Weakly supervised and multi-dimensional feature coupled seawater backflow grading method

Abstract

The application discloses a weakly supervised and multi-dimensional feature coupled seawater backflow grading method, and belongs to the technical field of natural disaster monitoring. The method first performs weakly supervised sampling based on elevation, spectrum and radar scattering and other physical discrimination rules to automatically generate a high-confidence seed sample set. Secondly, multi-dimensional feature tensors are constructed by coupling optical and radar multi-source features, a model is trained by using the sample set, and noise-resistant extraction is performed in combination with a multi-dimensional space and a semantic mask to obtain a disaster-affected range. Finally, a three-dimensional feature vector space including biological, chemical and physical dimensions is constructed, an ecological background is adaptively adjusted in weight, and a dynamic percentile model is used to output a grading result. The application realizes high-precision, automation and refined grading of seawater backflow disasters.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing and natural disaster monitoring technology, specifically involving a seawater intrusion classification method coupled with weak supervision and multidimensional features. Background Technology

[0002] Seawater intrusion often causes severe damage to coastal agriculture and ecosystems, and satellite remote sensing technology is currently the mainstream monitoring method. However, existing remote sensing monitoring technologies still face many technical limitations in practical applications: First, single data sources have observation blind spots, optical images are easily obscured by clouds and fog, and single synthetic aperture radar (SAR) lacks the ability to spectrally diagnose vegetation health and surface salinity; second, sudden disasters present the challenge of cold-start sampling, making it difficult to quickly obtain a large number of manually labeled samples after a disaster, resulting in traditional supervised classification models being unable to respond quickly; third, the complex urban-rural intermingling environment along the coast contains a large number of impermeable surfaces with high humidity or dark scattering, which can easily lead to false alarms due to misidentification of objects with the same spectrum; fourth, existing methods are mostly limited to simple binary extraction of submerged / unsubmerged data, and often use fixed empirical thresholds, ignoring the differences in the tolerance of different land cover types to salinity stress, and lacking adaptive quantitative classification capabilities. Therefore, there is an urgent need for a seawater intrusion classification and assessment method that couples optical and radar multi-source features, has a high degree of automation, and can adapt to background differences. Summary of the Invention

[0003] The purpose of this invention is to overcome the aforementioned deficiencies of existing technologies and provide a seawater intrusion classification method coupled with weak supervision and multi-dimensional features. Specifically, addressing the difficulties in obtaining samples in the early stages of sudden disasters and the challenges of model cold start, this invention aims to automatically generate high-confidence disaster samples by introducing a weakly supervised sampling mechanism driven by physical prior rules. Simultaneously, to solve the problem of low extraction accuracy caused by homospectral foreign objects in complex coastal environments, this invention aims to construct an extraction framework that deeply couples optical and radar multi-source features and multi-spatial semantic noise resistance, thereby significantly improving the robustness and accuracy of intrusion range monitoring. Finally, addressing the pain points of traditional disaster assessment mechanisms being rigid and lacking quantitative methods, this invention aims to establish a three-dimensional feature vector assessment space and dynamic percentile model that takes into account the ecological background, in order to achieve objective, automated, and refined classification and damage assessment of seawater intrusion disasters.

[0004] The technical solution of this invention is an automatic classification and assessment method for seawater backflow based on weak supervision of physical rules and coupling of multidimensional features, comprising the following steps:

[0005] S1: Based on features such as gravitational potential energy, temporal spectrum, and radar scattering, discrimination constraints are constructed to automatically filter multi-source remote sensing images and generate a backflow positive sample set. and background negative sample set High-confidence seed sample set

[0006] S2: Extract the optical dynamic change features and radar backscattering features from multi-source remote sensing images, and construct a multi-dimensional feature tensor based on these features. Multidimensional feature tensors The data is input into a seed sample set to train the model for prediction. Then, spatial noise reduction is achieved by sequentially combining morphological opening operations, elevation filtering operators, and non-permeable surface semantic masks to obtain the preliminary range of seawater intrusion.

[0007] S3: Preliminary scope of seawater intrusion For any affected pixel within the dataset, extract its Normalized Difference Vegetation Index (NDVI) difference, post-disaster salinity index, and post-disaster surface water index features to construct a biomass-based matrix. Chemical Dimensions and physical dimensions The three-dimensional normalized feature vector; S4: Extract semantic information about land cover, dynamically assign weight coefficients based on the land cover type of the pixel, and perform a weighted calculation with the three-dimensional normalized feature vector to obtain a comprehensive damage index that takes into account ecological baseline correction. An adaptive statistical percentile model is used to classify the comprehensive damage index and output the disaster classification results as mild, moderate or severe. The specific implementation process of S1 includes the following steps: Based on gravitational potential energy, time-series spectral data, and radar scattering characteristics, discrimination rules are constructed to automatically filter multi-source remote sensing data and generate a high-confidence seed sample set. It consists of a backflow of positive sample sets. and background negative sample set composition:

[0008] Among them, the positive sample set satisfy:

[0009] negative sample set satisfy:

[0010] in, Represents pixels in an image; The gravitational potential energy constraint condition is defined as follows: ,in For digital elevation model (DEM) values, A preset elevation threshold is used to filter low-lying areas; The temporal spectral abrupt change constraint condition is defined as follows: Used to capture signals of vegetation damage, where the subscript This represents the baseline time-phase image before the disaster occurred, with the subscript... This represents the temporal images observed after a disaster occurred; For non-permanent water body constraints; For optical extreme humidity confinement, defined as ; For radar dark pixel constraints, defined as ; For safety elevation constraints ; Restraint for healthy vegetation ;in, This is the threshold for the decline in the vegetation index, used to determine the degree of vegetation damage; This is the surface extreme moisture threshold, used to determine soil moisture content; This is the radar scattering intensity threshold, used to determine water body signals; For safety elevation thresholds; The threshold for healthy vegetation is defined; all of the above thresholds are preset based on historical disaster data and statistical distribution characteristics of the study area.

[0011] The specific implementation process of S2 includes the following steps: Extract optical and radar features that reflect the dynamic process of disasters, and construct a multidimensional feature tensor. The specific calculation formula is as follows:

[0012] in, Indicates band stacking operation; The multispectral raw reflectance bands of post-disaster images (including visible, near-infrared, and short-wave infrared bands) are used to provide basic spectral information of ground features; The difference in normalized vegetation index between pre-disaster and post-disaster images; Characteristics of the surface moisture index difference; Salt characteristics representing soil salinity extracted from post-disaster images; Improved water body index features for characterizing water body information; , This is the radar backscattering coefficient.

[0013] The multidimensional feature tensor After inputting the data into the classification model for prediction, the initial prediction results are obtained. The preliminary range of backflow is obtained by sequentially combining morphological opening operation, secondary elevation filtering operator and non-permeable surface semantic mask. :

[0014] in, The initial prediction results of the classifier; This represents the morphological opening operation; For a quadratic filtering operator based on a digital elevation model, it is defined as when The value is 0 if the condition is met, and 1 otherwise. It is a preset elevation limit threshold (or altitude limit) used to forcibly exclude high-altitude areas; A binary mask constructed based on high-resolution land cover data, used to forcibly remove interference from impermeable surfaces such as urban buildings and roads.

[0015] The specific implementation process of S3 includes the following steps: Regarding the preliminary scope of the backflow Any affected pixel Establish a three-dimensional normalized feature vector that includes biological, chemical, and physical dimensions. :

[0016] in, This is the normalization function; , , These are the normalized biological, chemical, and physical dimensions, respectively. Preliminary range of the backflow Any cell within; , , The pixels are respectively Characteristics of the normalized difference vegetation index, post-disaster salinity index, and post-disaster surface moisture index.

[0017] The specific implementation process of S4 includes the following steps: Calculate pixels The comprehensive damage index that takes into account the ecological baseline correction :

[0018] in , , Based on pixels The weighting coefficient is dynamically assigned to the land cover type; when the pixel For areas belonging to general farmland, the first set of weights is used; when the pixel For areas belonging to salt-tolerant vegetation or wetlands, the second set of weights is used; wherein, the biological dimension weight coefficient in the second set of weights is... Less than the first group, and the chemical dimension weight coefficient in the second group weight configuration The value is greater than the first group; finally, an adaptive statistical threshold is used for classification to obtain the classification result. :

[0019] in, The comprehensive damage index is within the initial range of the backflow. The first percentile, It is the first percentiles (of which) ).

[0020] This invention discloses a seawater intrusion classification method coupled with weak supervision and multidimensional features. This method relies on a complete automated assessment process. First, it automatically generates disaster samples through a weakly supervised sampling mechanism driven by physical prior rules, effectively overcoming the cold-start problems of difficult manual annotation and delayed model response in the early stages of sudden disasters. Second, in the range extraction stage, it deeply integrates optical and radar multi-source features and combines multidimensional space and semantic noise reduction strategies to effectively eliminate false alarm interference from foreign objects of the same spectrum in complex coastal environments, improving the accuracy of intrusion range extraction. Finally, in the disaster damage assessment stage, it breaks through the traditional binary extraction mode, constructing an assessment vector space including biological, chemical, and physical dimensions, and introducing adaptive weights based on ecological background and dynamic percentile thresholds, ultimately achieving an objective, quantitative, and refined classification assessment of seawater intrusion disasters. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the seawater backflow classification method coupled with weak supervision and multidimensional features according to the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] This invention discloses a seawater backflow classification method coupled with weak supervision and multidimensional features, which specifically includes the following steps: S1. Based on gravitational potential energy, time-series spectral data, and radar scattering characteristics, a discrimination rule is constructed to automatically filter multi-source remote sensing data and generate a high-confidence seed sample set. It consists of a backflow of positive sample sets. and background negative sample set composition:

[0026] Among them, the positive sample set satisfy:

[0027] negative sample set satisfy:

[0028] in, Represents pixels in an image; The gravitational potential energy constraint condition is defined as follows: ,in For digital elevation model (DEM) values, A preset elevation threshold is used to filter low-lying areas; The temporal spectral abrupt change constraint condition is defined as follows: Used to capture signals of vegetation damage, where the subscript This represents the baseline time-phase image before the disaster occurred, with the subscript... This represents the temporal images observed after a disaster occurred; For non-permanent water body constraints; For optical extreme humidity confinement, defined as ; For radar dark pixel constraints, defined as ; For safety elevation constraints ; Restraint for healthy vegetation ;in, This is the threshold for the decline in the vegetation index, used to determine the degree of vegetation damage; This is the surface extreme moisture threshold, used to determine soil moisture content; This is the radar scattering intensity threshold, used to determine water body signals; For safety elevation thresholds; The threshold for healthy vegetation is defined; all of the above thresholds are preset based on historical disaster data and statistical distribution characteristics of the study area.

[0029] S2. Extract optical and radar features reflecting the dynamic process of disasters and construct a multidimensional feature tensor. The specific calculation formula is as follows:

[0030] in, Indicates band stacking operation; The multispectral raw reflectance bands of post-disaster images (including visible, near-infrared, and short-wave infrared bands) are used to provide basic spectral information of ground features; The difference in normalized vegetation index between pre-disaster and post-disaster images; Characteristics of the surface moisture index difference; Salt characteristics representing soil salinity extracted from post-disaster images; Improved water body index features for characterizing water body information; , This is the radar backscattering coefficient.

[0031] The multidimensional feature tensor After inputting the data into the classification model for prediction, the initial prediction results are obtained. The preliminary range of backflow is obtained by sequentially combining morphological opening operation, secondary elevation filtering operator and non-permeable surface semantic mask. :

[0032] in, The initial prediction results of the classifier; This represents the morphological opening operation; For a quadratic filtering operator based on a digital elevation model, it is defined as when The value is 0 if the condition is met, and 1 otherwise. It is a preset elevation limit threshold (or altitude limit) used to forcibly exclude high-altitude areas; A binary mask constructed based on high-resolution land cover data, used to forcibly remove interference from impermeable surfaces such as urban buildings and roads.

[0033] S3, Regarding the preliminary scope of the backflow Any affected pixel Establish a three-dimensional normalized feature vector that includes biological, chemical, and physical dimensions. :

[0034] in, This is the normalization function; , , These are the normalized biological, chemical, and physical dimensions, respectively. Preliminary range of the backflow Any cell within; , , The pixels are respectively Characteristics of the normalized difference vegetation index, post-disaster salinity index, and post-disaster surface moisture index.

[0035] S4, Calculate pixels The comprehensive damage index that takes into account the ecological baseline correction :

[0036] in, It is a physical quantity that characterizes the intensity of composite damage to affected pixels under a specific ecological background. This index deeply integrates the vegetation damage status diagnosed by optical spectroscopy, the degree of salt accumulation, and the extremely wet surface state retrieved by radar and infrared bands. The higher the value, the more thorough the vegetation destruction and the more severe the soil salinization stress experienced by the area after seawater intrusion and receding. By calculating this index, this invention successfully transforms multi-source heterogeneous remote sensing observation signals into quantifiable and comparable physical disaster damage indicators; , , Based on pixels The weighting coefficient is dynamically assigned to the land cover type; when the pixel For areas classified as general farmland, the first set of weights is used, for example... , , When pixel For areas with salt-tolerant vegetation or wetlands, the second set of weights is used, for example... , 6, That is, in response to the salt tolerance background, the system adaptively reduces the weight of biological features. This also correspondingly increased the weight of the chemical dimension feature representing salt accumulation. This effectively avoids the underestimation of disaster damage caused by differences in vegetation salt tolerance; finally, adaptive statistical thresholds are used for classification and discrimination to obtain the classification results. :

[0037] in, The comprehensive damage index is within the initial range of the backflow. The first percentile, It is the first percentiles (of which) ).

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for classifying seawater backflow coupled with weak supervision and multidimensional features, characterized in that, include: S1: Based on features such as gravitational potential energy, temporal spectrum, and radar scattering, discrimination constraints are constructed to automatically filter multi-source remote sensing images and generate a backflow positive sample set. and background negative sample set High-confidence seed sample set ; S2: Extract the optical dynamic change features and radar backscattering features from multi-source remote sensing images, and construct a multi-dimensional feature tensor based on these features. Multidimensional feature tensors The data is input into a seed sample set to train the model for prediction. Then, spatial noise reduction is achieved by sequentially combining morphological opening operations, elevation filtering operators, and non-permeable surface semantic masks to obtain the preliminary range of seawater intrusion. ; S3: Preliminary scope of seawater intrusion For any affected pixel within the dataset, extract its Normalized Difference Vegetation Index (NDVI) difference, post-disaster salinity index, and post-disaster surface water index features to construct a biomass-based matrix. Chemical Dimensions and physical dimensions The three-dimensional normalized feature vector; S4: Extract semantic information about land cover, dynamically assign weight coefficients based on the land cover type of the pixel, and perform a weighted calculation with the three-dimensional normalized feature vector to obtain a comprehensive damage index that takes into account ecological baseline correction. The adaptive statistical percentile model is used to classify the comprehensive damage index and output the disaster classification results as mild, moderate or severe.

2. The seawater backflow classification method coupled with weak supervision and multidimensional features according to claim 1, characterized in that: Based on gravitational potential energy, time-series spectral data, and radar scattering characteristics, discrimination rules are constructed to automatically filter multi-source remote sensing data and generate a high-confidence seed sample set. Including the backflow of positive sample sets and background negative sample set : Among them, the positive sample set satisfy: negative sample set satisfy: in, Represents pixels in an image; The gravitational potential energy constraint condition is defined as follows: ,in For digital elevation model (DEM) values, Preset elevation threshold; The temporal spectral abrupt change constraint condition is defined as follows: subscript This represents the baseline time-phase image before the disaster occurred, with the subscript... This represents the temporal images observed after a disaster occurred; For non-permanent water body constraints; For optical extreme humidity confinement, defined as ; For radar dark pixel constraints, defined as ; For safety elevation constraints ; Restraint for healthy vegetation ; The threshold for the decline in vegetation index; The extreme wet threshold for the Earth's surface; The radar scattering intensity threshold; For safety elevation thresholds; The threshold for healthy vegetation is defined; all of the above thresholds are preset based on historical disaster data and statistical distribution characteristics of the study area.

3. The seawater backflow classification method coupled with weak supervision and multidimensional features according to claim 1, characterized in that: Extract optical and radar features that reflect the dynamic process of disasters, and construct a multidimensional feature tensor. The specific calculation formula is as follows: in, Indicates band stacking operation; The original reflectance band of the multispectral imagery after the disaster; The difference in normalized vegetation index between pre-disaster and post-disaster images; Characteristics of the surface moisture index difference; Salt characteristics representing soil salinity extracted from post-disaster images; Improved water body index features for characterizing water body information; , This refers to the radar backscattering coefficient; The multidimensional feature tensor The sample data is input into the seed sample set to train the model for prediction, and the initial prediction results are obtained. The preliminary range of backflow is obtained by sequentially combining morphological opening operation, secondary elevation filtering operator and non-permeable surface semantic mask. : in, The initial prediction results of the classifier; This represents the morphological opening operation; For a quadratic filtering operator based on a digital elevation model, it is defined as when The value is 0 if the condition is met, and 1 otherwise. It is a preset elevation limit threshold; A binary mask constructed based on high-resolution land cover data.

4. The seawater backflow classification method coupled with weak supervision and multidimensional features according to claim 1, characterized in that: Regarding the preliminary scope of the backflow Any affected pixel within Establish a three-dimensional normalized feature vector that includes biological, chemical, and physical dimensions. : in, This is the normalization function; , , These are the normalized biological, chemical, and physical dimensions, respectively. Preliminary range of the backflow Any affected pixel within; , , The pixels are respectively Characteristics of the normalized difference vegetation index, post-disaster salinity index, and post-disaster surface moisture index.

5. The seawater intrusion classification method coupled with weak supervision and multidimensional features according to claim 1, characterized in that: Calculate pixels The comprehensive damage index that takes into account the ecological baseline correction : in , , Based on pixels The weighting coefficient is dynamically assigned to the land cover type; when the pixel For areas belonging to general farmland, the first set of weights is used; when the pixel For areas belonging to salt-tolerant vegetation or wetlands, the second set of weights is used; wherein, the biological dimension weight coefficient in the second set of weights is... Less than the first group, and the chemical dimension weight coefficient in the second group weight configuration The value is greater than the first group; finally, an adaptive statistical threshold is used for classification to obtain the classification result. : in, The comprehensive damage index is within the initial range of the backflow. The first percentile, It is the first percentiles, among which .

Images