares fitting method i .

[0133] Appendix Figure 4 is a partial enlarged view of the SIF results (original spatial resolution) retrieved from the radiance data before and after denoising. Spatially, it is obvious that the difference between the SIF results retrieved from the denoised radiance data between Enteromorpha and seawater is more obvious than that before denoising, and the stripe phenomenon has been significantly improved.

[0134] 5. Select an appropriate moving window and use the moving average method to perform spectral smoothing on the reflectance image data. Aiming at the problem of spectral jitter in the UAV reflectance data, different moving window sizes are set for multiple experimental comparisons. In the present invention, a better spectral smoothing effect is obtained when the moving window is set to 5. Statistically analyze the spectral information of different ground objects in the reflectance image data, select 162 bands with better data quality in the range of 446.330–805.968 nm including the red, green, and near-infrared bands, perform band cropping on the three preprocessed reflectance image data, and then use the moving average method with a moving window of 5 to perform spectral smoothing on the cropped reflectance data R M×N×162 .

[0135] Select bands to calculate the Normalized Difference Vegetation Index (NDVI), Floating Algae Index (FAI), and Virtual Baseline Floating Macroalgae Height (VB_FAH):

[0136]

[0137] where: λ is the wavelength value of the band selected for calculating the reflectance index, and R is the reflectance corresponding to the selected band.

[0138] UAV hyperspectral data is a kind of continuous high-resolution narrow-band spectral information. Due to the spectral continuity, a method of dynamic band selection is used to select the bands for calculating the index, that is:

[0139]

[0140] λ NIR = λ NIR1 or λ NIR2

[0141] According to the selected wavelength bands, the indices are customarily named as FAI-NIR, NDVI / VB_FAH-NIR1, and NDVI / VB_FAH-NIR2 respectively.

[0142] Based on the OTSU algorithm, threshold segmentation is performed on the reflectance index to extract the range of Enteromorpha prolifera. To address the problem of missed detection of small patches of Enteromorpha prolifera, ENVI is then used for visual interpretation to modify small areas.

[0143] 6. For the calculated reflectance index and the SIF inversion results, pixel averaging is performed in a 3×3 window in space. According to the detected range of Enteromorpha prolifera, the reflectance index and SIF pixels within the Enteromorpha prolifera area are extracted, totaling 19,429 pixels. The correlation between the extracted reflectance index of Enteromorpha prolifera and SIF is statistically analyzed, and the correlation coefficient R of linear regression is calculated 2 and the significance test P-value to comparatively analyze the differences in the correlation between the index and SIF before and after denoising.

[0144] Based on the above method, the linear R of the reflectance index in the Enteromorpha prolifera area and SIF before and after denoising 2 is shown in Table 1:

[0145] Table 1

[0146]

[0147] For NDVI and VB_FAH, there are two choices for NIR. The correlations between NDVI-NIR2 and VB_FAH-NIR1 and SIF are relatively good;

[0148] The R of the reflectance index and the SIF inverted from the denoised radiance image 2 has a significant improvement compared to that before denoising. The R of FAI-NIR and SIF before and after denoising 2 reaches 0.4965 and 0.7731. The R of NDVI-NIR2 and SIF before and after denoising 2 reaches 0.5047 and 0.7163. The R of VB_FAH-NIR1 and SIF before and after denoising 2 reaches 0.4925 and 0.7750, indicating that the high and low distribution of SIF values in space is in good agreement with the strength of the reflectance index, and can effectively reflect the distribution and density of Enteromorpha prolifera.

[0149] Appendix Figure 5The reflectance index results of the Enteromorpha region extracted in the present invention and the SIF results retrieved before and after denoising of the radiance image. (a) shows NDVI-NIR2, FAI-NIR, and VB_FAH-NIR1 of Enteromorpha after spectral smoothing of the reflectance data using the moving average method. (b) and (c) show the SIF results of Enteromorpha retrieved before and after denoising of the radiance image. Spatially, the consistency between the high and low distribution of the SIF values after denoising and the strength of the reflectance index is better than that before denoising.

[0150] Appendix Figure 6 The reflectance index of Enteromorpha extracted in the present invention ( Figure 5 (a)) and the scatter plots of the correlation between the SIF pixel points ( Figure 5 (b), Figure 5 (c)) before and after denoising. Figure 6 (a), Figure 6 (b), Figure 6 (c) are the scatter plots of the linear correlation between NDVI-NIR2, FAI-NIR, VB_FAH-NIR1 of Enteromorpha and the SIF before and after denoising, respectively. It can be seen more intuitively that the scatter plot of the reflectance index and the SIF retrieved after denoising of the radiance image is more concentrated, and the R 2 significantly improves.

[0151] In summary, in the above embodiments of the present application, by providing a method for remotely sensing fluorescence extraction of marine Enteromorpha based on UAV hyperspectral, including: collecting Enteromorpha images using a UAV-borne hyperspectral imager (Nano-Hyperspec); preprocessing the UAV hyperspectral remote sensing image data; performing band cropping and denoising on the preprocessed UAV radiance image, and determining the optimal parameter combination of the denoising algorithm through multiple experimental comparisons; analyzing the retrieved results of SIF through multiple experimental comparisons, and combining the characteristics of the UAV radiance spectral data of Enteromorpha pixels to select the sensitive bands with the best SIF retrieval effect for Enteromorpha; retrieving the SIF of marine Enteromorpha by combining the Fraunhofer dark line filling principle and the principal component analysis method; performing spectral smoothing on the UAV hyperspectral reflectance image using the moving average method with a suitable moving window, dynamically selecting bands to calculate NDVI, FAI, and VB_FAH, and extracting the Enteromorpha range based on the OTSU algorithm threshold segmentation; statistically analyzing the correlation between the reflectance index of the Enteromorpha region and the SIF pixel points. The results show that the denoising strategy and SIF retrieval process proposed in the present application can obtain relatively reliable SIF extraction results for Enteromorpha. The SIF results retrieved from the denoised radiance data are significantly different between Enteromorpha and seawater in terms of space compared to before denoising; SIF is significantly correlated with the reflectance index, and the R 2 between NDVI-NIR2, FAI-NIR, and VB_FAH-NIR1 and the SIF after denoising reach approximately 0.72, 0.77, and 0.78 respectively, compared to the R before denoising2 (Approximately 0.50, 0.50, and 0.49) has been greatly improved. The high and low distribution of SIF values in space is in good agreement with the strength of the reflectance index, and it can effectively reflect the distribution and density of Enteromorpha prolifera. The method flow of this application realizes the reliable inversion of the chlorophyll fluorescence of Enteromorpha prolifera in the ocean based on airborne hyperspectral images.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on UAV hyperspectral, characterized in that, Including: Collecting original image data of Enteromorpha prolifera on the sea surface; Performing correction processing on the original image data of Enteromorpha prolifera on the sea surface to obtain initial radiance image data and reflectance image data; Performing denoising processing on the initial radiance image data to obtain denoised radiance image data; Using the denoised radiance image data, the initial radiance image data and the reflectance image data, and combining the Fraunhofer dark line filling principle and the principal component analysis method to invert the solar-induced chlorophyll fluorescence of Enteromorpha prolifera.

2. The method for remotely sensing and extracting fluorescence of Enteromorpha prolifera in the ocean based on drone hyperspectral according to claim 1, wherein The performing denoising processing on the initial radiance image data includes: Constructing a denoising model based on the representative coefficient total variation regularization method; Using the alternating direction multiplier method to solve the denoising model.

3. The method for remotely sensing and extracting fluorescence of Enteromorpha prolifera in the ocean based on drone hyperspectral according to claim 2, wherein The denoising model: s.t. Y = UV T + E + S, V T V = I In the formula: U is the representative coefficient matrix, V is the endmember matrix; E represents the system Gaussian noise, S represents the sparse noise; λ, β and τ are trade-off parameters for balancing the weights of each item; Y represents the cropped radiance image data matrix.

4. The method for remotely sensing and extracting fluorescence of Enteromorpha prolifera in the ocean based on UAV hyperspectral according to claim 2, wherein, The using the alternating direction multiplier method to solve the denoising model includes: Introducing an augmented Lagrangian function to solve the denoising model.

5. The method for remotely sensing and extracting the fluorescence of Enteromorpha prolifera in the ocean based on drone hyperspectral according to claim 1, wherein, The performing correction processing on the original image data of Enteromorpha prolifera on the sea surface includes: Radiance correction, geometric correction and reflectance correction.

6. The method for remotely sensing and extracting fluorescence of Enteromorpha prolifera in the ocean based on drone hyperspectral according to claim 5, wherein The radiance correction includes: Converting the DN value of the original image data of Enteromorpha prolifera on the sea surface into radiance data to obtain the radiance data of Enteromorpha prolifera on the sea surface; The relationship between the DN value and the radiance: DN = L1 * G * t exp + DF Among them, DN is the original light intensity value recorded by the hyperspectral imager, dimensionless; L1 is the radiance value; G is the camera gain; t exp is the integration time of the photosensitive element; DF is the dark current noise.

7. A method for remotely sensing and extracting fluorescence of Enteromorpha prolifera in the ocean based on drone hyperspectral according to claim 5, characterized in that The geometric correction includes: Based on the radiance data of Enteromorpha prolifera on the sea surface, by automatically matching each frame of image data with its corresponding attitude information, longitude and latitude, altitude and time information, and modifying the altitude compensation of each image data, obtaining the radiance data of Enteromorpha prolifera on the sea surface with geographical information.

8. The method for remotely sensing and extracting fluorescence of Enteromorpha prolifera in the ocean based on drone hyperspectral according to claim 1, wherein, The using the denoised radiance image data, the initial radiance image data and the reflectance image data, and combining the Fraunhofer dark line filling principle and the principal component analysis method to invert the solar-induced chlorophyll fluorescence of Enteromorpha prolifera includes: According to the spectral characteristics of Enteromorpha prolifera, presetting the sensitive band of solar-induced chlorophyll fluorescence of marine Enteromorpha prolifera; Calculating the apparent reflectance smoothing value based on the principal component analysis method; Calculating the solar incident radiance data through the reflectance image data and the denoised radiance image data; Using the apparent reflectance smoothing value, the sensitive band of solar-induced chlorophyll fluorescence of marine Enteromorpha prolifera and the solar incident radiance data, and combining the Fraunhofer dark line filling principle to obtain the solar-induced chlorophyll fluorescence value corresponding to the original image data of Enteromorpha prolifera on the sea surface.

Images