Improved semi-analytical algorithm for the inversion of seawater inherent optical properties

Abstract

The application provides an improved semi-analytical algorithm for inverting seawater inherent optical parameters. The method generates an inclusive synthetic dataset containing 200,000 data sets based on IOCCG report 5, and constructs an improved semi-analytical algorithm model (DQAA model) based on the QAA semi-analytical algorithm and a three-level deep learning model. The first-level deep learning model is fused to invert the particle backscattering coefficient b bp (λ), the second-level deep learning model introduces the ultraviolet band R rs (360)+ R rs (visible) to invert the key parameters ζ and ε, and the third-level deep learning model inverts the non-algal particle absorption coefficient a d (443) technology, and outputs the colored dissolved organic matter absorption coefficient a g (443) inversion result, and further obtains the phytoplankton absorption coefficient a ph (443). After being verified by the IOCCG simulation dataset and the NOMAD measured dataset, the method realizes low error (RMSD<0.13 m ‑1 , MARD<0.51) inversion, fully utilizes the UV band, and can effectively separate the water body phytoplankton, colored dissolved organic matter and inorganic suspended particle absorption coefficients in water color remote sensing, and has strong universality and robustness.

Description

Technical Field

[0001] This invention relates to the field of ocean color remote sensing, and in particular to a method for inverting the absorption coefficients of phytoplankton, colored dissolved organic matter, and inorganic suspended particles from ocean color using an improved semi-analysis algorithm. Background Technology

[0002] The light-absorbing components in marine waters are mainly composed of phytoplankton, colored dissolved organic matter (CDOM), and non-algal particulate matter (NAP). The absorption coefficients of these three components (i.e., ...) a ph (λ) a g (λ) a d (λ) is not only a core element determining the inherent optical properties (IOPs) of water bodies, but also a key parameter characterizing marine biogeochemical cycles, primary productivity, and ecological environment evolution. Phytoplankton, as primary producers in the ocean, dominate carbon fixation and food web structure; CDOM regulates photochemical processes by absorbing ultraviolet and visible light radiation and serves as an important tracer of terrestrial input and water pollution; while non-algal particulate matter is closely related to sediment transport and estuarine and coastal dynamic processes. These three factors jointly determine the distribution of the underwater light field and the depth of the light transmittance, directly affecting the energy flow and material cycle of marine ecosystems (Tranviket et al., 2009; Wang et al., 2021). Therefore, achieving high-precision, large-scale synchronous remote sensing inversion of these key water color parameters is of irreplaceable importance for a deeper understanding of global climate change response, precise assessment of marine carbon budget, and monitoring of water environmental quality, and is also the core research goal in the current field of marine water color remote sensing.

[0003] Currently, methods for retrieving IOPs based on ocean color remote sensing are mainly divided into two categories: empirical algorithms and semi-analytical algorithms. Empirical algorithms rely on the statistical relationship between water body components and remote sensing data, are easily limited by regional environmental conditions, and have poor applicability. Semi-analytical algorithms (such as the quasi-analytical algorithm QAA and the GSM algorithm) are based on radiative transfer theory, reducing unknowns by simplifying model parameters, and have wider applicability. However, their core steps rely on empirical formulas (such as the separation of IOPs by the absorbance ratio of the 410nm and 440nm bands in QAA). a ph and a dg It has the following limitations: Insufficient utilization of the ultraviolet band: Traditional semi-analysis algorithms (such as QAA) mainly rely on the visible light band (410nm and above), while the UV band (350-400nm) is underutilized. a g Much highera ph It is separation a g It is a key source of information on absorption by other components. Although new-generation satellite sensors (such as Sentinel-3 / OLCI, SGLI, PACE) include the UV band, existing improved algorithms (such as QAA_UV) still have shortcomings. Liu et al. (2021) only verified the effectiveness of QAA_UV in clean water bodies. In nearshore Class II water bodies, the absorption coefficient of non-algal particulate matter is a key factor. a d and a g Spectral overlap, a dg The inversion error remains significant, and QAA_UV is consistent with QAA, neither of which has been corrected. a dg Further divided into a d and a g Unable to achieve a g Independent inversion (Wang et al., 2021).

[0004] Error propagation through empirical formulas: Key parameters in semi-analytical algorithms (such as the spectral index Y, proportionality coefficient ζ, and ε in QAA) are inverted using empirical formulas, making them susceptible to the influence of water body type, leading to error accumulation, especially in nearshore Class II water bodies. a d (Non-algal particulate matter absorption) and a g Their spectral shapes are similar, making them difficult to separate effectively.

[0005] The integration of deep learning and semi-analysis algorithms is insufficient: existing studies mostly use deep learning alone to directly invert IOPs, lacking integration with the physical mechanisms of semi-analysis algorithms, resulting in poor model interpretability and sensitivity to noise in the input data. Summary of the Invention

[0006] The present invention aims to at least solve one of the technical problems existing in the prior art, and provides a method for inverting the absorption coefficients of phytoplankton, colored dissolved organic matter and inorganic suspended particles in ocean water color using an improved semi-analysis algorithm.

[0007] In a first aspect, the present invention provides a method for inverting the absorption coefficients of phytoplankton, colored dissolved organic matter, and inorganic suspended particles from ocean color using an improved semi-analysis algorithm, comprising the following steps: Based on the IOCCG report 5 method, using the phytoplankton uptake coefficient a ph(440) Generate a spectral dataset containing intrinsic optical parameters (IOPs) and remotely sensed reflectance for the driving variables. R rs 200,000 sets of data (λ) are used to construct an inclusive synthetic dataset; Based on the aforementioned inclusive synthetic dataset, a three-level deep learning model is constructed: the first-level deep learning model uses the aforementioned remote sensing reflectance. R rs (λ) is the input direct inversion output particulate backscattering coefficient. b bp (λ); The second-level deep learning model introduces the ultraviolet band. R rs (360), with a remote sensing reflectance of 360 nm. R rs (λ) represents the key parameters ζ and ε that are input to the inversion output; the third-level deep learning model uses the particulate backscattering coefficients inverted from the first-level deep learning model. b bp (555) Combined with the total absorption coefficient of anhydrous a nw (443) R rs (360) R rs (443) R rs (555) R rs (670) is the input, and the output is the absorption coefficient of non-algae particles at 443 nm. a d (443); Based on the QAA semi-analysis algorithm and the aforementioned three-level deep learning model, an improved semi-analysis algorithm model (DQAA model) is constructed. First, the total absorption coefficient of non-algal particulate matter-colored dissolved organic matter is calculated using the QAA semi-analysis algorithm combined with ζ and ε output from the second-level deep learning model. a dg (443), and then based on the formula a g (443)= a dg (443)- a d (443) The absorption coefficient of colored dissolved organic matter at 443 nm was obtained by inversion. a g (443), using formula a ph (443)= a (443)- a dg (443)-a w (443), the absorption coefficient of phytoplankton at 443 nm was obtained. a ph (443); Based on the IOCCG simulation dataset and the NOMAD test dataset, the improved semi-analysis algorithm model (DQAA model) was evaluated and validated using multivariate statistical indicators.

[0008] In some possible embodiments, the phytoplankton absorption coefficient a ph The value range of (440) is 0.001-1m. -1 The inclusive synthetic dataset covers clean ocean waters and Class II coastal waters, and the remote sensing reflectance... R rs The range of (555) is 6.0*10. -4 -0.059 sr -1 Phytoplankton absorption coefficient a ph The range of (443) is 1.3*10. -3 -1.38 m -1 Colored dissolved organic matter absorption coefficient a g The range of (443) is 3.9*10. -4 -8.05 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 1.3*10. -4 -0.80 m -1 Particle backscattering coefficient b bp The range of (443) is 4.6*10. -5 -0.71m -1 Total absorption coefficient a The range of (443) is 6.8*10. -3 -10.08 m -1 .

[0009] In some possible embodiments, the remote sensing reflectance R rs The spectral range of (λ) is 350-800nm, with a wavelength interval of 1nm, and includes at least the ultraviolet band of 360nm and the visible light bands of 410nm, 443nm, 490nm, 555nm, and 670nm; 80% of the inclusive synthetic dataset is used to train the three-level deep learning model, and 20% is used to validate the three-level deep learning model.

[0010] In some possible embodiments, the first-level deep learning model, the second-level deep learning model, and the third-level deep learning model are all constructed using the Keras model (Chollet) framework, and all include an input layer, a hidden layer, and an output layer. The hidden layer consists of three layers: the first layer has 256 neurons, the second layer has 128 neurons, and the third layer has 16 neurons.

[0011] In some possible embodiments, the activation function of the three-level deep learning model is the Rectified Linear Unit (ReLU) function, the optimization function of the three-level deep learning model is the Adaptive Moment Estimation (Adam) function, and the learning rate is set to 2×10. -5 Training stops when the loss function converges and iteration stops.

[0012] In some possible embodiments, the IOCCG simulation dataset is constructed based on the International Organization for Harmony of Ocean Color (IOCCG) dataset (IOCCG 2006), with a spectral range extended to the ultraviolet band of 360 nm, and contains 500 sets of intrinsic optical parameters (IOPs) spectral data, including phytoplankton absorption coefficients. a ph The range of (443) is 5.0*10. -3 -0.42 m -1 , a g The range of (443) is 2.5*10. -3 -2.37 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 5.6*10. -4 -0.40m -1 , b bp The range of (443) is 6.4*10. -4 -0.13 m -1 The measured dataset is the NASA Bio-Optical Oceanography Algorithm Dataset (NOMAD), downloaded from the SeaBASS website. After screening, 942 spectra were retained for analysis. The measured dataset includes in-situ measurements. R rs (λ) and overlapping IOPs, where the phytoplankton absorption coefficient a ph The range of (443) is 2.0*10. -3 -1.48m -1 , a gThe range of (443) is 5.4*10. -4 -1.12 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 1.5*10. -5 -7.49 m -1 , R rs The range of (555) is 6.4*10. -4 -0.040 sr -1 .

[0013] In some possible embodiments, the multivariate statistical indicators include root mean square difference (RMSD), mean absolute relative difference (MARD), mean bias, and coefficient of determination. R 2 The improved semi-analysis algorithm model (DQAA model) is applied to the verification process using NOMAD measured data and satisfies the following requirements. a ph The RMSD of (443) is 0.13 m -1 MARD=0.40 R 2 =0.51, a d The RMSD of (443) is 0.086 m. -1 MARD=0.51 R 2 =0.65, a g The RMSD of (443) is 0.11 m -1 MARD=0.46 R 2 =0.72.

[0014] In some possible embodiments, the validation of the improved semi-analysis algorithm model (DQAA model) further includes applying the improved semi-analysis algorithm model (DQAA model) to SeaWiFS satellite remote sensing data: via UVISR. dl Model extrapolation of SeaWiFS data R rs (360) Quality control was performed on the satellite data, retaining data with a quality assurance (QA) score > 0.8 and excluding data containing special markers (l2_flags). The satellite remote sensing reflectance was extracted using the 3×3 pixel median method and matched with NOMAD measured data within a ±5-hour time window. The matching verification results were as follows: a ph The RMSD of (443) is 0.078 m.-1 MARD=0.58 R 2 =0.59, a d The RMSD of (443) is 0.074 m. -1 MARD=0.68 R 2 =0.41, a g The RMSD of (443) is 0.11 m -1 MARD=0.61 R 2 =0.27.

[0015] In a second aspect, the present invention provides an electronic device, comprising: One or more processors; A storage unit is provided for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the improved semi-analysis algorithm described above for inverting the inherent optical parameters of seawater.

[0016] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it can realize the inversion of the inherent optical parameters of seawater according to the improved semi-analysis algorithm described above.

[0017] The improved semi-analysis algorithm of this invention is used to invert the intrinsic optical parameters of seawater, and its beneficial effects are as follows: The DQAA model of this invention replaces the empirical formula for QAA with a three-level deep learning model. a ph (443) a d (443) a g (443) The inversion rate is significantly lower than that of traditional QAA_CDOM and Dong2013, among which a g (443) Inversion of RMSD (from 0.18 m on the NOMAD dataset) -1 Reduced to 0.11m -1 MARD was significantly reduced (from 0.51 to 0.46), especially at low concentrations. a g (443) (<0.05m) -1 In the scenario, R 2The inversion accuracy was significantly improved, increasing from 0.46 to 0.72. The DQAA model of this invention fully leverages the value of the UV band by introducing… R rs (360) Enhancement a ph and a dg The separation accuracy is high, and the model's resistance to noise in the ultraviolet band is superior to that in the visible light band, providing technical support for the application of next-generation satellite UV data. This invention's DQAA model is the first to achieve inversion using a deep learning model. a d (443), to achieve a dg Towards a d and a g The analytical separation solves the problem that traditional algorithms cannot independently invert the results. a g The problem is that the inclusive synthetic dataset of this invention covers clean waters up to coastal waters, and the model shows stable inversion capability in different sea areas around the world (equatorial Pacific, Gyre region, estuary). It has outstanding anti-interference capability against non-critical band noise, which improves universality and robustness to a certain extent. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of an example electronic device used to retrieve the inherent optical parameters of seawater using an improved semi-analysis algorithm according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of an improved semi-analysis algorithm for retrieving the inherent optical parameters of seawater, according to another embodiment of the present invention.

[0021] Figure 3 Used in this invention R rs (λ) spectral diagram, ( a Inclusive synthetic datasets R rs (λ) Example spectrum; b The simulation data provided by IOCCG R rs (λ) Example spectrum;c The NOMAD dataset provides real-world testing. R rs (λ) Example spectrum.

[0022] Figure 4 This invention uses a three-level deep learning model to estimate a ph (443) a d (443) a g (443) flowchart.

[0023] Figure 5 The DQAA model derived for this invention b bp (λ) and IOCCG simulation dataset b bp The comparison results between (λ) are shown in the figure.

[0024] Figure 6 The DQAA model derived for this invention a (λ) and IOCCG simulation dataset a The comparison results between (λ) are shown in the figure.

[0025] Figure 7 (a) is calculated for the QAA_CDOM of this invention. a ph (443), (b) a d (443), (c) a g (443), Dong2013's calculation of (d) a ph (443), (e) a d (443), (f) a g (443), DQAA calculation of (g) a ph (443), (h) a d (443), (i) a g (443) Comparison results with the IOCCG dataset.

[0026] Figure 8 The DQAA model derived for this invention a (λ) and NOMAD simulation dataset a The comparison results between (λ) are shown in the figure.

[0027] Figure 9(a) is calculated for the QAA_CDOM of this invention. a ph (443), (b) a d (443), (c) a g (443), Dong2013's calculation of (d) a ph (443), (e) a d (443), (f) a g (443), DQAA calculation of (g) a ph (443), (h) a d (443), (i) a g (443) Comparison results with the NOMAD dataset.

[0028] Figure 10 (a) is the result of the inversion of SeaWiFS and DQAA in this invention. a (410), (b) a (443), (c) a (490), (d) a ph (443), (e) a d (443), (f) a g The comparison results between (443) are shown in the figure. Detailed Implementation

[0029] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Figure 1 This is a schematic diagram of an example electronic device used to retrieve the inherent optical parameters of seawater using an improved semi-analysis algorithm according to an embodiment of the present invention. Figure 1 As shown, the electronic device 100 includes one or more processors 110, one or more storage devices 120, one or more input devices 130, one or more output devices 140, etc., and these components are interconnected via a bus system 150 and / or other forms of connection mechanisms. It should be noted that... Figure 1The components and structures of the electronic devices shown are merely exemplary and not limiting; other components and structures may be used as needed.

[0031] The processor 110 may be a central processing unit (CPU), or may be a processing unit consisting of multiple processing cores, or other forms of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 100 to perform desired functions.

[0032] Storage device 120 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, which a processor may execute to implement the client functions (implemented by the processor) in the embodiments of this disclosure described below, and / or other desired functions. Various applications and various data may also be stored in the computer-readable storage medium, such as various data used and / or generated by the applications.

[0033] The input device 130 may be a device used by a user to input commands, and may include one or more of a keyboard, mouse, microphone, and touch screen.

[0034] The output device 140 can output various information (such as images or sounds) to the outside (e.g., a user) and may include one or more of a display, a speaker, etc.

[0035] Figure 2 This is a schematic diagram of an improved semi-analysis algorithm for retrieving inherent optical parameters of seawater, according to another embodiment of the present invention. Figure 2 As shown, an improved semi-analysis algorithm for retrieving intrinsic optical parameters of seawater includes the following steps: Step S201: Based on the method of IOCCG report 5 (IOCCG-OCAG 2003; IOCCG 2006) (see https: / / ioccg.org / wp-content / uploads / 2016 / 03 / lee-data.pdf for specific formulas), using the phytoplankton absorption coefficient... a ph (440) Generate a spectral dataset containing intrinsic optical parameters (IOPs) and remotely sensed reflectance for the driving variables. R rsWe construct an inclusive synthetic dataset from 200,000 sets of data (λ).

[0036] In some embodiments, such as Figure 3 The Rrs(λ) spectra used in this invention are shown in the following figures: (a) an example Rrs(λ) spectrum from an inclusive synthetic dataset; (b) an example Rrs(λ) spectrum from simulated data provided by IOCCG; and (c) an example Rrs(λ) spectrum from a NOMAD dataset. Phytoplankton uptake coefficient. a ph The value range of (440) is 0.001-1 m. -1 The inclusive synthetic dataset covers clean ocean waters and Class II coastal waters ( ); R rs (670) As a criterion for determining water body type), remote sensing reflectance R rs The range of (555) is 6.0*10. -4 -0.059 sr -1 Phytoplankton absorption coefficient a ph The range of (443) is 1.3*10. -3 -1.38 m -1 Colored dissolved organic matter absorption coefficient a g The range of (443) is 3.9*10. -4 -8.05 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 1.3*10. -4 -0.80 m -1 Particle backscattering coefficient b bp The range of (443) is 4.6*10. -5 -0.71 m -1 Total absorption coefficient a The range of (443) is 6.8*10. -3 -10.08 m -1 Remote sensing reflectance R rs The spectral range of (λ) is 350-800 nm, with a wavelength interval of 1 nm, and includes at least the ultraviolet band of 360 nm and the visible light bands of 410 nm, 443 nm, 490 nm, 555 nm, and 670 nm; 80% of the inclusive synthetic dataset is used to train the Level 3 deep learning model, and 20% is used to validate the Level 3 deep learning model; among which, the inclusive synthetic dataset... R rs (λ) Spectrum (see λ) Figure 3 ( a ).

[0037] Specifically, in step S201, during the process of generating the inclusive synthetic dataset, random parameters are set to ensure that the synthetic dataset covers a wide range of water body types. The generation of these constraint random values ​​follows the IOCCG-OCAG (2003).

[0038] Step S202: Based on the inclusive synthetic dataset, improvements are made to the three-step empirical formula involved in the semi-analysis algorithm QAA, thereby constructing a three-level deep learning model: The first-level deep learning model uses remote sensing reflectance... R rs (λ) is the input direct inversion output particulate backscattering coefficient. b bp (λ); The second-level deep learning model introduces the ultraviolet band. R rs (360), with a remote sensing reflectance of 360 nm. R rs (λ) represents the key parameters ζ and ε that are input to the inversion output; the third-level deep learning model uses the particulate backscattering coefficients inverted from the first-level deep learning model. b bp (555) Combined with the total absorption coefficient of anhydrous a nw (443) R rs (360) R rs (443) R rs (555) R rs (670) is the input, and the output is the absorption coefficient of non-algae particles at 443 nm. a d (443).

[0039] In some embodiments, such as Figure 4 The invention shown uses a three-level deep learning model to estimate... a ph (443) a d (443) a g(443) Flowchart. The first-level, second-level, and third-level deep learning models are all constructed using the Keras (Chollet) framework and all contain an input layer, hidden layers, and an output layer. The hidden layers consist of three layers: the first layer has 256 neurons, the second layer has 128 neurons, and the third layer has 16 neurons. The activation function of the third-level deep learning model is the Rectified Linear Unit (ReLU) function, and the optimization function is the Adaptive Moment Estimation (Adam). The learning rate is set to 2 × 10⁻⁶. -5 Training stops when the loss function (root-mean-square error, RMSE) converges and iteration stops.

[0040] Step S203: Based on the QAA semi-analysis algorithm and the three-level deep learning model, construct an improved semi-analysis algorithm model (DQAA model). First, calculate the total absorption coefficient of non-algal particulate matter-colored dissolved organic matter using ζ and ε output by the QAA semi-analysis algorithm combined with the output of the second-level deep learning model. a dg (443), and then based on the formula a g (443)= a dg (443)- a d (443) The absorption coefficient of colored dissolved organic matter at 443 nm was obtained by inversion. a g (443), using formula a ph (443)= a (443)- a dg (443)- a w (443), the absorption coefficient of phytoplankton at 443 nm was obtained. a ph (443).

[0041] Specifically, the QAA in step S203 was developed by Lee et al. (2002) to derive the inherent optical properties of optical deepwater. Its inversion process consists of two consecutive steps: the first part involves selecting a reference wavelength. Then, the backscattering coefficient and absorption coefficient of particulate matter at each wavelength are estimated more accurately using a semi-analytical model formula. In the second part, the absorption coefficients of phytoplankton, yellow matter, and debris are calculated using the total absorption coefficient from the first part. Currently, QAA has reached its 6th version (https: / / ioccg.org / wp-content / uploads / 2020 / 11 / qaa_v6_202011.pdf). The specific formulas are shown in Table 1 below.

[0042] Table 1 QAA Data Inversion Algorithm ; ; Furthermore, the QAA model in b bp (λ) During the inversion process, Lee recommends selecting a reference wavelength of 55x or 670 nm based on the water body type, and using an exponential decay model. b bp (λ) is calculated during this process. a (λ0) and the spectral slope Y lead to further propagation of errors; while the DQAA model of this invention uses a three-level deep learning model. R rs (λ) calculation b bp (λ) can effectively avoid the judgment of water body type and the error propagation of two-step empirical formulas, and the DQAA model inversion b bp The result of (λ) is as follows Figure 5 As shown, Figure 5 (a)-(d) demonstrate b bp (410) b bp (443) b bp (490) b bp The inversion results of (555) are shown in Table 2 for accuracy evaluation indicators. (DQAA model inversion results) b bp (λ) shows good agreement with the simulation data, RMSD < 0.0058m -1 MARD < 0.1, bias < -0.0018 m -1 , R 2 >0.96.

[0043] Table 2. DQAA applied to IOCCG simulation dataset inversion b bp Statistical table after (λ) ; Furthermore, such as Figure 6 The DQAA model of the present invention shown is derived from... a (λ) and IOCCG simulation dataset a The comparison results between (λ) show that for a (555) in 0.06 - 0.99 m -1 DQAA model inversion of water bodies within the range a (λ) value and simulation data a The data in (λ) show good agreement, with very small data point dispersion. RMSD < 0.13m. -1 MARD < 0.11 R 2 >0.95, see Table 3 for accuracy evaluation indicators.

[0044] Table 3 shows the application of DQAA to IOCCG simulation dataset inversion. a (410) a (443) a (490) a Statistical table after (555) ; Furthermore, Zhu et al. (2011) developed QAA_CDOM, which utilizes... b bp (λ) and a d Establish empirical relationships between (λ) a g The inversion of (λ). Dong et al. (2013) established a ph (443) a d (443) a g The inversion formula of (443) (named Dong2013). This invention compares the DQAA model and QAA_CDOM, Dong2013 with simulation data. a ph (443) a d (443) a g (443) Compared with the data, the inversion results are as follows Figure 7 As shown. Figure 7 ( a )-( f As shown in Table 4, QAA_CDOM for aph (443) a d (443) a g The inversion result of (443) has an RMSD < 0.70m. -1 MARD < 0.83 R 2 >0.46. Dong2013 for a ph (443) a d (443) a g The inversion result of (443) is better than that of QAA_CDOM, where RMSD < 0.53m -1 MARD < 1.57 R 2 >0.25, DQAA for a ph (443) a d (443) a g The inversion results of (443) are the best, with RMSD < 0.45m. -1 MARD < 0.60 R 2 >0.71. This indicates that DQAA has good consistency in the simulated dataset.

[0045] Table 4. Applications of QAA_CDOM, Dong2013, and DQAA to IOCCG Simulation Dataset Inversion a ph (443), a d (443), a g (443) Statistical table ; Furthermore, this invention applies the DQAA model to NOMAD data. a (λ) The inversion accuracy is compared. The results are as follows: Figure 8 The DQAA model shown in (a)-(d) is derived from a (λ) and NOMAD simulation dataset a The comparison results between (λ) revealed an underestimation phenomenon in the inversion (see Table 5), with RMSD < 0.24m. -1 MARD < 0.25 R 2 >0.75 compared to NOMAD's measured data, a(410) The underestimation of the results of the inversion is even more serious, such as Figure 8 As shown in (a), RMSD = 0.24m -1 MARD = 0.25 R 2 =0.80. For example... Figure 8 As shown in (b) and (c), the DQAA model inversion results show good consistency in a(443) and a(490). Figure 8 As shown in (d), compared with the NOMAD data, the DQAA model performs better in... a (555) The inversion results are good, with RMSD = 0.047m. -1 MARD=0.18 R 2 =0.75.

[0046] Table 5. DQAA applied to NOMAD inversion a (410) a (443) a (490) a Statistical table after (555) ; Furthermore, a comparison of the QAA_CDOM, Dong2013, and DQAA algorithms for... a ph (443) a d (443) a g The inversion result of (443) is as follows: Figure 9 As shown in (a)-(i) and Table 6, the DQAA inversion accuracy is the best in the NOMAD dataset. The DQAA model in... a ph (443) The accuracy index of the inversion result RMSD = 0.13m -1 MARD = 0.40 R 2 =0.51, a d (443) The accuracy index of the inversion result RMSD = 0.086 m -1 MARD = 0.51 R 2 =0.65, a g The accuracy index of the inversion result (443) is RMSD = 0.11 m. -1 MARD=0.46 R 2=0.72. QAA_CDOM used IOCCG, NOMAD, Hudson, Mississippi, and Neponset data (Zhu et al. 2014) in its model parameter tuning, which helps improve the inversion accuracy of the IOCCG simulation data and the NOMAD data. Furthermore, QAA_CDOM in... a d (443) During the inversion, different j1 and j2 were obtained by fitting IOCCG simulation data and NOMAD data. In fact, this is crucial for the final... a g The influence of (443) is very significant. For example, Zhu et al. (2011) suggested j1=10.51, j2=1.56, and the inversion obtained using these parameters... a g (443) Compared with NOMAD data, MARD > 0.9. Dong2013 used both NOMAD data and measured data from the East China Sea for model fitting, therefore Dong2013 performed poorly on IOCCG data, but had higher inversion accuracy on NOMAD data. In contrast, the DQAA model has a wider range of training data coverage and better applicability.

[0047] Table 6 shows the application of QAA_CDOM, Dong2013, and DQAA to NOMAD dataset inversion. a ph (443), a d (443), a g (443) Statistical table ; Step S204: Based on the IOCCG simulation dataset and the NOMAD test dataset, train and validate the improved semi-analysis algorithm model (DQAA model) using multivariate statistical indicators.

[0048] In some embodiments, the IOCCG simulation dataset in step S204 is constructed based on the International Organization for Harmony of Ocean Color (IOCCG) dataset (IOCCG 2006). The original IOCCG dataset has a spectral range of 400-800 nm (wavelength increment of 10 nm) and a solar zenith angle of 30°. The spectral range of this invention is extended to the ultraviolet band of 360 nm (data extrapolation method is from Wang et al. (2021)), and includes 500 sets of intrinsic optical parameters (IOPs) spectral data. The data range of the IOCCG simulation data is shown in Table 7. R rs (λ) Spectrum (see λ) Figure 3 (b The actual dataset used was the NASA Bio-Optics Oceanography Algorithm Dataset (NOMAD, version 2.a), a publicly available, high-quality, in-situ bio-optics dataset used for ocean color algorithm development and satellite data product validation activities (Werdell and Bailey 2005b). It was downloaded from the SeaBASS website (http: / / seabass.gsfc.nasa.gov / ) on June 16, 2020. The original downloaded data contained 4559 spectra, which were filtered (incomplete records and low-quality data were excluded, and the remote sensing reflectance Rrs(λ) and CDOM absorption coefficient were ensured to be consistent). a g (443) After matching, 942 spectra were retained for analysis. The measured dataset includes in-situ measurements. R rs (λ) and the overlapping IOPs (including particulate absorption coefficient, CDOM absorption coefficient, detrital absorption coefficient, and backscattering coefficient), where R rs The range of (555) is 6.4*10. -4 -0.040 sr -1 The data range for NOMAD simulations is shown in Table 7. R rs (λ) Spectrum (see λ) Figure 3 ( c ).

[0049] Table 7. Datasets for model training, validation, and testing. R rs (555) (with) R rs (Taking 555 as an example) Statistical description table (coefficient of variation (CV) is the ratio of standard deviation to mean) ; In some embodiments, the multivariate statistical indicators in step S204 include root mean square difference (RMSD), mean absolute relative difference (MARD), mean bias, and coefficient of determination. R 2 The improved semi-analysis algorithm model (DQAA model) was applied to the verification process using NOMAD measured data and met the following requirements. a ph The RMSD of (443) is 0.13 m -1 MARD=0.40 R 2 =0.51, a d The RMSD of (443) is 0.086 m. -1MARD=0.51 R 2 =0.65, a g The RMSD of (443) is 0.11 m -1 MARD=0.46 R 2 =0.72.

[0050] Specifically, the calculation formulas for multivariate statistical indicators are as follows: Root mean square difference (RMSD): , Mean Absolute Relative Difference (MARD): , Mean bias: , Coefficient of determination (R) 2 ): , In the above formula, N represents the number of samples. This represents the inversion result (estimated value) in the data. This represents the measured results (true values) in the data.

[0051] In some embodiments, such as Figure 10 As shown in Table 8, the verification of the improved semi-analysis algorithm model (DQAA model) in step S204 also includes applying the improved semi-analysis algorithm model (DQAA model) to SeaWiFS satellite remote sensing data: via UVISR dl Model extrapolation of SeaWiFS data R rs (360) Quality control was performed on the satellite data, retaining data with a Quality Assurance (QA) score > 0.8 and excluding data containing special markers (l2_flags). The satellite remote sensing reflectance was extracted using the 3×3 pixel median method and matched with NOMAD measured data within a ±5-hour time window, matching 206 data points. The results were then compared... a (410) a (443) a (490) a ph (443) a d (443) a g (443) Figure 10 (a)-(f)), the verification result is RMSD<0.32m -1 MARD < 0.68 R 2 >0.27, specifically, the matching verification result.a ph The RMSD of (443) is 0.078 m. -1 MARD=0.58 R 2 =0.59, a d The RMSD of (443) is 0.074 m. -1 MARD=0.68 R 2 =0.41, a g The RMSD of (443) is 0.11 m -1 MARD=0.61 R 2 =0.27.

[0052] Table 8. DQAA Applications in SeaWiFS Satellite Data Retrieval a (410) a (443) a (490) a ph (443) a d (443) a g (443) Statistical table ; The method for inverting the absorption coefficient of colored dissolved organic matter from ocean color using an improved semi-analysis algorithm, as described in this invention, has the following beneficial effects: The DQAA model of this invention replaces the empirical formula for QAA with a three-level deep learning model. a ph (443) a d (443) a g (443) The inversion rate is significantly lower than that of traditional QAA_CDOM and Dong2013, among which a g (443) Inversion of RMSD (from 0.18 m on the NOMAD dataset) -1 Reduced to 0.11m -1 MARD was significantly reduced (from 0.51 to 0.46), especially at low concentrations. a g (443) (<0.05m) -1 In the scenario, R 2The inversion accuracy was significantly improved, increasing from 0.46 to 0.72. The DQAA model of this invention fully leverages the value of the UV band by introducing… R rs (360) Enhancement a ph and a dg The separation accuracy is high, and the model's resistance to noise in the ultraviolet band is superior to that in the visible light band, providing technical support for the application of next-generation satellite UV data. This invention's DQAA model is the first to achieve inversion using a deep learning model. a d (443), to achieve a dg Towards a d and a g The analytical separation solves the problem that traditional algorithms cannot independently invert the results. a g The problem is that the inclusive synthetic dataset of this invention covers clean waters up to coastal waters, and the model shows stable inversion capability in different sea areas around the world (equatorial Pacific, Gyre region, estuary). It has outstanding anti-interference capability against non-critical band noise, which improves universality and robustness to a certain extent.

[0053] In another aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables the implementation of the system described above.

[0054] The computer-readable medium may be included in the apparatus, device, or system disclosed herein, or it may exist independently.

[0055] The computer-readable storage medium may be any tangible medium that contains or stores a program, and may be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, optical fibers, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0056] The computer-readable storage medium may also include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code, specific examples of which include, but are not limited to, electromagnetic signals, optical signals, or any suitable combination thereof.

[0057] Parameter description: All parameter values ​​mentioned in this manual are not subjective assumptions, but rather a comprehensive result of the application scenario's security / efficiency requirements, industry standards and specifications, and industry practice experience thresholds. In actual applications, the parameters will be fine-tuned according to the relevant scenarios.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions 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. An improved semi-analysis algorithm for retrieving intrinsic optical parameters of seawater, characterized in that, Includes the following steps: Based on the IOCCG report 5 method, using the phytoplankton uptake coefficient a ph (440) Generate a spectral dataset containing intrinsic optical parameters (IOPs) and remotely sensed reflectance for the driving variables. R rs 200,000 sets of data (λ) are used to construct an inclusive synthetic dataset; Based on the aforementioned inclusive synthetic dataset, a three-level deep learning model is constructed: the first-level deep learning model uses the aforementioned remote sensing reflectance. R rs (λ) is the input direct inversion output particulate backscattering coefficient. b bp (λ); The second-level deep learning model introduces the ultraviolet band. R rs (360), with a remote sensing reflectance of 360 nm. R rs (λ) represents the key parameters ζ and ε that are input to the inversion output; the third-level deep learning model uses the particulate backscattering coefficients inverted from the first-level deep learning model. b bp (555) Combined with the total absorption coefficient of anhydrous a nw (443) R rs (360) R rs (443) R rs (555) R rs (670) is the input, and the output is the absorption coefficient of non-algae particles at 443 nm. a d (443); Based on the QAA semi-analysis algorithm and the aforementioned three-level deep learning model, an improved semi-analysis algorithm model (DQAA model) is constructed. First, the total absorption coefficient of non-algal particulate matter-colored dissolved organic matter is calculated using the QAA semi-analysis algorithm combined with ζ and ε output from the second-level deep learning model. a dg (443), and then based on the formula a g (443)= a dg (443)- a d (443) The absorption coefficient of colored dissolved organic matter at 443 nm was obtained by inversion. a g (443), using formula a ph (443)= a (443)- a dg (443)- a w (443), the absorption coefficient of phytoplankton at 443 nm was obtained. a ph (443); Based on the IOCCG simulation dataset and the NOMAD test dataset, the improved semi-analysis algorithm model (DQAA model) was evaluated and validated using multivariate statistical indicators.

2. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The phytoplankton absorption coefficient a ph The value range of (440) is 0.001-1 m. -1 The inclusive synthetic dataset covers clean ocean waters and Class II coastal waters, and the remote sensing reflectance... R rs The range of (555) is 6.0*10. -4 -0.059 sr -1 Phytoplankton absorption coefficient a ph The range of (443) is 1.3*10. -3 -1.38 m -1 Colored dissolved organic matter absorption coefficient a g The range of (443) is 3.9*10. -4 -8.05 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 1.3*10. -4 -0.80 m -1 Particle backscattering coefficient b bp The range of (443) is 4.6*10. -5 -0.71 m -1 Total absorption coefficient a The range of (443) is 6.8*10. -3 -10.08 m -1 .

3. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The remote sensing reflectance R rs The spectral range of (λ) is 350-800nm, with a wavelength interval of 1nm, and includes at least the ultraviolet band of 360nm and the visible light bands of 410nm, 443nm, 490nm, 555nm, and 670nm; 80% of the inclusive synthetic dataset is used to train the three-level deep learning model, and 20% is used to validate the three-level deep learning model.

4. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The first-level deep learning model, the second-level deep learning model, and the third-level deep learning model are all constructed using the Keras model (Chollet) framework, and all contain an input layer, a hidden layer, and an output layer. The hidden layer consists of three layers: the first layer has 256 neurons, the second layer has 128 neurons, and the third layer has 16 neurons.

5. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The activation function of the three-level deep learning model is the Rectified Linear Unit (ReLU) function, and the optimization function is the Adaptive Moment Estimation (Adam) function. The learning rate is set to 2×10⁻⁶. -5 Training stops when the loss function converges and iteration stops.

6. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The IOCCG simulation dataset is based on the International Organization for Harmony of Ocean Color (IOCCG) dataset (IOCCG 2006), with a spectral range extended to the ultraviolet band of 360 nm. It contains spectral data of 500 intrinsic optical parameters (IOPs), including phytoplankton absorption coefficients. a ph The range of (443) is 5.0*10. -3 -0.42 m -1 , a g The range of (443) is 2.5*10. -3 -2.37 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 5.6*10. -4 -0.40m -1 , b bp The range of (443) is 6.4*10. -4 -0.13 m -1 The measured dataset is the NASA Bio-Optical Oceanography Algorithm Dataset (NOMAD), downloaded from the SeaBASS website. After screening, 942 spectra were retained for analysis. The measured dataset includes in-situ measurements. R rs (λ) and overlapping IOPs, where the phytoplankton absorption coefficient a ph The range of (443) is 2.0*10. -3 -1.48 m -1 , a g The range of (443) is 5.4*10. -4 -1.12 m -1 Inorganic suspended particulate matter absorption coefficient a d The range of (443) is 1.5*10. -5 -7.49 m -1 , R rs The range of (555) is 6.4*10. -4 -0.040 sr -1 .

7. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The multivariate statistical indicators include root mean square difference (RMSD), mean absolute relative difference (MARD), mean bias, and coefficient of determination. R 2 The improved semi-analysis algorithm model (DQAA model) was applied to the verification process using NOMAD measured data and met the following requirements. a ph The RMSD of (443) is 0.13 m -1 MARD=0.40 R 2 =0.51, a d The RMSD of (443) is 0.086 m. -1 MARD=0.51 R 2 =0.65, a g The RMSD of (443) is 0.11 m -1 MARD=0.46 R 2 =0.

72.

8. The improved semi-analysis algorithm according to claim 1 for inverting inherent optical parameters of seawater, characterized in that, The validation of the improved semi-analysis algorithm model (DQAA model) also includes applying the improved semi-analysis algorithm model (DQAA model) to SeaWiFS satellite remote sensing data: via UVISR. dl Model extrapolation of SeaWiFS data R rs (360) Quality control was performed on the satellite data, retaining data with a Quality Assurance (QA) score > 0.8 and excluding data containing special markers (l2_flags). The satellite remote sensing reflectance was extracted using the 3×3 pixel median method and matched with NOMAD measured data within a ±5-hour time window. The matching verification results were then verified. a ph The RMSD of (443) is 0.078 m. -1 MARD=0.58 R 2 =0.59, a d The RMSD of (443) is 0.074 m. -1 MARD=0.68 R 2 =0.41, a g The RMSD of (443) is 0.11 m -1 MARD=0.61 R 2 =0.

27.

9. An electronic device, characterized in that, include: One or more processors; A storage unit for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the improved semi-analysis algorithm according to any one of claims 1 to 8 for inverting the inherent optical parameters of seawater.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can implement the improved semi-analysis algorithm according to any one of claims 1 to 8 for inverting the inherent optical parameters of seawater.

Images