Processing method, device and equipment for remote sensing prediction model of water bloom in deep and large reservoirs

Abstract

The present application provides a processing method, device and processing equipment for a remote sensing prediction model of water bloom in deep and large reservoirs, which is used to innovatively use a long short-term memory network based on a self-attention mechanism for remote sensing prediction of water bloom. The model integrates the self-attention mechanism and the long short-term memory network, and can more effectively identify and capture key features in long time series data. Therefore, the remote sensing prediction model of water bloom in deep and large reservoirs obtained by training the model training scheme built by the long short-term memory network based on the self-attention mechanism in this application has achieved a solution effect of significantly improving the accuracy and reliability in the remote sensing prediction of water bloom, which can provide important decision-making basis and technical support for early water bloom warning and later water bloom prevention and control intervention in deep and large reservoirs, thereby effectively promoting the refined management of regional water resources and the progress of water ecological protection.

Description

Technical Field

[0001] The present application relates to the field of water quality monitoring, and specifically to a processing method, device and processing equipment for a remote sensing prediction model of water blooms in deep and large reservoirs. Background Art

[0002] Compared with ordinary lakes and small and medium-sized reservoirs, deep and large reservoirs are famous for their vast water area and huge storage capacity. They play a key role in regional water resources management and undertake important functions such as water supply, flood control, and irrigation.

[0003] The water levels of deep and large reservoirs are more significantly affected by human regulation, resulting in large fluctuations in water levels. During the process of water level fluctuations, the formation of a drawdown zone promotes the exchange of materials between the drawdown zone of the reservoir and the water body and surrounding land. For example, excess nitrogen, phosphorus and other nutrients accumulated in the drawdown zone are washed into the reservoir. This exchange not only changes the water quality of the reservoir, but also has a profound impact on the ecological balance and the occurrence of algal blooms.

[0004] This complex exchange of substances and ecological impacts is particularly evident in the backwater areas of rivers entering deep and large reservoirs. Due to the combined effects of their unique geographical location and reservoir operation, these areas exhibit unique hydrodynamic characteristics, making them a high-incidence zone for eutrophication and algal blooms. Unlike traditional rivers, which have high flow rates, rapid water turnover, and difficulty accumulating nutrients, the backwater areas of deep and large reservoirs, after impoundment, are affected by the reservoir's water support, resulting in slower flow rates, sediment deposition, and the accumulation of nutrients.

[0005] In addition, changes in surrounding human activities and meteorological conditions have also increased the risk of algal blooms in backwater areas, such as agricultural non-point source pollution and domestic sewage discharge, which continuously inject excessive nitrogen, phosphorus and other nutrients into the water body.

[0006] Therefore, the outbreak of algal blooms in tributaries not only threatens the overall water quality and function of the reservoir, affects water supply security, and reduces the availability of water resources, but may also cause long-term destructive impacts on the ecological environment. Therefore, it is particularly important to monitor and prevent algal blooms in the backwater areas of rivers entering deep and large reservoirs.

[0007] Remote sensing technology has become a powerful tool for algal bloom monitoring and early warning due to its wide coverage, fast monitoring speed and powerful dynamic monitoring capabilities.

[0008] However, the inventors of this application found that when using relevant neural network models to predict algal blooms based on remote sensing images, there are still certain defects in the prediction accuracy from the actual performance, which affects the development of deep and large reservoir management work. Summary of the Invention

[0009] The present application provides a processing method, device and processing equipment for a remote sensing prediction model of water bloom in deep and large reservoirs, which is used to innovatively use a long short-term memory network (Key-Attention Network - Long Short-Term Memory Network, KAN-LSTM) based on a self-attention mechanism for remote sensing prediction of water bloom. The model integrates the self-attention mechanism and the long short-term memory network, and can more effectively identify and capture key features in long time series data. Therefore, the remote sensing prediction model of water bloom in deep and large reservoirs obtained by training the model training scheme built by the long short-term memory network based on the self-attention mechanism in this application has achieved a solution effect of significantly improving the accuracy and reliability of water bloom remote sensing prediction work, which can provide important decision-making basis and technical support for early water bloom warning and later water bloom prevention and control intervention in deep and large reservoirs, thereby effectively promoting the refined management of regional water resources and the progress of water ecological protection.

[0010] In the first aspect, the present application provides a method for processing a remote sensing prediction model of algal blooms in deep and large reservoirs, the method comprising:

[0011] For target deep and large reservoirs, obtain sample multispectral satellite images;

[0012] Based on the preset algal bloom characterization indicators, the indicator values ​​of each location of the sample multispectral satellite image are calculated;

[0013] Combining the index values ​​and index thresholds of each location in the sample multispectral satellite image, determine whether there is algal bloom at each location in the multispectral satellite image, and form an algal bloom binary distribution layer corresponding to the sample multispectral satellite image;

[0014] Based on sample multispectral satellite images and binary distribution layers of algal blooms, the cumulative effect of preset indicators and the multi-day average of preset indicators on algal blooms were quantified. The preset indicators included water quality parameters, reservoir hydrological rhythms, and meteorological parameters.

[0015] Based on the quantitative results, the target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms are screened out from the preset indicators through redundancy analysis method.

[0016] The indicator data corresponding to the target indicator and the water bloom binary distribution layer are configured as training samples to train the water bloom remote sensing prediction model of Shenda Reservoir. The water bloom remote sensing prediction model of Shenda Reservoir is specifically a long short-term memory network based on the self-attention mechanism. The water bloom remote sensing prediction model of Shenda Reservoir is configured with 7 output layers, and each output layer corresponds to the prediction output of different days in the next 7 days. The water bloom remote sensing prediction model of Shenda Reservoir is used to predict whether there will be water bloom in the next 7 days based on the indicator data input into the model corresponding to the target indicator.

[0017] In a second aspect, the present application provides a processing device for a remote sensing prediction model of algal blooms in deep and large reservoirs, the device comprising:

[0018] An acquisition unit is used to acquire sample multispectral satellite images of a target deep and large reservoir;

[0019] A calculation unit, configured to calculate an index value of each position of a sample multispectral satellite image based on a preset algal bloom characterization index;

[0020] a determination unit, configured to determine whether algal bloom exists at each location in the multispectral satellite image by combining the index value and the index threshold of each location in the sample multispectral satellite image, and to form an algal bloom binary distribution layer corresponding to the sample multispectral satellite image;

[0021] A quantification unit is used to quantify the impact of the cumulative effect of preset indicators on the algal bloom situation and the impact of the multi-day average of the preset indicators on the algal bloom situation based on the sample multispectral satellite image and the algal bloom binary distribution layer, wherein the preset indicators include water quality parameters, reservoir hydrological rhythms and meteorological parameters;

[0022] A screening unit is used to screen target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms from preset indicators through a redundancy analysis method based on the quantitative results;

[0023] The training unit is used to configure the indicator data corresponding to the target indicator and the water bloom binary distribution layer with training samples, and train the water bloom remote sensing prediction model of the Shenda Reservoir. The water bloom remote sensing prediction model of the Shenda Reservoir is specifically a long short-term memory network based on the self-attention mechanism. The water bloom remote sensing prediction model of the Shenda Reservoir is configured with 7 output layers, each output layer corresponds to the prediction output of different days in the next 7 days. The water bloom remote sensing prediction model of the Shenda Reservoir is used to predict whether there will be water bloom in the next 7 days based on the indicator data input into the model corresponding to the target indicator.

[0024] In a third aspect, the present application provides a processing device comprising a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method provided in the first aspect of the present application or any possible implementation of the first aspect of the present application is executed.

[0025] In a fourth aspect, the present application provides a computer-readable storage medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the method provided in the first aspect of the present application or any possible implementation of the first aspect of the present application.

[0026] From the above content, it can be concluded that this application has the following beneficial effects:

[0027] Aiming at the goal of remote sensing prediction of water blooms in deep and large reservoirs, this application innovatively uses long short-term memory networks based on self-attention mechanisms for remote sensing prediction of water blooms. This model integrates the self-attention mechanism and the long short-term memory network, which can more effectively identify and capture key features in long time series data. Therefore, the water bloom remote sensing prediction model for deep and large reservoirs trained by the model training scheme built by the long short-term memory network based on the self-attention mechanism in this application has achieved the effect of significantly improving the accuracy and reliability of water bloom remote sensing prediction work, which can provide important decision-making basis and technical support for early water bloom warning and later water bloom prevention and control intervention in deep and large reservoirs, thereby effectively promoting the refined management of regional water resources and the progress of water ecological protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 A flowchart of a processing method for the remote sensing prediction model of water bloom in deep and large reservoirs in this application;

[0030] Figure 2 A structural diagram of a processing device for the remote sensing prediction model of water bloom in deep and large reservoirs in this application;

[0031] Figure 3 This is a structural diagram of the processing equipment for this application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0033] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. The naming or numbering of steps in this application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can be changed in the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0034] The division of modules in this application is a logical division. In actual application, there may be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between modules can be electrical or other similar forms, which are not limited in this application. Moreover, the modules or submodules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed into multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application.

[0035] Before introducing the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs provided by this application, the background content involved in this application is first introduced.

[0036] The processing method, device and computer-readable storage medium of the remote sensing prediction model of water bloom in deep and large reservoirs provided in this application can be applied to processing equipment for innovatively using the long short-term memory network based on the self-attention mechanism for remote sensing prediction of water bloom. The model integrates the self-attention mechanism and the long short-term memory network, and can more effectively identify and capture key features in long time series data. Therefore, the remote sensing prediction model of water bloom in deep and large reservoirs obtained by training the model training scheme built by the long short-term memory network based on the self-attention mechanism in this application has achieved the effect of significantly improving the accuracy and reliability of remote sensing prediction of water bloom, which can provide important decision-making basis and technical support for early water bloom warning and later water bloom prevention and control intervention in deep and large reservoirs, thereby effectively promoting the refined management of regional water resources and the progress of water ecological protection.

[0037] The processing method for the remote sensing prediction model for algal blooms in deep and large reservoirs mentioned in this application can be executed by a processing device for the remote sensing prediction model for algal blooms in deep and large reservoirs, or by various types of processing devices such as a server, physical host, or user equipment (UE) that integrates the processing device for the remote sensing prediction model for algal blooms in deep and large reservoirs. The processing device for the remote sensing prediction model for algal blooms in deep and large reservoirs can be implemented using hardware or software. The UE can specifically be a terminal device such as a smartphone, tablet computer, laptop computer, desktop computer, or personal digital assistant (PDA). The processing device can be configured as a device cluster.

[0038] In specific applications, considering that the purpose of this application is mainly to train the remote sensing prediction model of water bloom in deep and large reservoirs based on existing data, the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs in this application, or the processing equipment equipped with the corresponding application service of the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs in this application, usually only needs to meet the required data processing capabilities, and the specific equipment type and equipment deployment form are relatively flexible.

[0039] If it also involves the collection of source data, namely multispectral satellite images, or the practical application of remote sensing prediction models for water blooms in deep and large reservoirs, further adaptive adjustments to the processing equipment will be required.

[0040] As an example, the processing device may specifically include a first device that performs a training task of a remote sensing prediction model for water blooms in deep and large reservoirs in a laboratory environment and a second device that performs an actual application of the remote sensing prediction model for water blooms in deep and large reservoirs on site.

[0041] It is easy to see that the processing equipment can be adjusted in terms of specific equipment type and equipment deployment form according to the application scenarios and data processing links involved in actual situations.

[0042] Next, we will introduce the processing method of the remote sensing prediction model for algal blooms in deep and large reservoirs provided in this application.

[0043] First, see Figure 1 , Figure 1 A flow chart of a processing method for a remote sensing prediction model of water bloom in a deep and large reservoir provided by the present application is shown. The processing method for a remote sensing prediction model of water bloom in a deep and large reservoir provided by the present application may specifically include the following steps S101 to S106:

[0044] Step S101, acquiring sample multispectral satellite images for a target deep and large reservoir;

[0045] It can be understood that the model training scheme involved in the present application is to configure training samples specifically based on the target deep and large reservoir to predict whether algal bloom has occurred, so as to train a deep and large reservoir algal bloom remote sensing prediction model that is highly adaptable to the target deep and large reservoir.

[0046] Among them, the target deep and large reservoir can be any reservoir that meets the criteria for determining a deep and large reservoir. The sample multispectral satellite image of the target deep and large reservoir can be an actual image, such as a related satellite telemetry data product in the past, or an image obtained through corresponding data processing based on the actual image of the target deep and large reservoir, or an image directly obtained for the target deep and large reservoir. These are all possible under actual circumstances.

[0047] As an example, multispectral satellite images of satellite telemetry data products such as HJ-2A / B, Sentinel-2, and GF series (GF1, GF1-B, GF1-C, GF1-D, and GF6) can be obtained.

[0048] It can be seen that the sample multispectral satellite images can also have the characteristics of multiple sources or different data sources in actual operation. Therefore, further adjustments can be made based on the differences between different data sources to form fused multi-source remote sensing images to adapt to the model input after the actual application planned by the subsequent deep and large reservoir water bloom remote sensing prediction model.

[0049] In addition, it can be understood that the remote sensing prediction model for water bloom in deep and large reservoirs to be trained in this application involves time series characteristics in the prediction process. Therefore, for sample multispectral satellite images, it usually involves a larger time range and is a continuous multispectral satellite image.

[0050] As an example, continuous multispectral satellite images with a time span of 3 years can be used.

[0051] Step S102, calculating the index value of each position of the sample multispectral satellite image based on the preset algal bloom characterization index;

[0052] It can be understood that, corresponding to the current image, the model training process involves corresponding annotations or true values, that is, whether algal bloom occurs or whether algal bloom exists, which can be determined in combination with preset algal bloom characterization indicators.

[0053] In this way, the specific index values ​​of each position of the sample multispectral satellite image under the preset water bloom characterization index can be calculated, where the position is usually in pixels. Correspondingly, what this application aims to achieve is the water bloom remote sensing prediction effect at a pixel-by-pixel or pixel granularity.

[0054] In this way, the detailed index values ​​at each location of the target deep and large reservoir will lay the foundation for the subsequent production of the binary distribution layer of water bloom.

[0055] Step S103, combining the index value and index threshold of each position of the sample multispectral satellite image, determining whether there is an algal bloom at each position of the multispectral satellite image, and forming an algal bloom binary distribution layer corresponding to the sample multispectral satellite image;

[0056] It can be understood that the index value quantifies the degree of algae in the water body, and whether it is consistent with the actual situation of whether algal bloom has occurred needs to be divided in combination with the corresponding threshold (which can be denoted as K).

[0057] In this regard, based on the index values ​​of each position of the sample multispectral satellite image obtained previously, combined with the index threshold, it can be determined whether there is algal bloom at each position of the multispectral satellite image. The determination result of whether there is algal bloom at each position of the multispectral satellite image itself corresponds to a binary processing, that is, the presence of algal bloom and the absence of algal bloom. In this case, the application can continue to produce a corresponding algal bloom binary distribution layer that reflects the overall situation.

[0058] Among them, the binary distribution layer of water bloom is convenient for display and recognition processing by the model. It can be used as the corresponding label or true value involved in the model training process.

[0059] Step S104, based on the sample multispectral satellite image and the algal bloom binary distribution layer, quantifying the impact of the cumulative effect of preset indicators on the algal bloom situation and the impact of the multi-day average of the preset indicators on the algal bloom situation, wherein the preset indicators include water quality parameters, reservoir hydrological rhythms, and meteorological parameters;

[0060] It is understandable that this application believes that in actual situations, different algal bloom characterization indicators may have different contributions to the algal bloom remote sensing prediction work of different deep and large reservoirs. Therefore, it is possible to screen them through the importance specially designed by this application to obtain the algal bloom remote sensing prediction target that is suitable for the target deep and large reservoir.

[0061] Under this design, this application can target the preset indicators in the preliminary stage. On the basis of sample multispectral satellite images, on the one hand, it can quantify the impact of the cumulative effect of the preset indicators on the algal bloom situation, and on the other hand, it can quantify the impact of the multi-day average of the preset indicators on the algal bloom situation. Among them, the multi-day average corresponds to the configuration of the subsequent model, which can involve the average value of 2 days, 3 days,..., 7 days.

[0062] In this section, we can see that this application not only discloses the three major parameter indicators designed by this application that can contribute to the remote sensing prediction of algal blooms (namely, water quality parameters, reservoir hydrological rhythms and meteorological parameters), but also gives the two specific aspects in which these parameter indicators can contribute to the remote sensing prediction of algal blooms (namely, cumulative effect and the impact of multi-day averages).

[0063] For the former, different water quality parameters and different meteorological parameters are understandable and generally fall within the scope of existing technologies. In specific operations, existing indicators can be directly used, or they can be optimized (increased, reduced, replaced, etc.) based on the existing indicators, or self-developed novel indicators can be used. As for the reservoir hydrological rhythm, this application introduces a new indicator that reflects the unique laws of hydrological changes of the Shenzhen-Da Reservoir itself, which will help to provide a more adaptive and accurate contribution to the remote sensing prediction of algal blooms.

[0064] As an example, water quality parameters may specifically include total phosphorus, total nitrogen, chemical oxygen demand, chlorophyll a, etc., and meteorological parameters may specifically include temperature, air pressure, etc.

[0065] For the latter, this application does not directly capture the mapping relationship between changes in basic indicators and changes in algal bloom conditions as in conventional solutions / thinking, but instead configures cumulative effects and multi-day averages between the two, so as to infer the algal bloom conditions more delicately and accurately from a deeper perspective.

[0066] Step S105: Based on the quantification results, target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms are screened from preset indicators using a redundancy analysis method;

[0067] After deeply quantifying the impact of different aspects on hydrological prediction, we can, based on the corresponding quantitative results, delve into the response relationship between different indicator variables in the preset indicators in the initial stage and the water bloom binary distribution layer, and determine the relative importance of different indicator variables in the preset indicators in the initial stage for the hydrological prediction of this target deep and large reservoir. In this way, under the specific redundant analysis method adopted, we can screen out multiple indicators that meet the preset relative importance conditions, such as relative importance greater than a threshold or relative importance ranking within a preset sequence, and output them as target indicators. This target indicator corresponds to the model input content to be configured for the subsequent deep and large reservoir water bloom remote sensing prediction model.

[0068] Step S106, configure the indicator data corresponding to the target indicator and the water bloom binary distribution layer with training samples, and train the water bloom remote sensing prediction model of the deep large reservoir, wherein the water bloom remote sensing prediction model of the deep large reservoir is specifically a long short-term memory network based on the self-attention mechanism, and the water bloom remote sensing prediction model of the deep large reservoir is configured with 7 output layers, each output layer corresponds to the prediction output of different days in the next 7 days, and the water bloom remote sensing prediction model of the deep large reservoir is used to predict whether there will be water bloom in the next 7 days based on the indicator data input by the model corresponding to the target indicator.

[0069] It can be understood that this application, based on artificial intelligence (AI) technology, uses corresponding machine learning models to configure the remote sensing prediction model of water bloom in deep and large reservoirs. It specifically introduces a novel long short-term memory network based on the self-attention mechanism, namely KAN-LSTM, to configure the remote sensing prediction model of water bloom in deep and large reservoirs.

[0070] Specifically, faced with complex time series data, traditional machine learning models often find it difficult to explore the deep relationships between time series factors. Although recurrent neural networks, especially LSTM (Long Short-Term Memory Network), have performed well in the field of time series prediction and are widely used in monitoring and early warning of algal blooms, the inventors of this application found that they may ignore early key information when processing long time series data, which will limit the model's comprehensive grasp of global information, thereby affecting the accuracy of remote sensing prediction of algal blooms.

[0071] In order to break through the above limitations, this application introduces a hybrid prediction model for the first time in the field of algal bloom remote sensing prediction, namely the long short-term memory network (KAN-LSTM) based on the self-attention mechanism. This new model combines the self-attention mechanism and can more effectively identify and capture key features in long time series data. It can not only finely process local features in the time series and effectively capture the time dependence of algal bloom occurrence, but also pay comprehensive attention to global information and use the self-attention mechanism of key variables to improve prediction performance, thereby gaining a deeper understanding of the complex mechanism of algal bloom occurrence and significantly improving the accuracy and reliability of predictions.

[0072] For the long short-term memory network based on the self-attention mechanism introduced into the remote sensing prediction of algal blooms in this application, this application sets the time step of its model input layer to 7 and the model output layer to 7 neurons, that is, 7 output layers, corresponding to the remote sensing prediction of algal blooms in the next 1-7 days, and specifically uses the sigmoid activation function to output 0 or 1 to indicate whether algal blooms occur or not (usually 0 indicates no algal bloom and 1 indicates algal bloom. This setting corresponds to the labeling method of the previous algal bloom binary distribution layer).

[0073] The model training process usually includes the following:

[0074] In each round of model training, a training sample is input into the model, allowing the model to carry out the corresponding water bloom remote sensing prediction processing and realize forward propagation. Then, based on the model output results, the loss function is calculated in combination with the water bloom binary distribution layer (true value), and the model parameters are optimized according to the loss function calculation results to realize reverse propagation. In this way, through a large number of rounds of training and meeting the model training requirements such as training time, number of training times, and prediction accuracy, the model training can be completed, and a deep and large reservoir water bloom remote sensing prediction model that can be put into practical use is obtained.

[0075] It is understandable that for specific model training schemes and loss functions used in the training process, you can adopt existing schemes, further optimize them based on the existing schemes, or adopt novel self-developed schemes.

[0076] As an example, during the training process, the cross-validation method can be used to divide the training set and the validation set, and key indicators such as accuracy, precision, recall, F1 score, confusion matrix, etc. can be used to evaluate the model performance to better demonstrate the model performance.

[0077] In addition, for the remote sensing prediction model of water blooms in deep and large reservoirs, in addition to directly inputting specific data sets corresponding to target indicators, it can also be configured with the processing capability to extract specific data sets corresponding to target indicators from multispectral satellite images. This corresponds to the two model input methods that can be adopted in actual situations.

[0078] from Figure 1 It can be seen from the shown embodiments that, for the remote sensing prediction target of water bloom in deep and large reservoirs, the present application innovatively uses the long short-term memory network based on the self-attention mechanism for remote sensing prediction of water bloom. The model integrates the self-attention mechanism and the long short-term memory network, which can more effectively identify and capture key features in long time series data. Therefore, the remote sensing prediction model of water bloom in deep and large reservoirs obtained by training the model training scheme built by the long short-term memory network based on the self-attention mechanism in this application has achieved the effect of significantly improving the accuracy and reliability of remote sensing prediction of water bloom, which can provide important decision-making basis and technical support for early water bloom warning and later water bloom prevention and control intervention in deep and large reservoirs, thereby effectively promoting the refined management of regional water resources and the progress of water ecological protection.

[0079] Continue to the above Figure 1 The various steps of the illustrated embodiment and possible embodiments thereof in practical applications are explained in detail.

[0080] As an exemplary embodiment, step S101 acquires sample multispectral satellite images for a target deep and large reservoir, which may specifically include:

[0081] 1) For target deep and large reservoirs, obtain first-sample multispectral satellite imagery that meets requirements regarding image quality, cloud cover, and watershed coverage;

[0082] It can be understood that the purpose here is to select multispectral satellite images with good image quality, less cloud cover and as comprehensive coverage of the study basin as possible during the acquisition of source data.

[0083] 2) performing preprocessing including radiometric calibration and atmospheric correction on the first sample multispectral satellite image to obtain a second sample multispectral satellite image;

[0084] It can be understood that the purpose here is to further effectively improve the image quality through preprocessing such as radiation calibration and atmospheric correction. The preprocessing that can be adopted can usually directly use the existing scheme. Of course, the possibility of adopting the optimization scheme of the existing scheme or the possibility of adopting the novel scheme developed independently is not ruled out.

[0085] 3) Using the pre-configured standard multispectral satellite image as a reference, georeferencing the second sample multispectral satellite image to obtain a sample multispectral satellite image.

[0086] It can be understood that the purpose here is to use standard multispectral satellite images as a reference for georeferencing and accurately adjust the image position to obtain a more accurate high-quality image dataset. This ensures the spatial accuracy of the image data and provides a solid foundation for subsequent remote sensing prediction and analysis of algal blooms.

[0087] Among them, the preset standard multispectral satellite image can be either a selected existing satellite observation data product that meets high-quality requirements, such as Sentinel-2 satellite imagery, or a multispectral satellite image processed on the basis of existing satellite observation data products, or a multispectral satellite image directly produced at the theoretical level. These can all be flexibly configured according to specific needs in actual situations.

[0088] Thus, this application, through the scheme setting of the embodiments herein, designs and provides a practical implementation scheme for how to obtain high-quality sample multispectral satellite images of target deep and large reservoirs.

[0089] Furthermore, the present application may also involve a super-resolution link to increase the resolution of the sample multispectral satellite images obtained in the initial stage to a high level. This corresponds to the situation in which there is no need to directly obtain high-level images as sample multispectral satellite images and as model input after the model is put into actual use, which can effectively reduce the cost of solution deployment.

[0090] Correspondingly, as an exemplary embodiment, between step S101 and step S102, the method of the present application may further include:

[0091] The sample multispectral satellite image is subjected to super-resolution processing to increase the resolution of the sample multispectral satellite image to the resolution of a standard multispectral satellite image, wherein the temporal resolution of the standard multispectral satellite image is 24 hours (h) or 1 day, and the spatial resolution is 10 meters (m).

[0092] It can be seen from the embodiments herein that the algal bloom remote sensing prediction effect to be achieved in this application specifically involves a high spatiotemporal resolution of 24 hours and 10 meters.

[0093] Furthermore, the super-resolution processing additionally introduced in this application can also be completed by the corresponding machine learning model, so as to complete the model processing task efficiently and accurately through the powerful processing performance of the AI ​​model.

[0094] Specifically, as an exemplary embodiment, super-resolution processing of a sample multispectral satellite image may include:

[0095] The sample multispectral satellite image is super-resolution processed through the super-resolution processing model, wherein the super-resolution processing model specifically adopts the RealESR-GAN model;

[0096] The training process of the super-resolution processing model includes the following processing:

[0097] Using standard multispectral satellite images as a benchmark, low-resolution images are paired with corresponding standard multispectral satellite images to form high and low spatial resolution training sets;

[0098] The super-resolution processing model is trained using high and low spatial resolution training sets. During the training process, the generator and discriminator in the model are continuously optimized through an adversarial process. The quality of the generated images is evaluated using corresponding visual evaluation indicators and quantitative indicators until the training requirements are met.

[0099] It can be seen that in the embodiment here, the present application introduces a Real Enhanced Super Resolution Generative Adversarial Network (RealESR-GAN) to perform the super-resolution processing designed by the present application. After the training of the Real Enhanced Super Resolution Generative Adversarial Network is completed, the generator therein is used to perform high-performance super-resolution processing on the image input to the network, thereby constructing a sample multispectral satellite remote sensing image with high temporal resolution (1 day) and high spatial resolution (10 meters).

[0100] In addition, it can be understood that the sample multispectral satellite remote sensing images obtained in this application as mentioned above can themselves involve different data sources, that is, multi-source remote sensing images. In the super-resolution processing process here, it is obvious that while effectively improving the temporal and spatial resolution, it also helps to promote the fusion of images from different data sources and obtain a multi-source remote sensing image dataset with high temporal / spatial resolution.

[0101] In addition, for the indicator threshold used in step S103, it can be understood that, in actual applications, for the target deep and large reservoir this time, it can be a pre-adapted fixed threshold, or a threshold adapted according to actual conditions within a selected time range, or a threshold adapted in real time when the indicator threshold is used, so as to achieve the best timeliness, promote more accurate binarization processing of water bloom occurrence, and obtain a water bloom binary distribution layer reflecting deeper conditions.

[0102] Furthermore, for the latter solution setting, as an exemplary embodiment, the method of the present application may also include:

[0103] Obtain algae density data measured synchronously from ground monitoring points and sample multispectral satellite images;

[0104] Based on the algae density data and the index values ​​of each location in the sample multispectral satellite images, a scatter fitting model was constructed, and the index threshold suitable for the current situation was determined in combination with the algae density evaluation algal bloom degree classification standard in the specification.

[0105] It can be noted that this involves the collection of new data, namely the algae density data (ground-measured data) measured synchronously by ground monitoring points deployed at deep and large reservoirs. The detection instruments involved can usually directly use existing / general equipment.

[0106] It can be seen that in the embodiments herein, the present application provides a set of indicator threshold processing solutions that are easy to operate and can obtain high adaptability.

[0107] In addition, for the multispectral satellite images and algae density data involved in the above plan, in terms of details, at least one of further parameter inversion, spatial interpolation and resampling can be performed to ensure that data from different aspects can be aligned and matched to better carry out the corresponding data processing above.

[0108] After completing the training of the deep and large reservoir water bloom remote sensing prediction model adapted for the target deep and large reservoir, it is obvious that it can be put into specific practical applications.

[0109] Correspondingly, as an exemplary embodiment, the method of the present application may further include:

[0110] According to the needs of algal bloom remote sensing prediction, obtain multispectral satellite images of the target deep and large reservoir to be predicted;

[0111] Input the preset multispectral satellite images into the remote sensing prediction model of water bloom in deep and large reservoirs;

[0112] Extract the water bloom remote sensing prediction results output by the water bloom remote sensing prediction model for deep and large reservoirs.

[0113] It can be understood that the multispectral satellite images to be predicted can be manually entered by staff according to needs, or can be obtained by staff manually triggering the system to initiate an algal bloom remote sensing prediction task, or can be obtained by the system according to a preset algal bloom remote sensing prediction task or a real-time autonomously generated algal bloom remote sensing prediction task. The specific acquisition of the multispectral satellite images corresponds to the previous sample multispectral satellite images.

[0114] In this way, after obtaining the multispectral satellite image to be predicted that needs to be processed, it can be input into the deep and large reservoir water bloom remote sensing prediction model to carry out the corresponding water bloom remote sensing prediction processing. After the processing is completed, the deep and large reservoir water bloom remote sensing prediction model will output the corresponding water bloom remote sensing prediction results. At this time, the water bloom remote sensing prediction results can be extracted to meet the on-site or remote water bloom monitoring needs of the target deep and large reservoir, thereby providing important decision-making basis and technical support for the early water bloom warning and later water bloom prevention and control intervention of the target deep and large reservoir, thereby effectively promoting the refined management of regional water resources and the progress of water ecological protection.

[0115] In this case, the remote sensing prediction results of algal blooms may also involve corresponding data application processing, such as local storage, remote storage, result display, output of prediction completion prompts, or further data analysis. It is understandable that the specific data application processing can be flexibly adjusted according to pre-configured and real-time configured data application strategies / rules.

[0116] The above is an introduction to the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs provided in this application. In order to facilitate better implementation of the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs provided in this application, this application also provides a processing device of the remote sensing prediction model of water bloom in deep and large reservoirs from the perspective of functional modules.

[0117] See Figure 2 , Figure 2 This is a structural diagram of a processing device for a remote sensing prediction model of water bloom in a deep reservoir in this application. In this application, the processing device 200 for a remote sensing prediction model of water bloom in a deep reservoir may specifically include the following structure:

[0118] An acquisition unit 201 is used to acquire sample multispectral satellite images for a target deep and large reservoir;

[0119] The calculation unit 202 is used to calculate the index value of each position of the sample multispectral satellite image based on the preset algal bloom characterization index;

[0120] The determination unit 203 is configured to determine whether algal bloom exists at each location in the multispectral satellite image by combining the index value and the index threshold of each location in the sample multispectral satellite image, and to form an algal bloom binary distribution layer corresponding to the sample multispectral satellite image;

[0121] A quantification unit 204 is configured to quantify the impact of the cumulative effect of preset indicators on the algal bloom situation and the impact of the multi-day average of the preset indicators on the algal bloom situation based on the sample multispectral satellite image and the algal bloom binary distribution layer, wherein the preset indicators include water quality parameters, reservoir hydrological rhythms, and meteorological parameters;

[0122] The screening unit 205 is used to screen out target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms from preset indicators through a redundancy analysis method based on the quantified results;

[0123] The training unit 206 is used to configure the indicator data corresponding to the target indicator and the water bloom binary distribution layer with training samples, and train the water bloom remote sensing prediction model of the deep-sea reservoir. The water bloom remote sensing prediction model of the deep-sea reservoir is specifically a long short-term memory network based on the self-attention mechanism. The water bloom remote sensing prediction model of the deep-sea reservoir is configured with 7 output layers, each output layer corresponds to the prediction output of different days in the next 7 days. The water bloom remote sensing prediction model of the deep-sea reservoir is used to predict whether there will be water bloom in the next 7 days based on the indicator data input into the model corresponding to the target indicator.

[0124] In an exemplary embodiment, the acquiring unit 201 is specifically configured to:

[0125] For target deep and large reservoirs, obtain first-sample multispectral satellite imagery that meets requirements regarding image quality, cloud cover, and watershed coverage;

[0126] performing preprocessing including radiometric calibration and atmospheric correction on the first sample multispectral satellite image to obtain a second sample multispectral satellite image;

[0127] The second sample multispectral satellite image is georeferenced using a pre-configured standard multispectral satellite image as a reference to obtain a sample multispectral satellite image.

[0128] In another exemplary embodiment, the apparatus further includes a processing unit 207 configured to:

[0129] The sample multispectral satellite image is super-resolution processed to increase the resolution of the sample multispectral satellite image to the resolution of a standard multispectral satellite image, where the temporal resolution of the standard multispectral satellite image is 24 hours and the spatial resolution is 10 meters.

[0130] In another exemplary embodiment, the processing unit 207 is specifically configured to:

[0131] The sample multispectral satellite image is super-resolution processed through the super-resolution processing model, wherein the super-resolution processing model specifically adopts the RealESR-GAN model;

[0132] The training process of the super-resolution processing model includes the following processing:

[0133] Using standard multispectral satellite images as a benchmark, low-resolution images are paired with corresponding standard multispectral satellite images to form high and low spatial resolution training sets;

[0134] The super-resolution processing model is trained using high and low spatial resolution training sets. During the training process, the generator and discriminator in the model are continuously optimized through an adversarial process. The quality of the generated images is evaluated using corresponding visual evaluation indicators and quantitative indicators until the training requirements are met.

[0135] In another exemplary embodiment, the determining unit 203 is further configured to:

[0136] Obtain algae density data measured synchronously from ground monitoring points and sample multispectral satellite images;

[0137] Based on the algae density data and the index values ​​of each location in the sample multispectral satellite images, a scatter fitting model was constructed, and the index threshold suitable for the current situation was determined in combination with the algae density evaluation algal bloom degree classification standard in the specification.

[0138] In another exemplary embodiment, the apparatus further includes an application unit 208, configured to:

[0139] According to the needs of algal bloom remote sensing prediction, obtain multispectral satellite images of the target deep and large reservoir to be predicted;

[0140] Input the preset multispectral satellite images into the remote sensing prediction model of water bloom in deep and large reservoirs;

[0141] Extract the water bloom remote sensing prediction results output by the water bloom remote sensing prediction model for deep and large reservoirs.

[0142] This application also provides a processing device from the perspective of hardware structure, see Figure 3 , Figure 3The schematic diagram of the structure of the processing device of the present application is shown. Specifically, the processing device of the present application may include a processor 301, a memory 302 and an input / output device 303. The processor 301 is used to execute the computer program stored in the memory 302 to implement the following Figure 1 The steps of the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs in the corresponding embodiment; or, when the processor 301 is used to execute the computer program stored in the memory 302, the following is realized. Figure 2 The memory 302 is used to store the functions of each unit in the embodiment corresponding to the processor 301. Figure 1 The computer program required for the processing method of the remote sensing prediction model of water bloom in deep and large reservoirs in the corresponding embodiment.

[0143] For example, the computer program may be divided into one or more modules / units, one or more of which are stored in the memory 302 and executed by the processor 301 to complete the present application. One or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in a computer device.

[0144] The processing device may include, but is not limited to, a processor 301, a memory 302, and an input / output device 303. Those skilled in the art will appreciate that the illustrations are merely examples of processing devices and do not limit the processing device. The processing device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the processing device may also include a network access device, a bus, etc., and the processor 301, the memory 302, the input / output device 303, etc. are connected via a bus.

[0145] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the processing device and connects various parts of the entire device using various interfaces and lines.

[0146] Memory 302 can be used to store computer programs and / or modules. Processor 301 implements various functions of the computer device by running or executing computer programs and / or modules stored in memory 302 and accessing data stored in memory 302. Memory 302 may primarily include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, and the data storage area may store data generated based on the use of the processing device. Furthermore, memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0147] When the processor 301 is used to execute the computer program stored in the memory 302, it can specifically implement the following functions:

[0148] For target deep and large reservoirs, obtain sample multispectral satellite images;

[0149] Based on the preset algal bloom characterization indicators, the indicator values ​​of each location of the sample multispectral satellite image are calculated;

[0150] Combining the index values ​​and index thresholds of each location in the sample multispectral satellite image, determine whether there is algal bloom at each location in the multispectral satellite image, and form an algal bloom binary distribution layer corresponding to the sample multispectral satellite image;

[0151] Based on sample multispectral satellite images and binary distribution layers of algal blooms, the cumulative effect of preset indicators and the multi-day average of preset indicators on algal blooms were quantified. The preset indicators included water quality parameters, reservoir hydrological rhythms, and meteorological parameters.

[0152] Based on the quantitative results, the target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms are screened out from the preset indicators through redundancy analysis method.

[0153] The indicator data corresponding to the target indicator and the water bloom binary distribution layer are configured as training samples to train the water bloom remote sensing prediction model of Shenda Reservoir. The water bloom remote sensing prediction model of Shenda Reservoir is specifically a long short-term memory network based on the self-attention mechanism. The water bloom remote sensing prediction model of Shenda Reservoir is configured with 7 output layers, and each output layer corresponds to the prediction output of different days in the next 7 days. The water bloom remote sensing prediction model of Shenda Reservoir is used to predict whether there will be water bloom in the next 7 days based on the indicator data input into the model corresponding to the target indicator.

[0154] Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working process of the processing device, processing equipment and corresponding units of the deep and large reservoir water bloom remote sensing prediction model described above can refer to the following: Figure 1 The description of the processing method of the remote sensing prediction model for water bloom in deep and large reservoirs in the corresponding embodiment will not be repeated here.

[0155] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0156] To this end, the present application provides a computer-readable storage medium, which stores a plurality of instructions, which can be loaded by a processor to execute the present application as follows: Figure 1 The steps of the processing method of the remote sensing prediction model of the water bloom in the deep reservoir in the corresponding embodiment, the specific operation can be referred to as follows Figure 1 The description of the processing method of the remote sensing prediction model for water bloom in deep and large reservoirs in the corresponding embodiment will not be repeated here.

[0157] The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0158] Due to the instructions stored in the computer readable storage medium, the present application can be executed as follows: Figure 1 The steps of the processing method of the remote sensing prediction model of the water bloom in the deep reservoir in the corresponding embodiment, therefore, the present application can be realized as follows Figure 1 The beneficial effects that can be achieved by the processing method of the remote sensing prediction model for water bloom in deep and large reservoirs in the corresponding embodiment are detailed in the previous description and will not be repeated here.

[0159] The above is a detailed introduction to the processing method, device, processing equipment and computer-readable storage medium of the remote sensing prediction model of water bloom in deep and large reservoirs provided by this application. Specific examples are used in this article to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the core idea of ​​this application; at the same time, for technical personnel in this field, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this application.

Claims

1. A processing method for a remote sensing prediction model of algal bloom in deep and large reservoirs, characterized in that: The method comprises: For target deep and large reservoirs, obtain sample multispectral satellite images; Calculating the index value of each position of the sample multispectral satellite image based on the preset algal bloom characterization index; Determine whether algal bloom exists at each location of the multispectral satellite image by combining the index value and index threshold of each location of the sample multispectral satellite image, and form an algal bloom binary distribution layer corresponding to the sample multispectral satellite image; Based on the sample multispectral satellite image and the algal bloom binary distribution layer, quantify the impact of the cumulative effect of preset indicators on the algal bloom situation and the impact of the multi-day average of the preset indicators on the algal bloom situation, wherein the preset indicators include water quality parameters, reservoir hydrological rhythms, and meteorological parameters; Based on the quantitative results, target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms are screened out from the preset indicators through a redundancy analysis method; The indicator data corresponding to the target indicator and the water bloom binary distribution layer are configured as training samples to train the water bloom remote sensing prediction model of deep and large reservoirs, wherein the water bloom remote sensing prediction model of deep and large reservoirs is specifically a long short-term memory network based on the self-attention mechanism, and the water bloom remote sensing prediction model of deep and large reservoirs is configured with 7 output layers, each of which corresponds to the prediction output of different days in the next 7 days. The water bloom remote sensing prediction model of deep and large reservoirs is used to predict whether the water bloom situation will exist in the next 7 days based on the indicator data input into the model corresponding to the target indicator.

2. The method according to claim 1, characterized in that For the target deep and large reservoir, the sample multispectral satellite image is obtained, including: For the target deep and large reservoir, obtain a first sample of multispectral satellite images that meet the requirements of image quality, cloud coverage, and watershed coverage; performing preprocessing including radiometric calibration and atmospheric correction on the first sample multispectral satellite image to obtain a second sample multispectral satellite image; The second sample multispectral satellite image is geo-referenced using a pre-configured standard multispectral satellite image as a reference to obtain the sample multispectral satellite image.

3. The method according to claim 2, characterized in that The method further comprises: Super-resolution processing is performed on the sample multispectral satellite image to increase the resolution of the sample multispectral satellite image to the resolution of the standard multispectral satellite image, wherein the temporal resolution of the standard multispectral satellite image is 24 hours and the spatial resolution is 10 meters.

4. The method according to claim 3, characterized in that Performing the super-resolution processing on the sample multispectral satellite image includes: Performing super-resolution processing on the sample multispectral satellite image through a super-resolution processing model, wherein the super-resolution processing model specifically adopts a RealESR-GAN model; The training process of the super-resolution processing model includes the following processing contents: Taking the standard multispectral satellite image as a benchmark, pair the low-resolution image with the corresponding standard multispectral satellite image to form a high- and low-spatial-resolution training set; The super-resolution processing model is trained using the high and low spatial resolution training sets. During the training process, the generator and discriminator in the model are continuously optimized through an adversarial process, and the quality of the generated image is evaluated using corresponding visual evaluation indicators and quantitative indicators until the training requirements are met.

5. The method according to claim 1, wherein The method further comprises: Acquiring algae density data measured synchronously by ground monitoring points and multispectral satellite images of the sample; Based on the algae density data and the index values ​​of each location of the sample multispectral satellite image, a scatter fitting model is constructed, and the index threshold adapted to the current situation is determined in combination with the bloom degree grading standard for algae density evaluation in the specification.

6. A processing device for a remote sensing prediction model of algal bloom in deep and large reservoirs, characterized in that: The device comprises: An acquisition unit is used to acquire sample multispectral satellite images of a target deep and large reservoir; A calculation unit, configured to calculate an index value of each position of a sample multispectral satellite image based on a preset algal bloom characterization index; a determination unit, configured to determine whether algal bloom exists at each location in the multispectral satellite image by combining the index value and the index threshold of each location in the sample multispectral satellite image, and to form an algal bloom binary distribution layer corresponding to the sample multispectral satellite image; A quantification unit is used to quantify the impact of the cumulative effect of preset indicators on the algal bloom situation and the impact of the multi-day average of the preset indicators on the algal bloom situation based on the sample multispectral satellite image and the algal bloom binary distribution layer, wherein the preset indicators include water quality parameters, reservoir hydrological rhythms and meteorological parameters; A screening unit is used to screen target indicators that meet the relative importance requirements for remote sensing prediction of algal blooms from preset indicators through a redundancy analysis method based on the quantitative results; The training unit is used to configure the indicator data corresponding to the target indicator and the water bloom binary distribution layer with training samples, and train the water bloom remote sensing prediction model of the Shenda Reservoir. The water bloom remote sensing prediction model of the Shenda Reservoir is specifically a long short-term memory network based on the self-attention mechanism. The water bloom remote sensing prediction model of the Shenda Reservoir is configured with 7 output layers, each output layer corresponds to the prediction output of different days in the next 7 days. The water bloom remote sensing prediction model of the Shenda Reservoir is used to predict whether there will be water bloom in the next 7 days based on the indicator data input into the model corresponding to the target indicator.

7. A processing device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 5 when calling the computer program in the memory.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to any one of claims 1 to 5.

Images