Marine environmental change monitoring method, device and medium based on remote sensing imagery

Abstract

The present invention relates to the field of image processing, and discloses a method, device, and medium for monitoring marine environmental changes based on remote sensing images. The method intuitively displays risk distribution in the form of pseudo-color images, providing a powerful tool for decision support. The method includes obtaining the original sea surface temperature image and performing fractional differential enhancement processing to generate an enhanced temperature gradient map, and then extracting the front topology feature matrix, extracting the chlorophyll concentration map using multispectral remote sensing images, constructing a multi-band phase coherence field, generating an ecological feature tensor field, generating a multi-scale fused image, extracting feature contour lines based on the image, calculating the curvature gradient and generating a thermal map, and outputting a dynamic risk map of the marine environment. The present invention integrates multi-source remote sensing data, and through multi-scale analysis and feature enhancement, realizes comprehensive, accurate, and real-time monitoring of marine environmental changes, providing strong technical support for marine environmental management, disaster warning, and ecological protection.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a method, device and medium for monitoring marine environmental changes based on remote sensing images. Background Art

[0002] With global climate change and intensified human activities, the marine environment is undergoing unprecedented dynamic changes, including abnormal sea surface temperatures, imbalanced marine ecosystems, and increased pollutant dispersion. These changes pose serious threats to the health of marine ecosystems, the sustainability of fishery resources, and the socioeconomic security of coastal areas. Timely and accurate monitoring of marine environmental changes is crucial for understanding ocean dynamics, assessing marine ecological risks, and developing scientific management strategies.

[0003] Deficiencies in the existing technology:

[0004] When processing remote sensing images, traditional methods often focus on the extraction of single features, ignoring the intrinsic connections and interactions between different features, making it difficult to fully reflect the complex changes in the marine environment.

[0005] Marine environmental changes have multi-scale characteristics, but existing technologies often lack effective multi-scale fusion strategies when processing multi-source remote sensing data, resulting in information loss or redundancy, affecting the accuracy and reliability of monitoring results.

[0006] Existing technologies mostly focus on static or quasi-static marine environment monitoring, and lack the ability to monitor dynamic changes in the marine environment in real time and conduct risk assessments, making it difficult to meet the needs of rapid response and decision support.

[0007] Therefore, we propose a method, device and medium for monitoring marine environmental changes based on remote sensing images to solve the above problems. Summary of the Invention

[0008] The present invention provides a method, device and medium for monitoring marine environmental changes based on remote sensing images, which intuitively displays risk distribution in the form of pseudo-color images and provides a powerful tool for decision support.

[0009] The first aspect of the present invention provides a method for monitoring marine environmental changes based on remote sensing images, which includes: obtaining an original sea surface temperature image, performing fractional differential enhancement processing on the original image, strengthening the temperature gradient characteristics, and generating an enhanced temperature gradient map; based on the enhanced temperature gradient map, analyzing the connected domain structure and the ring structure in the gradient map, and outputting a front topology feature matrix; extracting chlorophyll concentration maps of each band based on the multispectral remote sensing image, calculating the local phase field and constructing a coherence relationship, and generating a multi-band phase coherence field; performing fusion processing based on the front topology feature matrix and the multi-band phase coherence field, and outputting an ecological characteristic tensor field; obtaining a high-resolution ocean turbidity image, combining it with the ecological characteristic tensor field, performing multi-scale fusion reconstruction, and generating a multi-scale fused image; extracting image feature contour lines based on the multi-scale fused image, calculating the curvature gradient and generating a thermal map, and obtaining a dynamic risk map of the marine environment.

[0010] Optionally, in a first implementation method of the first aspect of the present invention, it includes: obtaining an original sea surface temperature image through a satellite remote sensing system, dividing the image pixel value by the maximum temperature value of the image to generate a dimensionless SST image; based on the dimensionless SST image, determining the fractional order α based on the ocean turbulence scaling law, calculating the fractional differential response along the longitudinal and latitudinal directions respectively, and generating a longitudinal and / or latitudinal differential response map; based on the longitudinal and / or latitudinal differential response map, performing Euclidean norm fusion on the differential responses in the two directions, and outputting a fused gradient feature map; based on the fused gradient feature map, applying the hyperbolic tangent function nonlinearly to enhance the weak gradient features to generate an enhanced temperature gradient map.

[0011] Optionally, in a second implementation method of the first aspect of the present invention, it includes: based on the enhanced temperature gradient map, extracting multi-level level set curves in the continuous function space to generate a level set curve family; based on the level set curve family, calculating the Euler characteristic of each level set curve, counting the number of homotopy equivalence classes, and outputting the connected domain topological invariant; based on the level set curve family, calculating the curvature integral of each curve, identifying closed curves whose curvature integral exceeds the ring threshold, and generating a ring structure identification field; based on the connected domain topological invariant and the ring structure identification field, integrating the topological invariants along the time dimension, spatially aggregating the ring structure density, and outputting the front topological feature matrix.

[0012] Optionally, in a third implementation method of the first aspect of the present invention, it includes: based on multispectral remote sensing images, inverting the chlorophyll concentration of each band based on optical absorption characteristics to generate a multi-band chlorophyll concentration atlas; based on the multi-band chlorophyll concentration atlas, performing a two-dimensional Hilbert transform on each band concentration image, extracting the phase component of the analytical signal, and outputting a multi-band local phase field; based on the multi-band local phase field, calculating the phase difference field between any two bands, quantifying the spatial coherence through complex exponential summation, and generating an inter-band phase coherence matrix; based on the inter-band phase coherence matrix, performing sliding averaging along the time dimension to enhance stability based on the output multi-band phase coherence field.

[0013] Optionally, in a fourth implementation method of the first aspect of the present invention, it includes: based on the front topological characteristic matrix, expanding the matrix elements along the spatial dimension, retaining the continuity of the time dimension, and generating a topological characteristic vector sequence; based on the multi-band phase coherent field, separating the real and imaginary parts of the coherent field, stacking the components along the band dimension, and outputting a phase coherent component cube; based on the topological characteristic vector sequence and the phase coherent component cube, aligning the characteristic vectors and the coherent components according to the spatial position, performing a tensor outer product operation to construct a high-dimensional feature space, and generating an original characteristic tensor field; based on the original characteristic tensor field, determining the dominant scale based on the ocean turbulence energy spectrum, truncating the non-dominant scale characteristic components, and outputting the ecological characteristic tensor field.

[0014] Optionally, in a fifth implementation method of the first aspect of the present invention, it includes: based on the high-resolution ocean turbidity image, selecting a wavelet basis function that matches the ocean turbulence scale, performing continuous wavelet multi-scale decomposition, and generating a turbidity wavelet coefficient set; based on the ecological characteristic tensor field, mapping the feature tensor to the wavelet scale space, retaining the scale component corresponding to the turbulent inertia sub-region, and outputting a scale-constrained feature tensor; based on the turbidity wavelet coefficient set and the scale-constrained feature tensor, performing coefficient fusion in the corresponding scale space, enhancing feature consistency through a phase-preserving algorithm, and generating a modulated wavelet coefficient set; based on the modulated wavelet coefficient set, performing a continuous wavelet inverse transform, restoring the spatial domain image data, and outputting a multi-scale fused image.

[0015] Optionally, in a sixth implementation method of the first aspect of the present invention, it includes: based on the multi-scale fusion image, setting a continuous threshold interval based on the ocean front intensity, extracting contour line clusters in the continuous function space, and generating a feature contour line set; based on the feature contour line set, parameterizing the arc length of each contour line, calculating the distribution of curvature along the curve, and outputting a continuous curvature field; based on the continuous curvature field, calculating the spatial gradient modulus of the curvature field, applying the ocean front propagation model to perform physical constraint enhancement, and generating a gradient intensity field; based on the gradient intensity field, mapping the gradient intensity to the HSV color space, superimposing the geographic coordinates to generate a pseudo-color image, and outputting a dynamic risk map of the marine environment.

[0016] The second aspect of the present invention provides a marine environment change monitoring device based on remote sensing images, which includes: an acquisition module for acquiring the original sea surface temperature image, performing fractional differential enhancement processing on the original image, strengthening the temperature gradient characteristics, and generating an enhanced temperature gradient map; a matrix module for analyzing the connected domain structure and ring structure in the gradient map based on the enhanced temperature gradient map, and outputting the front topology feature matrix; a setting module for extracting chlorophyll concentration maps of each band based on the multispectral remote sensing image, calculating the local phase field and constructing a coherence relationship, and generating a multi-band phase coherence field; a tensor module for performing fusion processing based on the front topology feature matrix and the multi-band phase coherence field, and outputting an ecological characteristic tensor field; a fusion module for acquiring high-resolution ocean turbidity images, combining them with the ecological characteristic tensor field, performing multi-scale fusion reconstruction, and generating a multi-scale fused image; an allocation module for extracting image feature contour lines based on the multi-scale fused image, calculating the curvature gradient and generating a thermal map, and obtaining a dynamic risk map of the marine environment.

[0017] A third aspect of the present invention provides a computer-readable storage medium, which stores instructions that, when executed on a computer, enable the computer to execute the above-mentioned method for monitoring marine environmental changes based on remote sensing images.

[0018] The mechanism of the present invention is as follows:

[0019] Through the three-in-one architecture of physical mechanism embedding → continuous domain operation → tensor cascade: physical drive replaces data training with fluid mechanics / ecological dynamics equations; continuous processing of the entire process avoids discretization errors (from differentiation to curvature calculation); tensor fusion topological features and non-orthogonal coupling of phase coherent fields; realize the paradigm shift of marine remote sensing monitoring from data fitting to physical modeling, and achieve a breakthrough improvement in accuracy and efficiency without the premise of machine learning.

[0020] Beneficial effects:

[0021] Deeply integrate multi-source remote sensing data such as sea surface temperature, chlorophyll concentration, and ocean turbidity, and process the sea surface temperature image through fractional differential enhancement, significantly enhancing the temperature gradient characteristics;

[0022] The continuous wavelet multi-scale decomposition and inverse transform technology is used to achieve the effective fusion and reconstruction of information at different scales, fully considering the multi-scale characteristics of marine environmental changes;

[0023] By extracting the frontal topological characteristic matrix and multi-band phase coherence field, the complex topological structure and phase coherence relationship in the changing ocean environment are revealed;

[0024] The frontal topological characteristics and multi-band phase coherence information were subjected to tensor outer product operations to construct a high-dimensional ecological characteristic tensor field.

[0025] By extracting image feature contour lines, calculating curvature gradients and generating heat maps, real-time monitoring and risk assessment of dynamic changes in the marine environment are achieved, and the risk distribution is intuitively displayed in the form of pseudo-color images. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of an embodiment of a method for monitoring marine environmental changes based on remote sensing images in an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of another embodiment of a method for monitoring marine environmental changes based on remote sensing images in an embodiment of the present invention;

[0028] Figure 3 Schematic diagram of an embodiment of a marine environment change monitoring device based on remote sensing images in an embodiment of the present invention;

[0029] Figure 4 Schematic diagram of an embodiment of a marine environment change monitoring device based on remote sensing images in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The embodiments of the present invention provide a method, device and medium for monitoring marine environmental changes based on remote sensing images, which intuitively displays risk distribution in the form of pseudo-color images, providing a powerful tool for decision support. The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 In one embodiment of the present invention, a method for monitoring marine environmental changes based on remote sensing images includes:

[0032] 101. Temperature gradient feature enhancement processing: Original input: original sea surface temperature image acquired by satellite remote sensing; fractional differential enhancement processing is performed on the original image to strengthen the temperature gradient feature; and an enhanced temperature gradient map is generated;

[0033] It is understandable that the execution subject of the present invention can be a marine environment change monitoring device based on remote sensing images, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0034] It should be noted that the input data used was a thermal infrared image of the waters surrounding the Qinshan Nuclear Power Plant acquired by the Landsat-9 satellite (spatial resolution 100m, timed at 10:25 on August 14, 2022, under summer low tide conditions). The original sea surface temperature (SST) ranged from 31.0–38.0°C, with a weak temperature gradient near the outlet (approximately 0.5°C / km), making it difficult to directly identify the warm discharge diffusion boundary.

[0035] Fractional derivative enhancement step, temperature inversion and normalization:

[0036] The SST is inverted using the radiation transfer equation method to generate a temperature matrix (x,y). To eliminate the dimension effect, the temperature field is normalized:

[0037]

[0038] in =31.0℃, =38.0℃ (measured value from nuclear power plant drain outlet).

[0039] Fractional differential operator design: Use Riemann-Liouville fractional differential definition, order =0.8 (empirically optimized value), enhancing the non-local gradient feature. The operator expression is:

[0040]

[0041] in is the spatial step length (100m), is the neighborhood radius (5 pixels). The operator is dimensionless by using the gamma function.

[0042] Directional gradient synthesis:

[0043] Calculate the fractional gradient in the x and y directions respectively and , the total gradient amplitude is:

[0044]

[0045] This step intensified the warm drainage front (a weak gradient of 0.2 °C at 0.8 km southeast of the drainage outlet).

[0046] Output data: Generate enhanced temperature gradient map.

[0047] 102. Frontal topology feature extraction: Input: enhanced temperature gradient map output by 101; analyze the connected domain structure and ring structure in the gradient map; output the frontal topology feature matrix;

[0048] The input data is the enhanced temperature gradient map output from step 101, with a spatial resolution of 100 meters, covering a 10 km × 10 km area surrounding the Qinshan Nuclear Power Plant. The enhanced temperature gradient ranges from 0 to 1.2°C / km, with the nuclear power plant's thermal discharge diffusion zone exhibiting a high-gradient zone (gradient > 0.8°C / km) and the background sea area gradient < 0.3°C / km.

[0049] Topological feature extraction steps, connected domain structure analysis:

[0050] Binarization segmentation: setting gradient threshold =0.5℃ / km (empirical value), will enhance the gradient Figure 2 The connected domain of the nuclear power drainage front area is 1.8 km 2 (about 1800 pixels), the background noise area is less than 0.05km 2 (about 50 pixels).

[0051] Edge tracking and endpoint connection: Detect all connected edge points in the binary graph, marking endpoints (number of fronts in the neighborhood = 1) and bifurcation points (number of fronts in the neighborhood ≥ 3). Twelve endpoints and three bifurcation points were detected in the main nuclear power front zone.

[0052] Connect adjacent endpoints: If the endpoint distance d is ≤ 300m (3 pixels) and the front direction difference is ≤ 30°, the front is merged into a continuous front. After connection, the number of endpoints is reduced to 6, and the number of broken fronts is reduced by 40%.

[0053] Ring structure detection: Vortex region identification: Calculation of the curvature integral of all closed loops in the connected domain:

[0054]

[0055] in is the tangent direction angle of the front, An annular structure was detected on the southeast side of the nuclear power drainage front, and the curvature integral =5.2 (dimensionless), corresponding to the center of the thermal plume vortex.

[0056] Topological feature matrix construction:

[0057] Define the feature matrix Each row represents a front object, including: connected domain area (km 2): Main front area 1.8km 2 Number of endpoints: 6; Number of bifurcations: 3; Number of closed loops: 1 (vortex ring); Average curvature: 0.15 rad / km for the main front and 0.28 rad / km for the vortex ring;

[0058] Matrix example:

[0059]

[0060] The first row is the main belt of the nuclear power front, and the second row is the background noise.

[0061] Output and effect, topological feature matrix: quantifies the continuity (few endpoints) and vortex characteristics (closed loops) of the front structure, effectively distinguishing warm drainage fronts (large area, containing ring structures) from noise (small area, no rings).

[0062] 103. Multispectral phase field construction: Original input: multispectral remote sensing image; extract chlorophyll concentration map of each band, calculate local phase field and construct coherence relationship; generate multi-band phase coherence field;

[0063] It should be noted that the input data is: Landsat-9 multispectral image (spatial resolution 100m, 10:25 on August 14, 2022), including blue-green light (450–515nm), red edge (705–745nm) and near-infrared (NIR, 850–880nm) bands, covering the 10km×10km sea area around the Qinshan Nuclear Power Plant.

[0064] Chlorophyll concentration map extraction, radiation correction and inversion: perform radiation calibration and atmospheric correction on multispectral images to eliminate the influence of aerosol scattering (using the FLAASH model).

[0065] Based on the red edge and NIR band reflectance, the normalized difference chlorophyll index (NDCI) is calculated:

[0066]

[0067] in is the reflectance (dimensionless). The fitting relationship between NDCI and measured chlorophyll concentration is: ( =0.87), generating a chlorophyll concentration map (range: 0.5–15.0 μg / L).

[0068] Local phase field calculation, phase consistency modeling: Perform a two-dimensional analytical wavelet transform (using the Morlet wavelet basis) on the chlorophyll concentration map of each band to extract the local phase :

[0069]

[0070] in is the wavelet kernel, Indicates convolution operation. Phase value range Radians quantify the direction of local structure in the spatial distribution of chlorophyll (abrupt changes in phase angle at the edge of vortices).

[0071] Coherence relationship construction, multi-band cross-spectrum analysis: calculation of cross-spectral coherence coefficients of blue-green light, red edge, and NIR band phase fields (i,j are band indices):

[0072]

[0073] in The blue-green-NIR coherence coefficient in the vortex southeast of the nuclear power plant outlet (coordinates [6.2 km, 3.8 km]) reaches 0.92, while that in the background sea area is only 0.15–0.3.

[0074] Output and effects, multi-band phase coherence field: three-dimensional tensor (1000×1000 pixels×3 bands), each pixel contains a dimensionless coherence coefficient value.

[0075] Ecological significance: High coherence area ( >0.8) indicates the chlorophyll aggregation vortex driven by warm water discharge, and the phase angle distribution reveals the direction of plume diffusion. <0.3) corresponds to background turbulence, reflecting natural tidal disturbances.

[0076] 104. Ecological feature fusion: Input 1: the frontal topological feature matrix output by 102; Input 2: the multi-band phase coherence field output by 103; Fusion the topological features with the phase coherence field; Output the ecological feature tensor field;

[0077] It should be noted that the input data is: front topology feature matrix (102 outputs): describing the topological structure of the temperature front in the sea area around the Qinshan Nuclear Power Plant, with a matrix dimension of 20×5 (20 front objects × 5 features). Parameter example of the main front object (nuclear power plant thermal discharge): connected domain area 1.8km 2 , 6 endpoints, 3 bifurcation points, 1 closed loop, and an average curvature of 0.15 rad / km.

[0078] Multi-band phase coherence field (103 outputs): Chlorophyll multispectral phase coherence tensor (1000×1000 pixels×3 bands), the blue-green light-NIR coherence coefficient in the southeast vortex area of ​​the nuclear power outlet (coordinates [6.2km, 3.8km]) reaches 0.92.

[0079] Feature fusion step, manifold embedding of topological features: Local linear embedding (LLE) is used to reduce the topological matrix to a 3D latent space, preserving the nonlinear relationship of the front structure:

[0080]

[0081] The weight matrix By minimizing the reconstruction error After dimensionality reduction, the eigenvector of the main front object is [0.82, -0.31, 0.47] (dimensionless).

[0082] Fusion of phase coherence field and topological latent feature outer product:

[0083] The topological latent eigenvector Perform tensor outer product operation with the phase coherence field P to construct a high-order ecological feature tensor:

[0084] (i,j≤1000;k≤3)

[0085] in For the The topological latent characteristic component of the band. After fusion, the tensor value of the nuclear vortex region increases to 0.92×0.82=0.75 (dimensionless), while that of the background sea area decreases to 0.2×(-0.31)=-0.06.

[0086] Non-negative tensor decomposition optimization:

[0087] Redundant information is compressed by non-negative tensor decomposition (NTF), and the decomposition formula is:

[0088]

[0089] Where R=8 is the rank (empirical value), is the core weight. =0.68 (vortex area), =0.05 (background noise).

[0090] Output and effect, ecological characteristic tensor field , combining topological structure and spectral coherence:

[0091] Nuclear power temperature drainage front area: high tensor values ​​(>0.7) clearly mark the vortex boundary, and the deviation from the closed loop position of the temperature front is <50m.

[0092] Background sea area: Low tensor value (<0.1) suppresses random turbulence noise and improves the signal-to-noise ratio by 4.2 times.

[0093] Table: Comparison of key areas before and after feature fusion:

[0094]

[0095] 105. Cross-scale image reconstruction: Original input: high-resolution ocean turbidity image; Input: ecological characteristic tensor field output by 104; Multi-scale fusion reconstruction of the characteristic tensor field and the turbidity image; Generate a multi-scale fused image;

[0096] It should be noted that the input data: high-resolution ocean turbidity image: turbidity image acquired synchronously by the Sentinel-2 satellite (spatial resolution 10m, August 14, 2022). The turbidity range of the sea area around the Qinshan Nuclear Power Plant is 0.5-25.0 NTU, and the warm drainage diffusion area shows a high turbidity belt distribution (>15 NTU).

[0097] Ecological characteristic tensor field (104 outputs): dimension (1000×1000 pixels×8 feature channels), the tensor value of the nuclear vortex area reaches 0.75, and the background sea area is <0.05.

[0098] Reconstruction step, tensor field decomposition and feature alignment:

[0099] Perform non-negative tensor decomposition (NTF) on the ecological tensor field and extract the core eigenvectors (rank R=3):

[0100]

[0101] in is the weight (nuclear power front area =0.68, background area =0.05).

[0102] The decomposed eigenvector V is bilinearly registered with the turbidity image to eliminate the scale difference (100m→10m), and the position error is <0.5 pixel.

[0103] Multi-scale Transformer feature fusion: Design a cross-scale hierarchical Transformer (CHT) module, including: Shallow feature extraction: 3×3 convolution maps the turbidity image to a high-dimensional space (64 channels).

[0104] Multi-scale residual attention: stacking 4 Transformer groups, each containing a cross-scale self-attention (CSA) module, to calculate the spatial weights of features at different scales:

[0105] (d=64, dimensionless)

[0106] Where Q, K are query / key vectors, and vortex boundary response is enhanced (weight > 0.8).

[0107] Feature refinement: Channel-wise attention (CCA) weighted fusion of multi-scale features suppresses background noise (increasing the signal-to-noise ratio by 3.1×).

[0108] Feature fusion and upsampling:

[0109] Fuse the ecological feature vector V with the Transformer output by tensor outer product to generate fusion features :

[0110] (p,q≤10000;r≤64)

[0111] in Transformer output. The fused feature intensity in the nuclear power front region reaches 38.5 (dimensionless), while in the background region it is <2.0.

[0112] Sub-pixel convolution upsampling: Reconstructs a 10m resolution fused image, preserving turbidity texture and ecological feature boundaries.

[0113] 106. Dynamic Risk Visualization: Input: Multi-scale fused image output from 105; extract image feature contours, calculate curvature gradients and generate heat maps; publish dynamic risk maps of the marine environment.

[0114] The input data is a multi-scale fused image (10m resolution, 10,000×10,000 pixels) generated in step 105, covering the waters surrounding the Qinshan Nuclear Power Plant. This image incorporates information about the warm water front (turbidity >15 NTU), chlorophyll eddies (ecological signature tensor >0.7), and the background waters (turbidity <2 NTU, tensor <0.1).

[0115] Visualization implementation steps, image feature contour extraction: The Canny edge detection algorithm is used to extract the gradient amplitude of the fused image and generate a contour vector map. The vortex area southeast of the nuclear power plant outlet (coordinates [6.2km, 3.8km]) forms a closed contour ring with a length of 2.8km and a spacing density of 15 lines / km. 2 (Background sea area <2 lines / km 2 ), marking the warm drainage diffusion boundary and ecologically sensitive areas.

[0116] Curvature gradient calculation: Calculate the curvature κ along the contour sampling points. The formula is:

[0117]

[0118] Where f is the pixel gray value (dimensionless), is a gradient operator. The curvature gradient of the nuclear vortex boundary is 0.45 km. -1(Standard deviation ±0.12), only 0.08 km in the background sea area -1 (Standard deviation ±0.03), quantifying the frontal deformation intensity.

[0119] Thermal map generation and risk grading:

[0120] Map the curvature gradient to thermal values

[0121] ( =0.85 km -1 is the empirical threshold), generate the RGB thermal map:

[0122] High-risk area (red): H>0.8 (curvature gradient >0.4 km -1 ), covering the nuclear power drainage front and the vortex core, with an area of 1.5 km 2 ;

[0123] Medium-risk area (yellow): 0.4 < H ≤ 0.8, indicating the plume diffusion transition zone, with an area of 3.2 km 2 ;

[0124] Low-risk area (blue): H ≤ 0.4, corresponding to the background sea area.

[0125] Output the dynamic risk map of the marine environment: Spatial accuracy: Under a 10m resolution, the positioning error of the warm water discharge front boundary <20m (compared with on-site measurements); Risk grading effectiveness: The coincidence rate of the high-risk area with the turbidity anomaly area (>18.2 NTU) and the high-value area of ecological characteristics (>0.75) reaches 92%, and the false alarm rate in the background sea area <5%;

[0126] Release form: Overlay satellite base maps on the WebGIS platform, supporting real-time zooming and risk value querying.

[0127] Table: Corresponding relationship between risk grading and curvature gradient:

[0128]

[0129] In an embodiment of the present invention, fractional-order differential enhancement processing is adopted, and the Riemann-Liouville fractional-order differential definition is selected. The temperature gradient characteristics, especially the non-local gradient characteristics, are enhanced through specific orders and operator design; the front topological characteristics are comprehensively extracted through connected domain structure analysis and ring structure detection, and a topological feature matrix containing multiple feature information is constructed; based on multispectral remote sensing images, chlorophyll concentration maps are extracted, and the local phase field is calculated and the coherence relationship is constructed through two-dimensional analytical wavelet transform and cross-spectral analysis to generate a multi-band phase coherence field; local linear embedding (LLE) is used to reduce the dimensionality of the topological features, and the topological latent features are fused with the phase coherence field through tensor outer product operation, and non-negative tensor decomposition optimization is used to construct high-order ecological characteristics. Tensor field; decompose the ecological characteristic tensor field, perform bilinear registration with the high-resolution ocean turbidity image, use the cross-scale hierarchical Transformer (CHT) module for multi-scale feature fusion, and reconstruct the high-resolution fused image through sub-pixel convolution upsampling; extract image feature contour lines, calculate the curvature gradient and generate a heat map to realize dynamic risk classification and visualization of the marine environment; it can intuitively display the marine environmental risk areas with high spatial accuracy and good risk classification effectiveness, providing an intuitive and accurate basis for marine environmental management and decision-making, supporting real-time scaling and risk value query, and improving the efficiency and accuracy of risk management.

[0130] See also Figure 2 Another embodiment of the method for monitoring marine environmental changes based on remote sensing images in the embodiment of the present invention includes:

[0131] 201. Temperature gradient feature enhancement: Original input: original sea surface temperature image acquired by satellite remote sensing; fractional differential enhancement processing is performed on the original image to enhance the temperature gradient feature; and an enhanced temperature gradient map is generated;

[0132] Specifically, the original input is obtained as follows: Source: original sea surface temperature (2020T) images directly obtained from satellite remote sensing systems; the original 2020T image matrix is ​​obtained;

[0133] Dimensionless normalization input: original 2020T image matrix; divide the image pixel value by the maximum temperature value of the image; generate a dimensionless 2020T image (pixel value range [0,1]);

[0134] Anisotropic fractional differentials: Input: dimensionless 2020T image; (a) Determine the fractional order α (0.5<α<1.5) based on the ocean turbulence scaling law; (b) Calculate the fractional differential response along the longitudinal and latitudinal directions;

[0135] Generate warp and / or latitudinal differential response maps;

[0136] Gradient feature fusion: Input: longitudinal and / or latitudinal differential response map; perform Euclidean norm fusion on the differential responses in the two directions and output a fused gradient feature map;

[0137] Feature enhancement output: Input: fused gradient feature map; applying hyperbolic tangent function nonlinearly to enhance weak gradient features; generating enhanced temperature gradient map (input to 202 front topology feature extraction).

[0138] It should be noted that the original input data source is the sea surface temperature (SST) image of the East China Sea region on August 1, 2023 (spatial resolution 1 km) obtained by the SLSTR sensor of the Sentinel-3 satellite.

[0139] Example of the original data matrix (3×3 local grid, unit: °C):

[0140] | 28.5 29.1 28.7 |

[0141] | 27.8 28.9 29.4 |

[0142] | 26.5 27.2 28.0 |

[0143] Dimensionless normalization processing is performed to calculate the maximum temperature value of the entire image: =31.6℃ (obtained from actual data). Normalization formula: . Normalized matrix (retain 3 decimal places):

[0144] | 0.902 0.921 0.908 |

[0145] | 0.880 0.915 0.930 |

[0146] | 0.839 0.861 0.886 |

[0147] Anisotropic fractional differentials, determining order :Based on the analysis of turbulence energy spectrum in the East China Sea, we select =1.2 (conforms to 0.5< <1.5 turbulence scaling law requirement).

[0148] Meridional differential (along the latitude direction): using the Grunwald-Letnikov fractional differential definition:

[0149]

[0150] Calculate the differential response of the central pixel (2,2) along the longitudinal direction (row direction) (take m=2 neighborhood) is -0.108;

[0151] Latent differential (along the longitude direction): Calculate the response of the central pixel (2,2) along the latitudinal (column direction) in the same way is 0.072;

[0152] Gradient feature fusion, Euclidean norm fusion longitude / latitude differential response: G= = ;

[0153] The gradient intensity of the central pixel after fusion is 0.130.

[0154] Feature enhancement output, nonlinear enhancement: Apply the hyperbolic tangent function (tanh) to strengthen weak gradient features:

[0155]

[0156] Enhancement effect: Original gradient strength 0.130 → enhanced to 0.572 (a 340% increase). Weak gradient regions (G < 0.05) are amplified, while strong gradient regions (G > 0.3) are saturated to around 1.0, preserving nonlinear boundary features.

[0157] 202. Frontal topology feature extraction: Input: enhanced temperature gradient map output from 201; analyze the connected domain structure and ring structure in the gradient map; output the frontal topology feature matrix;

[0158] Specifically, the level set continuous extraction method includes the following steps: input: the enhanced temperature gradient map generated by 201; extracting multi-level level set curves in the continuous function space; generating a level set curve family (a set of closed curves with different gradient thresholds);

[0159] Connected domain structure analysis: Input: level set curve family; (a) Calculate the Euler characteristic of each level set curve; (b) Count the number of homotopy equivalence classes; Output the connected domain topological invariant (quantized gradient peak cluster structure);

[0160] Ring structure detection: Input: level set curve family; (a) calculate the curvature integral of each curve; (b) identify closed curves whose curvature integral exceeds the ring threshold; generate a ring structure identification field (marking the location of vortex / front ring);

[0161] Construction of the spatiotemporal feature matrix: Input 1: connected domain topological invariants; Input 2: ring structure identification field; (a) integration of topological invariants along the time dimension; (b) spatially aggregated ring structure density; Output: frontal topological feature matrix (input into 204 ecological feature fusion).

[0162] It should be noted that the input data for the East China Sea front topological feature extraction is: enhanced temperature gradient map (3×3 local example, dimensionless gradient intensity):

[0163] | 0.12 0.58 0.23 |

[0164] | 0.45 0.92 0.67 |

[0165] | 0.06 0.31 0.18 |

[0166] Level set continuous extraction, multi-level threshold setting: set the threshold interval {0.3, 0.6, 0.9} according to the gradient range [0, 1].

[0167] Level set curve generation: Threshold = 0.3: A closed curve encloses pixels with gradients ≥ 0.3 (the center of the example). Threshold = 0.6: The curve narrows to the center pixel (2,2) and its neighbors (1,2) and (2,3). Threshold = 0.9: Only the center pixel (2,2) is enclosed (a single-point closed curve). Output: A family of three-level level set curves.

[0168] Connected domain structure analysis and Euler characteristic calculation (quantified topology): A threshold of 0.3 indicates a simply connected domain (no holes), with an Euler number of χ = 1 (formula: χ = V - E + F, where V is vertices, E is edges, and F is faces). A threshold of 0.6 indicates a biconnected domain (split center region), with an Euler number of χ = 2.

[0169] Homotopy equivalence class statistics: Threshold 0.3 curve: 1 connected domain → 1 homotopy class. Threshold 0.6 curve: 2 connected domains → 2 homotopy classes. Output: Topological invariant matrix [ (0.3, χ=1, homotopy class=1), (0.6, χ=2, homotopy class=2) ].

[0170] Ring structure detection, curvature integral calculation (identification of vortex / frontal ring): parameterize the arc length of the threshold 0.6 curve (closed polygon) and calculate the curvature integral The curvature of the central pixel (2,2) is κ = 0.85 (high gradient), and the integral value is 2.71 (dimensionless). Ring threshold = 2.0 → If the threshold is exceeded, it is marked as a ring structure. Output: Ring structure identification field (binary matrix, the central pixel is marked as 1, and the rest are 0).

[0171] Construct a spatiotemporal feature matrix and integrate topological invariants: Calculate the mean Euler number along the time dimension (assuming multi-day data) (single-day data is omitted for the example). Annular density aggregation: Calculate the proportion of annular pixels within the spatial grid (density = 1 / 9 ≈ 0.111 in a 3×3 grid, for example). Output the frontal topological feature matrix.

[0172] 203. Multispectral Phase Field Construction: Original input: multispectral remote sensing image; extract chlorophyll concentration maps of each band, calculate local phase fields and construct coherence relationships; generate multi-band phase coherence fields;

[0173] Specifically, chlorophyll concentration inversion: original input: multispectral remote sensing image (visible light and near-infrared bands); inversion of chlorophyll concentration in each band based on optical absorption characteristics; generation of a multi-band chlorophyll concentration atlas (spatial dimension + band dimension);

[0174] Local phase field calculation: Input: multi-band chlorophyll concentration atlas; (a) perform a 2D Hilbert transform on each band concentration map; (b) extract the phase component of the analytical signal; output a multi-band local phase field (complex matrix);

[0175] Cross-band coherence relationship construction: Input: multi-band local phase field; (a) calculate the phase difference field between any two bands; (b) quantify the spatial coherence by complex exponential summation; generate the inter-band phase coherence matrix;

[0176] Integration of spatiotemporal coherence fields: Input: inter-band phase coherence matrix; Sliding average along the time dimension to enhance stability; Output: multi-band phase coherence field (input into 204 ecological feature fusion).

[0177] It should be noted that the multispectral image (visible and near-infrared bands) of the East China Sea region on August 1, 2023, was obtained by the SLSTR sensor of the Sentinel-3 satellite: Construction of the East China Sea chlorophyll multispectral phase field and inversion of chlorophyll concentration: Input data: SLSTR multispectral image (bands: B4 665nm, B5 708nm, B6 753nm, B7 778nm).

[0178] Inversion method: Based on the band absorption characteristics, the OC4Me algorithm is used to calculate the chlorophyll concentration:

[0179] = +

[0180] Output example (3×3 local grid, unit: mg / m 3 ):

[0181] | Band B4: 0.82 0.75 0.68 | | Band B5: 1.05 0.98 0.91 |

[0182] | Band B6: 1.12 1.24 1.36 | | Band B7: 0.93 0.87 0.81 |

[0183] Local phase field calculation, two-dimensional Hilbert transform: chlorophyll concentration map for each band Compute the analytical signal: hilbert ;

[0184] Phase component extraction: =arg ;

[0185] Example (center pixel B5 band): real part: 0.98 → imaginary part: 0.21 (after Hilbert transform); phase: =arctan(0.21 / 0.98)≈0.211 radians;

[0186] Cross-band coherence relationship construction, phase difference field calculation: Calculate the spatial phase difference for the band combination (B4-B5): = − ;

[0187] Coherence quantification: Calculate spatial coherence by summing complex exponentials:

[0188]

[0189] Example (B4-B5 combination): Phase difference matrix (3×3 local):

[0190] | 0.15 0.12 0.10 |

[0191] | 0.18 0.20 0.22 |

[0192] | 0.25 0.30 0.35 |

[0193] Coherence: ≈0.892 (high coherence indicates similar spatial distribution patterns of chlorophyll).

[0194] Integration of spatiotemporal coherence fields, sliding average enhancement: Integrate the band coherence matrix of 5 consecutive days (July 28-August 1) and take the average along the time dimension:

[0195]

[0196] Output example (B4-B5 integration value): 0.85 (after stability improvement).

[0197] 204. Ecological feature fusion: Input 1: frontal topological feature matrix output by 202; Input 2: multi-band phase coherence field output by 203; fuse the topological features with the phase coherence field; output the ecological feature tensor field;

[0198] Specifically, the topological feature vectorization: input: the front topological feature matrix output by 202; expand the matrix elements along the spatial dimension, retaining the continuity of the time dimension; generate a topological feature vector sequence (one-dimensional expression of space and time);

[0199] Phase coherent field reformatting: Input: Multi-band phase coherent field output by 203; (a) Separate the real and imaginary parts of the coherent field; (b) Stack the components along the band dimension; Output: Phase coherent component cube (real / imaginary dual-channel 3D data volume);

[0200] Tensor product fusion operation: Input 1: topological feature vector sequence; Input 2: phase coherent component cube; (a) align the feature vectors and coherent components according to spatial position; (b) perform tensor outer product operation to construct a high-dimensional feature space; generate the original feature tensor field;

[0201] Physically constrained dimensionality reduction: Input: original feature tensor field; (a) determine the dominant scale based on the ocean turbulence energy spectrum; (b) truncate the non-dominant scale feature components; output: ecological feature tensor field (input into 205 cross-scale image reconstruction).

[0202] It should be noted that the data for the East China Sea region acquired by the Sentinel-3 satellite on August 1, 2023 (input the frontal topology characteristic matrix from step 202 and the multi-band phase coherence field from step 203) are as follows: Input data, frontal topology characteristic matrix (3×3 spatial grid, dimensionless):

[0203] | 0.12 0.58 0.23 |

[0204] | 0.45 0.92 0.67 |

[0205] | 0.06 0.31 0.18 |

[0206] (The matrix element values ​​represent the strength of the topological invariant and the Euler characteristic is the normalized result);

[0207] Multi-band phase coherence field (3-band complex matrix, spatial 3×3 grid):

[0208] Real part Re and imaginary part Im of the coherent field in band B4-B5:

[0209] Re: | 0.85 0.78 0.72 | Im: | -0.32 -0.28 -0.25 |

[0210] Topological eigenvectorization expands the matrix elements along the spatial dimensions, preserving temporal continuity:

[0211] =[0.12,0.58,0.23,0.45,0.92,0.67,0.06,0.31,0.18] (one-dimensional space-time vector, length 9);

[0212] Phase coherent field renormalization, separation of real and imaginary coherent field parts, and stacking along the band:

[0213] Real cube Re_cube (band × space):

[0214] | B4-B5: 0.85 0.78 0.72 |

[0215] Imaginary cube Im_cube (band × space):

[0216] | B4-B5: -0.32 -0.28 -0.25 |

[0217] Output three-dimensional data volume: Phase_cube = [Re_cube; Im_cube] (real / imaginary dual channels).

[0218] Tensor product fusion operation, spatial alignment: topological vector Align with Phase_cube by grid position. Tensor outer product: construct high-dimensional feature space Z= Phase_cube;

[0219] Example of the outer product of the central pixel (2,2): =0.92×[0.78,−0.28]=[0.718,−0.258]; generate the original feature tensor field (9 spatial points × 2 channels × 1 band combination);

[0220] Physical constraint dimensionality reduction, turbulence energy spectrum constraint: based on the energy spectrum of the inertial sub-region of ocean turbulence E(k)∝ , retaining the scale range 10–100 km (corresponding to wave number k∈[ , ] ).

[0221] Scale truncation: Calculate the energy of each scale component of the tensor field , retain the dominant scale ( k = 5× ) corresponds to the component. Non-dominant scale ( k< ) components are set to zero, and >85% of the turbulent energy is retained after dimensionality reduction.

[0222] Output ecological characteristic tensor field (9×2×1 data volume after dimensionality reduction).

[0223] 205. Cross-scale image reconstruction: Original input: high-resolution ocean turbidity image; Input: ecological characteristic tensor field output from 204; Multi-scale fusion reconstruction of the characteristic tensor field and the turbidity image; Generate a multi-scale fused image;

[0224] Specifically, the turbidity image wavelet decomposition is as follows: original input: high-resolution (≤10m) ocean turbidity image; (a) wavelet basis functions matching the ocean turbulence scale are selected; (b) continuous wavelet multiscale decomposition is performed; and a turbidity wavelet coefficient set (scale-space 3D data volume) is generated.

[0225] Characteristic tensor scale projection: Input: ecological characteristic tensor field output by 204; (a) map the characteristic tensor to wavelet scale space; (b) retain the scale component corresponding to the turbulent inertial subregion; output scale-constrained characteristic tensor;

[0226] Wavelet domain feature modulation: Input 1: turbidity wavelet coefficient set; Input 2: scale-constrained feature tensor; (a) Coefficient fusion in the corresponding scale space; (b) Enhance feature consistency through phase-preserving algorithm; Generate a modulated wavelet coefficient set;

[0227] Continuous wavelet reconstruction: Input: modulated wavelet coefficient set; (a) perform continuous wavelet inverse transform; (b) restore spatial domain image data; output multi-scale fused image (input to 206 dynamic risk visualization).

[0228] It should be noted that the data for the East China Sea region on August 1, 2023, obtained by the SLSTR sensor of the Sentinel-3 satellite (the input is the ecological characteristic tensor field and high-resolution turbidity image output in step 204): Input data:

[0229] High-resolution turbidity image (3×3 local grid, spatial resolution 10m, dimensionless turbidity value):

[0230] | 0.15 | 0.38 | 0.22 |

[0231] | 0.42 | 0.91 | 0.65 |

[0232] | 0.08 | 0.29 | 0.17 |

[0233] Ecological characteristic tensor field (from step 204, 3×3 grid, real / imaginary dual channels):

[0234] Real part: [0.72, 0.85, 0.68];

[0235] Imaginary part: [-0.31, -0.45, -0.28];

[0236] Turbidity image wavelet decomposition, wavelet basis selection: Morlet wavelet basis function (similar to the ocean turbulence inertial sub-region scaling law) Multi-scale decomposition: Perform continuous wavelet transform on the turbidity image to generate a scale-space three-dimensional coefficient set. Scale range: 10m (detail layer), 100m (transition layer), 1km (background layer). Example of coefficients for the center pixel (2,2): 10m scale: =0.52 (high gradient feature); 100m scale: =0.31;

[0237] Characteristic tensor scale projection, turbulence energy spectrum constraint: retain the inertial sub-zone scale (10–100m), cut off large-scale circulation (>1km) and microscale turbulence (<10m). Tensor mapping: project the ecological characteristic tensor to the wavelet scale space: the real part is projected at the 100m scale value: =0.85× =0.79 (k is the wave number); imaginary projection value: =-0.45× =-0.42;

[0238] Wavelet domain feature modulation, coefficient fusion: fusion of turbidity coefficient and feature tensor at 100m scale space: = ( =0.7, =0.3); central pixel modulation coefficient: =0.7×0.31+0.3×0.79=0.454; Phase hold: imaginary phase angle θ=arg( ) is used to enhance the consistency of the vortex edge (Kuroshio frontal ring).

[0239] Continuous wavelet reconstruction and inverse transform restoration: Performs an inverse continuous wavelet transform on the modulated coefficient set to restore the spatial domain image. Reconstruction results (change in central pixel turbidity): Original: 0.91 → Reconstructed: 0.87; Feature enhancement: The gradient of the front edge increases by 40% (0.38 → 0.42), and the texture of the vortex core is clearer (curvature details are enhanced).

[0240] 206. Dynamic Risk Visualization: Input: Multi-scale fused image output from 205; extract image feature contours, calculate curvature gradients and generate heat maps; publish dynamic risk maps of the marine environment.

[0241] Specifically, continuous contour extraction: Input: multi-scale fused image output by 205; (a) set continuous threshold interval based on ocean front intensity; (b) extract contour clusters in continuous function space; generate characteristic contour set (parameterized curve set);

[0242] Curvature field calculation: Input: characteristic contour set; (a) arc length parameterization for each contour line; (b) calculate the distribution of curvature along the curve; output continuous curvature field (scalar field);

[0243] Curvature gradient generation: Input: continuous curvature field; (a) Calculate the spatial gradient modulus of the curvature field; (b) Apply the ocean front propagation model to enhance physical constraints; Generate the gradient intensity field (risk quantification field);

[0244] Dynamic risk heat map rendering: Input: gradient intensity field; (a) Map the gradient intensity to H20V color space; (b) Overlay geographic coordinates to generate a pseudo-color image; Output: dynamic risk map of the marine environment (final result).

[0245] It should be noted that the multi-scale fusion image of the East China Sea region acquired by the Sentinel-3 satellite on August 1, 2023 (input from step 205) is as follows: Input data:

[0246] Multi-scale fused image (3×3 local grid, dimensionless fusion eigenvalue):

[0247] | 0.18 0.52 0.25 |

[0248] | 0.48 0.95 0.73 |

[0249] | 0.12 0.35 0.21 |

[0250] Continuous contour extraction, threshold interval setting: Based on the front intensity range [0,1], set the gradient threshold to {0.3, 0.6, 0.9}. Contour generation (arc-length parameterized curve): Threshold 0.3: Closes the curve around pixels with a value ≥ 0.3 (covering the center area). Threshold 0.6: Constricts the curve to pixels (1,2), (2,2), and (2,3). Threshold 0.9: Only the center pixel (2,2) is enclosed (single-point curve).

[0251] Curvature field calculation, arc length parameterization: For threshold 0.6, the curve (polygon) is parameterized with arc length s.

[0252] Curvature formula: κ(s)= (r is the position vector); the curvature of the central pixel (2,2): κ = 0.82 (high curvature front). Full curve integral: =2.65 (dimensionless).

[0253] Curvature gradient generation, spatial gradient calculation:

[0254] = ;

[0255] Central pixel gradient modulus: =0.38.

[0256] Physical constraint enhancement: Application of front propagation model v∝ (v is the front moving speed), zooming in on the gradient of the high-risk area: =1.5× =0.57;

[0257] Dynamic risk heat map rendering, H20V color mapping: low risk (G<0.2): light blue (H=180°, V=80%); medium risk (0.2≤G<0.4): yellow (H=60°, V=90%); high risk (G≥0.4): dark red (H=0°, V=100%);

[0258] Center pixel rendering: G=0.57 → dark red (indicating the vortex core of the Kuroshio front).

[0259] Geographic overlay: Bind the pseudo-color heat map to the geographic coordinates of the East China Sea and mark the location of the vortex at 28.5°N and 125.2°E.

[0260] In an embodiment of the present invention, multi-dimensional information such as temperature gradient characteristics, front topology characteristics and multispectral phase field is comprehensively utilized, and a more comprehensive representation of marine environmental characteristics is constructed through complex feature extraction and fusion algorithms; the fractional order is determined based on the ocean turbulence scaling law, and the fractional differential response is calculated along the longitudinal and latitudinal directions respectively. This anisotropic processing method fully considers the anisotropic characteristics of the marine environment; through methods such as level set continuous extraction, connected domain structure analysis and ring structure detection, the enhanced temperature gradient map is subjected to topological feature analysis, the gradient peak cluster structure is quantified and the vortex / front ring position is identified; the chlorophyll concentration is inverted using multispectral remote sensing images, and the local phase field and cross-band coherence relationship are calculated to construct a multi-band phase coherence field; the dominant scale is determined based on the ocean turbulence energy spectrum, the characteristic tensor field is subjected to dimensionality reduction processing, and the characteristic tensor field and turbidity image are reconstructed by multi-scale fusion; by extracting image feature contour lines, calculating curvature gradients and generating heat maps, a visual display of the dynamic risks of the marine environment is achieved.

[0261] The above describes the method for monitoring ocean environment changes based on remote sensing images in an embodiment of the present invention. The following describes the apparatus for monitoring ocean environment changes based on remote sensing images in an embodiment of the present invention. Figure 3In one embodiment of the present invention, an embodiment of a marine environment change monitoring device based on remote sensing images includes: an acquisition module 301, used to acquire an original sea surface temperature image, perform fractional differential enhancement processing on the original image, enhance the temperature gradient characteristics, and generate an enhanced temperature gradient map; a matrix module 302, used to analyze the connected domain structure and ring structure in the gradient map based on the enhanced temperature gradient map, and output a front topology feature matrix; a setting module 303, used to extract chlorophyll concentration maps of each band based on the multispectral remote sensing image, calculate the local phase field and construct a coherence relationship, and generate a multi-band phase coherence field; a tensor module 304, used to perform fusion processing based on the front topology feature matrix and the multi-band phase coherence field, and output an ecological feature tensor field; a fusion module 305, used to acquire a high-resolution ocean turbidity image, combine it with the ecological feature tensor field, perform multi-scale fusion reconstruction, and generate a multi-scale fused image; an allocation module 306, used to extract image feature contour lines based on the multi-scale fused image, calculate the curvature gradient and generate a thermal map, and obtain a dynamic risk map of the marine environment.

[0262] In this embodiment of the present invention, multi-dimensional information, including sea surface temperature gradient characteristics (represented by enhancement processing and a matrix of frontal topology characteristics), multi-band chlorophyll concentration phase coherence fields, and high-resolution ocean turbidity images, is fused. Information from different dimensions reflects different aspects of the marine environment. This fusion approach can more comprehensively and accurately capture the changing characteristics of the marine environment. Compared with monitoring methods that rely on single-dimensional information, it significantly improves monitoring accuracy and reliability, helping to more accurately grasp the dynamic changes in the marine environment. The acquisition module uses fractional differential enhancement to process the raw sea surface temperature image to enhance the temperature gradient characteristics. Fractional differential, as an emerging signal processing method, can more effectively extract edge and detail information in images than traditional image processing methods, making temperature gradient characteristics more prominent. This helps to more clearly identify temperature variation areas in the marine environment, providing more accurate basic data for subsequent feature analysis and risk assessment. The fusion module performs multi-scale fusion reconstruction to generate a multi-scale fused image. Multi-scale fusion leverages the complementarity of image information at different scales, organically combining feature information from different scales and enhancing the image's ability to depict changes in the marine environment. This multi-scale fused imagery can more intuitively demonstrate changes in the marine environment at different spatial scales, providing richer and more comprehensive information for marine environmental monitoring and research. The allocation module extracts image feature contours, calculates curvature gradients, and generates heat maps, ultimately producing a dynamic marine environmental risk map. Curvature gradients reflect the changing trends and severity of marine environmental characteristics. By converting these into heat maps, they can visually demonstrate the risk levels of different areas within the marine environment. This intuitive risk assessment method helps decision-makers quickly understand changes in the marine environment and potential risks, allowing them to take timely countermeasures and management measures, thus possessing significant practical application value.

[0263] above Figure 3 The ocean environment change monitoring device based on remote sensing images in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The ocean environment change monitoring device based on remote sensing images in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0264] Figure 4 Schematic diagram of the structure of the ocean environment change monitoring device based on remote sensing images provided by an embodiment of the present invention. The ocean environment change monitoring device based on remote sensing images 400 may have relatively large differences due to different configurations or performances. The device 400 includes a transmitter 401, a receiver 402 and a processor 403. The processor 403 may also be a controller. Figure 4denoted as “controller / processor 403 ”. Optionally, the device 400 may further include a modem processor 405 , wherein the modem processor 405 may include an encoder 406 , a modulator 407 , a decoder 408 , and a demodulator 409 .

[0265] In one example, transmitter 401 conditions (e.g., performs analog-to-analog conversion, filtering, amplification, and frequency upconversion) the output samples and generates an uplink signal, which is transmitted via an antenna to an access network device. On the downlink, the antenna receives the downlink signal transmitted by the access network device. Receiver 402 conditions (e.g., performs filtering, amplification, frequency downconversion, and digitization) the signal received from the antenna and provides input samples. Within modem processor 405, encoder 406 receives traffic data and signaling messages to be transmitted on the uplink and processes them (e.g., formats, encodes, and interleaves them). Modulator 407 further processes (e.g., performs symbol mapping and modulation) the encoded traffic data and signaling messages and provides output samples. Demodulator 409 processes (e.g., demodulates) the input samples and provides symbol estimates. Decoder 408 processes (e.g., deinterleaves and decodes) the symbol estimates and provides decoded data and signaling messages for transmission to device 400. The encoder 406, modulator 407, demodulator 409, and decoder 408 can be implemented by the combined modem processor 405. These units perform processing based on the radio access technology (e.g., LTE and other evolved system access technologies) used by the radio access network. It should be noted that when the device 400 does not include the modem processor 405, the above functions of the modem processor 405 can also be performed by the processor 403.

[0266] Processor 403 controls and manages the actions of device 400, and is configured to execute the processing performed by device 400 in the above-described embodiments of the present disclosure. For example, processor 403 is also configured to execute the various steps of the sending device or receiving device in the above-described method embodiments, and / or other steps of the technical solutions described in the embodiments of the present disclosure.

[0267] Furthermore, the device 400 may further include a memory 404 , and the memory 404 is used to store program codes and data for the device 400 .

[0268] It is understandable that Figure 4 Only a simplified design of the device 400 is shown. In actual applications, the device 400 may include any number of transmitters, receivers, processors, modem processors, memories, etc., and all devices that can implement the embodiments of the present disclosure are within the scope of protection of the embodiments of the present disclosure.

[0269] The present invention also provides a marine environment change monitoring device based on remote sensing images. The marine environment change monitoring device based on remote sensing images includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the marine environment change monitoring method based on remote sensing images in the above-mentioned embodiments.

[0270] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the method for monitoring marine environmental changes based on remote sensing images.

[0271] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0272] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0273] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring marine environmental changes based on remote sensing images, characterized in that: The method for monitoring marine environmental changes based on remote sensing images includes: Obtain the original sea surface temperature image, perform fractional differential enhancement on the original image, strengthen the temperature gradient characteristics, and generate an enhanced temperature gradient map; According to the enhanced temperature gradient map, the connected domain structure and ring structure in the gradient map are analyzed, and the front topology feature matrix is ​​output; Extract chlorophyll concentration maps of each band from multispectral remote sensing images, calculate local phase fields and construct coherence relationships to generate multi-band phase coherence fields; Based on the front topological characteristic matrix and the multi-band phase coherence field, fusion processing is performed to output the ecological characteristic tensor field, including: According to the front topological characteristic matrix, the matrix elements are expanded along the spatial dimension, the continuity of the time dimension is retained, and a topological characteristic vector sequence is generated; According to the multi-band phase coherent field, the real part and the imaginary part of the coherent field are separated, the components are stacked along the band dimension, and the phase coherent component cube is output; Based on the topological eigenvector sequence and the phase coherent component cube, the eigenvectors and coherent components are aligned according to spatial positions, and the tensor outer product operation is performed to construct a high-dimensional feature space and generate the original feature tensor field; Based on the original characteristic tensor field, the dominant scale is determined based on the ocean turbulence energy spectrum, the non-dominant scale characteristic components are truncated, and the ecological characteristic tensor field is output; Obtain high-resolution ocean turbidity images, combine them with ecological characteristic tensor fields, perform multi-scale fusion reconstruction, and generate multi-scale fused images; Based on the multi-scale fusion image, the image feature contour lines are extracted, the curvature gradient is calculated and the thermal map is generated to obtain the dynamic risk map of the marine environment.

2. The method for monitoring marine environmental changes based on remote sensing images according to claim 1, characterized in that: include: The original sea surface temperature image obtained by the satellite remote sensing system is divided by the image pixel value by the maximum temperature value of the image to generate a dimensionless SST image; Based on the dimensionless SST image, the fractional order α is determined based on the ocean turbulence scaling law, and the fractional differential response is calculated along the longitudinal and latitudinal directions respectively to generate the longitudinal and / or latitudinal differential response maps; According to the longitudinal and / or latitudinal differential response maps, the differential responses in the two directions are fused using the Euclidean norm, and a fused gradient feature map is output; According to the fused gradient feature map, the hyperbolic tangent function is applied to nonlinearly enhance the weak gradient features to generate an enhanced temperature gradient map.

3. The method for monitoring marine environmental changes based on remote sensing images according to claim 1, characterized in that: include: According to the enhanced temperature gradient map, multi-level level set curves are extracted in the continuous function space to generate a level set curve family; According to the level set curve family, the Euler characteristic of each level set curve is calculated, the number of homotopy equivalence classes is counted, and the topological invariant of the connected domain is output; Based on the level set curve family, the curvature integral of each curve is calculated, and the closed curves whose curvature integral exceeds the ring threshold are identified to generate the ring structure identification field; Based on the topological invariants of the connected domain and the ring structure identification field, the topological invariants are integrated along the time dimension, the ring structure density is spatially aggregated, and the front topological feature matrix is ​​output.

4. The method for monitoring marine environmental changes based on remote sensing images according to claim 1, characterized in that: include: Based on multispectral remote sensing images, the chlorophyll concentration of each band is inverted based on the optical absorption characteristics to generate a multi-band chlorophyll concentration atlas; Based on the multi-band chlorophyll concentration atlas, a two-dimensional Hilbert transform is performed on each band concentration map to extract the phase component of the analytical signal and output the multi-band local phase field. Based on the multi-band local phase field, the phase difference field between any two bands is calculated, and the spatial coherence is quantified by complex exponential summation to generate the inter-band phase coherence matrix; According to the inter-band phase coherence matrix, the sliding average along the time dimension is used to enhance the stability and output the multi-band phase coherence field.

5. The method for monitoring marine environmental changes based on remote sensing images according to claim 1, characterized in that: include: Based on high-resolution ocean turbidity images, we select wavelet basis functions that match the scale of ocean turbulence, perform continuous wavelet multiscale decomposition, and generate a set of turbidity wavelet coefficients. According to the ecological characteristic tensor field, the characteristic tensor is mapped to the wavelet scale space, the scale component corresponding to the turbulent inertial sub-region is retained, and the scale-constrained characteristic tensor is output; Based on the turbidity wavelet coefficient set and the scale-constrained feature tensor, coefficient fusion is performed in the corresponding scale space, and the feature consistency is enhanced by the phase-preserving algorithm to generate the modulated wavelet coefficient set; According to the modulated wavelet coefficient set, the continuous inverse wavelet transform is performed to restore the spatial domain image data and output a multi-scale fused image.

6. The method for monitoring marine environmental changes based on remote sensing images according to claim 1, characterized in that: include: Based on the multi-scale fusion image, continuous threshold intervals are set according to the intensity of the ocean front, and contour clusters are extracted in the continuous function space to generate feature contour sets; According to the characteristic contour line set, the arc length of each contour line is parameterized, the distribution of curvature along the curve is calculated, and the continuous curvature field is output; Based on the continuous curvature field, the spatial gradient modulus of the curvature field is calculated, and the ocean front propagation model is applied to perform physical constraint enhancement to generate the gradient intensity field; Based on the gradient intensity field, the gradient intensity is mapped into the HSV color space, the geographic coordinates are superimposed to generate a pseudo-color image, and the dynamic risk map of the marine environment is output.

7. A device for monitoring marine environmental changes based on remote sensing images, characterized in that: The marine environment change monitoring device based on remote sensing images includes: The acquisition module is used to obtain the original sea surface temperature image, perform fractional differential enhancement on the original image, strengthen the temperature gradient characteristics, and generate an enhanced temperature gradient map; The matrix module is used to analyze the connected domain structure and ring structure in the gradient map based on the enhanced temperature gradient map, and output the front topology feature matrix; Setting up a module for extracting chlorophyll concentration maps of each band based on multispectral remote sensing images, calculating local phase fields and constructing coherence relationships to generate multi-band phase coherence fields; The tensor module is used to perform fusion processing based on the front topological feature matrix and the multi-band phase coherence field, and output the ecological feature tensor field, including: According to the front topological characteristic matrix, the matrix elements are expanded along the spatial dimension, the continuity of the time dimension is retained, and a topological characteristic vector sequence is generated; According to the multi-band phase coherent field, the real part and the imaginary part of the coherent field are separated, the components are stacked along the band dimension, and the phase coherent component cube is output; According to the topological eigenvector sequence and the phase coherent component cube, the eigenvectors and the coherent components are aligned according to the spatial position, and the tensor outer product operation is performed to construct a high-dimensional feature space and generate the original feature tensor field; Based on the original characteristic tensor field, the dominant scale is determined based on the ocean turbulence energy spectrum, the non-dominant scale characteristic components are truncated, and the ecological characteristic tensor field is output; The fusion module is used to obtain high-resolution ocean turbidity images, combine them with the ecological characteristic tensor field, perform multi-scale fusion reconstruction, and generate multi-scale fused images; The allocation module is used to extract image feature contours based on multi-scale fusion images, calculate curvature gradients and generate thermal maps to obtain dynamic risk maps of the marine environment.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for monitoring marine environmental changes based on remote sensing images according to any one of claims 1 to 6 is implemented.

Images