Physical prior and spatio-temporal evolution fused remote sensing image ocean green tide monitoring method and system

Abstract

The invention relates to the technical field of remote sensing monitoring, in particular to a remote sensing image ocean green tide monitoring method and system fusing physical prior and spatio-temporal evolution. The method comprises the following steps: acquiring a multi-modal remote sensing monitoring image; performing multi-modal feature extraction on the acquired image, wherein the multi-modal feature extraction comprises spectral reflectivity feature extraction, ocean dynamics feature extraction and feature alignment and unified representation; establishing a physical prior of a green tide characteristic wave band by using an ocean optical radiation transmission model; constructing a dynamic space-time diagram based on the extracted multi-modal features to obtain a node global feature vector and a dynamic adjacency matrix; carrying out adaptive graph convolution feature coding based on physical prior and a dynamic space-time diagram; through fusion of multi-spectral images of multiple platforms such as satellites and unmanned aerial vehicles and ocean dynamic data and combination of atmospheric correction and wave band resampling, consistency processing and high-precision extraction of multi-source features are realized, and comprehensiveness and reliability of green tide feature recognition are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing monitoring, in particular to a remote sensing image marine green tide monitoring method and system fusing physical priori and spatio-temporal evolution. BACKGROUND

[0002] As an important form of marine ecological disaster, green tide is often caused by the outbreak of large-area floating algae such as Enteromorpha, and has the characteristics of strong suddenness, fast diffusion speed and wide coverage. Green tide can lead to depletion of marine dissolved oxygen, deterioration of water quality, loss of fishery resources and damage to coastal economy, and has become a major challenge to marine environmental safety monitoring and ecological management.

[0003] At present, marine green tide monitoring and early warning mainly rely on multispectral remote sensing images. By obtaining the reflectivity information of the sea area in visible light, near-infrared and other multi-band, the distribution and evolution process of marine green tide are identified. Traditional green tide detection methods can be divided into three categories: first, the detection method based on spectral index, which uses the reflectivity difference between green tide and surrounding water in near-infrared and visible light bands to realize detection by constructing green tide sensitive index (such as normalized vegetation index NDVI, floating algae detection index FAI, normalized green algae index NGI, etc.). This method is simple to implement and efficient to calculate, but it relies on fixed thresholds and is difficult to adapt to changes in water turbidity, suspended particulate matter concentration and observation conditions in different sea areas, and the detection accuracy is significantly limited. Second, the detection method based on physical model, which uses marine radiation transfer model or water optical model, combined with the physical priori knowledge of green tide such as chlorophyll absorption peak and spectral reflectance characteristics, to derive the judgment rule of the existence of green tide. This method has strong physical interpretability, but the model construction is complex, the parameters are difficult to obtain, and the model accuracy is easily affected by different sensors, different times and variable marine dynamic conditions, and the generalization ability is insufficient. Third, the detection method based on data-driven, which uses machine learning or deep learning model to extract features and classify remote sensing images, to realize automatic detection of green tide. In recent years, convolutional neural networks, Transformers and other methods have made significant progress in intelligent analysis of remote sensing images, and can mine complex patterns from high-dimensional spectral and spatial features. However, deep learning methods have high dependence on large-scale high-quality labeled samples, and are prone to overfitting in small sample sea areas and complex observation conditions, and it is difficult to fuse marine dynamics and physical priori, resulting in a lack of spatio-temporal consistency and interpretability of the detection results.

[0004] Overall, the existing methods for detecting marine green tide still have obvious shortcomings in physical interpretability, temporal consistency, and cross-regional generalization ability: lack of physical prior constraints, most methods rely only on data statistical characteristics, ignoring the specific spectral absorption mechanism of green tide and the optical properties of marine water, resulting in insufficient robustness of the algorithm in different sea areas and different sensors; lack of spatio-temporal evolution modeling, the occurrence and spread of green tide are driven by multiple ocean dynamic factors such as ocean currents, wind fields, and tides, and existing methods are mostly single-time, static detection, which is difficult to depict the propagation path and life cycle evolution of green tide; lack of multi-source heterogeneous data fusion, multi-spectral images, historical time series data, and ocean dynamic models cannot be effectively integrated, resulting in limited detection accuracy and prediction ability.

[0005] Therefore, it is urgent to propose a multi-spectral remote sensing image marine green tide detection method that integrates physical prior and spatio-temporal evolution constraints. SUMMARY

[0006] In order to solve the problems of insufficient utilization of physical prior, weak spatio-temporal evolution modeling ability, and insufficient fusion of multi-source and multi-temporal information in the process of marine green tide remote sensing detection, the present application provides a remote sensing image marine green tide monitoring method and system that integrates physical prior and spatio-temporal evolution.

[0007] In the first aspect, the present application provides a remote sensing image marine green tide monitoring method that integrates physical prior and spatio-temporal evolution, which adopts the following technical solution: A remote sensing image marine green tide monitoring method that integrates physical prior and spatio-temporal evolution, comprising: Obtaining multi-modal remote sensing monitoring images; Performing multi-modal feature extraction on the obtained images, including spectral reflectance feature extraction, ocean dynamics feature extraction, and feature alignment and unified representation; Using a marine optical radiation transfer model to establish physical prior of green tide characteristic bands; Based on the extracted multi-modal features, constructing a dynamic spatio-temporal graph to obtain node global feature vectors and a dynamic adjacency matrix; Based on the physical prior and the dynamic spatio-temporal graph, performing adaptive graph convolution feature coding; Based on the physical prior and the time series Transformer, modeling the multi-modal features in time series; Performing marine green tide prediction by fusing the adaptive graph convolution feature coding and the time series modeling results; Outputting the prediction results.

[0008] Further, the multi-modal feature extraction on the obtained images includes spectral reflectance feature extraction, wherein the obtained multi-band original radiation value is R raw, first atmospheric correction and band resampling are performed to obtain apparent water reflectance R(λ), after obtaining the water reflectance, key vegetation indices are extracted for the green tide sensitive band; for the extraction of marine dynamics characteristics, the original marine dynamics data is R d , the interpolation and re-projection method is used to calculate the marine dynamics characteristic vector, at the same time, in order to dynamically depict the change trend of green tide with the dynamic condition, the multi-temporal sequence {X d t} is used to calculate the time sequence gradient; finally, a cross-platform alignment method based on adversarial learning is used, the high-resolution spectral features extracted by the unmanned aerial vehicle are X u , the low-resolution spectral features extracted by the satellite are X s , the multi-source features are unified and mapped to a common representation space by minimizing the alignment loss function, and the global feature vector is obtained, the global feature vector includes spectral information and marine dynamics prior.

[0009] Further, the physical prior of the green tide characteristic band established by the marine optical radiation transfer model includes introducing the marine optical radiation transfer model RTM, defining the theoretical reflectance vector of the green tide pixel in different bands R phy , constructing the theoretical prior feature of the typical spectrum of the green tide based on the RTM, in the process of deep model training, the theoretical prior curve and the model predicted spectrum are constrained, by minimizing the difference in shape, the prediction result still conforms to the physical law under different observation conditions, and the spectral shape consistency loss is: , Among them, R pred represents the reflectance predicted by the deep model, R phy represents the theoretical spectrum generated by the RTM; the linear mixing abundance constraint based on the end member theory represents the reflectance of each pixel as a linear combination of multi-class end member reflectance; according to the marine optical radiation transfer model, the theoretical reflectance curve and the prior abundance distribution of different end members are obtained; then in the deep model training, the prediction result is consistent with the theoretical combination result, and the linear mixing abundance constraint loss is introduced: , Among them, a phy represents the theoretical abundance distribution calculated by the RTM, a pred represents the theoretical abundance distribution obtained by deep prediction.

[0010] Further, the extracted multi-modal features are used to construct a dynamic spatio-temporal graph, including taking the remote sensing pixels as graph nodes to construct a dynamic spatio-temporal graph dynamically reflecting spatial diffusion relationship and temporal evolution law, wherein the remote sensing image is divided into N nodes, the geographic coordinates of each node are P i =( lon i , lat i ), the Euclidean distance between nodes is d ij =∥p i -p j ∥, the spatial similarity weight between nodes i and j is defined according to the Gaussian kernel function, and when the two nodes are blocked by the coastline or island topography, the shielding is performed through the mask matrix Bij; the ocean current driving weight and the wind field driving weight are calculated based on the marine dynamic process, and the ocean current and wind field effects are weighted and fused, a time evolution prior graph is constructed using historical outbreak data, and finally the geographic similarity, dynamic driving and historical prior are fused to construct a dynamic adjacency matrix at time t: , wherein α, β, δ are preset weights, A geo represents a spatial adjacency relationship matrix, A dyn represents a dynamic driving feature matrix, P t represents a time sequence transition probability matrix, and is normalized to obtain: , wherein, D t represents a diagonal matrix, and the output {A t} is a dynamic spatio-temporal graph sequence, which is a graph structure prior for deep spatio-temporal modeling.

[0011] Further, the physical prior and the dynamic spatio-temporal graph are used to perform adaptive graph convolution feature coding, including designing a feature encoder that propagates stably across scales on the dynamic graph, wherein the relationship vector of edge i-j at time t is , the relationship vector is mapped to the convolution kernel parameter of the edge by a small kernel generator MLP K , and the edge is conditionally aggregated on the dynamic adjacency A t , which is represented as: , wherein σ(·) is a nonlinear activation function, A t represents a dynamic adjacency matrix, B (l)is a self-connection transformation; then a spectrum domain learnable filtering and a time domain attention convolution are performed, wherein a polynomial approximation of Laplacian spectral filtering is adopted in the frequency domain to avoid eigen decomposition of the Laplace matrix, and high-order neighborhood information aggregation is realized; in the time domain, a time window H t - Tw+1 ,…, H t is calculated to capture the diffusion time lag and speed information, and finally the two representations obtained in the spectrum domain and the time domain are fused through a learnable gate t : , wherein pool(·) represents a global pooling operation, U represents a parameter matrix, sigma is a sigmoid, and is an element-wise multiplication.

[0012] Further, the adaptive graph convolution feature encoding based on the physical prior and dynamic spatio-temporal graph further comprises constructing a multi-scale propagation for modeling the regional diffusion pattern while preserving the boundary details at the pixel level, wherein a boundary-conditioned convolution and a spectrum-time domain filtering are applied to obtain a fine-scale representation , and fine-scale encoding is realized; a plurality of adjacent and dynamically similar pixels are aggregated into super nodes, a learnable allocation matrix S is used, and each row s i represents a soft coefficient of the pixel i allocation to M super nodes, and satisfies row normalization; a coarse-scale representation and adjacency are obtained by aggregation using S; then the fine-scale and the coarse-scale are aggregated to obtain a final representation with both fine details and coarse-scale consistency, and the uncertainty is weighted and fused, and there are differences in observation noise and physical interpretability between different pixels; to avoid interference of high-noise pixels on the overall loss, a loss function is set during the training process: , wherein represents an uncertainty probability parameter of the i-th pixel, represents a spectral consistency loss, represents an abundance constraint loss, represents a prediction task loss.

[0013] Further, the physical prior and time series Transformer are used to model the time series of multi-modal features, including a physical prior guided time series Transformer, which realizes high-precision modeling of the whole process of green tide by explicitly introducing spectral prior, chlorophyll concentration dynamic characteristics and ocean dynamics gradient constraints in the self-attention mechanism, wherein the self-attention mechanism based on the physical prior bias is used to improve the modeling ability of the green tide life cycle by explicitly introducing the chlorophyll concentration estimation and spectral gradient characteristics as a bias term in the attention calculation, and the physical prior bias B phy is obtained; after the physical prior guided Transformer encoding of the layer, the time series deep feature representation is obtained, and finally a residual correction mechanism based on the physical prior is introduced to calculate the residual between the predicted spectrum and the endmember mixed reconstruction spectrum by a linear mixing model: L wherein R phy represents the physical prior residual term, S represents the predicted spectrum, k represents the total number of spectral bands, k represents the physical weight coefficient; the residual is mapped back to the deep feature space for dynamic correction: wherein a is a learnable parameter, finally, the model obtains the features and the features based on the time series Transformer and adaptive convolution to predict the whole process of the occurrence, development and disappearance of the multi-temporal green tide.

[0014] Further, the green tide is predicted by fusing the adaptive graph convolution feature encoding and the time series modeling result, including fusing the feature vectors generated by the graph adaptive convolution and the physical prior guided time series Transformer, combining the spatial diffusion and time series evolution characteristics, and jointly modeling in the probability space, wherein the graph adaptive convolution network is responsible for modeling the spatial diffusion process of the green tide, the PP-Transformer extracts the evolution law in the long time series, the spatial diffusion prediction extracted by the graph adaptive convolution network is set as , the output time series prediction is , and the fusion prediction probability map P f is represented as: wherein a represents an adaptive fusion coefficient, sigma (·) is a Sigmoid activation function, H and W represent the height and width of the remote sensing image respectively, and the multi-source result fusion is used to capture the local spatial diffusion mode and global time series evolution trend of the green tide. ​​​​

[0015] Further, the marine green tide prediction by fusing adaptive graph convolution feature encoding and time series modeling results also includes introducing an adaptive prediction correction method based on prediction uncertainty, by dynamically adjusting the contribution of different models in high uncertainty areas, first, the prediction variance of the graph adaptive convolution network and the physically prior guided time series Transformer at each pixel position is calculated, to quantify the uncertainty of the prediction value in multiple sampling or multi-model prediction, according to the uncertainty adaptive allocation of weights, finally get the adaptive corrected prediction map: , Where, P (A) represents the original prediction probability output by the graph adaptive convolution network, P (T) represents the original prediction probability output by the physically prior guided time series Transformer, P f represents the fusion prediction probability map; introduce green tide risk index, comprehensive consideration of green tide coverage area, chlorophyll concentration gradient and historical evolution mode, realize dynamic risk classification and early warning, wherein the green tide coverage mask M is determined by threshold θ: , Where, M ij represents the green tide coverage label of pixel (i,j), P c (i,j) represents the value of the fusion prediction probability map at pixel (i,j), and the green tide risk index is defined as: , Where, λ1, λ2, λ3 represent the weighting coefficients of the risk index, A represents the green tide coverage area, ∇C represents the chlorophyll concentration gradient, H represents the historical evolution trend factor.

[0016] The second aspect is a remote sensing image marine green tide monitoring system fusing physical prior and spatio-temporal evolution, comprising: A data acquisition module configured to acquire multi-modal remote sensing monitoring images; A feature extraction module configured to perform multi-modal feature extraction on the acquired images, including spectral reflectance feature extraction, ocean dynamics feature extraction, and feature alignment and unified representation; A physical prior module configured to establish a physical prior of green tide characteristic bands using a marine optical radiation transfer model; A space-time module configured to construct a dynamic space-time graph based on the extracted multi-modal features, and obtain a node global feature vector and a dynamic adjacency matrix; An encoding module configured to perform adaptive graph convolution feature encoding based on the physical prior and the dynamic space-time graph; a modeling module configured to perform time-series modeling on the multi-modal features based on physical priors and a time-series Transformer; a prediction module configured to perform marine green tide prediction by fusing adaptive graph convolution feature encoding and time-series modeling results; and output the prediction results.

[0017] In a third aspect, the present application provides a computer-readable storage medium having stored therein a plurality of instructions adapted to be loaded and executed by a processor of a terminal device to implement the marine green tide monitoring method based on fusion of physical priors and spatio-temporal evolution of remote sensing images.

[0018] In a fourth aspect, the present application provides a terminal device comprising a processor and a computer-readable storage medium, the processor being configured to implement the instructions, and the computer-readable storage medium being configured to store a plurality of instructions adapted to be loaded and executed by the processor to implement the marine green tide monitoring method based on fusion of physical priors and spatio-temporal evolution of remote sensing images.

[0019] To sum up, the present application has the following beneficial technical effects: By fusing multi-platform multispectral images such as satellites and unmanned aerial vehicles and marine dynamic data, combining atmospheric correction and band resampling, the consistency processing and high-precision extraction of multi-source features are realized, and the comprehensiveness and reliability of green tide feature recognition are significantly improved; the marine optical radiation transfer model is introduced to construct physical priors such as red band absorption valley, near-infrared reflection peak, and green band enhancement, and by means of spectral shape consistency loss and linear mixing abundance constraint, the physical knowledge is embedded into the deep model, effectively improving the detection robustness across scenes and time phases; at the same time, based on the adaptive graph convolution network and the time-series Transformer guided by the physical priors, the system can dynamically model the spatio-temporal evolution features of the green tide, realize the accurate prediction of the occurrence, diffusion and extinction process of the green tide, and provide high-confidence green tide detection results and early warning information by combining risk index modeling, thereby providing strong support for marine ecological environment monitoring and emergency decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 FIG. 1 is a schematic diagram of a marine green tide monitoring method based on fusion of physical priors and spatio-temporal evolution of remote sensing images according to an embodiment of the present application; Figure 2 FIG. 2 is a schematic diagram of ACC and F1 comparison of green tide anomaly detection models according to an embodiment of the present application; Figure 3 FIG. 3 is a schematic diagram of accuracy rate comparison of green tide anomaly detection models under different Dropout ratios according to an embodiment of the present application; Figure 4 FIG. 4 is a schematic diagram of spatial distribution heat of green tide anomaly detection according to an embodiment of the present application; DETAILED DESCRIPTION

[0021] The application will be further described in detail below with reference to the accompanying drawings.

[0022] Embodiment 1 With reference Figure 1 , a fusion of physical prior and spatio-temporal evolution of the remote sensing image marine green tide monitoring method of the embodiment includes: S1. Multi-modal feature extraction module In the marine green tide monitoring process, the remote sensing images collected by different platforms (satellite, unmanned aerial vehicle, etc.) have significant differences in observation angle, spatial resolution, band range, etc. In order to realize high-precision description of the distribution of green tide, it is necessary to extract water reflectivity, vegetation index, marine dynamics and other multi-source features from multi-platform, multi-spectral, multi-temporal remote sensing images, and to uniformly represent them. The core goal of this module is to improve the robustness and generalization ability of green tide detection through the fusion expression of spectral, spatial, and dynamic features, and to provide a reliable feature basis for subsequent physical prior modeling and spatio-temporal evolution prediction. Specifically, it includes the following three parts: 1) Spectral reflectance feature extraction, different bands in multi-spectral remote sensing data have significant sensitivity to green tide biomass changes. Let the multi-band original radiation value obtained from satellite or unmanned aerial vehicle image be R raw , first perform atmospheric correction and band resampling to obtain apparent water reflectivity R(λ): , Among them, L (λ) represents the band radiance received by the sensor, d represents the distance correction factor, E s ( lambda ) represents the solar irradiance at the top of the atmosphere; theta s represents the solar zenith angle. The water reflectivity R includes all the original information of the bands, and for the green tide sensitive bands (blue, green, red, near-infrared), the key vegetation indexes NDVI , NDWI such as: , Among them, R NIR , R RED , R GREEN are the near-infrared, red and green band reflectances, respectively.

[0023] 2) Marine dynamics feature extraction, the occurrence and diffusion process of green tide is strongly driven by marine dynamic environment, including sea surface temperature, ocean current velocity, wind field and other parameters. Let the original marine dynamics data be Rd , using the interpolation and reprojection method, the ocean dynamics eigenvector is defined as: , in, T s represents the sea surface temperature, V c 、 theta c Represents the speed and direction of ocean currents, V w 、 theta w In order to dynamically depict the trend of green tide changing with dynamic conditions, the multi-phase sequence {X d t}Calculate the temporal gradient: , 3) Feature alignment and unified representation. Due to the differences in spatial resolution, spectral range, and observation angle between satellite and UAV images, direct feature fusion will lead to a decrease in green tide detection performance. This paper adopts a cross-platform alignment method based on adversarial learning. The high-resolution image extracted by the UAV is extracted through a convolutional network and then the spectral features obtained by splicing it with the vegetation coefficients NDVI and NDWI are as follows: X u The spectral characteristics of the low-resolution image extracted by the satellite after the convolutional network feature extraction and the vegetation coefficient NDVI and NDWI are obtained: X s , achieve multi-platform feature consistency by minimizing the alignment loss function: , in, f θ is the feature mapping function based on the convolution-Transformer hybrid encoder. Multi-source features are uniformly mapped to a common representation space to obtain the global feature vector: , This eigenvector contains both spectral information and ocean dynamics priors, providing a unified input for subsequent physical constraint modeling and spatiotemporal evolution prediction.

[0024] S2. Physical Prior Modeling and Constraint Module In green tide detection, a pure data-driven deep learning model is vulnerable to sensor noise, observation condition changes, and multi-scene differences, resulting in insufficient generalization ability. To this end, this module introduces a marine optical radiation transfer model (RTM) to establish a physical prior for the green tide feature band and convert it into a physical constraint term for the deep detection network to guide the subsequent deep learning model training. Specifically, it includes the following three parts: 1) Physical prior spectrum modeling. Green tide shows significant spectral characteristics in water remote sensing reflectance curves, with a significant absorption valley in the red band, a clear reflection peak in the near-infrared band, and enhanced reflection characteristics in the green band. To embed this physical prior of spectral shape in the deep model, we introduce a marine optical radiation transfer model (RTM): , where θs represents the solar incidence angle, θv represents the sensor observation angle, C chl represents the chlorophyll concentration, C CDOM represents the colored dissolved organic matter concentration, C TSM represents the suspended matter concentration, represents the geometric correction factor, which is used to correct the influence of the sun and sensor angles on reflectance. a(λ) represents the total absorption coefficient, a water (λ), a CDOM (λ), a TSM (λ) represents the specific absorption coefficient of the corresponding component, and b(λ) represents the backscattering coefficient, which is determined by the water body itself.

[0025] Define the theoretical reflectance vector of green tide pixels in different bands R phy : , where, f RTM represents the marine optical radiation transfer model.

[0026] 2) Spectral shape consistency constraint. Based on the marine optical radiation transfer model (RTM), the theoretical prior feature of the typical spectrum of green tide is constructed. During the training of the deep model, the theoretical prior curve and the model predicted spectrum are constrained, and the difference between the two in shape is minimized to make the prediction results still comply with the physical law under different observation conditions. Spectral shape consistency loss: , where, R pred represents the reflectance predicted by the deep model, R phy represents the theoretical spectrum generated by RTM.

[0027] 3) Linear mixture abundance constraint. In spectral remote sensing, endmember refers to the most representative spectral component, and endmember abundance refers to the proportion of a certain type of endmember in a pixel. Based on the linear mixture abundance constraint of endmember theory, the reflectivity of each pixel is expressed as a linear combination of the reflectivity of multiple types of endmembers.

[0028] , wherein, a i represents endmember abundance, E i represents endmember reflectivity, K represents the number of endmembers.

[0029] The theoretical reflectivity curve of different endmembers is obtained according to the marine optical radiation transfer model, and the predicted prior abundance distribution is obtained by inversion of the linear mixture model a phy : , wherein, R obs represents the reflectivity vector of the pixel at different wavebands, E RTM represents the endmember theoretical spectrum matrix generated by RTM, a = [a1, a2, a3, …, a k ] represents the endmember abundance vector of the pixel, which is a K-dimensional vector, wherein each a k represents the proportion of a certain type of endmember in the pixel.

[0030] In the depth model training, the prediction result is consistent with the theoretical combination result, and the linear mixture abundance constraint loss is introduced: , wherein, a phy represents the theoretical abundance distribution calculated by the linear mixture model, a pred represents the theoretical abundance distribution obtained by depth prediction. In the subsequent depth green tide detection network, the two constraints are optimized together with the main task loss L cls as a regularization term.

[0031] S3. Dynamic spatiotemporal graph construction module In multispectral remote sensing of ocean monitoring, the occurrence and evolution of green tide are driven by multiple factors, including geographical proximity, ocean current transport, wind field effect, and historical outbreak rules. It is difficult to fully depict the dynamic diffusion process of green tide by relying on single-time satellite images or local observation data. Therefore, the core goal of this module is to construct a dynamic spatio-temporal graph that can dynamically reflect the spatial diffusion relationship and temporal evolution rule by taking remote sensing pixels or regions as graph nodes, and provide graph structure prior constraints for subsequent spatio-temporal depth models.

[0032] 1) Spatial adjacency relationship modeling. In the scenario of ocean remote sensing monitoring, the geographical adjacency determines the possibility of spatial diffusion of green tide. Let the remote sensing image be divided into N nodes, the geographical coordinates of each node be P i =( lon i , lat i ), and the Euclidean distance between nodes be d ij =∥p i -p j ∥. First, define the spatial similarity weight between nodes i and j according to the Gaussian kernel function: , wherein is the scale parameter for controlling the range of adjacent action. When two nodes are blocked by coastlines, islands, and other topography, the mask matrix B ij is used for shielding: , wherein B ij =0 indicates that there is physical obstruction, B ij =1 indicates that the open sea is connected.

[0033] 2) Fusion of dynamic driving features. The diffusion of green tide not only depends on the spatial distance, but also is significantly affected by the ocean dynamic process, including the ocean current field and the wind field. Let the node i have the ocean current velocity vector u t (i) at time t , the wind velocity vector be w t (i) , and the displacement vector of node i pointing to j be r ij =p j -p i . The calculation formula of the ocean current driving weight is: , wherein currControlling the attenuation effect of ocean current on distance.

[0034] The wind field driving weight calculation formula: , Where λ wind Control the driving range of wind direction.

[0035] Weighted fusion of ocean current and wind field effect: , Where, β , gamma Adjustable weight.

[0036] 3) Time evolution prior modeling. The occurrence, diffusion and extinction of green tide have significant time sequence rules. Therefore, a time evolution prior graph is constructed using historical outbreak data, and the time dependence of the dynamic adjacency matrix is introduced. Let the historical multi-time green tide distribution sequence be { Y 1, Y 2, …, Y T}, and the number of state transitions between nodes N i→j is obtained. The time transition probability matrix is: , To adapt to the sudden event and real-time update demand, exponential smoothing is adopted:

[0037] Where, represents the transition statistics based on the latest window, and η ∈ (0, 1) is the smoothing coefficient.

[0038] 4) Dynamic adjacency matrix generation. The geographical similarity, dynamic driving and historical prior are fused to construct the dynamic adjacency matrix at time t: , Where α, β, δ are learnable or preset weights, A geo represents the spatial adjacency relationship matrix, A dyn represents the dynamic driving feature matrix, P t represents the time transition probability matrix. Normalize : , Where, D t represents the diagonal matrix, and the output {A t} is the dynamic spatio-temporal graph sequence, which is used as the graph structure prior for subsequent deep spatio-temporal modeling.

[0039] S4. Adaptive graph convolutional feature encoder module After the completion of multi-modal feature extraction and the construction of node global feature vectors and dynamic adjacency matrix {At}, the goal of the adaptive graph convolutional feature encoder module is to design a feature encoder that can be physically interpretable, spatiotemporal consistent, and cross-scale robust propagation on dynamic graphs, providing high-quality node representations for subsequent green tide detection and time series prediction.

[0040] 1) Set the edge conditioning kernel, and the dynamic adjacency matrix has provided a relationship description for each edge (such as geographical proximity, ocean current and wind field projection, historical transition probability, etc.). Let the relationship vector of edge i-j at time t be: , wherein, represents the spatial adjacency relationship, represents the dynamic driving feature, represents the temporal evolution relationship. Through a small kernel generator MLP K Map the relationship vector to the convolution kernel parameters of the edge: , wherein, represents the weight parameter. In the dynamic adjacency A t Perform edge-conditioned aggregation (including self-connection residual term) on: , wherein, σ(·) is a nonlinear activation function, A t represents the dynamic adjacency matrix, B (l) is a self-connection transformation. This mechanism allows different physical situations to have differentiated propagation kernels.

[0041] 2) Spectrum-time domain integration: The goal of spectrum-time domain integration is to simultaneously suppress high-frequency noise in the spatial spectrum domain and improve large-scale consistency, and to capture the green tide diffusion speed and life cycle dependence in the time domain, so that spatial propagation and temporal evolution complement each other. First, perform a spectrum domain learnable filter, then perform a time domain attention / convolution, and finally integrate the two through a learnable gate to form a unified representation. In the spectral domain, we use a polynomial approximation of the Laplacian spectral filter to avoid eigenvalue decomposition of the Laplacian matrix, thereby efficiently implementing high-order neighborhood information aggregation: , wherein, h t is the node feature matrix at time t, is the normalized graph Laplacian, T k (·) is the k order Chebyshev polynomial, which controls the receptive field size of the spectral filter,theta k (t) Spectral filter coefficients are learnable (tuned over time to adapt to non-stationarity). In time domain, we compute an attentional temporal representation for a time window { H t - Tw+1 ,…, H t} to capture the diffused temporal lag and velocity information: , where, W Q , W K , W V is a linear mapping matrix, d q is a dimension scaling factor for keys / queries, T w is the length of the time window. The two representations in spectral and time domains are fused through a learnable gating t : , where pool(·) denotes a global pooling operation, U represents a parameter matrix, σ is sigmoid, and is element-wise multiplication. The gating can be a scalar, a vector, or a vector per node / channel, thus flexibly determining the weight allocation between spectral and time domains.

[0042] 3) Multi-scale propagation. Multi-scale propagation is used to model the regional (super-pixel) diffusion pattern while preserving the pixel-level boundary details, thus balancing local fineness and global consistency. It is implemented in three steps: fine-scale encoding, coarse-scale aggregation, and coarse-fine fusion. We apply edge-conditioned convolution and spectral-time domain filtering to obtain the fine-scale representation , which is mainly responsible for the fine depiction of local boundaries. To obtain the regional-level semantics, we aggregate several neighboring and dynamically similar pixels into super-nodes using a learnable assignment matrix S, where each row s i represents the soft coefficients of assigning pixel i to M super-nodes, satisfying row normalization (softmax). The coarse-scale representation is obtained by aggregation with S and adjacency: , The fine-scale and coarse-scale are aggregated to obtain the final representation that has both details and coarse-scale consistency: , 4) Uncertainty weighted fusion. Different pixels have different observation noise and physical interpretability. In order to avoid the interference of high-noise pixels on the overall loss and consider the spectral consistency constraint and abundance constraint loss, the loss function is set during training: , in, Representative i The uncertainty probability parameter of each pixel is obtained by model learning. represents the loss of spectral consistency, represents the abundance constraint loss, Represents the prediction task loss.

[0043] S5. Temporal Transformer Module Guided by Physical Priors During the occurrence, spread, and disappearance of marine green tides, their spatiotemporal evolution is influenced by a variety of factors, including sea surface temperature (SST), wind speed, tidal flow velocity, ocean eddy dynamics, and changes in chlorophyll concentration. Multi-temporal and multispectral remote sensing observations can provide rich spectral, radiometric, and dynamic characteristics. However, due to the phased, non-stationary, and time-dependent life cycle of green tides, it is often difficult to accurately predict the occurrence and development of green tides by relying solely on static feature extraction. Therefore, this module proposes a physical prior-guided temporal transformer. By explicitly introducing spectral priors, chlorophyll concentration dynamic characteristics, and ocean dynamic gradient constraints in the self-attention mechanism, it achieves high-precision modeling of the entire green tide process.

[0044] Introducing temporal self-attention with physical prior bias: Traditional Transformer ignores the unique spectral and dynamic evolution laws of green tides when modeling long sequence dependencies. We propose a self-attention mechanism guided by physical priors. By explicitly introducing chlorophyll concentration estimation and spectral gradient features as bias terms in the attention calculation, we improve the modeling ability of the green tide life cycle and introduce physical prior bias B. phy , and get the improved formula: , Among them, B phy It consists of two parts: chlorophyll concentration difference bias and spectral reflectance gradient bias. Q, K, and V represent query, key, and value matrices, respectively. d Represents dimension.

[0045] Multi-stage prediction and residual correction, after L After layer-by-layer physical prior-guided Transformer encoding, we obtain the deep feature representation of the time series: , To further improve the prediction accuracy, we introduce a residual correction mechanism based on physical priors. First, we calculate the residual between the predicted spectrum and the endmember-mixed reconstructed spectrum using a linear mixing model: , where R phy represents the physical prior residual term, S represents the predicted spectrum, k represents the total number of spectral bands, k represents the physical weight coefficient. Then, the residual is mapped back to the deep feature space for dynamic correction: , where α is a learnable parameter. Finally, the model predicts the entire process of green tide occurrence, development, and disappearance based on the features obtained by the time-series Transformer and the features obtained by the adaptive convolution.

[0046] S6. Green tide detection and warning module In green tide prediction and warning, due to the existence of noise in multi-source remote sensing data, cloud cover, and ocean dynamics uncertainty, single model prediction has limitations. We propose a green tide detection and warning module to achieve high-precision prediction through multi-source prediction result fusion, adaptive prediction correction based on uncertainty perception, and green tide risk index modeling and warning. First, we fuse the feature vectors generated by the graph adaptive convolution and the physical prior guided time-series Transformer, combine spatial diffusion and time-series evolution features, and jointly model in the probability space. Second, we introduce uncertainty perception correction to improve prediction robustness and physical reasonableness. Finally, we build a green tide risk index model based on the fusion results, covering area, chlorophyll concentration gradient, and historical evolution patterns to generate risk levels and warning information.

[0047] 1) Multi-source prediction result fusion To fully utilize the complementarity of different models in feature extraction, we propose a multi-source prediction fusion method based on graph adaptive convolution network and physical prior guided time-series Transformer. The graph adaptive convolution network is responsible for modeling the spatial diffusion process of green tide, and the PP-Transformer extracts the evolution law in long time series. Finally, we jointly predict and model in the probability space. Let the spatial diffusion prediction extracted by the graph adaptive convolution network be , the time-series prediction output be , and the fusion prediction probability map P f be represented as: , where, a denotes the adaptive fusion coefficient, which is dynamically learned by the prediction accuracy on the validation set, and is the Sigmoid activation function that maps the fusion result to the probability space, and H, W represent the height and width of the remote sensing image, respectively. Through multi-source result fusion, both the local spatial diffusion pattern and the global temporal evolution trend of green tide can be captured, significantly improving the overall prediction accuracy and generalization ability.

[0048] 2) Uncertainty-aware adaptive prediction correction Remote sensing data can introduce large uncertainties due to cloud cover, observation gaps, and the complexity of physical processes, leading to deviations in the prediction results in some local areas. Therefore, an adaptive prediction correction method based on prediction uncertainty is introduced to improve the prediction robustness by dynamically adjusting the contribution of different models in high-uncertainty areas. First, the prediction variance of the graph adaptive convolution network and the physically prior-guided temporal Transformer at each pixel position is calculated: , where Var(·) represents the variance operator, which is used to quantify the uncertainty of the prediction value in multiple samplings or multi-model predictions. According to the uncertainty adaptive allocation of weights: , where w (A) represents the weight of the graph adaptive convolution network model in the final fusion prediction. w (T) represents the weight of the physically prior-guided temporal Transformer model in the final fusion prediction. U (A) , U (T) represents the prediction variance of the corresponding model, and the greater the uncertainty, the lower the weight.

[0049] The final adaptive corrected prediction map is obtained: , where P (A) represents the original prediction probability output by the graph adaptive convolution network, P (T) represents the original prediction probability output by the physically prior-guided temporal Transformer, P f represents the fusion prediction probability map.

[0050] 3) Green tide risk index modeling and early warning After obtaining the high-precision green tide prediction map Pc, the green tide risk index is further introduced to comprehensively consider the green tide coverage area, chlorophyll concentration gradient, and historical evolution pattern, achieving dynamic risk classification and early warning. Let the prediction probability map be P c and determine the green tide coverage mask M by the threshold θ: , where M ij represents the green tide coverage label of pixel (i,j), P c (i,j) represents the value of the fusion prediction probability map at pixel (i,j), and θ represents the prediction probability threshold.

[0051] The green tide risk index is defined as: , where λ1, λ2, λ3 represent the weighted coefficients of the risk index, which are learned through the validation set, A represents the green tide coverage area (statistically obtained through the mask M), ∇C represents the chlorophyll concentration gradient, reflecting the trend of green tide intensity change, H represents the historical evolution trend factor. According to the numerical range of GCRI, different risk levels are set.

[0052] Experimental verification: To verify the effectiveness of the green tide spatio-temporal modeling and early warning system proposed in this study based on multi-source remote sensing, unmanned aerial vehicle images, and physical prior modeling, this paper constructs an experimental platform for field scene simulation and multi-source data driving in a typical nearshore green tide high-risk sea area. Satellite remote sensing images, low-altitude unmanned aerial vehicle high-resolution images, and multi-modal data such as temperature, salinity, dissolved oxygen, pH value, and chlorophyll concentration from ocean dynamic observation buoys, totaling 12,500 samples, completely cover the green tide evolution period (occurrence-diffusion-development-remission), and have rich spatial and temporal dynamic change characteristics. To enhance the adaptability of the experiment to the actual green tide monitoring and emergency warning needs, complex environmental variables such as typhoon disturbance, cloud cover, remote sensing image quality degradation, ocean current mutation, and partial monitoring node data missing are introduced in the design process to comprehensively test the robustness and generalization ability of the model under high uncertainty scenarios.

[0053] The comparative method selects the current representative green tide detection and prediction models, including the basic Transformer, the green tide detection and prediction model GCN-LSTM combined with graph convolution and time series modeling, ASTGCN combined with graph attention mechanism, ST-Transformer based on spatio-temporal attention mechanism, Cross-ModalMatching Transformer (CMMT) based on cross-modal feature alignment, and the AGCN-PPTransformer joint modeling method proposed in this paper. All models are compared under the same data set division (training set: validation set: test set = 6:2:2), consistent optimization strategy and training rounds to ensure the fairness and comparability of the evaluation results.

[0054] The performance evaluation indicators include accuracy (ACC), F1-Score, spatial positioning error, early warning lead time, false positive rate, and model inference delay. The model is evaluated from three dimensions: detection accuracy, timeliness, and deployment feasibility. The experimental results are shown in Figures 2, 3, and Table 1. The proposed method is significantly better than the comparative models in all indicators, fully verifying its effectiveness and advantages in multi-modal feature fusion, physical prior modeling, and green tide evolution trend prediction tasks.

[0055] Table 1 Comparison of data of different methods in six indicators Model name ACC F1 Early warning lead time Positioning error False alarm rate Inference delay Transformer 81.9% 82.3% 1.6 days 11.4 km 9.8% 2.5 min ASTGCN 86.4% 85.0% 2.1 days 10.5 km 8.1% 3.2 min GCN-LSTM 85.5% 84.1% 1.8 days 13.5 km 8.9% 4.0 min CMMT 87.4% 86.8% 2.4 days 11.7 km 7.2% 4.6 min ST-Transformer 83.2% 80.7% 1.5 days 12.1 km 10.4% 3.0 min The method of the present application 91.5% 89.9% 3.1 days 6.3 km 5.1% 3.4 min From Figure 2 , Figure 3 and Table 1, it can be seen that the traditional Transformer, ASTGCN, MMFN, CMMT and ConvLSTM methods have certain prediction ability in the green tide anomaly detection task, but there are still obvious shortcomings in the key performance dimensions. Transformer has certain advantages in global modeling, which can model long-term dependencies in time series, but its ability to model regional spatial details is weak, resulting in large positioning errors and a false positive rate of up to 9.8%. ASTGCN introduces a graph structure modeling approach, which better integrates time series and spatial factor information, and performs better than Transformer in early warning lead time and accuracy, but its static graph structure cannot adapt to the dynamic nature of the rapid evolution of green tide. GCN-LSTM and CMMT improve prediction accuracy through time series mechanisms and multi-modal fusion mechanisms, with F1 values of 84.1% and 86.8%, but the fusion method is relatively shallow, lacking cross-modal consistency alignment and semantic enhancement, resulting in a high false positive rate and a significant increase in inference delay. Transformer has certain advantages in time series modeling, but due to the lack of explicit spatial structure expression, the prediction accuracy is low, with the lowest ACC and F1 values (81.9% and 82.3%).

[0056] In contrast, the method of the present application is based on multi-source multi-spectral feature extraction, physical prior constraint, dynamic spatio-temporal graph modeling, adaptive graph convolution feature coding and physical prior guided time series Transformer, and fully integrates multi-source information such as satellite remote sensing images, unmanned aerial vehicle high-resolution images and ocean dynamic observations, to realize higher level spatial consistency, temporal sensitivity and physical prior constraint modeling capability. In the experiment, the method is superior to other comparative methods in six evaluation indexes: the accuracy reaches 91.5%, the F1 value reaches 89.9%, the green tide early warning lead (3.1 days) and the spatial positioning accuracy (error 6.3 km) are both significantly ahead, the false positive rate is reduced to 5.1%, and the reasoning delay is controlled within 3.4 minutes. The above results fully verify the practicability of the method in green tide detection, boundary identification and evolution trend prediction, and have good engineering deployment prospect and marine emergency response value.

[0057] In order to verify the robustness and generalization ability of the model under different data Dropout ratios, the system tests the accuracy changes of multiple models and presents the results as shown in Figure 3 The figure shows the trend of the accuracy curve of each model with the change of the Dropout ratio (0% - 40%). It can be seen that the accuracy curve corresponding to the method of the present application is always above other comparative models, and can maintain a high performance level even at a high Dropout ratio, especially when the accuracy is still high (about 78%) at a Dropout ratio of 40, and the downward trend is relatively flat during the whole change process. The experimental results show that the method of the present application performs excellently in resisting overfitting and maintaining prediction stability, which helps to improve the tolerance of the model to data noise and reliability in practical applications.

[0058] In order to verify the spatial anomaly distribution perception ability of the model, the system constructs a spatial heat map of the probability of green tide occurrence, and the results are shown in Figure 4 In the figure, different color depths correspond to the probability of green tide occurrence in different monitoring areas, and the deeper the color, the higher the risk of the area. It can be seen that the abnormal distribution predicted by the method of the present application has obvious spatial aggregation, and the high-risk areas are concentrated in the nearshore sea area affected by human activities, the spatial boundary is clear, and the change trend conforms to the historical monitoring data. The results show that the method not only has high prediction accuracy, but also can effectively locate the abnormal area of green tide in the spatial scale, providing fine reference basis for subsequent early warning and response measures.

[0059] Embodiment 2 The embodiment provides a remote sensing image marine green tide monitoring system fusing physical prior and spatio-temporal evolution, comprising: A data acquisition module configured to acquire multi-modal remote sensing monitoring images; a feature extraction module configured to perform multi-modal feature extraction on the acquired images, including spectral reflectance feature extraction, ocean dynamics feature extraction, and feature alignment and unified representation; a physical prior module configured to establish physical priors for green tide feature bands using a marine optical radiation transfer model; a spatio-temporal module configured to construct a dynamic spatio-temporal graph based on the extracted multi-modal features, and obtain a node global feature vector and a dynamic adjacency matrix; an encoding module configured to perform adaptive graph convolution feature encoding based on the physical priors and the dynamic spatio-temporal graph; a modeling module configured to perform time series modeling on the multi-modal features based on the physical priors and a time series Transformer; a prediction module configured to perform marine green tide prediction by fusing the adaptive graph convolution feature encoding and the time series modeling results; and output the prediction results.

[0060] A computer-readable storage medium, wherein a plurality of instructions are stored, the instructions being adapted to be loaded and executed by a processor of a terminal device to implement the remote sensing image marine green tide monitoring method fusing physical priors and spatio-temporal evolution.

[0061] A terminal device, comprising a processor and a computer-readable storage medium, the processor being configured to implement instructions; and the computer-readable storage medium being configured to store a plurality of instructions, the instructions being adapted to be loaded and executed by the processor to implement the remote sensing image marine green tide monitoring method fusing physical priors and spatio-temporal evolution.

[0062] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, any equivalent changes made in the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for monitoring marine green tide in remote sensing images by fusing physical priori and spatio-temporal evolution, characterized in that, The method comprises the following steps: acquiring multi-modal remote sensing monitoring images; performing multi-modal feature extraction on the acquired images, including spectral reflectance feature extraction, ocean dynamics feature extraction, and feature alignment and unified representation; establishing physical priors of green tide feature bands using an ocean optical radiation transfer model; constructing a dynamic spatio-temporal graph based on the extracted multi-modal features to obtain node global feature vectors and a dynamic adjacency matrix; performing adaptive graph convolution feature coding based on the physical priors and the dynamic spatio-temporal graph; performing time series modeling on the multi-modal features based on the physical priors and a time series Transformer; performing green tide prediction by fusing the adaptive graph convolution feature coding and the time series modeling results; outputting the prediction results. 2.The method according to claim 1, wherein, The multi-modal feature extraction on the acquired image comprises spectral reflectance feature extraction, wherein the acquired multi-band original radiation value is R raw , first, atmospheric correction and band resampling are performed to obtain apparent water reflectance R(λ), after obtaining the water reflectance, key vegetation indices are extracted for green tide sensitive bands; for marine dynamics feature extraction, the original marine dynamics data is R d , an interpolation and re-projection method is used to calculate the marine dynamics feature vector, at the same time, in order to dynamically depict the change trend of the green tide with the dynamic condition, a multi-time sequence {X d t} is used to calculate the time sequence gradient; finally, a cross-platform alignment method based on adversarial learning is used, wherein the high-resolution spectral feature extracted by the unmanned aerial vehicle is X u , the low-resolution spectral feature extracted by the satellite is X s , the multi-platform features are unified to a common representation space by minimizing the alignment loss function to realize the consistency of the multi-platform features, and the global feature vector is obtained, wherein the global feature vector comprises spectral information and marine dynamics prior. 3.The method of claim 2, wherein, The physical prior of the green tide characteristic wave band is established by using a marine optical radiation transfer model, including introducing a marine optical radiation transfer model RTM, defining a theoretical reflectivity vector of a green tide pixel in different wave bands R phy Theoretical prior characteristics of a typical green tide spectrum are constructed based on the RTM, and in the depth model training process, the theoretical prior curve and the model predicted spectrum are constrained, the prediction result is still consistent with the physical law under different observation conditions by minimizing the difference in shape, and the spectral shape consistency loss is: ， wherein, R pred representative reflectance predicted by the depth model, R phy representative theoretical spectrum generated by RTM; based on the linear mixing abundance constraint of endmember theory, the reflectance of each pixel is represented as the linear combination of multi-class endmember reflectance; the theoretical reflectance curves of different endmembers and their prior abundance distribution are obtained according to the ocean optical radiation transfer model; then in the depth model training, the predicted result is consistent with the theoretical combination result, and the linear mixing abundance constraint loss is introduced: ， wherein, a phy represents the theoretical abundance distribution calculated by RTM, a pred represents the theoretical abundance distribution predicted by depth.

4. The method according to claim 3, wherein, The extracted multi-modal feature is used to construct a dynamic space-time graph, including taking a remote sensing pixel as a graph node to construct a dynamic space-time graph dynamically reflecting spatial diffusion relationship and time evolution law, wherein the remote sensing image is divided into N nodes, the geographic coordinates of each node are P i =( lon i , lat i ), the Euclidean distance between nodes is d ij =∥p i -p j ∥, the spatial similarity weight between nodes i and j is defined according to the Gaussian kernel function, when two nodes are blocked by the coastline or island topography, the shielding is performed through a mask matrix Bij; the ocean current driving weight and the wind field driving weight are calculated based on the marine dynamic process, and the ocean current and wind field effects are weighted and fused, a time evolution prior graph is constructed using historical outbreak data, and finally the geographic similarity, dynamic driving and historical prior three factors are fused to construct a dynamic adjacency matrix at time t: ， wherein α, β, δ are preset weights, A geo represents a spatial adjacency relation matrix, A dyn represents a dynamics driving feature matrix, P t represents a time sequence transition probability matrix, and is normalized to obtain: ， where, D t represents a diagonal matrix, and the output {A t} is a dynamic spatio-temporal graph sequence, as a graph structure prior for deep spatio-temporal modeling.

5. The method according to claim 4, wherein, The adaptive graph convolution feature coding based on the physical prior and the dynamic space-time graph includes designing a feature encoder that is robust in scale propagation on a dynamic graph, and the relationship vector of edge i-j at time t is generated by a small nuclear generator MLP K The relationship vector is mapped to the convolution kernel parameter of the edge, and the edge is conditioned and aggregated on the dynamic adjacency A t , which is represented as: ， where σ(·) is a nonlinear activation function, A t denotes the dynamic adjacency matrix, B (l) is a self-connection transformation; then spectral domain learnable filtering and time domain attention convolution are performed, where polynomial approximation of Laplacian spectral filtering is adopted in the frequency domain to avoid eigen-decomposition of the Laplacian matrix, and high-order neighborhood information aggregation is realized; in the time domain, an attention-based time sequence representation is calculated for a time window { H t Tw+1 , H t to capture the diffusion time delay and speed information, and finally the two representations obtained in the spectral domain and the time domain are fused through a learnable gate γ t fusion:​ ， wherein pool(·) represents a global pooling operation, U represents a parameter matrix, σ is a sigmoid, and is an element-wise multiplication.

6. The method according to claim 5, wherein, The adaptive graph convolution feature coding based on the physical prior and the dynamic space-time graph further comprises constructing a multi-scale propagation for modeling a regional-level diffusion mode while preserving a pixel-level boundary detail, wherein a boundary-conditioned convolution and a spectral-time domain filtering are applied to obtain a fine-scale representation , and a fine-scale coding is realized; a plurality of adjacent and dynamically similar pixels are aggregated into super nodes, a learnable allocation matrix S is used, and each row s i represents a soft coefficient of pixel allocation to M super nodes, and satisfies row normalization i , a coarse-scale representation and adjacency are obtained by aggregation using S; and the fine-scale and the coarse-scale are aggregated to obtain a final representation having both details and coarse-scale consistency, while the uncertainty is weighted and fused, and there are differences in observation noise and physical interpretability between different pixels, so as to avoid the interference of high-noise pixels on the overall loss in the training process, and a loss function is set ， wherein, represents an uncertainty probability parameter of the i-th pixel, represents a spectral consistency loss, represents an abundance constraint loss, represents a prediction task loss.

7. The method for monitoring marine green tides using remote sensing images that integrates physical priors with spatiotemporal evolution according to claim 6, characterized in that: The method performs temporal modeling of multimodal features based on physical priors and temporal transformers, including using a temporal transformer guided by physical priors to explicitly introduce spectral priors, chlorophyll concentration dynamic characteristics and ocean dynamic gradient constraints in the self-attention mechanism to achieve high-precision modeling of the entire green tide process. The self-attention mechanism guided by physical priors improves the modeling ability of the green tide life cycle by explicitly introducing chlorophyll concentration estimation and spectral gradient characteristics as bias terms in the attention calculation, and introduces physical prior bias B. phy , get the improved formula; after L After the Transformer encoding guided by the physical prior layer, the deep feature representation of the time series is obtained. Finally, the residual correction mechanism based on the physical prior is introduced to calculate the residual between the predicted spectrum and the endmember mixing reconstructed spectrum through the linear mixture model: ， wherein R phy represents the physical prior residual term, S represents the predicted spectrum, k represents the total number of spectral bands, k represents the physical weight coefficient; the residual is mapped back to the deep feature space for dynamic correction: ， where a is a learnable parameter, finally, the model gets the features based on the time series Transformer and the features based on the adaptive convolution The whole process of the occurrence, development and disappearance of the green tide is predicted.

8. The method according to claim 7, wherein, The marine green tide prediction by fusing adaptive graph convolution feature coding and time series modeling results comprises fusing feature vectors generated by the graph adaptive convolution and the physically prior guided time series Transformer, combining spatial diffusion and time evolution characteristics, and jointly modeling in a probability space, wherein the graph adaptive convolution network is responsible for modeling the diffusion process of the green tide in space, the PP-Transformer extracts the evolution law in a long time sequence, and the spatial diffusion prediction extracted by the graph adaptive convolution network is , the output time series prediction is , and the fusion prediction probability map P f is represented as: ， wherein α represents an adaptive fusion coefficient, σ(·) is a Sigmoid activation function, H and W represent the height and width of the remote sensing image, respectively, and the multi-source result fusion is performed to simultaneously capture the local spatial diffusion pattern and the global time series evolution trend of the green tide.

9. The method according to claim 8, wherein, The method of predicting the green tide in the sea by fusing the adaptive graph convolution feature coding and the time series modeling results further comprises introducing an adaptive prediction correction method based on prediction uncertainty, and by dynamically adjusting the contribution degree of different models in a high uncertainty area, first, the prediction variance of the graph adaptive convolution network and the time series Transformer guided by the physical priors at each pixel position is calculated, which is used to quantify the uncertainty of the prediction value in multiple sampling or multi-model prediction, and the weight is adaptively assigned according to the uncertainty, and finally the prediction map after adaptive correction is obtained: ， where P (A) represent the raw prediction probability output by the representative graph adaptive convolutional network, P (T) represent the raw prediction probability output by the physical prior guided time series Transformer, P f represent the fused prediction probability map; introduce a green tide risk index, comprehensively consider the green tide coverage area, chlorophyll concentration gradient and historical evolution mode, realize dynamic risk classification and early warning, wherein the green tide coverage mask M is determined by threshold θ: ， where M ij a green tide coverage label, P c (i,j) represents the value of the fused prediction probability map at pixel (i,j), and the green tide risk index is defined as: ， Wherein, λ1, λ2, λ3 represent the weighted coefficients of risk index, A represents the green tide coverage area, ∇C represents the chlorophyll concentration gradient, H represent the historical evolution trend factor.

10. A remote sensing image marine green tide monitoring system fusing physical priori and space-time evolution, characterized in that, The method comprises the following steps: a data acquisition module configured to acquire multi-modal remote sensing monitoring images; a feature extraction module configured to perform multi-modal feature extraction on the acquired images, including spectral reflectance feature extraction, ocean dynamics feature extraction, and feature alignment and unified representation; a physical prior module configured to establish physical priors of green tide feature bands using an ocean optical radiation transfer model; a spatio-temporal module configured to construct a dynamic spatio-temporal graph based on the extracted multi-modal features to obtain node global feature vectors and a dynamic adjacency matrix; an encoding module configured to perform adaptive graph convolution feature coding based on the physical priors and the dynamic spatio-temporal graph; a modeling module configured to perform time series modeling on the multi-modal features based on the physical priors and a time series Transformer; a prediction module configured to perform green tide prediction by fusing the adaptive graph convolution feature coding and the time series modeling results; and output the prediction results.

Images