A method and system for predicting marine water pollution based on remote sensing images

Abstract

The present invention discloses a method and system for predicting marine water pollution based on remote sensing images, comprising: obtaining a first remote sensing image corresponding to a target sea area, performing pollution analysis on the target sea area according to the first remote sensing image, obtaining sea surface monitoring data and a first time series remote sensing image corresponding to the target sea area when it is determined that water pollution occurs in the target sea area, extracting a first pollution distribution image in the first time series remote sensing image, obtaining the pollutant concentration corresponding to each first pollution distribution image through a pre-trained pollutant concentration inversion model according to the sea surface monitoring data and the first pollution distribution image, and predicting the pollution degree corresponding to the target sea area through a pre-trained marine pollution degree prediction model according to the pollutant concentration and the first pollution distribution image, thereby improving the accuracy and efficiency of the prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of marine pollution monitoring, and in particular to a method and system for predicting marine water pollution based on remote sensing images. Background Art

[0002] With the increasing global attention paid to marine environmental pollution, marine pollution monitoring and prediction has become an important research direction. Traditional marine pollution monitoring and prediction has long relied mainly on manual sampling for chemical analysis to monitor various harmful substances. This method is not only time-consuming and labor-intensive, but also unable to monitor and predict pollution in distant sea areas in a timely manner, and is greatly restricted by the environment, climate, and cost.

[0003] With the rapid development of technology, researchers control drones to collect image data of the ocean, and then use models to process the image data to identify whether the sea area in the image is polluted and the degree of pollution. Although this pollution monitoring method can monitor marine pollution in distant sea areas in real time, the models used are often computationally intensive and have high requirements for computing equipment. Furthermore, due to the diversity of factors affecting marine pollution, the method using image recognition cannot accurately obtain detailed statistical data on water pollutants in the ocean, and thus cannot predict the degree of pollution caused by the water pollutants to the current sea area or bring certain errors to the pollution prediction results caused by water pollutants. Summary of the invention

[0004] In order to solve the above technical problems, the present invention discloses a method and system for predicting ocean water pollution based on remote sensing images, which are used to improve the accuracy of predicting the degree of ocean water pollution.

[0005] In order to achieve the above object, the present invention discloses a method for predicting ocean water pollution based on remote sensing images, comprising:

[0006] Acquire a first remote sensing image corresponding to the target sea area, and extract spectral features and pixel features corresponding to the first remote sensing image by using a preset band separation algorithm;

[0007] According to the spectral characteristics and the pixel characteristics, pollution analysis is performed on the target sea area through big data retrieval to determine whether water pollution occurs in the target sea area;

[0008] When it is determined that water pollution occurs in the target sea area, sea surface monitoring data and a first time series remote sensing image corresponding to the target sea area are obtained, and edge detection is performed on each second remote sensing image in the first time series remote sensing image to extract a first pollution distribution image corresponding to each second remote sensing image;

[0009] Performing atmospheric correction processing on each of the first pollution distribution images to obtain a corresponding remote sensing reflectance, and inputting the remote sensing reflectance and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain a pollutant concentration corresponding to each of the first pollution distribution images, so as to mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image;

[0010] Inputting the second time series remote sensing image into a pre-trained marine pollution degree prediction model, extracting the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fusing the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image;

[0011] According to the pollution trend characteristics, the pollution degree corresponding to the target sea area is predicted by a classifier preset in the marine pollution degree prediction model, and early warning information is generated according to the pollution degree.

[0012] The present invention discloses a method for predicting ocean water pollution based on remote sensing images. In the prediction method, a first remote sensing image corresponding to a target sea area is firstly obtained, and spectral features and pixel features corresponding to the first remote sensing image are extracted, so that according to the spectral features and the pixel features, it is determined by a preset big data search whether water pollution currently occurs in the target sea area, so that the pollution situation of the ocean can be monitored in real time. After it is determined that water pollution occurs in the ocean, sea surface monitoring data and time series remote sensing images corresponding to the target sea area are obtained, and edge detection is performed on the time series remote sensing images to obtain only the first pollution distribution image where water pollution occurs in the ocean. In this way, the processing amount of later data is reduced by intercepting the continuous images, the efficiency of monitoring is improved, and it is convenient to extract the spatial spread characteristics of pollution in real time according to the edge detection results of the pollution distribution image in the later stage, so as to improve the accuracy of prediction.

[0013] After obtaining a first pollution distribution image corresponding to only polluted seawater, atmospheric processing is performed on the first pollution distribution image to obtain remote sensing reflectivity, so as to reduce the impact of atmospheric reasons on the acquisition of the remote sensing image and improve the accuracy of data used to predict the degree of pollution. Furthermore, based on the fact that the degree of pollution in the ocean depends not only on the pollution itself but also on the type of ocean, the present invention uses a pollutant concentration inversion model to invert the pollution concentration of the remote sensing reflectivity corresponding to the first pollution distribution image and the sea surface monitoring data corresponding to the target sea area, so as to obtain the pollutant concentration corresponding to each pollution distribution image.

[0014] Finally, the marine pollution degree prediction model pre-constructed in the present invention extracts the spatial spread characteristics corresponding to the image area and the concentration spread characteristics corresponding to the concentration change trend in the second time series remote sensing image, so as to predict the pollution degree by integrating the above characteristics, and issue an early warning based on the predicted pollution degree, thereby improving the accuracy of the prediction.

[0015] As a preferred example, the extracting spectral features and pixel features corresponding to the first remote sensing image by using a preset band separation algorithm includes:

[0016] Performing minimum noise separation change processing on the first remote sensing image, extracting pixel values ​​of each band corresponding to the first remote sensing image, and calculating the square sum of the pixel values ​​of each band to generate first spectral images corresponding to different bands;

[0017] Integrate the first spectral images corresponding to different bands to obtain a spectral image corresponding to the first remote sensing image;

[0018] A normalized water body index corresponding to the target sea area is calculated according to the first remote sensing image, and pixels of the first remote sensing image are classified according to the normalized water body index to divide the first remote sensing image into a clean water image and a sewage image.

[0019] The present invention is based on the fact that in the process of different types of pollutants diffusing in the ocean, the remote sensing images corresponding to the polluted seawater have different band pixels and color characteristics due to the molecular composition of the pollutants. Based on this, the present invention obtains the pixel values ​​of each band of the first remote sensing image to extract the corresponding spectral image to generate the corresponding spectral characteristics. Furthermore, by calculating the normalized water body index of the first remote sensing image, the first remote sensing image is pixel classified to obtain the sewage image suspected of being polluted in the remote sensing image.

[0020] As a preferred example, the pollution analysis of the target sea area is performed through big data retrieval to determine whether water pollution occurs in the target sea area, including:

[0021] Through big data retrieval, the spectral image and remote sensing image corresponding to each polluted water body are obtained to construct a polluted water body data set;

[0022] Calculate the similarity between the spectral image and the sewage image corresponding to the first remote sensing image and the polluted water body dataset to obtain the similarity between the first remote sensing image and each type of polluted water body;

[0023] The similarity is compared with a preset similarity threshold to determine whether water pollution occurs in the target sea area and obtain the type of the polluted water according to the comparison result.

[0024] The present invention obtains the spectral image and remote sensing image corresponding to each type of polluted water body through big data retrieval, so as to construct a comparative data set for pollution judgment, so as to calculate the similarity between the spectral image and the sewage image corresponding to the first remote sensing image and the comparative data set respectively, and then determine whether water pollution occurs in the current target sea area according to the similarity, thereby improving the detection efficiency.

[0025] As a preferred example, the performing edge detection on each second remote sensing image in the first time series remote sensing image to extract the first pollution distribution image corresponding to each second remote sensing image includes:

[0026] Splitting the first time series remote sensing image into a plurality of second remote sensing images according to the time series, and performing grayscale processing on each of the second remote sensing images, so as to perform preliminary segmentation on each second remote sensing image after grayscale processing by using a preset grayscale threshold, and generate an initial pollution distribution image corresponding to each second remote sensing image;

[0027] Performing pixel segmentation on each of the initial pollution distribution images, and extracting texture features and color distribution features corresponding to each of the initial pollution distribution images according to the pixel segmentation result;

[0028] According to the texture features and the color distribution features, a clustering algorithm is used to perform image segmentation on each of the initial pollution distribution images, and the results of the image segmentation are regionally merged to generate a first pollution distribution image corresponding to each of the initial pollution distribution images.

[0029] The present invention is based on the fact that when the pollutants diffuse in the water body, the grayscale values ​​in the corresponding images are different because the same volume of seawater has different molecular capacities, and the spread of pollutants will affect the texture of the sea surface itself. Therefore, the present invention extracts the grayscale values, pixel features, and texture features in the second remote sensing image to perform image segmentation, thereby providing segmentation accuracy and thus improving the accuracy of pollution degree prediction.

[0030] As a preferred example, the step of inputting the remote sensing reflectivity and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain the pollutant concentration corresponding to each of the first pollution distribution images includes:

[0031] According to the type of the polluted water body, determining a target pollutant concentration inversion model corresponding to the target sea area from a set of pre-trained pollutant concentration inversion models;

[0032] According to the remote sensing reflectivity and the sea surface monitoring data, an initial concentration inversion result corresponding to the first pollution distribution image is outputted through a first branch network preset in the target pollutant concentration inversion model; wherein the sea surface monitoring data includes ocean type data and marine ecological data;

[0033] Performing one-hot encoding processing on the sea surface monitoring data through a second branch network preset in the target pollutant concentration inversion model to obtain a sparse matrix, so as to output a first concentration inversion result corresponding to the first pollution distribution image according to the sparse matrix, the remote sensing reflectivity and the initial concentration inversion result;

[0034] Performing dimensionality reduction processing on the sea surface monitoring data by a third branch network preset in the target pollutant concentration inversion model to obtain a dense matrix, so as to output a second concentration inversion result corresponding to the first pollution distribution image according to the dense matrix, the remote sensing reflectivity and the initial concentration inversion result;

[0035] The first concentration inversion result and the second concentration inversion result are weighted by the fourth branch network preset in the target pollutant concentration inversion model to obtain the pollutant concentration corresponding to the first pollution distribution image.

[0036] The present invention first determines the corresponding pollutant concentration inversion model according to the type of pollutant identified in the aforementioned step, and then uses the pollutant concentration inversion model to process the remote sensing reflectivity and the sea surface monitoring data so as to deduce the concentration corresponding to the pollutant in each pollution distribution image. In the concentration inversion process, the inversion model is used to perform different processing on the sea surface monitoring data so as to finally integrate the processing results to perform concentration inversion, thereby performing concentration inversion based on the ocean and the pollution source itself, thereby improving the accuracy of the prediction.

[0037] As a preferred example, the extracting of the spatial spread characteristics and the concentration spread characteristics corresponding to the second time series remote sensing image by the marine pollution degree determination model includes:

[0038] The first convolution layer preset in the marine pollution degree determination model performs spatial separation and data separation on the second time series remote sensing image to obtain spatial information data and concentration information data corresponding to the second time series remote sensing image;

[0039] The second convolution layer preset in the marine pollution degree determination model performs convolution processing on the spatial information data and the concentration information data respectively, so as to obtain the spatial spread characteristics corresponding to the spatial information data and the concentration spread characteristics corresponding to the concentration information data.

[0040] The present invention improves the characteristics of the space covered by pollutants and the corresponding concentration characteristics during the spread of pollutants in the target sea area, so that the degree of pollution can be determined in combination with the characteristics of the covered space and the corresponding concentration characteristics, thereby improving the accuracy of the degree prediction.

[0041] As a preferred example, the fusing of the spatial spread feature and the concentration spread feature to obtain the pollution trend feature corresponding to the second time series remote sensing image includes:

[0042] Performing dimensionality up-processing and feature fusion on the spatial spread feature and the concentration spread feature according to the fully connected layer preset in the marine pollution degree prediction model, so as to obtain the initial pollution trend feature corresponding to the second time series remote sensing image;

[0043] The initial pollution trend characteristics are averaged according to a preset average function to obtain the pollution trend characteristics corresponding to the second time series remote sensing image.

[0044] The present invention obtains the pollution trend characteristics of the pollutants spreading in the target sea area by integrating the concentration spread characteristics and spatial coverage characteristics of the pollutants, so as to make a degree prediction based on the pollution trend characteristics and improve the accuracy of the prediction.

[0045] As a preferred example, predicting the pollution degree corresponding to the target sea area by using a classifier preset in the marine pollution degree prediction model includes:

[0046] Predicting the pollution degree of the pollution trend characteristics respectively by using a plurality of different decision trees preset in the classifier, and obtaining a first pollution degree corresponding to each decision tree;

[0047] The plurality of first pollution levels are averaged through a preset linear regression calculation function to obtain the pollution level corresponding to the target sea area.

[0048] The present invention utilizes different pollution levels corresponding to different decision trees to determine the pollution level value corresponding to the pollution trend feature, and then performs average processing to improve the accuracy of the model output.

[0049] On the other hand, the present invention discloses a marine water pollution prediction system based on remote sensing images, including a feature extraction module, a pollution determination module, an image segmentation module, an image calibration module, an image processing module and a pollution prediction module;

[0050] The feature extraction module is used to obtain a first remote sensing image corresponding to the target sea area, and extract spectral features and pixel features corresponding to the first remote sensing image through a preset band separation algorithm;

[0051] The pollution determination module is used to perform pollution analysis on the target sea area through big data retrieval according to the spectral characteristics and the pixel characteristics, and determine whether water pollution occurs in the target sea area;

[0052] The image segmentation module is used to obtain sea surface monitoring data and a first time series remote sensing image corresponding to the target sea area after determining that water pollution occurs in the target sea area, and perform edge detection on each second remote sensing image in the first time series remote sensing image to extract a first pollution distribution image corresponding to each second remote sensing image;

[0053] The image calibration module is used to perform atmospheric correction processing on each of the first pollution distribution images to obtain a corresponding remote sensing reflectance, and input the remote sensing reflectance and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain the pollutant concentration corresponding to each of the first pollution distribution images, so as to mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image;

[0054] The image processing module is used to input the second time series remote sensing image into a pre-trained marine pollution degree prediction model, extract the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fuse the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image;

[0055] The pollution prediction module is used to predict the pollution degree corresponding to the target sea area according to the pollution trend characteristics through a classifier preset in the marine pollution degree prediction model, and generate early warning information according to the pollution degree.

[0056] The present invention discloses a marine water pollution prediction system based on remote sensing images. In the prediction method, a first remote sensing image corresponding to a target sea area is firstly obtained, and spectral features and pixel features corresponding to the first remote sensing image are extracted, so that according to the spectral features and the pixel features, it is determined by a preset big data search whether water pollution currently occurs in the target sea area, so that the pollution situation of the ocean can be monitored in real time. After it is determined that water pollution occurs in the ocean, sea surface monitoring data and time series remote sensing images corresponding to the target sea area are obtained, and edge detection is performed on the time series remote sensing images to obtain only the first pollution distribution image of water pollution occurring in the ocean. In this way, the processing amount of later data is reduced by intercepting the continuous images, the efficiency of monitoring is improved, and it is convenient to extract the spatial spread characteristics of pollution in real time according to the edge detection results of the pollution distribution image in the later stage, so as to improve the accuracy of prediction.

[0057] After obtaining a first pollution distribution image corresponding to only polluted seawater, atmospheric processing is performed on the first pollution distribution image to obtain remote sensing reflectivity, so as to reduce the impact of atmospheric reasons on the acquisition of the remote sensing image and improve the accuracy of data used to predict the degree of pollution. Furthermore, based on the fact that the degree of pollution in the ocean depends not only on the pollution itself but also on the type of ocean, the present invention uses a pollutant concentration inversion model to invert the pollution concentration of the remote sensing reflectivity corresponding to the first pollution distribution image and the sea surface monitoring data corresponding to the target sea area, so as to obtain the pollutant concentration corresponding to each pollution distribution image.

[0058] Finally, the marine pollution degree prediction model pre-constructed in the present invention extracts the spatial spread characteristics corresponding to the image area and the concentration spread characteristics corresponding to the concentration change trend in the second time series remote sensing image, so as to predict the pollution degree by integrating the above characteristics, and issue an early warning based on the predicted pollution degree, thereby improving the accuracy of the prediction.

[0059] As a preferred example, the feature extraction module includes a spectrum extraction unit and a pixel classification unit;

[0060] The spectrum extraction unit is used to perform minimum noise separation change processing on the first remote sensing image, extract the pixel values ​​of each band corresponding to the first remote sensing image, and calculate the square sum of the pixel values ​​of each band to generate first spectrum images corresponding to different bands; integrate the first spectrum images corresponding to different bands to obtain the spectrum image corresponding to the first remote sensing image;

[0061] The pixel classification unit is used to calculate the normalized water body index corresponding to the target sea area based on the first remote sensing image, and to perform pixel classification on the first remote sensing image according to the normalized water body index, so as to divide the first remote sensing image into a clean water image and a sewage image.

[0062] The present invention is based on the fact that in the process of different types of pollutants diffusing in the ocean, the remote sensing images corresponding to the polluted seawater have different band pixels and color characteristics due to the molecular composition of the pollutants. Based on this, the present invention obtains the pixel values ​​of each band of the first remote sensing image to extract the corresponding spectral image to generate the corresponding spectral characteristics. Furthermore, by calculating the normalized water body index of the first remote sensing image, the first remote sensing image is pixel classified to obtain the sewage image suspected of being polluted in the remote sensing image. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 : A schematic diagram of a flow chart of a method for predicting ocean water pollution based on remote sensing images disclosed in an embodiment of the present invention;

[0064] Figure 2: is a schematic diagram of the structure of a marine water pollution prediction system based on remote sensing images disclosed in an embodiment of the present invention;

[0065] Figure 3 : A flow chart of a method for predicting ocean water pollution based on remote sensing images disclosed in another embodiment of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Embodiment 1

[0068] This embodiment discloses a method for predicting ocean water pollution based on remote sensing images. For the specific implementation process of the prediction method, please refer to Figure 1 , mainly including steps 101 to 106, the steps mainly include:

[0069] Step 101: Acquire a first remote sensing image corresponding to a target sea area, and extract spectral features and pixel features corresponding to the first remote sensing image by using a preset band separation algorithm.

[0070] In this embodiment, this step mainly includes: performing minimum noise separation change processing on the first remote sensing image, extracting the pixel values ​​of each band corresponding to the first remote sensing image, and calculating the square sum of the pixel values ​​of each band to generate the first spectral images corresponding to different bands; integrating the first spectral images corresponding to different bands to obtain the spectral image corresponding to the first remote sensing image; calculating the normalized water body index corresponding to the target sea area based on the first remote sensing image, and performing pixel classification on the first remote sensing image according to the normalized water body index to divide the first remote sensing image into a clean water image and a sewage image.

[0071] In this embodiment, this step is based on the fact that different types of pollutants diffuse in the ocean, and the remote sensing images corresponding to the polluted seawater have different band pixels and color characteristics due to the molecular composition of the pollutants. Based on this, the pixel values ​​of each band of the first remote sensing image are obtained to extract the corresponding spectral image to generate the corresponding spectral characteristics. Furthermore, by calculating the normalized water body index of the first remote sensing image, the first remote sensing image is pixel classified to obtain the sewage image suspected of being polluted in the remote sensing image.

[0072] Step 102: According to the spectral characteristics and the pixel characteristics, pollution analysis is performed on the target sea area through big data retrieval to determine whether water pollution occurs in the target sea area.

[0073] In this embodiment, this step mainly includes: obtaining the spectral image and remote sensing image corresponding to each type of polluted water body through big data retrieval, and constructing a polluted water body data set; calculating the similarity between the spectral image and sewage image corresponding to the first remote sensing image and the polluted water body data set, and obtaining the similarity between the first remote sensing image and each type of polluted water body; comparing the similarity with a preset similarity threshold, so as to determine whether water pollution occurs in the target sea area and obtain the type of the polluted water body according to the comparison result.

[0074] In this embodiment, this step obtains the spectral image and remote sensing image corresponding to each type of polluted water body through big data retrieval, so as to construct a comparative data set for pollution judgment, so as to calculate the similarity between the spectral image and sewage image corresponding to the first remote sensing image and the comparative data set, and then determine whether water pollution occurs in the current target sea area based on the similarity, thereby improving the efficiency of detection.

[0075] Step 103: When it is determined that water pollution occurs in the target sea area, the sea surface monitoring data and the first time series remote sensing image corresponding to the target sea area are obtained, and edge detection is performed on each second remote sensing image in the first time series remote sensing image to extract the first pollution distribution image corresponding to each second remote sensing image.

[0076] In this embodiment, this step mainly includes: splitting the first time series remote sensing image into several second remote sensing images according to the time series, and performing grayscale processing on each of the second remote sensing images, so as to use a preset grayscale threshold to perform preliminary segmentation on each second remote sensing image after grayscale processing, and generate an initial pollution distribution image corresponding to each second remote sensing image; performing pixel segmentation on each of the initial pollution distribution images, and extracting texture features and color distribution features corresponding to each of the initial pollution distribution images according to the results of pixel segmentation; according to the texture features and the color distribution features, using a clustering algorithm to perform image segmentation on each of the initial pollution distribution images, and performing regional merging on the results after the image segmentation, to generate a first pollution distribution image corresponding to each of the initial pollution distribution images.

[0077] In this embodiment, this step is based on the fact that when the pollutant diffuses in the water body, the grayscale values ​​in the corresponding images are different due to the different molecular capacities of seawater of the same volume, and the spread of pollutants will affect the texture of the sea surface itself. Therefore, image segmentation is performed by extracting the grayscale values, pixel features, and texture features in the second remote sensing image to provide segmentation accuracy, thereby improving the accuracy of pollution degree prediction.

[0078] Step 104: Perform atmospheric correction processing on each of the first pollution distribution images to obtain the corresponding remote sensing reflectance, and input the remote sensing reflectance and the sea surface monitoring data into the pre-trained pollutant concentration inversion model to obtain the pollutant concentration corresponding to each of the first pollution distribution images, so as to mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image.

[0079] In this embodiment, this step mainly includes: according to the type of the polluted water body, determining the target pollutant concentration inversion model corresponding to the target sea area from the pre-trained pollutant concentration inversion model set; according to the remote sensing reflectivity and the sea surface monitoring data, outputting the initial concentration inversion result corresponding to the first pollution distribution image through the first branch network preset in the target pollutant concentration inversion model; wherein the sea surface monitoring data includes ocean type data and marine ecological data; performing one-hot encoding processing on the sea surface monitoring data through the second branch network preset in the target pollutant concentration inversion model to obtain a sparse matrix, so as to output the first concentration inversion result corresponding to the first pollution distribution image according to the sparse matrix, the remote sensing reflectivity and the initial concentration inversion result; performing dimensionality reduction processing on the sea surface monitoring data through the third branch network preset in the target pollutant concentration inversion model to obtain a dense matrix, so as to output the second concentration inversion result corresponding to the first pollution distribution image according to the dense matrix, the remote sensing reflectivity and the initial concentration inversion result; performing weighted processing on the first concentration inversion result and the second concentration inversion result through the fourth branch network preset in the target pollutant concentration inversion model to obtain the pollutant concentration corresponding to the first pollution distribution image.

[0080] In this embodiment, this step first determines the corresponding pollutant concentration inversion model according to the type of pollutant identified in the previous step, and then uses the pollutant concentration inversion model to process the remote sensing reflectivity and the sea surface monitoring data to deduce the concentration corresponding to the pollutant in each pollution distribution image. During the concentration inversion process, the inversion model is used to perform different processing on the sea surface monitoring data, so that the processing results are finally integrated to perform concentration inversion, thereby performing concentration inversion based on the ocean and the pollution source itself, thereby improving the accuracy of the prediction.

[0081] Step 105: Input the second time series remote sensing image into a pre-trained marine pollution degree prediction model, extract the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fuse the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image.

[0082] In this embodiment, this step mainly includes: performing spatial separation and data separation on the second time series remote sensing image through the first convolution layer preset in the marine pollution degree determination model, so as to obtain the spatial information data and concentration information data corresponding to the second time series remote sensing image; performing convolution processing on the spatial information data and the concentration information data respectively through the second convolution layer preset in the marine pollution degree determination model, so as to obtain the spatial spread characteristics corresponding to the spatial information data and the concentration spread characteristics corresponding to the concentration information data.

[0083] Furthermore, according to the fully connected layer preset in the marine pollution degree prediction model, the spatial spread characteristics and the concentration spread characteristics are subjected to dimensionality upgrading and feature fusion to obtain the initial pollution trend characteristics corresponding to the second time series remote sensing image; the initial pollution trend characteristics are averaged according to a preset averaging function to obtain the pollution trend characteristics corresponding to the second time series remote sensing image.

[0084] In this embodiment, this step improves the characteristics of the space covered by pollutants and the corresponding concentration characteristics during the spread of pollutants in the target sea area, so that by integrating the concentration spread characteristics and spatial coverage characteristics of pollutants, the pollution trend characteristics of pollutants spreading in the target sea area are obtained, so that the degree of prediction can be made based on the pollution trend characteristics, thereby improving the accuracy of the prediction.

[0085] Step 106: According to the pollution trend characteristics, the pollution degree corresponding to the target sea area is predicted by a classifier preset in the marine pollution degree prediction model, and warning information is generated according to the pollution degree.

[0086] In this embodiment, this step mainly includes: predicting the pollution degree of the pollution trend characteristics respectively by using several different decision trees preset in the classifier, and obtaining the first pollution degree corresponding to each decision tree; averaging several first pollution degrees through a preset linear regression calculation function to obtain the pollution degree corresponding to the target sea area.

[0087] In this embodiment, this step uses different pollution levels corresponding to different decision trees to determine the pollution level value corresponding to the pollution trend feature, and then performs average processing to improve the accuracy of the model output.

[0088] On the other hand, this embodiment also discloses a marine water pollution prediction system based on remote sensing images. For the specific structure of the prediction system, please refer to Figure 2 , including a feature extraction module 201, a pollution determination module 202, an image segmentation module 203, an image calibration module 204, an image processing module 205 and a pollution prediction module 206.

[0089] The feature extraction module 201 is used to obtain a first remote sensing image corresponding to the target sea area, and extract spectral features and pixel features corresponding to the first remote sensing image through a preset band separation algorithm.

[0090] The pollution determination module 202 is used to perform pollution analysis on the target sea area through big data retrieval according to the spectral characteristics and the pixel characteristics, and determine whether water pollution occurs in the target sea area.

[0091] The image segmentation module 203 is used to obtain the sea surface monitoring data and the first time series remote sensing image corresponding to the target sea area after determining that water pollution occurs in the target sea area, and perform edge detection on each second remote sensing image in the first time series remote sensing image to extract the first pollution distribution image corresponding to each second remote sensing image.

[0092] The image calibration module 204 is used to perform atmospheric correction processing on each of the first pollution distribution images to obtain the corresponding remote sensing reflectivity, and input the remote sensing reflectivity and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain the pollutant concentration corresponding to each of the first pollution distribution images, so as to mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image.

[0093] The image processing module 205 is used to input the second time series remote sensing image into a pre-trained marine pollution degree prediction model, extract the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fuse the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image.

[0094] The pollution prediction module 206 is used to predict the pollution degree corresponding to the target sea area according to the pollution trend characteristics through a classifier preset in the marine pollution degree prediction model, and generate warning information according to the pollution degree.

[0095] In this embodiment, the feature extraction module 201 includes a spectrum extraction unit and a pixel classification unit.

[0096] The spectral extraction unit is used to perform minimum noise separation change processing on the first remote sensing image, extract the pixel values ​​of each band corresponding to the first remote sensing image, and calculate the square sum of the pixel values ​​of each band to generate first spectral images corresponding to different bands; integrate the first spectral images corresponding to different bands to obtain the spectral image corresponding to the first remote sensing image.

[0097] The pixel classification unit is used to calculate the normalized water body index corresponding to the target sea area based on the first remote sensing image, and to perform pixel classification on the first remote sensing image according to the normalized water body index, so as to divide the first remote sensing image into a clean water image and a sewage image.

[0098] Embodiment 2

[0099] This embodiment further discloses a method for predicting ocean water pollution based on remote sensing images. For details, please refer to Figure 3 , mainly including steps 301 to 305, wherein the steps are mainly:

[0100] Step 301: Acquire a first remote sensing image corresponding to the target sea area, and perform spectral feature enhancement processing on the first remote sensing image based on a band separation algorithm to calculate the spectral values ​​of each band corresponding to the first remote sensing image, and then extract the spectral features and pixel features corresponding to the first remote sensing image according to the spectral values ​​of each band.

[0101] In this embodiment, this step mainly includes: performing minimum noise separation change processing on the first remote sensing image to extract the corresponding pixel values ​​of each band, and calculating the square sum of the pixel values ​​of each band to generate a spectral image corresponding to the first remote sensing image, and at the same time calculating the normalized water body index corresponding to the target sea area, and classifying the pixels of the first remote sensing image according to the normalized water body index, so as to divide the first remote sensing image into a clean water image and a sewage image.

[0102] In this embodiment, a band separation algorithm is used to enhance the spectral features corresponding to the water body in the target sea area, so as to facilitate the extraction of spectral features based on the enhanced spectral features. Specifically, principal component analysis method, minimum noise separation change processing and other methods can be used. Furthermore, after using the band separation algorithm to enhance the spectral features in the first remote sensing image, the square sum of the pixel values ​​of each band is calculated to generate a spectral image corresponding to the pixel values ​​of each band, and the spectral images are normalized to reduce interference in the spectral image. Finally, the spectral images of different bands are added and integrated to generate a spectral image corresponding to the first remote sensing image.

[0103] Furthermore, a normalized water index corresponding to the first remote sensing image is calculated. Specifically, a clustering algorithm, such as a linear iterative clustering method, a K-MEANS clustering algorithm, etc., is used to segment the first remote sensing image into a number of pixels, and then the normalized water index corresponding to each pixel is calculated. Pixels of the first remote sensing image are classified according to the normalized water index and a preset index threshold, and the first remote sensing image is divided into a clean water image and a sewage image.

[0104] Specifically, the calculation method of the normalized water index is:

[0105] ;

[0106] Among them, the is the pixel currently being calculated, is a pixel point in the currently calculated pixel, are all the pixel points in the currently calculated pixel, is the green band value of pixel i in the currently calculated pixel, is the mid-infrared band value of pixel point i in the currently calculated pixel, The normalized water index corresponding to the pixel currently calculated is set as follows: wherein a threshold is set, and the pixel is judged to correspond to a clean water image or a sewage image according to the threshold and the normalized water index.

[0107] Step 302: Calculate the similarity between the spectral features and pixel features and the remote sensing images and spectral features corresponding to each type of polluted water body in the pre-constructed comparison data set, so as to determine whether the target sea area is polluted and the type of pollution based on the similarity.

[0108] In this embodiment, this step mainly includes: obtaining the spectral image and remote sensing image corresponding to each type of polluted water body through big data retrieval, and constructing a polluted water body data set.

[0109] Specifically, in this embodiment, remote sensing images corresponding to each type of polluted water body occurring in the target sea area are obtained through big data retrieval, and spectral images corresponding to the remote sensing images are extracted according to the same spectral feature extraction method, so as to construct pollution data corresponding to each type of polluted water body according to the spectral image and the remote sensing image, and then construct a comparative data set according to the pollution data and the type of the polluted water body.

[0110] Furthermore, similarity calculation is performed between the spectral image and the sewage image corresponding to the first remote sensing image and the polluted water body dataset to obtain the similarity between the first remote sensing image and each type of polluted water body.

[0111] Specifically, the pixel overlap of the spectral image and sewage image corresponding to the first remote sensing image is calculated with the spectral image and remote sensing image corresponding to each polluted water body in the comparison data set, so as to calculate the similarity between the first remote sensing image and each polluted water body based on the pixel overlap.

[0112] The similarity is compared with a preset similarity threshold to determine whether water pollution occurs in the target sea area and obtain the type of the polluted water according to the comparison result.

[0113] Specifically, a similarity threshold is set. When the similarity between the first remote sensing image and each type of polluted water body is less than the similarity threshold, it is determined that no water pollution has occurred in the target sea area. When the similarity between the first remote sensing image and several types of polluted water bodies is greater than the similarity threshold, the type of polluted water body corresponding to the highest similarity value is the type of pollution occurring in the target sea area.

[0114] Step 303: After determining that water pollution occurs in the target sea area, obtain the sea surface monitoring data and the first time series remote sensing image corresponding to the target sea area, and perform edge detection on each second remote sensing image in the first time series remote sensing image to extract the first pollution distribution image corresponding to each second remote sensing image.

[0115] In this embodiment, the step mainly includes: first, splitting the first time series remote sensing image into several second remote sensing images according to the time series, and performing grayscale processing on each of the second remote sensing images, so as to use a preset grayscale threshold to perform preliminary segmentation on each second remote sensing image after grayscale processing, and generate an initial pollution distribution image corresponding to each second remote sensing image.

[0116] Specifically, each of the second remote sensing images is first subjected to noise reduction processing to obtain a denoised remote sensing image, and then the denoised remote sensing image is subjected to grayscale processing to generate a corresponding grayscale image, and the grayscale value corresponding to each pixel in the grayscale image is calculated, and then the grayscale value of the grayscale image corresponding to the target sea area in a clear water state is pre-saved as the grayscale threshold, and the second remote sensing image is preliminarily segmented according to the pixel points whose grayscale values ​​are greater than the grayscale threshold to generate a corresponding initial pollution distribution image.

[0117] Secondly, the initial pollution distribution image is segmented into superpixels, and the corresponding texture features and color distribution features are extracted based on the results of the superpixel segmentation. Specifically, the texture of the superpixel segmentation result is removed by a filtering method based on a total variation model to obtain the color distribution features, and the superpixel segmentation result is filtered by a short-time windowed Fourier transformer to obtain the texture features.

[0118] Finally, feature fusion techniques such as linear fusion technology and Bayesian fusion are used to fuse the texture features and the color distribution features to generate fusion features for image segmentation. Then, based on the fusion features, a clustering algorithm is used to perform image segmentation on each of the initial pollution distribution images, and the results of the image segmentation are regionally merged to generate a first pollution distribution image corresponding to each of the initial pollution distribution images.

[0119] Step 304: Calculate the remote sensing reflectivity corresponding to each of the first pollution distribution images according to a preset atmospheric correction algorithm, and input the remote sensing reflectivity and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to output the pollutant concentration corresponding to each of the first pollution distribution images through the pollutant concentration inversion model.

[0120] In this embodiment, the steps are mainly as follows: first, the atmospheric correction algorithm is used to perform atmospheric correction on the first pollution distribution image to obtain the remote sensing reflectivity corresponding to each of the first pollution distribution images, and then the remote sensing reflectivity and the sea surface monitoring data are input into the pre-trained pollutant concentration inversion model.

[0121] Secondly, in selecting the pollutant concentration inversion model, the corresponding pollutant concentration inversion model is selected from the trained pollutant concentration inversion model set according to the type of pollutant, and then the concentration of each of the first pollution distribution images is inverted. Specifically, in the inversion process, the first branch network preset in the pollutant concentration inversion model is first used to perform preliminary concentration inversion to generate an initial concentration inversion result.

[0122] Then, based on the initial concentration inversion result, the remote sensing reflectivity, the sea surface monitoring data and the initial concentration inversion result are input into the second branch network and the third branch network of the target pollutant concentration inversion model to determine the target pollutant concentration corresponding to the first pollution distribution image.

[0123] Wherein, a linear regression algorithm is used in the second branch network to perform one-hot encoding processing on the sea surface monitoring data to obtain a sparse matrix, and determine the corresponding first concentration inversion result according to the remote sensing reflectivity, the sparse matrix and each initial concentration inversion result. Wherein, the sea surface monitoring data is processed by one-hot encoding through a one-hot encoding layer to obtain a sparse matrix, and then the first concentration inversion result is determined according to the remote sensing reflectivity, the sparse matrix and each initial concentration inversion result through a linear regression layer.

[0124] In the third branch network, multiple composite nonlinear functions are used to reduce the dimension of the sea surface monitoring data to obtain a dense matrix, and the second concentration inversion result is determined according to the remote sensing reflectivity, the dense matrix and each initial concentration inversion result. The sea surface monitoring data is reduced in dimension by an embedding layer to obtain a dense matrix, and then the second concentration inversion result is determined by a hidden layer according to the remote sensing reflectivity, the dense matrix and each initial concentration inversion result.

[0125] Finally, the first concentration inversion result and the second concentration inversion result are weighted by the fourth branch network preset in the target pollutant concentration inversion model to obtain the target concentration inversion result.

[0126] Thus, the pollutant concentration corresponding to each of the first pollution distribution images is outputted through the pollutant concentration inversion model.

[0127] Step 305: Mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image, and input the second time series remote sensing image into a pre-trained ocean pollution degree prediction model, predict the pollution degree corresponding to the target sea area through the ocean pollution degree determination model, and generate early warning information based on the pollution degree.

[0128] In this embodiment, this step is specifically as follows: extracting the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fusing the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image, and then predicting the pollution degree corresponding to the target sea area based on the pollution trend characteristics through the classifier preset in the marine pollution degree prediction model, and generating early warning information based on the pollution degree.

[0129] Specifically, in this embodiment, the second time series remote sensing image is input into the first convolution layer of a preset marine pollution degree prediction model, the network structure of the marine pollution degree prediction model includes a first convolution layer, a second convolution layer and a fully connected layer, the second time series remote sensing image is input into the first convolution layer, and the second time series remote sensing image is subjected to image separation and data separation by using the spatiotemporal separation technology preset in the first convolution layer to obtain spatial information data and concentration information data corresponding to the second time series remote sensing image, wherein the first convolution layer is composed of multiple convolution kernels and multiple fully connected layers, and the fully connected layer is connected after each convolution block.

[0130] The second convolution layer preset in the marine pollution degree determination model performs convolution processing on the spatial information data and the concentration information data respectively. The convolution layer obtains feature information of the spatial information data and the concentration information data, and uses several convolution kernels to convolve the data of the previous layer. The result is output as a two-dimensional vector. After the convolution operation, several two-dimensional outputs can be obtained, and the spatial spread characteristics and the concentration spread characteristics can be output. The spatial spread characteristics and the concentration spread characteristics can be upgraded and feature fused using a fully connected layer to obtain the initial pollution trend characteristics corresponding to the second time series remote sensing image. The initial pollution trend characteristics corresponding to the second time series remote sensing image are calculated by the mean average function so that the obtained characteristics are not affected and are more universal.

[0131] Furthermore, the contribution of each decision tree is adjusted using the random forest algorithm used in the classifier so that the pollution trend characteristics are respectively input into each decision tree to predict the pollution degree, and multiple prediction results are obtained, and then the final prediction result is obtained by voting or averaging.

[0132] The present embodiment discloses a method for predicting ocean water pollution based on remote sensing images. In the prediction method, a first remote sensing image corresponding to a target sea area is first obtained, and spectral features and pixel features corresponding to the first remote sensing image are extracted, so that according to the spectral features and the pixel features, it is determined by a preset big data search whether water pollution currently occurs in the target sea area, so that the pollution situation of the ocean can be monitored in real time. After it is determined that water pollution occurs in the ocean, sea surface monitoring data and time series remote sensing images corresponding to the target sea area are obtained, and edge detection is performed on the time series remote sensing images to obtain only a first pollution distribution image of water pollution occurring in the ocean. This not only reduces the amount of data processing in the later stage and improves the monitoring efficiency by intercepting the continuous images, but also facilitates the real-time extraction of spatial spread features of pollution according to the edge detection results of the pollution distribution image in the later stage, thereby improving the accuracy of prediction.

[0133] After obtaining a first pollution distribution image corresponding to only polluted seawater, atmospheric processing is performed on the first pollution distribution image to obtain remote sensing reflectivity, so as to reduce the impact of atmospheric reasons on the acquisition of the remote sensing image and improve the accuracy of data used to predict the degree of pollution. Furthermore, based on the fact that the degree of pollution in the ocean depends not only on the pollution itself but also on the type of ocean, the present invention uses a pollutant concentration inversion model to invert the pollution concentration of the remote sensing reflectivity corresponding to the first pollution distribution image and the sea surface monitoring data corresponding to the target sea area, so as to obtain the pollutant concentration corresponding to each pollution distribution image.

[0134] Finally, the marine pollution degree prediction model pre-constructed in the present invention extracts the spatial spread characteristics corresponding to the image area and the concentration spread characteristics corresponding to the concentration change trend in the second time series remote sensing image, so as to predict the pollution degree by integrating the above characteristics, and issue an early warning based on the predicted pollution degree, thereby improving the accuracy of the prediction.

[0135] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting marine water pollution based on remote sensing images, characterized in that: include: Acquire a first remote sensing image corresponding to the target sea area, and extract spectral features and pixel features corresponding to the first remote sensing image by using a preset band separation algorithm; According to the spectral characteristics and the pixel characteristics, pollution analysis is performed on the target sea area through big data retrieval to determine whether water pollution occurs in the target sea area; When it is determined that water pollution occurs in the target sea area, sea surface monitoring data and a first time series remote sensing image corresponding to the target sea area are obtained, and edge detection is performed on each second remote sensing image in the first time series remote sensing image to extract a first pollution distribution image corresponding to each second remote sensing image; Performing atmospheric correction processing on each of the first pollution distribution images to obtain a corresponding remote sensing reflectance, and inputting the remote sensing reflectance and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain a pollutant concentration corresponding to each of the first pollution distribution images, so as to mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image; Inputting the second time series remote sensing image into a pre-trained marine pollution degree prediction model, extracting the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fusing the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image; According to the pollution trend characteristics, the pollution degree corresponding to the target sea area is predicted by a classifier preset in the marine pollution degree prediction model, and early warning information is generated according to the pollution degree.

2. The method for predicting ocean water pollution based on remote sensing images according to claim 1, characterized in that: The extracting spectral features and pixel features corresponding to the first remote sensing image by using a preset band separation algorithm includes: Performing minimum noise separation change processing on the first remote sensing image, extracting pixel values ​​of each band corresponding to the first remote sensing image, and calculating the square sum of the pixel values ​​of each band to generate first spectral images corresponding to different bands; Integrate the first spectral images corresponding to different bands to obtain a spectral image corresponding to the first remote sensing image; A normalized water body index corresponding to the target sea area is calculated according to the first remote sensing image, and pixels of the first remote sensing image are classified according to the normalized water body index to divide the first remote sensing image into a clean water image and a sewage image.

3. The method for predicting ocean water pollution based on remote sensing images according to claim 2 is characterized in that: The pollution analysis of the target sea area through big data retrieval to determine whether water pollution occurs in the target sea area includes: Through big data retrieval, the spectral image and remote sensing image corresponding to each polluted water body are obtained to construct a polluted water body data set; Calculate the similarity between the spectral image and the sewage image corresponding to the first remote sensing image and the polluted water body dataset to obtain the similarity between the first remote sensing image and each type of polluted water body; The similarity is compared with a preset similarity threshold to determine whether water pollution occurs in the target sea area and obtain the type of the polluted water according to the comparison result.

4. The method for predicting ocean water pollution based on remote sensing images according to claim 1 is characterized in that: The performing edge detection on each second remote sensing image in the first time series remote sensing image to extract a first pollution distribution image corresponding to each second remote sensing image includes: Splitting the first time series remote sensing image into a plurality of second remote sensing images according to the time series, and performing grayscale processing on each of the second remote sensing images, so as to perform preliminary segmentation on each second remote sensing image after grayscale processing by using a preset grayscale threshold, and generate an initial pollution distribution image corresponding to each second remote sensing image; Performing pixel segmentation on each of the initial pollution distribution images, and extracting texture features and color distribution features corresponding to each of the initial pollution distribution images according to the pixel segmentation result; According to the texture features and the color distribution features, a clustering algorithm is used to perform image segmentation on each of the initial pollution distribution images, and the results of the image segmentation are regionally merged to generate a first pollution distribution image corresponding to each of the initial pollution distribution images.

5. The method for predicting ocean water pollution based on remote sensing images according to claim 3 is characterized in that: The step of inputting the remote sensing reflectivity and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain the pollutant concentration corresponding to each of the first pollution distribution images includes: According to the type of the polluted water body, determining a target pollutant concentration inversion model corresponding to the target sea area from a set of pre-trained pollutant concentration inversion models; According to the remote sensing reflectivity and the sea surface monitoring data, an initial concentration inversion result corresponding to the first pollution distribution image is outputted through a first branch network preset in the target pollutant concentration inversion model; wherein the sea surface monitoring data includes ocean type data and marine ecological data; Performing one-hot encoding processing on the sea surface monitoring data through a second branch network preset in the target pollutant concentration inversion model to obtain a sparse matrix, so as to output a first concentration inversion result corresponding to the first pollution distribution image according to the sparse matrix, the remote sensing reflectivity and the initial concentration inversion result; Performing dimensionality reduction processing on the sea surface monitoring data by a third branch network preset in the target pollutant concentration inversion model to obtain a dense matrix, so as to output a second concentration inversion result corresponding to the first pollution distribution image according to the dense matrix, the remote sensing reflectivity and the initial concentration inversion result; The first concentration inversion result and the second concentration inversion result are weighted by the fourth branch network preset in the target pollutant concentration inversion model to obtain the pollutant concentration corresponding to the first pollution distribution image.

6. The method for predicting ocean water pollution based on remote sensing images according to claim 1, characterized in that: The extracting of the spatial spread characteristics and the concentration spread characteristics corresponding to the second time series remote sensing image by the marine pollution degree determination model includes: The first convolution layer preset in the marine pollution degree determination model performs spatial separation and data separation on the second time series remote sensing image to obtain spatial information data and concentration information data corresponding to the second time series remote sensing image; The second convolution layer preset in the marine pollution degree determination model performs convolution processing on the spatial information data and the concentration information data respectively, so as to obtain the spatial spread characteristics corresponding to the spatial information data and the concentration spread characteristics corresponding to the concentration information data.

7. The method for predicting ocean water pollution based on remote sensing images according to claim 6 is characterized in that: The fusing of the spatial spread feature and the concentration spread feature to obtain the pollution trend feature corresponding to the second time series remote sensing image includes: Performing dimensionality up-processing and feature fusion on the spatial spread feature and the concentration spread feature according to the fully connected layer preset in the marine pollution degree prediction model, so as to obtain the initial pollution trend feature corresponding to the second time series remote sensing image; The initial pollution trend characteristics are averaged according to a preset average function to obtain the pollution trend characteristics corresponding to the second time series remote sensing image.

8. The method for predicting ocean water pollution based on remote sensing images according to claim 1, characterized in that: The method of predicting the pollution degree corresponding to the target sea area by using a classifier preset in the marine pollution degree prediction model includes: Predicting the pollution degree of the pollution trend characteristics respectively by using a plurality of different decision trees preset in the classifier, and obtaining a first pollution degree corresponding to each decision tree; The plurality of first pollution levels are averaged through a preset linear regression calculation function to obtain the pollution level corresponding to the target sea area.

9. A marine water pollution prediction system based on remote sensing images, characterized in that: It includes a feature extraction module, a pollution determination module, an image segmentation module, an image calibration module, an image processing module and a pollution prediction module; The feature extraction module is used to obtain a first remote sensing image corresponding to the target sea area, and extract spectral features and pixel features corresponding to the first remote sensing image through a preset band separation algorithm; The pollution determination module is used to perform pollution analysis on the target sea area through big data retrieval according to the spectral characteristics and the pixel characteristics, and determine whether water pollution occurs in the target sea area; The image segmentation module is used to obtain sea surface monitoring data and a first time series remote sensing image corresponding to the target sea area after determining that water pollution occurs in the target sea area, and perform edge detection on each second remote sensing image in the first time series remote sensing image to extract a first pollution distribution image corresponding to each second remote sensing image; The image calibration module is used to perform atmospheric correction processing on each of the first pollution distribution images to obtain a corresponding remote sensing reflectance, and input the remote sensing reflectance and the sea surface monitoring data into a pre-trained pollutant concentration inversion model to obtain the pollutant concentration corresponding to each of the first pollution distribution images, so as to mark each of the first pollution distribution images according to the pollutant concentration to generate a second time series remote sensing image; The image processing module is used to input the second time series remote sensing image into a pre-trained marine pollution degree prediction model, extract the spatial spread characteristics and concentration spread characteristics corresponding to the second time series remote sensing image through the marine pollution degree determination model, and fuse the spatial spread characteristics and concentration spread characteristics to obtain the pollution trend characteristics corresponding to the second time series remote sensing image; The pollution prediction module is used to predict the pollution degree corresponding to the target sea area according to the pollution trend characteristics through a classifier preset in the marine pollution degree prediction model, and generate early warning information according to the pollution degree.

10. The marine water pollution prediction system based on remote sensing images according to claim 9, characterized in that: The feature extraction module includes a spectrum extraction unit and a pixel classification unit; The spectrum extraction unit is used to perform minimum noise separation change processing on the first remote sensing image, extract the pixel values ​​of each band corresponding to the first remote sensing image, and calculate the square sum of the pixel values ​​of each band to generate first spectrum images corresponding to different bands; integrate the first spectrum images corresponding to different bands to obtain the spectrum image corresponding to the first remote sensing image; The pixel classification unit is used to calculate the normalized water body index corresponding to the target sea area based on the first remote sensing image, and to perform pixel classification on the first remote sensing image according to the normalized water body index, so as to divide the first remote sensing image into a clean water image and a sewage image.

Images