A cyanobacterial bloom prediction method based on multi-modal perception, electronic equipment and medium

Abstract

This application provides a method, electronic device, and medium for predicting cyanobacterial blooms based on multimodal sensing. The method includes: acquiring water quality monitoring data; identifying risk areas; acquiring visual image data of the risk areas; acquiring patch data of each sub-region within the risk areas; identifying all sub-regions suspected of having a risk of cyanobacterial blooms based on the patch data; acquiring hyperspectral data and multibeam data of the risk sub-regions; obtaining fused data according to a preset fusion strategy; obtaining a first cyanobacterial concentration in the risk sub-regions based on the fused data; obtaining a second cyanobacterial concentration and cyanobacterial diffusion rate in the risk areas after a first preset time period; and determining whether a cyanobacterial bloom will occur based on the second cyanobacterial concentration and cyanobacterial diffusion rate. This application improves the overall accuracy and reliability of the prediction.

Description

Technical Field

[0001] This invention relates to the field of cyanobacterial bloom monitoring technology, and more specifically, to a cyanobacterial bloom prediction method, electronic device, and medium based on multimodal sensing. Background Technology

[0002] Currently, most cyanobacterial bloom monitoring adopts a conventional technical approach based on satellite macro-monitoring, combined with meso-level auxiliary information, and relying on predictive models for comprehensive analysis. This approach has several significant shortcomings. Satellite remote sensing can only perform preliminary inversion of water quality parameters and lacks a precise risk screening and judgment mechanism based on water quality monitoring data. It cannot efficiently screen out high-risk waters for cyanobacterial blooms, resulting in an excessively large and untargeted meso-level monitoring scope, increasing monitoring costs and data processing burden. Furthermore, even if image data of risk areas is acquired, current technologies lack systematic processing methods and have not developed scientific methods for sub-region division and patch feature quantification, making it difficult to accurately extract cyanobacterial patch data and support [further analysis]. While existing models can identify risk sub-regions, cross-scale data fusion is often limited to simple feature splicing or weighting, lacking a cross-scale bridging mechanism for the dynamic processes of cyanobacterial aggregation and diffusion. This fails to fully capture the spatiotemporal patterns of cyanobacterial growth and evolution. Furthermore, most existing prediction models do not combine patch data with current cyanobacterial concentrations for collaborative prediction, resulting in insufficient model generalization and prediction accuracy. Significant errors occur in the early stages of algal blooms and during rapid diffusion phases. Moreover, after outputting concentration data, the models rely solely on a single concentration index to determine the occurrence of algal blooms, lacking a collaborative judgment logic for concentration and diffusion rate. This ignores the accelerating effect of diffusion rate on algal blooms, ultimately leading to overall prediction accuracy and reliability failing to meet practical application requirements. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, electronic device and medium for predicting cyanobacterial blooms based on multimodal perception, which can improve the problem that the overall accuracy and reliability of prediction cannot meet the needs of practical applications.

[0004] To achieve the above technical objectives, the technical solution adopted in this application is as follows:

[0005] In a first aspect, embodiments of this application provide a method for predicting cyanobacterial blooms based on multimodal sensing, the method comprising:

[0006] Acquire water quality monitoring data for all water bodies based on satellite remote sensing;

[0007] Based on the water quality monitoring data of all the aforementioned water areas, all water areas suspected of having a risk of cyanobacterial blooms are identified as risk areas.

[0008] Obtain visual image data of the risk area;

[0009] Based on the visual image data, patch data for each sub-region of the risk area is obtained, wherein the risk area is divided into multiple sub-regions;

[0010] Based on the patch data, all sub-regions in the risk area that represent the risk of suspected cyanobacterial blooms are obtained, and these are the risk sub-regions.

[0011] Acquire hyperspectral data and multibeam data of the risk sub-region. The hyperspectral data includes a first reflectance and a second reflectance, the first reflectance and the second reflectance have different characteristic bands, and the multibeam data includes the scattering intensity of suspended matter.

[0012] According to the preset fusion strategy, the hyperspectral data, multibeam data, and water quality monitoring data of the risk sub-region are fused to obtain fused data;

[0013] Based on the fused data, the first cyanobacteria concentration in the risk sub-region is obtained;

[0014] The first cyanobacteria concentration and the corresponding patch data are input into the first preset model after training. The first preset model after training outputs the second cyanobacteria concentration and cyanobacteria diffusion rate of the risk area after a first preset time.

[0015] Based on the second cyanobacterial concentration and the cyanobacterial diffusion rate, the prediction results for whether a cyanobacterial bloom will occur are determined.

[0016] In conjunction with the first aspect, the water quality monitoring data includes the cyanobacteria index and chlorophyll a remote sensing data;

[0017] Based on the water quality monitoring data of all the aforementioned water areas, all water areas suspected of having a risk of cyanobacterial blooms are obtained, including:

[0018] Based on the water quality monitoring data, the first time series data of the cyanobacteria index in each water body over the past second preset time period is obtained, and the second time series data of the chlorophyll a remote sensing data in each water body over the past second preset time period is obtained.

[0019] When the water body meets the first preset condition, and a specified number of other adjacent water bodies also meet the first preset condition, the water body is a risk area. The first preset condition is: the first time series data corresponding to the water body indicates that the cyanobacteria index has been increasing over the past second preset time period; the second time series data indicates that the chlorophyll a remote sensing data has been increasing over the past second preset time period; and the phase difference between the first time series data and the second time series data is less than or equal to a phase threshold.

[0020] In conjunction with the first aspect, the patch data includes patch spread rate and patch density;

[0021] The step of obtaining patch data in the risk area based on the visual image data includes:

[0022] Based on the visual image data, the patches in the risk area are obtained;

[0023] Based on all the patches, obtain the patch density and patch diffusion rate of each sub-region, wherein the patch density is obtained based on the number of patches in the risk sub-region, and the patch diffusion rate is used to characterize the amount of change in the patch density of the risk sub-region over the past second preset time period.

[0024] In conjunction with the first aspect, obtaining the patches of the risk area based on the visual image data includes:

[0025] Obtain the red-edge band reflectance and near-infrared band reflectance of each pixel in the visual image data;

[0026] Based on the red-edge band reflectance and near-infrared band reflectance, an evaluation value is obtained for each pixel, and the evaluation value is used to characterize the density of cyanobacteria in the corresponding pixel.

[0027] Based on all the evaluation values, determine all connected components in the visual image data that represent suspected patches;

[0028] Obtain the green band reflectance and near-infrared band reflectance of each of the connected domains;

[0029] Remove the connected components that satisfy the second preset condition to obtain the patch. The second preset condition is: the green band reflectance is less than or equal to the first green wave threshold, and the near-infrared band reflectance is less than or equal to the first infrared wave threshold, or the green band reflectance is greater than or equal to the second green wave threshold, and the near-infrared band reflectance is greater than or equal to the second infrared wave threshold, wherein the first green wave threshold is less than the second green wave threshold, and the first infrared wave threshold is less than the second infrared wave threshold.

[0030] In conjunction with the first aspect, the step of obtaining all sub-regions in the risk area that represent the potential risk of cyanobacterial blooms based on the patch data includes:

[0031] Based on the patch density and patch diffusion rate, a first coupling index is determined according to a first preset model;

[0032] When the first coupling index is greater than the first preset coupling value, the corresponding sub-region is the risk sub-region;

[0033] The first preset model is:

[0034]

[0035] Indicates the first coupling index;

[0036] Indicates the weight of the degree of clustering;

[0037] Indicates the synergistic enhancement coefficient;

[0038] and These represent the normalized patch density and patch diffusion rate, respectively.

[0039] In conjunction with the first aspect, before inputting the first cyanobacteria concentration and corresponding patch data into the trained first preset model, and before outputting the second cyanobacteria concentration and cyanobacteria diffusion rate of the risk area after a first preset time period through the trained first preset model, the method further includes:

[0040] Acquire multiple mapping data of the risk area, including first historical cyanobacteria concentration, historical patch data, and second historical cyanobacteria concentration and historical cyanobacteria diffusion rate after a first preset time.

[0041] Using the first historical cyanobacteria concentration and historical patch data as inputs, and the corresponding second historical cyanobacteria concentration and historical cyanobacteria diffusion rate as outputs, a first preset model is trained to obtain the trained first preset model.

[0042] In conjunction with the first aspect, the fused data obtained by fusing the hyperspectral data, multibeam data, and water quality monitoring data of the risk sub-region according to a preset fusion strategy includes:

[0043] The first confidence level of the hyperspectral data is determined based on the first reflectance and the second reflectance, as well as the minimum threshold and the maximum threshold of the historical cyanobacterial characteristic band ratio of the pre-stored risk sub-region.

[0044] The second confidence level of the multibeam data is determined based on the scattering intensity of the suspended object and the first reflectivity.

[0045] The third confidence level of the water quality monitoring data was determined based on the cyanobacteria index and chlorophyll a remote sensing data.

[0046] Based on the first confidence level, a first weight of the hyperspectral data is determined; based on the second confidence level, a second weight of the multibeam data is determined; and based on the third confidence level, a third weight of the water quality monitoring data is determined.

[0047] The first weight, the second weight, and the third weight, along with the water quality monitoring data, hyperspectral data, and multibeam data, are substituted into the fusion model to obtain the fused data.

[0048] In conjunction with the first aspect, the prediction result of whether a cyanobacterial bloom will occur based on the second cyanobacterial concentration and the cyanobacterial diffusion rate includes:

[0049] Based on the second cyanobacteria concentration and cyanobacteria diffusion rate, the second coupling index is determined according to the second preset model;

[0050] When the second coupling index is greater than the second preset coupling value, a prediction result is obtained to characterize whether a cyanobacterial bloom will occur; otherwise, a prediction result is obtained that a cyanobacterial bloom will not occur.

[0051] The second preset model is:

[0052]

[0053] This represents the historical maximum concentration of cyanobacteria in the aforementioned water area;

[0054] This represents the historical maximum diffusion rate of cyanobacteria in the aforementioned waters;

[0055] This is the second coupling index;

[0056] Indicates the rate of cyanobacteria diffusion;

[0057] Indicates the concentration of the second cyanobacteria;

[0058] This represents the normalized concentration of the second cyanobacteria;

[0059] This is the preset critical threshold for cyanobacterial bloom concentration;

[0060] This is the diffusion rate exponential enhancement coefficient.

[0061] Secondly, embodiments of this application provide an electronic device comprising a processor and a memory coupled together, wherein the memory stores a computer program, and when the computer program is executed by the processor, the electronic device performs the method described in the first aspect.

[0062] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the method described in the first aspect.

[0063] The invention employing the above technical solution has the following advantages:

[0064] In the technical solution provided in this application, water quality monitoring data of all water areas collected by satellite remote sensing are first obtained. Based on this data, water areas suspected of having the risk of cyanobacterial blooms are screened out as risk areas. Then, visual image data of the risk areas are obtained, and the risk areas are divided into multiple sub-regions. Based on the visual image data, patch data of each sub-region is obtained, and then risk sub-regions are screened out based on the patch data. Subsequently, the first cyanobacterial concentration of the risk sub-region is obtained, and the first cyanobacterial concentration and the corresponding patch data are input into a first preset model after training. The model outputs the second cyanobacterial concentration and cyanobacterial diffusion rate of the risk area after a first preset time period. Finally, the bloom prediction result is determined based on the second cyanobacterial concentration and cyanobacterial diffusion rate. This solution narrows the monitoring scope and reduces monitoring costs and data processing burden by precisely targeting monitoring from the entire water area to risk areas and then to risk sub-regions. Through sub-region division and patch quantification, it accurately extracts cyanobacterial patch data, improving the accuracy of risk sub-region identification. Utilizing a pre-set model, it achieves synergistic prediction of patch data and cyanobacterial concentration, capturing the spatiotemporal patterns of cyanobacterial growth and diffusion, improving model generalization ability and prediction accuracy, especially reducing prediction errors in the early stages and rapid diffusion phases of algal blooms. By synergistically determining concentration and diffusion rate, the algal bloom prediction results are more closely aligned with actual ecological processes, improving the overall accuracy and reliability of predictions and meeting practical application needs.

[0065] In the technical solution proposed in this application, multimodal data from satellite remote sensing, hyperspectral imaging, and multibeam bathymetry are fused to obtain the first cyanobacteria concentration. Satellite remote sensing data is used to screen risky water areas across the entire region. Hyperspectral imaging accurately characterizes the spectral features of surface cyanobacteria, while multibeam bathymetry compensates for the limitation of hyperspectral imaging, which can only detect surface water, covering the spatial dimensions of cyanobacteria growth and distribution. Furthermore, the fusion weights are dynamically allocated based on the confidence levels of each modality, avoiding the subjective impact of manually set fixed weights on prediction accuracy. The weight allocation is adaptively adjusted according to the reliability of each modality's data in different water areas and observation scenarios, giving higher fusion weights to modalities that more reliably characterize cyanobacteria, thus improving the accuracy of multimodal feature fusion. Based on this, the first cyanobacteria concentration obtained from the fusion features integrates multimodal data, avoiding fusion errors caused by single-modal data bias or fixed weights, and improving the model's prediction accuracy in the early and rapid diffusion stages of algal blooms. Attached Figure Description

[0066] This application can be further illustrated by the non-limiting embodiments given in the accompanying drawings. It should be understood that the following drawings only illustrate some embodiments of this application and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained from these drawings without any inventive effort.

[0067] Figure 1 A flowchart of a cyanobacterial bloom prediction method based on multimodal sensing provided in this application embodiment.

[0068] Figure 2 This is a sub-flowchart of S140 provided in an embodiment of this application. Detailed Implementation

[0069] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that similar or identical parts are referred to by the same reference numerals in the drawings or description. Implementations not shown or described in the drawings are forms known to those skilled in the art. In the description of this application, terms such as "first" and "second" are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0070] Please refer to Figure 1 This application provides a method for predicting cyanobacterial blooms based on multimodal sensing, which can be applied to electronic devices and whose steps can be executed or implemented by the electronic device. The electronic device can be, but is not limited to, personal computers, smartphones, and other electronic devices. The method for predicting cyanobacterial blooms based on multimodal sensing may include the following steps:

[0071] S110, acquire water quality monitoring data for all water areas based on satellite remote sensing;

[0072] S120, Based on the water quality monitoring data of all the water areas, obtain all water areas suspected of having a risk of cyanobacterial blooms, which are designated as risk areas;

[0073] S130, acquire visual image data of the risk area;

[0074] S140, Based on the visual image data, obtain patch data for each sub-region of the risk area, wherein the risk area is divided into multiple sub-regions;

[0075] S150, Based on the patch data, obtain all sub-regions in the risk area that represent the risk of suspected cyanobacterial blooms, which are the risk sub-regions;

[0076] S160, acquire hyperspectral data and multibeam data of the risk sub-region, the hyperspectral data includes a first reflectance and a second reflectance, the first reflectance and the second reflectance have different characteristic bands, and the multibeam data includes the scattering intensity of suspended matter.

[0077] S170, according to the preset fusion strategy, the hyperspectral data, multibeam data, and water quality monitoring data of the risk sub-region are fused to obtain fused data;

[0078] S180, Based on the fused data, obtain the first cyanobacteria concentration in the risk sub-region;

[0079] S190, input the first cyanobacteria concentration and the corresponding patch data into the trained first preset model, and output the second cyanobacteria concentration and cyanobacteria diffusion rate of the risk area after the first preset time through the trained first preset model.

[0080] S1100, based on the second cyanobacterial concentration and the cyanobacterial diffusion rate, determines the prediction result of whether a cyanobacterial bloom will occur.

[0081] In the above implementation, firstly, water quality monitoring data of all water areas collected by satellite remote sensing is acquired. Based on the water quality monitoring data of the entire water area, water areas suspected of having a risk of cyanobacterial blooms are screened and identified as risk areas. Then, visual image data of the risk areas is acquired. Based on the visual image data, patch data of each sub-region within the risk area is obtained. Based on the patch data, risk sub-regions suspected of having a risk of cyanobacterial blooms are determined within the risk area. Subsequently, the first cyanobacterial concentration of the risk sub-region is obtained. The first cyanobacterial concentration and the corresponding patch data are input into a first preset model that has been trained. The model outputs the second cyanobacterial concentration and the cyanobacterial diffusion rate of the risk area after a first preset time. Finally, the prediction result of whether a cyanobacterial bloom will occur is determined based on the second cyanobacterial concentration and the cyanobacterial diffusion rate. This technical solution combines macroscopic screening by satellite remote sensing with mesoscopic analysis of visual images to achieve step-by-step accurate positioning of risky water areas and risk sub-regions. It adopts a prediction model that integrates cyanobacterial concentration and patch characteristics, and combines concentration and diffusion rate to comprehensively determine the possibility of bloom occurrence, which can improve the rationality and accuracy of cyanobacterial bloom prediction.

[0082] The steps of the cyanobacterial bloom prediction method based on multimodal sensing will be described in detail below:

[0083] In S110, this embodiment acquires a wide range of water quality monitoring data through Sentinel-3OLCI, with particular attention to remote sensing data of cyanobacteria index (CI-cy) and chlorophyll a (Chl-a). The specific implementation process is as follows: First, select a satellite remote sensing data source that meets the monitoring needs of cyanobacteria in aquatic areas; download remote sensing image data covering all waters in the target area through the official satellite remote sensing data download platform to ensure that the temporal and spatial resolution of the images matches the monitoring requirements; preprocess the downloaded remote sensing images, successively completing radiometric correction, geometric correction, cloud pollution removal, and water boundary extraction; based on the preprocessed remote sensing images, use a mature water quality parameter inversion model (such as the empirical model for chlorophyll a concentration inversion) to invert the core water quality monitoring parameters of all waters pixel by pixel (key parameters include chlorophyll a concentration, turbidity, and eutrophication index, all of which are key parameters characterizing cyanobacteria growth); finally, screen and integrate the inverted water quality data, remove outliers and invalid data, and organize them into a standardized dataset to form water quality monitoring data covering all waters with complete parameters, providing basic data support for subsequent risk area screening (S120).

[0084] In S120, water quality monitoring data can identify all water bodies suspected of having a risk of cyanobacterial blooms. However, this data is primarily derived from satellite remote sensing, yielding indirect indicators such as chlorophyll a concentration and eutrophication index. These indicators only reflect the relative levels of algal biomass and eutrophication, failing to directly distinguish cyanobacteria from other planktonic algae. Furthermore, it cannot rule out abnormalities caused by suspended solids or environmental disturbances. Therefore, only a preliminary assessment of the possibility of cyanobacterial blooms in these water bodies is possible, classifying them as areas suspected of having a risk of cyanobacterial blooms. Based on this, S120 can specifically include the following solutions:

[0085] Based on the water quality monitoring data, the first time series data of the cyanobacteria index in each water body over the past second preset time period is obtained, and the second time series data of the chlorophyll a remote sensing data in each water body over the past second preset time period is obtained.

[0086] When the water body meets the first preset condition, and a specified number of other adjacent water bodies also meet the first preset condition, the water body is a risk area. The first preset condition is: the first time series data corresponding to the water body indicates that the cyanobacteria index has been increasing over the past second preset time period; the second time series data indicates that the chlorophyll a remote sensing data has been increasing over the past second preset time period; and the phase difference between the first time series data and the second time series data is less than or equal to a phase threshold.

[0087] The second preset duration is set according to the growth and evolution cycle of cyanobacteria (usually 7 to 14 days, adapting to the time span from initial aggregation to initial spread of cyanobacteria); the phase threshold is set according to the water type (usually 1 to 2 days, used to determine the growth synchronicity of two time series data); the specified number is set according to the spatial distribution density of water areas (usually 2 to 3, adapting to the spatial correlation of adjacent water areas).

[0088] In this embodiment, based on the acquired water quality monitoring data of the entire water body, the first time-series data of the cyanobacteria index and the second time-series data of chlorophyll a remote sensing data within the past second preset time period are extracted for each water body. The cyanobacteria index can directly reflect the change in the cyanobacteria content of the water body, and chlorophyll a, as a common indicator of algae, can help verify the growth of cyanobacteria. The combination of the two types of indicators can initially eliminate the abnormal interference of a single indicator. Subsequently, a first preset condition judgment is made for each water body. Through trend analysis, it is determined whether the cyanobacteria index represented by the first time-series data and the chlorophyll a remote sensing data represented by the second time-series data have both maintained an increase within the second preset time period. The phase difference between the two types of time-series data is calculated and it is determined whether it is less than or equal to the first preset condition judgment. A preset phase threshold is used. Since cyanobacterial growth synchronously drives an increase in chlorophyll a concentration, the growth rates of the two should be highly synchronized. If the phase difference exceeds the limit, abnormal indicators caused by non-cyanobacterial factors can be ruled out. Based on the premise that a single water body meets the first preset condition, it is further checked whether a specified number of adjacent water bodies also meet the condition. Because cyanobacterial blooms have spatial diffusion characteristics, the synchronous increase of indicators in adjacent water bodies can rule out isolated anomalies caused by local environmental fluctuations, thus improving the reliability of the judgment. Water bodies that meet all the above conditions are considered risk areas with suspected cyanobacterial bloom risk. This judgment relies solely on satellite remote sensing inversion data and cannot completely rule out non-cyanobacterial interference factors. It can only determine the possibility of bloom occurrence and is therefore defined as a suspected risk area.

[0089] In this embodiment, the cyanobacteria index is a targeted indicator that directly characterizes the content of cyanobacteria in the water. Its continuous increase means that the biomass of cyanobacteria is constantly accumulating, which is a prerequisite for cyanobacterial blooms. Chlorophyll a is a common indicator for all algae (including cyanobacteria). As an important type of algae, the growth and reproduction of cyanobacteria directly leads to a synchronous increase in the concentration of chlorophyll a in the water. Therefore, the continuous increase of chlorophyll a can be used as an auxiliary verification indicator for cyanobacteria growth. The combination of the two can preliminarily rule out the abnormality of a single indicator caused by non-cyanobacteria factors. Phase difference determination mechanism: There is an intrinsic correlation between the growth of cyanobacteria index and chlorophyll a. That is, cyanobacteria growth (increase in cyanobacteria index) will directly lead to an increase in chlorophyll a concentration. Ideally, the growth rhythm of the two should be basically synchronized, with a very small phase difference. If the phase difference is too large, it indicates that the increase of chlorophyll a may not be caused by cyanobacteria growth (such as the reproduction of other phytoplankton, interference from suspended solids in the water, etc.). Therefore, by setting a phase threshold, the accuracy of cyanobacteria growth determination can be improved and misjudgments can be reduced.

[0090] The occurrence and spread of cyanobacterial blooms exhibit a clear spatial correlation. Adjacent water bodies typically have similar hydrological conditions (such as water flow and temperature) and eutrophication levels. If cyanobacteria begin to grow continuously in one water body, they easily spread to adjacent water bodies, leading to a synchronous increase in cyanobacteria index and chlorophyll a in those adjacent water bodies. To ensure that the water bodies meeting the criteria are accompanied by a specified number of adjacent water bodies that also meet the criteria, it is necessary to exclude abnormal indicators caused by local environmental fluctuations (such as localized sewage discharge or short-term water temperature changes) in a single water body. This ensures that the identified risk areas are regions where cyanobacteria are showing continuous growth and a spreading trend, rather than isolated, accidental anomalies.

[0091] The above judgment is based solely on time-series data retrieved from satellite remote sensing, without combining data from on-site sampling and microscopic observation. It cannot completely rule out interference from non-cyanobacterial factors (such as other algae and water impurities), nor can it confirm the specific species of cyanobacteria (such as toxic and non-toxic cyanobacteria). It can only preliminarily determine the possibility of cyanobacterial blooms in the water area, and therefore it is called a "water area suspected of having a risk of cyanobacterial blooms". This provides a target for subsequent visual image monitoring and patch extraction at the mesoscopic level, and further verifies the authenticity of the risk.

[0092] In S130, after identifying the risky water area, detailed visual image data of the risk area needs to be provided by the drone. This visual image data can be images or multi-frame video. If it is video, the video is decomposed into multiple consecutive image frames, which are then processed in S140. It is understandable that when the water area is large, a complete image of the water area can be obtained by stitching together multiple visual image data.

[0093] In S140, patch data for each sub-region within the risk area is obtained through visual image data. Compared to satellite remote sensing water quality monitoring data, visual image data can more accurately present the specific distribution morphology of cyanobacteria within the risk area. Patch data is a quantitative representation of the aggregation state, distribution range, and density of cyanobacteria in each sub-region, serving as crucial data connecting the macro-risk area and the meso-risk sub-region determination. Water quality data acquired by satellite remote sensing can only reflect the overall trend of cyanobacteria changes in the entire risk area, failing to distinguish local differences in cyanobacteria within each sub-region (e.g., some sub-regions have cyanobacteria aggregation, while others have no cyanobacteria or sparse cyanobacteria). However, extracting patch data for each sub-region from visual image data allows for precise quantification of cyanobacteria aggregation characteristics in each sub-region. This provides reliable and targeted data support for subsequent screening of risk sub-regions based on patch data, obtaining the first cyanobacteria concentration in risk sub-regions, and inputting patch data into a preset model for prediction. Simultaneously, it can exclude areas within the sub-region with no or sparse cyanobacteria, avoiding redundancy in subsequent monitoring and calculations, and ensuring the accuracy and targeting of subsequent risk assessment and prediction.

[0094] When segmenting a risk area into sub-regions, the complete water area and boundaries of the risk area are first determined based on the visual image data of the risk area with registered spatial coordinates. Then, a regularized uniform grid segmentation method is used to divide the entire risk area into several independent units with non-overlapping and uniform ranges according to a preset spatial size. Each independent unit is a sub-region. Alternatively, the natural geographical features of the water area can be combined with natural boundaries such as bays, shoals, and water flow sections to assist in optimizing the segmentation, ensuring that the sub-regions match the actual water body spatial structure. Each sub-region is represented by a unique ID combined with a corresponding spatial coordinate range. The ID is used to distinguish different sub-regions, and the spatial coordinate range is used to clarify the specific location of each sub-region within the risk area, realizing the standardized identification and positioning of sub-regions. Subsequently, the patch data of each sub-region is bound to the ID and coordinates of that sub-region.

[0095] Therefore, the patch data in this embodiment includes patch diffusion rate and patch density. After extracting cyanobacterial patches based on visual image data of the risk area, patch density refers to the total number of cyanobacterial patches contained in the spatial range of a single sub-region as a statistical unit. The number of patches directly quantifies the spatial density of cyanobacterial patches in the sub-region. Patch diffusion rate refers to the extraction of time-series patch density data of a single sub-region over a second preset time period in the past. The change in patch density during this period is calculated to quantitatively characterize the dynamic rate of aggregation and diffusion of cyanobacterial patches in the sub-region over the second preset time period.

[0096] Based on this, such as Figure 2 As shown, S140 specifically includes the following steps:

[0097] S141: Obtain the red-edge band reflectance and near-infrared band reflectance of each pixel in the visual image data;

[0098] S142: Based on the red-edge band reflectance and near-infrared band reflectance, obtain an evaluation value for each pixel, the evaluation value being used to characterize the density of cyanobacteria in the corresponding pixel;

[0099] S143: Based on all the evaluation values, determine all connected components in the visual image data that represent suspected patches;

[0100] S144: Obtain the green band reflectance and near-infrared band reflectance of each of the connected domains;

[0101] S145: Remove the connected components that satisfy the second preset condition to obtain the patch. The second preset condition is: the green band reflectance is less than or equal to the first green wave threshold, and the near-infrared band reflectance is less than or equal to the first infrared wave threshold, or the green band reflectance is greater than or equal to the second green wave threshold, and the near-infrared band reflectance is greater than or equal to the second infrared wave threshold, wherein the first green wave threshold is less than the second green wave threshold, and the first infrared wave threshold is less than the second infrared wave threshold.

[0102] In S141, the first device used to acquire visual images of the risk area is a drone or satellite platform equipped with a multispectral imaging sensor. This sensor can simultaneously capture multi-channel spectral data, including red-edge band reflectance and near-infrared band reflectance. When imaging the ground, it collects the raw spectral radiation values ​​of surface water bodies point by point, with pixels as the smallest unit. Subsequently, the raw visual image data is preprocessed with radiometric calibration and geometric registration to convert the dimensionless DN values ​​output by the sensor into true physical radiance values, ensuring that the red-edge band reflectance, near-infrared band reflectance, and image spatial pixels are accurately matched one-to-one without spatial offset. Finally, through a spectral separation algorithm, the red-edge band reflectance layer and the near-infrared band reflectance layer are independently separated from the preprocessed multispectral image. All pixel units are traversed and the corresponding red-edge band reflectance spectral value and near-infrared band reflectance spectral value of each pixel are read. The pixel spatial coordinates are bound and matched with the two band values, and finally, the red-edge band reflectance and near-infrared band reflectance data of all pixels in the visual image are acquired.

[0103] The evaluation value is calculated in S142 as follows:

[0104]

[0105] This represents a fixed offset of 0.5, used to map the exponential response range to the [0,1] interval, simplifying subsequent threshold segmentation calculations. Due to the strong red edge reflection and weak near-infrared reflection of cyanobacteria, the UAV-CI value is significantly higher than that of normal water (the red edge and near-infrared reflectivity of normal water are close, and the UAV-CI value is close to 0.5), thus achieving strong differentiation of the spectral characteristics between cyanobacteria and the background.

[0106] RE represents the reflectivity of the red-edge band, and NIR represents the reflectivity of the near-infrared band.

[0107] This indicates the evaluation value.

[0108] In S143, based on the calculated evaluation values ​​of all pixels, an evaluation value threshold is first set according to the cyanobacteria identification standard. The visual image is then converted into a binary image containing only suspected patch pixels and background water pixels. Pixels with evaluation values ​​higher than the threshold are marked as suspected patch pixels, and the remaining pixels are marked as background pixels. Subsequently, an eight-neighborhood connected component analysis algorithm is used to traverse and detect the binary image row by row and pixel by pixel. Suspected patch pixels that are spatially adjacent and connected are aggregated and classified to form independent pixel sets. Each pixel set is a connected component. Finally, connected components with extremely small area noise caused by water impurities and light interference are removed, and connected components that conform to the size range of cyanobacteria patches are retained. Finally, all connected components representing suspected patches in the visual image data are determined.

[0109] In S144, the sum of the green band reflectance and the sum of the near-infrared band reflectance of all pixels in each connected domain are calculated.

[0110] In S145, the purpose of setting the second preset condition is to eliminate false connected regions caused by environmental interference during the cyanobacterial patch extraction process, ensuring that the final patches are real cyanobacterial aggregation areas. Among them, connected regions with low reflectivity in both the green and near-infrared bands mostly correspond to water shadows, dark areas with greater water depth, or exposed bottom areas, and are not formed by cyanobacterial aggregation. Connected regions with high reflectivity are mostly interference areas such as strong light reflection from the water surface, bright areas of shallow bottom sediment, or floating suspended matter, and are also not real cyanobacterial patches. Through this dual threshold combination condition, the above two types of typical false connected regions can be effectively filtered out, improving the accuracy and reliability of cyanobacterial patch extraction, and providing real and effective basic data for subsequent patch data calculation and algal bloom prediction.

[0111] In this embodiment, in S150, the determination of the risk sub-region can be achieved based on the following method, specifically:

[0112] Based on the patch density and patch diffusion rate, a first coupling index is determined according to a first preset model; when the first coupling index is greater than a first preset coupling value, the corresponding sub-region is the risk sub-region.

[0113] The first preset model is:

[0114]

[0115] Indicates the first coupling index;

[0116] Indicates the weight of the degree of clustering;

[0117] Indicates the synergistic enhancement coefficient;

[0118] and These represent the normalized patch density and patch diffusion rate, respectively.

[0119] First, this embodiment measures the patch density ( (reflecting the degree of static aggregation of cyanobacteria) and the rate of patch diffusion ( The two methods (one reflecting the dynamic spread trend of cyanobacteria) were normalized to eliminate the dimensional differences between them, ensuring fair participation in the calculation at the same scale; secondly, through weighting... It can be flexibly adapted to the monitoring characteristics of different water areas (such as still water areas with severe eutrophication, which can improve...). With a greater focus on plaque aggregation, highly mobile waters can reduce [the problem]. (With a greater focus on the speed of diffusion), to achieve linear integration of fundamental risks.

[0120] The interaction term indicates that the risk of cyanobacterial blooms is not simply the sum of patch density and diffusion rate. When patch density is high and diffusion rate is fast, cyanobacteria spread rapidly to surrounding areas, and the risk of blooms is non-linearly amplified. This interaction term is represented by the synergistic enhancement coefficient. This avoids the shortcomings of relying solely on linear weighting, which cannot fully reflect the combined risks of both factors, and significantly improves the sensitivity to identifying high-risk sub-regions.

[0121] The first coupling index CBRCI obtained can simultaneously quantify the static aggregation degree, dynamic diffusion trend and synergistic enhancement effect of cyanobacterial patches. When CBRCI exceeds the preset threshold, it can accurately screen out risk sub-regions that have the characteristics of "high aggregation + fast diffusion", providing reliable targets for subsequent cyanobacterial concentration prediction and algal bloom early warning.

[0122] The first preset coupling value is based on historical monitoring data of the complete cyanobacterial growth cycle of the target water area (including normalized patch density, patch diffusion rate and corresponding real algal bloom labels verified in the field). First, the CBRCI value of all sub-regions is calculated and ROC curves are plotted in combination with the real algal bloom labels. The CBRCI value corresponding to the point with the largest Youden index is selected as the initial threshold. Then, it is manually fine-tuned according to the water function (such as drinking water source areas, the threshold is lowered to reduce underreporting, and landscape water bodies, the threshold can be raised to reduce false alarms) and seasonal and hydrological conditions. Finally, the underreporting rate and false alarm rate are statistically analyzed by actual monitoring. The values ​​are iteratively adjusted until both meet the acceptable range of business operations. The final empirical statistical threshold is determined to be suitable for the characteristics of the water area and to balance the sensitivity and specificity of risk identification.

[0123] In S160, after determining the risk sub-region, it is necessary to collect hyperspectral data and multibeam data of the risk sub-region based on the unmanned vessel in the risk sub-region. The multibeam data represents the underwater three-dimensional detection data collected by the shipborne multibeam bathyphone sonar system, and the hyperspectral data represents the water surface detection data collected by the shipborne hyperspectral sensor. The multibeam data includes the scattering intensity of suspended objects, and the hyperspectral data includes the first reflectance and the second reflectance. The first reflectance and the second reflectance have different characteristic bands. In this embodiment, the first reflectance and the second reflectance are the water reflectance at a wavelength of 620 nanometers and the water reflectance at a wavelength of 680 nanometers, respectively.

[0124] Based on this, S170 specifically includes the following steps:

[0125] S171: Determine the first confidence level of the hyperspectral data based on the first reflectance and the second reflectance, as well as the minimum threshold and the maximum threshold of the historical cyanobacterial characteristic band ratio of the pre-stored risk sub-region.

[0126] S172: Determine the second confidence level of the multibeam data based on the scattering intensity of the suspended object and the first reflectivity;

[0127] S173: Determine the third confidence level of the water quality monitoring data based on the cyanobacteria index and chlorophyll a remote sensing data;

[0128] S174: Based on the first confidence level, determine the first weight of the hyperspectral data; based on the second confidence level, determine the second weight of the multibeam data; and based on the third confidence level, determine the third weight of the water quality monitoring data.

[0129] S175: Substitute the first weight, the second weight, and the third weight, along with the water quality monitoring data, hyperspectral data, and multibeam data, into the fusion model to obtain the fused data.

[0130] In S171, the first confidence level is calculated based on the following formula:

[0131]

[0132] The first reflectance is the water reflectance at a wavelength of 620nm collected by a hyperspectral sensor (a characteristic peak of phycocyanin, a core identification index of cyanobacteria).

[0133] The second reflectance refers to the water reflectance at a wavelength of 680 nm collected by a hyperspectral sensor (a characteristic peak of chlorophyll a, an indicator of total algal concentration).

[0134] This represents the reflectance ratio of characteristic wavelengths of cyanobacteria, used to distinguish cyanobacteria from other algae.

[0135] This represents the minimum threshold for the ratio of historical cyanobacterial characteristic bands, obtained from statistical analysis of historical cyanobacterial samples. It is the lower limit for determining the ratio of cyanobacteria.

[0136] This represents the maximum threshold for the pre-stored historical cyanobacteria characteristic band ratio, which is obtained from the statistics of historical measured cyanobacteria samples and is the upper limit for determining the ratio of cyanobacteria.

[0137] clamp() is a cutoff function that restricts the calculation result to the interval [0,1] (ratios less than 1). When the value is 0, it is higher than 0. Take 1 at the time.

[0138] This represents the first confidence level of the hyperspectral data, with a value of [0,1]. A higher value indicates that the spectral characteristics are more consistent with the typical characteristics of cyanobacteria.

[0139] In S172, the second confidence level is calculated based on the following formula:

[0140]

[0141] Used to represent the scattering intensity of suspended matter, the acoustic scattering intensity of particles (including cyanobacterial cells) in water collected by multibeam echo sounders, reflecting the distribution of underwater / mid-water algae.

[0142] Corr() is used to represent the Pearson correlation coefficient, and is calculated... and The degree of linear correlation between the two is measured by taking the absolute value of the correlation.

[0143] clamp(·) is a truncation function that restricts the result to [0,1] (the absolute value of the correlation coefficient is naturally in the [0,1] interval; this step is to ensure a consistent format).

[0144] This represents the second confidence level of the multibeam data, with a value of [0,1]. A higher value indicates a stronger correlation between the acoustic scattering characteristics and the spectral characteristics of cyanobacteria, and thus a more reliable data.

[0145] In S173, the third confidence level can be calculated based on the following formula:

[0146]

[0147] CI stands for Cyanobacteria Index, a macroscopic index of cyanobacteria distribution intensity retrieved from satellite remote sensing.

[0148] Chl-a represents chlorophyll a remote sensing data, an index of total algal concentration in water bodies retrieved from satellite remote sensing.

[0149] This represents the normalized cyanobacteria index / chlorophyll a concentration, mapped to the [0,1] interval to eliminate dimensional differences.

[0150] This represents the minimum / maximum historical cyanobacteria index, used for normalization.

[0151] This represents the minimum / maximum historical chlorophyll a concentration, used for normalization.

[0152] Consist represents the consistency index, with values ​​ranging from [0,1]. A larger value indicates that the consistency between CI and Chl-a in characterizing cyanobacteria is greater.

[0153] Signif represents the significance index, with values ​​ranging from [0,1]. A larger value indicates that the CI is significantly higher than that of waters without cyanobacteria, and the higher the probability of the presence of cyanobacteria.

[0154] This represents the pre-stored background threshold for the cyanobacteria index in cyanobacteria-free waters, obtained from long-term cyanobacteria-free samples.

[0155] This represents the third confidence level of water quality monitoring data, with a value of [0,1]. A higher value indicates that the satellite remote sensing data is more reliable in characterizing cyanobacteria.

[0156] In S174, the first weight, second weight, and third weight are determined based on the following formula:

[0157]

[0158] This represents the sum of the three confidence levels, used for normalizing the weights.

[0159] This represents the first weight (hyperspectral data weight), which is adaptively assigned to the fusion weight of the hyperspectral modes.

[0160] This represents the second weight (multi-beam data weight), which is adaptively assigned to the fusion weights of the multi-beam modes.

[0161] This represents the third weight (water quality monitoring data weight), which is adaptively assigned to the fusion weights of the satellite remote sensing modes.

[0162] in, This ensures the scale consistency of the fused features.

[0163] In S175, the fused data is determined through a fusion model, which specifically includes the following formula:

[0164]

[0165] This represents the normalized hyperspectral features, mapped to the [0,1] interval.

[0166] The normalized multibeam characteristics are mapped to the [0,1] interval.

[0167] This represents the normalized satellite remote sensing features, balancing the contribution of the cyanobacteria index and chlorophyll a.

[0168] The minimum / maximum ratios of historical cyanobacterial characteristics are used for hyperspectral feature normalization.

[0169] The minimum / maximum values ​​of historical suspended object scattering intensity are used for multi-beam feature normalization.

[0170] This represents the fused data, which is the result of multimodal fusion integrating three types of modal features, and is used for the subsequent calculation of the first cyanobacteria concentration.

[0171] In S180, the concentration of the first cyanobacteria is calculated according to the following formula:

[0172]

[0173] The first cyanobacterial concentration represents the actual concentration level of cyanobacteria in the current risk sub-region, and the unit is consistent with the measured cyanobacterial concentration (e.g., μg / L or cell number / mL).

[0174] 'a' represents the linear inversion slope coefficient, which is derived from historically measured cyanobacteria concentrations and corresponding fusion characteristics. The value was obtained by least squares fitting, reflecting the change in cyanobacteria concentration when the fusion characteristics change by 1 unit.

[0175] b represents the linear inversion intercept coefficient, which is also obtained by fitting historical measured data, and represents the value when the fused features... =0 (no cyanobacteria characteristics), the baseline value of cyanobacteria concentration (usually close to 0, which can be adjusted according to the background noise of the water area).

[0176] This indicates the trigger threshold for cyanobacteria concentration. At that time, it was considered that there was no risk of blue-green algae, and the concentration was set to 0. This indicates the saturation threshold of cyanobacteria concentration. At that time, it was considered that the concentration had reached the upper limit of observation, and the highest historical measured concentration was taken. .

[0177] This represents the maximum historically measured concentration of cyanobacteria, used to limit the upper limit of concentration prediction under high fusion characteristics and avoid numerical overflow.

[0178] The first pre-defined model mentioned in S190 is a deep learning model, and its architecture includes:

[0179] The MLP branch is used to receive the first cyanobacteria concentration and the corresponding patch data, and output the spatial feature vector; the L-TCN branch is used to receive the first cyanobacteria concentration and the corresponding patch data, and output the temporal feature vector; the fusion layer is used to receive and splice the spatial feature vector and the temporal feature vector to obtain the spliced ​​feature vector; the fully connected layer is used to output the second cyanobacteria concentration and the cyanobacteria diffusion rate based on the spliced ​​feature vector.

[0180] In this embodiment, the first preset model adopts a dual-branch spatiotemporal fusion architecture, consisting of a parallel MLP spatial feature extraction branch, an L-TCN temporal feature extraction branch, a feature fusion layer, and a dual-task fully connected output layer. The MLP spatial feature branch is a multilayer perceptron structure that receives static spatial features such as the first cyanobacteria concentration and patch density of the current risk sub-region as input. These features are processed through an input mapping layer, two fully connected hidden layers with ReLU activation functions, a batch normalization layer, and a random deactivation layer to fully exploit the correlation between cyanobacteria concentration and patch spatial distribution, ultimately outputting a fixed-dimensional spatial feature vector. The L-TCN temporal feature branch is a temporal convolutional network structure that incorporates a long short-term memory mechanism, receiving the first cyanobacteria concentration within the past second preset time period of the risk sub-region. The system takes a time series consisting of concentration, patch density, and patch diffusion rate as input, captures time-dependent features through multiple dilated causal convolutional layers, and combines LSTM long-term memory units to enhance the extraction of time-series change patterns. After time-series normalization, it outputs a fixed-dimensional time-series feature vector. The fusion layer is a feature splicing layer that splices and fuses spatial and time-series feature vectors in the feature dimension to obtain a spliced ​​feature vector that combines static spatial aggregation characteristics and dynamic time-series diffusion patterns. The fully connected layer adopts a multi-layer fully connected network structure with dual-task output. Taking the spliced ​​feature vector as input, it outputs the prediction results independently in two paths after hidden layer feature mapping. One path outputs the second cyanobacteria concentration in the risk area after the first preset time period, and the other path outputs the cyanobacteria diffusion rate of the corresponding time period.

[0181] Therefore, the training of the first preset model needs to be completed before executing S190. The training method for the first preset model is as follows:

[0182] Multiple mapping data of the risk area are obtained, including first historical cyanobacteria concentration, historical patch data, and second historical cyanobacteria concentration and historical cyanobacteria diffusion rate after a first preset time. The first historical cyanobacteria concentration and historical patch data are used as inputs, and the corresponding second historical cyanobacteria concentration and historical cyanobacteria diffusion rate are used as outputs to train a first preset model and obtain the trained first preset model.

[0183] This embodiment first constructs multiple sets of mapping datasets required for model training. Based on historical data from long-term monitoring of the target water area, multiple sets of mapping data corresponding to each sub-region within the risk area are extracted. Each set of mapping data includes two parts: input features and supervision labels. The input features are the first historical cyanobacteria concentration and historical patch data (historical patch density and historical patch diffusion rate time-series data) of the sub-region. The supervision labels are the second historical cyanobacteria concentration and historical cyanobacteria diffusion rate after a first preset time period, corresponding to the input features. All mapping data are standardized and preprocessed to eliminate the differences in data units of different features. They are then divided into training, validation, and test sets according to a preset ratio to complete the training data preparation. Subsequently, the network parameters of the first preset model are initialized. The training set's first historical cyanobacteria concentration and historical patch data are input into the model. Spatial feature vectors are extracted through the MLP branch, and temporal feature vectors are extracted through the L-TCN branch. After being concatenated by a fusion layer, the predicted second cyanobacteria concentration and cyanobacteria diffusion rate are output by a fully connected layer. The total loss function is calculated based on the model's predicted values ​​and supervision labels. An adaptive moment estimator (Adam) is used for backpropagation to update the network parameters of the MLP branch, L-TCN branch, fusion layer, and fully connected layer layer by layer. During training, after each iteration, validation set data is input into the model to evaluate its performance. An early stopping strategy is used to avoid model overfitting. The iteration continues until the total loss function converges and the validation set performance is stable, finally obtaining the first preset model after training.

[0184] The loss function for training the first preset model is:

[0185] The second cyanobacteria concentration predicted by the model. This represents the actual second-historical concentration of cyanobacteria. The model predicts the rate of cyanobacteria spread. This represents the actual historical rate of cyanobacteria spread. These are dual-task weighting coefficients, used to balance the importance of the two prediction tasks; This is the L2 regularization term, used to suppress model overfitting. This represents the mean square error.

[0186] The loss function measures the errors in concentration prediction and velocity prediction using mean squared error (MSE) respectively. Combined with regularization terms to constrain model complexity, it achieves collaborative optimization of the two tasks, ensuring that the model can accurately output both cyanobacteria concentration and diffusion rate at the same time.

[0187] In S1100, a second coupling index is introduced to determine the predicted likelihood of cyanobacterial blooms. Specifically:

[0188] Based on the second cyanobacteria concentration and cyanobacteria diffusion rate, the second coupling index is determined according to the second preset model;

[0189] When the second coupling index is greater than the second preset coupling value, a prediction result is obtained to characterize whether a cyanobacterial bloom will occur; otherwise, a prediction result is obtained that a cyanobacterial bloom will not occur.

[0190] The second preset model is:

[0191]

[0192] This represents the historical maximum concentration of cyanobacteria in the aforementioned water area;

[0193] This represents the historical maximum diffusion rate of cyanobacteria in the aforementioned waters;

[0194] This is the second coupling index;

[0195] Indicates the rate of cyanobacteria diffusion;

[0196] Indicates the concentration of the second cyanobacteria;

[0197] This represents the normalized concentration of the second cyanobacteria;

[0198] This is the preset critical threshold for cyanobacterial bloom concentration;

[0199] This is the diffusion rate exponential enhancement coefficient.

[0200] In this embodiment, the formula for the exponential enhancement term The constant 0.5 in the value represents the normalized baseline bias. This bias is not set by fixing the statistical median (or background value) of the normal cyanobacteria diffusion rate in the long-term historical observation data of the target water area. When the value is 0.5, it represents the historical average baseline level of the current cyanobacteria expansion. At this time, the exponent term takes a value of 1, i.e. The system does not nonlinearly amplify the concentration risk; only when the diffusion rate deviates from the baseline value does the system initiate an exponential penalty for fast diffusion or a reduction in slow diffusion, thus enabling the model to have adaptive adjustment capabilities to suit the background characteristics of different water areas.

[0201] when When the concentration of cyanobacteria is extremely low, it is directly normalized to 0, reflecting the mechanism that "if the concentration does not reach the baseline level, there is no risk of outbreak even if the diffusion is fast".

[0202] when When the value is 0.5, the exponent term is 1. ;

[0203] when When the value is greater than 0.5, the exponential term increases exponentially with the diffusion rate, amplifying the accelerating effect of diffusion on algal blooms.

[0204] when When the value is less than 0.5, the exponential term decreases gradually as the diffusion rate decreases, thus weakening the effect of diffusion.

[0205] In this embodiment, segmented normalization can filter out low-concentration, risk-free areas, reflect the risk increase of critical concentrations in gradients, and identify extreme risks of saturated high concentrations, thus avoiding interference from low concentrations or overflow of high concentration values. The nonlinear exponential term uses a diffusion rate of 0.5 as a boundary, with slight amplification during slow diffusion and exponential amplification during fast diffusion, capturing the high-risk synergistic effect of "high concentration + fast diffusion," enabling the second coupling index to simultaneously quantify the basic risk of concentration and the risk of diffusion gain. By comparing with the second preset coupling value, the sensitivity and specificity of the early warning can be balanced, ensuring that high-risk outbreak scenarios are not missed and low-risk false alarms are reduced, ultimately achieving accurate prediction of cyanobacterial blooms.

[0206] The Second Coupling Index (PCBRCI) is a quantitative indicator characterizing the comprehensive risk level of cyanobacterial blooms in a future (after a first preset time period) risk sub-region. It simultaneously integrates the basal accumulation risk of future cyanobacterial concentration and the dynamic catalytic risk of cyanobacterial diffusion: a piecewise normalized concentration term ( This characterizes whether the future cyanobacterial concentration reaches the baseline threshold for an algal bloom, and the risk gradient when the concentration approaches / exceeds the threshold (low concentration: no risk; critical concentration: risk gradually increases; high concentration: extreme risk), and the nonlinear diffusion term ( The value represents the catalytic amplification effect of the future cyanobacterial diffusion rate on the concentration risk. That is, the slower the diffusion, the weaker the amplification effect; the faster the diffusion, the exponentially higher the risk. Ultimately, the two are coupled, and the value of "precision" indicates that the higher the probability and severity of cyanobacterial blooms in the future in this sub-region.

[0207] This application provides an electronic device that may include a processing module and a memory. The memory stores a computer program, which, when executed by the processor, enables the electronic device to perform the corresponding steps in the aforementioned multimodal sensing-based cyanobacterial bloom prediction method.

[0208] In this embodiment, the processor can be an integrated circuit chip with signal processing capabilities. For example, the processor can be a central processing unit (CPU), 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, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0209] The memory can be, but is not limited to, random access memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, etc. In this embodiment, the memory can be used to store preset numbers, etc. Of course, the memory can also be used to store programs, which the processor executes after receiving an execution instruction.

[0210] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the electronic device described above can be referred to the corresponding steps in the aforementioned method, and will not be elaborated further here.

[0211] This application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to execute the cyanobacterial bloom prediction method based on multimodal sensing as described in the above embodiments.

[0212] Computer-readable storage media may be magnetic disks, optical disks, read-only memory, random access memory, flash memory, USB flash drives, hard disks, or solid-state drives, etc., and may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implement the methods shown in the above embodiments.

[0213] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned multimodal sensing-based cyanobacterial bloom prediction method. The computer program product may exist in a computer-readable storage medium in forms including, but not limited to, source files, executable files, and installation package files.

[0214] Based on the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by hardware or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, electronic device, or network device, etc.) to execute the methods described in the various implementation scenarios of this application.

[0215] In the embodiments provided in this application, it should be understood that the disclosed methods can also be implemented in other ways. The method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or part of code, which includes one or more executable instructions for implementing a specified logical function. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0216] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting cyanobacterial blooms based on multimodal sensing, characterized in that, The method includes: Acquire water quality monitoring data for all water bodies based on satellite remote sensing; Based on the water quality monitoring data of all the aforementioned water areas, all water areas suspected of having a risk of cyanobacterial blooms are identified as risk areas. Obtain visual image data of the risk area; Based on the visual image data, patch data for each sub-region of the risk area is obtained, wherein the risk area is divided into multiple sub-regions; Based on the patch data, all sub-regions in the risk area that represent the risk of suspected cyanobacterial blooms are obtained, and these are the risk sub-regions. Acquire hyperspectral data and multibeam data of the risk sub-region. The hyperspectral data includes a first reflectance and a second reflectance, the first reflectance and the second reflectance have different characteristic bands, and the multibeam data includes the scattering intensity of suspended matter. According to the preset fusion strategy, the hyperspectral data, multibeam data, and water quality monitoring data of the risk sub-region are fused to obtain fused data; Based on the fused data, the first cyanobacteria concentration in the risk sub-region is obtained; The first cyanobacteria concentration and the corresponding patch data are input into the first preset model after training. The first preset model after training outputs the second cyanobacteria concentration and cyanobacteria diffusion rate of the risk area after a first preset time. Based on the second cyanobacterial concentration and the cyanobacterial diffusion rate, the prediction results for whether a cyanobacterial bloom will occur are determined.

2. The method according to claim 1, characterized in that, The water quality monitoring data includes the cyanobacteria index and chlorophyll a remote sensing data. Based on the water quality monitoring data of all the aforementioned water areas, all water areas suspected of having a risk of cyanobacterial blooms are obtained, including: Based on the water quality monitoring data, the first time series data of the cyanobacteria index in each water body over the past second preset time period is obtained, and the second time series data of the chlorophyll a remote sensing data in each water body over the past second preset time period is obtained. When the water body meets the first preset condition, and a specified number of other adjacent water bodies also meet the first preset condition, the water body is a risk area. The first preset condition is: the first time series data corresponding to the water body indicates that the cyanobacteria index has been increasing over the past second preset time period; the second time series data indicates that the chlorophyll a remote sensing data has been increasing over the past second preset time period; and the phase difference between the first time series data and the second time series data is less than or equal to a phase threshold.

3. The method according to claim 1, characterized in that, The patch data includes patch spread rate and patch density; The step of obtaining patch data in the risk area based on the visual image data includes: Based on the visual image data, the patches in the risk area are obtained; Based on all the patches, obtain the patch density and patch diffusion rate of each sub-region, wherein the patch density is obtained based on the number of patches in the risk sub-region, and the patch diffusion rate is used to characterize the amount of change in the patch density of the risk sub-region over the past second preset time period.

4. The method according to claim 3, characterized in that, The step of obtaining the patches in the risk area based on the visual image data includes: Obtain the red-edge band reflectance and near-infrared band reflectance of each pixel in the visual image data; Based on the red-edge band reflectance and near-infrared band reflectance, an evaluation value is obtained for each pixel, and the evaluation value is used to characterize the density of cyanobacteria in the corresponding pixel. Based on all the evaluation values, determine all connected components in the visual image data that represent suspected patches; Obtain the green band reflectance and near-infrared band reflectance of each of the connected domains; Remove the connected components that satisfy the second preset condition to obtain the patch. The second preset condition is: the green band reflectance is less than or equal to the first green wave threshold, and the near-infrared band reflectance is less than or equal to the first infrared wave threshold, or the green band reflectance is greater than or equal to the second green wave threshold, and the near-infrared band reflectance is greater than or equal to the second infrared wave threshold, wherein the first green wave threshold is less than the second green wave threshold, and the first infrared wave threshold is less than the second infrared wave threshold.

5. The method according to claim 4, characterized in that, The step of obtaining all sub-regions in the risk area that represent the potential risk of cyanobacterial blooms based on the patch data includes: Based on the patch density and patch diffusion rate, a first coupling index is determined according to a first preset model; When the first coupling index is greater than the first preset coupling value, the corresponding sub-region is the risk sub-region; The first preset model is: ； Indicates the first coupling index; Indicates the weight of the degree of clustering; Indicates the synergistic enhancement coefficient; and These represent the normalized patch density and patch diffusion rate, respectively.

6. The method according to claim 1, characterized in that, Before inputting the first cyanobacteria concentration and corresponding patch data into the trained first preset model, and before outputting the second cyanobacteria concentration and cyanobacteria diffusion rate of the risk area after a first preset time period through the trained first preset model, the method further includes: Acquire multiple mapping data of the risk area, including first historical cyanobacteria concentration, historical patch data, and second historical cyanobacteria concentration and historical cyanobacteria diffusion rate after a first preset time. Using the first historical cyanobacteria concentration and historical patch data as inputs, and the corresponding second historical cyanobacteria concentration and historical cyanobacteria diffusion rate as outputs, a first preset model is trained to obtain the trained first preset model.

7. The method according to claim 2, characterized in that, The process involves fusing the hyperspectral data, multibeam data, and water quality monitoring data from the risk sub-regions according to a preset fusion strategy to obtain fused data, including: The first confidence level of the hyperspectral data is determined based on the first reflectance and the second reflectance, as well as the minimum threshold and the maximum threshold of the historical cyanobacterial characteristic band ratio of the pre-stored risk sub-region. The second confidence level of the multibeam data is determined based on the scattering intensity of the suspended object and the first reflectivity. The third confidence level of the water quality monitoring data was determined based on the cyanobacteria index and chlorophyll a remote sensing data. Based on the first confidence level, a first weight of the hyperspectral data is determined; based on the second confidence level, a second weight of the multibeam data is determined; and based on the third confidence level, a third weight of the water quality monitoring data is determined. The first weight, the second weight, and the third weight, along with the water quality monitoring data, hyperspectral data, and multibeam data, are substituted into the fusion model to obtain the fused data.

8. The method according to claim 1, characterized in that, The prediction results for determining whether a cyanobacterial bloom will occur based on the second cyanobacterial concentration and the cyanobacterial diffusion rate include: Based on the second cyanobacteria concentration and cyanobacteria diffusion rate, the second coupling index is determined according to the second preset model; When the second coupling index is greater than the second preset coupling value, a prediction result is obtained to characterize whether a cyanobacterial bloom will occur; otherwise, a prediction result is obtained that a cyanobacterial bloom will not occur. The second preset model is: ； This represents the historical maximum concentration of cyanobacteria in the aforementioned water area; This represents the historical maximum diffusion rate of cyanobacteria in the aforementioned waters; This is the second coupling index; Indicates the rate of cyanobacteria diffusion; Indicates the concentration of the second cyanobacteria; This represents the normalized concentration of the second cyanobacteria; This is the preset critical threshold for cyanobacterial bloom concentration; This is the diffusion rate exponential enhancement coefficient.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory coupled together, the memory storing a computer program that, when executed by the processor, causes the electronic device to perform the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 8.

Images