Multi-fusion seaweed field ecosystem observation method, system, equipment and medium

Abstract

The invention provides a multi-fusion seaweed field ecosystem observation method, system, device and medium, and belongs to the technical field of ecological monitoring, the method comprises the following steps: obtaining chlorophyll a concentration, seaweed canopy spectrum and three-dimensional biomass point cloud; aligning the chlorophyll a concentration of the target sea area with the seaweed canopy spectrum, and fusing the three-dimensional biomass point cloud to generate a three-dimensional biomass model; constructing an in-situ sampling network to monitor water quality parameters, benthic organism video streams and eDNA metagenome sequencing data, calibrating a three-dimensional biomass model, executing anomaly detection through a lightweight LSTM model, and identifying benthic organism species in real time through an improved YOLOv5 model; constructing a graph neural network, outputting a carbon sink prediction value, generating a brown tide early warning signal when the carbon sink prediction value is lower than a dynamic threshold value, optimizing a patrol path of the unmanned aerial vehicle based on reinforcement learning, and improving the sampling frequency of the water quality sensor. According to the invention, multi-fusion monitoring of the seaweed field is realized, the ecological condition is accurately evaluated, and abnormity is warned in advance.

Description

Technical Field

[0001] The present application belongs to the field of ecological monitoring technology, and specifically relates to a multi-integrated seaweed field ecosystem observation method, system, equipment and medium. Background Art

[0002] With the rapid development of the marine economy and the impact of global climate change, the marine ecological environment faces many severe challenges. Seaweed beds, as a key component of the marine ecosystem, are of great significance to the marine ecological environment. They not only maintain biodiversity and are related to fishery resources, but also purify water quality. Therefore, accurate and comprehensive monitoring of seaweed bed ecosystems is particularly critical.

[0003] Traditional seaweed farm ecosystem monitoring has the following limitations: First, although satellite remote sensing alone can obtain chlorophyll distribution over a large area, it is difficult to accurately analyze the detailed structure of algae communities; conventional water quality monitoring equipment can only conduct positioning monitoring of local waters and cannot fully grasp the overall scale and spatial layout of seaweed farms. Secondly, traditional monitoring methods mostly collect data on a regular basis, which cannot grasp the dynamic changes of seaweed farms in real time. For sudden ecological anomalies such as brown tides, early warnings cannot be issued in time, making it difficult to take precautions in advance and respond accurately. Finally, existing methods are mostly limited to the monitoring of a single ecological element, and fail to fully integrate multiple factors such as large seaweed communities, swimming animals, large benthic animals, water quality, and sedimentary environments. It is impossible to comprehensively assess the health status and development trends of seaweed farm ecosystems, and it is difficult to analyze the far-reaching impact of changes in marine ecological environment quality on seaweed farms. Summary of the Invention

[0004] In a first aspect, an embodiment of the present application provides a multi-fusion seaweed field ecosystem observation method, comprising the following steps: S1. Identify the target sea area and obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data using a hyperspectral imager mounted on an unmanned aerial vehicle, and obtain three-dimensional biomass point cloud data using a ship-borne lidar. S2. Align the chlorophyll a concentration data of the target sea area with the algae canopy spectral data using a U-Net network to generate 2D raster data. This 2D raster data is then fused with the 3D biomass point cloud data using a 3D convolutional neural network to generate a 3D biomass model. S3. Deploy water quality sensors in the target sea area to monitor water quality parameters in real time. Use an underwater robot equipped with a microscopic camera to collect benthic video streams. Deploy an eDNA sampler to collect eDNA metagenomic sequencing data. Generate an in situ validation dataset based on the water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. Use this in situ validation dataset to perform in situ validation and calibration of the 3D biomass model. S4. Edge nodes use a lightweight LSTM model to detect anomalies in water quality parameters and identify benthic species in real time using an improved YOLOv5 model. S5. Construct a graph neural network using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges, and output carbon sink predictions and key influencing factors through the graph neural network. S6. When the predicted carbon sink value falls below the dynamic threshold, a brown tide warning signal is generated. Reinforcement learning is then used to optimize the drone patrol path and automatically increase the sampling frequency of water quality sensors in abnormal target sea areas.

[0005] Furthermore, the specific steps of step S1 are as follows: S11. Obtaining a chlorophyll a concentration distribution map of the target sea area by using a satellite remote sensing sensor according to a preset first spatial resolution to generate a chlorophyll a spatial dataset; S12. Collecting spectral data of the seaweed canopy using a hyperspectral imager carried by the drone according to a preset second spatial resolution, a preset flight altitude, a set number of bands, and a set spectral range; S13. The shipborne lidar uses a preset wavelength, preset point cloud density, and preset vertical accuracy to scan the three-dimensional structure of the seaweed field, generating three-dimensional biomass point cloud data in a point cloud format that represents the vertical distribution of biomass.

[0006] Furthermore, the specific steps of step S2 are as follows: S21. Pre-build a U-Net network model and train it using a training set constructed using a historical chlorophyll a spatial dataset collected by satellite remote sensing sensors and a historical algae canopy spectral dataset collected by a hyperspectral imager carried by a drone. S22. Spatially align the chlorophyll a spatial dataset collected in real time by satellite remote sensing sensors with the algae canopy spectral data collected in real time by a hyperspectral imager carried by a drone using the trained U-Net network model to generate registered two-dimensional raster data. S23. Construct a 3D convolutional neural network model by sequentially connecting a 3D convolutional layer, a cross-modal attention layer, and a trilinear upsampling layer. Train the model using historically aligned 2D raster data and historically rasterized 3D biomass point cloud data. S24. After rasterizing the real-time three-dimensional biomass point cloud data, the real-time registered two-dimensional raster data is input into the trained 3D convolutional neural network model, and the three-dimensional biomass model stored in voxels of a set size is output.

[0007] Furthermore, the specific steps of step S3 are as follows: S31. Deploy a water quality sensor array in the target sea area to collect water quality parameters in real time and obtain time series data of water quality parameters, including dissolved oxygen, pH value, nutrients, turbidity, and chlorophyll a; S32. Scan benthic organisms in the middle layer of the target sea area using an underwater robot equipped with a microscopic camera along a spiral path, and output a video stream and key frame images; S33. Deploy eDNA samplers at a set density in the surface layer of the target sea area, collect water samples, perform on-site metagenomic sequencing, and output species composition data; S34. Compare the three-dimensional biomass point cloud data retrieved by the lidar with the species composition data observed by the eDNA sampler. If the deviation is greater than a set amplitude, calculate a calibration coefficient and use the calibration coefficient to calibrate the three-dimensional biomass point cloud data. S35. Correct the chlorophyll a spatial dataset collected by satellite remote sensing using water quality parameters collected by a water quality sensor array.

[0008] Furthermore, the specific steps of step S4 are as follows: S41. The water quality sensor is used as the first edge node, and a lightweight LSTM model is deployed on the first edge node. The water quality parameter time series data is input into the lightweight LSTM model for water quality anomaly detection, and the anomaly probability value is output; S42. Use the underwater robot as the second edge node, deploy an improved YOLOv5 model on the second edge node, and input the video stream of the benthic organism image into the improved YOLOv5 model to identify the benthic organism species.

[0009] Furthermore, the specific steps of step S5 are as follows: S51. Construct algae node features, animal node features, and environmental node features of the graph neural network based on the calibrated 3D biomass model, benthic organism identification results, and water quality parameters; S52. Establish edge relationships in graph neural networks based on the ecological relationships between node features; S53. Constructing the graph convolutional message passing layer of the graph neural network:

[0010] in, is the normalized degree; is the trainable weight matrix of layer l; σ is the ReLU activation function; S54. Construct carbon sink prediction branch and impact factor branch for graph neural network.

[0011] Furthermore, the specific steps of step S6 are as follows: S61. Calculate the moving average and standard deviation of the carbonization prediction value within the sampling period, and set the dynamic threshold value based on the moving average and standard deviation; S62. Compare the real-time carbonization prediction value with the dynamic threshold value, and generate a brown tide warning signal when the real-time carbon sink prediction value is less than the dynamic threshold value; S63. Construct a state space consisting of the drone's position, remaining battery power, and the carbon sink gradient tensor output by the graph neural network, and an action space consisting of the drone's heading angle increment and velocity increment. Then, construct a loss function and perform reinforcement learning on the drone's patrol path. S64. For the identified abnormal target sea area, increase the sampling frequency of the water quality sensor according to the preset amplitude.

[0012] In a second aspect, the present application also provides a multi-integrated seaweed field ecosystem observation system, including: The multi-scale data acquisition module is used to determine the target sea area, obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data through a hyperspectral imager mounted on a drone, and obtain three-dimensional biomass point cloud data through a ship-borne lidar; The multimodal data registration module is used to align the chlorophyll a concentration data of the target sea area with the spectral data of the seaweed canopy through the U-Net network to obtain two-dimensional raster data. The two-dimensional raster data is then fused with the three-dimensional biomass point cloud data through a 3D convolutional neural network to generate a three-dimensional biomass model. A dynamic in situ sampling network construction module is used to deploy water quality sensors in the target sea area to monitor water quality parameters in real time. An underwater robot equipped with a microscopic camera collects benthic video streams and deploys eDNA samplers to collect eDNA metagenomic sequencing data. An in situ validation dataset is generated based on water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. This in situ validation dataset is then used to perform in situ validation and calibration of the three-dimensional biomass model. An adaptive anomaly detection module, which uses a lightweight LSTM model at the edge node to detect anomalies in water quality parameters and identifies benthic species in real time using a modified YOLOv5 model; A multi-dimensional ecological assessment module is used to construct a graph neural network using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges. The graph neural network then outputs carbon sink predictions and key influencing factors. The dynamic early warning decision module is used to generate a brown tide warning signal when the carbon sink prediction value falls below the dynamic threshold. It then optimizes the drone patrol path based on reinforcement learning and automatically increases the sampling frequency of water quality sensors in abnormal target sea areas.

[0013] In a third aspect, an embodiment of the present application further provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, the steps of the multi-fusion seaweed field ecosystem observation method as described in the first aspect are implemented.

[0014] In a fourth aspect, an embodiment of the present application further provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-fusion seaweed field ecosystem observation method as described in the first aspect.

[0015] It can be seen from the above technical solutions that this application has the following advantages: The multi-integrated seaweed field ecosystem observation method, system, equipment and medium provided in this application integrate multi-source data from satellite remote sensing, drone hyperspectral imaging, and ship-borne lidar to comprehensively monitor the seaweed field ecosystem, evaluate biomass, species diversity and water quality, and analyze ecological relationships, predict carbon sink trends and provide early warning of ecological anomalies by constructing a graph neural network. Combined with edge computing and reinforcement learning optimization, it improves monitoring efficiency and accuracy, providing a basis for marine ecological protection, fishery resource management and environmental assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for the description. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 Schematic diagram of the process of the multi-fusion seaweed field ecosystem observation method of the present invention.

[0018] Figure 2 Schematic diagram of the multi-integrated seaweed field ecosystem observation system of the present invention. DETAILED DESCRIPTION

[0019] The various embodiments of the present disclosure will be described more fully below in detail in the specific steps of the multi-fusion seaweed field ecosystem observation method. The present disclosure can have various embodiments, and adjustments and changes can be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present disclosure to the specific embodiments disclosed herein, and instead, the present disclosure should be construed to encompass all adjustments, equivalents, and / or alternatives that fall within the spirit and scope of the various embodiments of the present disclosure.

[0020] For example, with the expansion of the marine industry and the intensification of climate change, the marine ecosystem is under tremendous pressure. Seaweed beds, as a key part of the marine ecosystem, play an important role in protecting biodiversity, supporting fishery resources, and purifying seawater. Accurate and comprehensive monitoring of seaweed bed ecology has become extremely critical. Traditional monitoring methods have obvious flaws: although satellite remote sensing can capture the distribution of chlorophyll over a large area, it cannot finely analyze the structure of algal communities; conventional water quality monitoring equipment can only conduct fixed-point monitoring of local waters, making it difficult to present a complete picture of the seaweed bed. Traditional monitoring is mostly conducted on a periodic basis and lacks real-time performance. It is difficult to capture sudden ecological anomalies, such as brown tides, and it is impossible to issue early warnings and respond accurately. In addition, existing methods often focus on a single ecological factor and fail to fully integrate multiple factors such as large seaweed communities, nematodes, large benthic animals, water quality, and sedimentary environment. This makes it difficult to comprehensively assess the health status and development trends of seaweed bed ecosystems, nor can it deeply analyze the long-term impact of changes in marine ecological environment quality on seaweed beds.

[0021] In response to the above problems, this embodiment provides a multi-integrated seaweed field ecosystem observation method, which realizes all-round, high-precision, real-time dynamic monitoring and evaluation of seaweed field ecosystems, improves monitoring efficiency and accuracy, provides support for marine ecological environment protection and seaweed field resource management, and can be widely used in marine ecological research, fishery resource management, environmental protection and other fields.

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] See also Figure 1 FIG. 1 is a flow chart of a multi-fusion seaweed field ecosystem observation method according to a specific embodiment, wherein the method comprises the following steps: S1. Identify the target sea area and obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data using a hyperspectral imager mounted on an unmanned aerial vehicle, and obtain three-dimensional biomass point cloud data using a ship-borne lidar. It should be noted that by using multiple means such as satellite remote sensing, drone-mounted hyperspectral imagers, and ship-borne lidar to identify target sea areas and obtain different types of data, it is possible to collect comprehensive information on seaweed beds from macro to micro, from two-dimensional to three-dimensional, covering multiple key elements of the seaweed bed ecosystem. This provides a data foundation for subsequent data fusion and analysis, makes up for the shortcomings of a single monitoring method, and improves the comprehensiveness and accuracy of monitoring. S2. Align the chlorophyll a concentration data of the target sea area with the algae canopy spectral data using a U-Net network to generate 2D raster data. This 2D raster data is then fused with the 3D biomass point cloud data using a 3D convolutional neural network to generate a 3D biomass model. It should be noted that by accurately matching and integrating different data sources at the spatial and feature levels, the advantages and potential information of each data are fully explored, so that the generated three-dimensional biological model not only has spatial structural information, but also reflects the spectral characteristics of seaweed. This provides support for the accurate assessment of seaweed field biomass distribution and ecosystem structure, and improves the accuracy and reliability of the model. S3. Deploy water quality sensors in the target sea area to monitor water quality parameters in real time. Use an underwater robot equipped with a microscopic camera to collect benthic video streams. Deploy an eDNA sampler to collect eDNA metagenomic sequencing data. Generate an in situ validation dataset based on the water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. Use this in situ validation dataset to perform in situ validation and calibration of the 3D biomass model. It should be noted that by obtaining in situ data from multiple levels, including the environment, individual organisms, and genetic material, the three-dimensional biomass model was verified and calibrated in situ, reducing model errors and improving the consistency between the model and the actual seaweed field conditions. At the same time, real-time water quality monitoring data facilitates timely monitoring of environmental changes in seaweed fields, providing accurate basic data for subsequent ecological assessments and early warnings. S4. Edge nodes use a lightweight LSTM model to detect anomalies in water quality parameters and identify benthic species in real time using an improved YOLOv5 model. It should be noted that through rapid and accurate detection of abnormal conditions in the seaweed field ecosystem and real-time monitoring of the composition of the biological community, combined with the lightweight model to ensure efficient operation on edge devices, computing resource consumption and data transmission delays are reduced, and the real-time nature and response speed of monitoring are improved. Problems can be discovered and alarms can be issued in the early stages of abnormalities, providing time for subsequent early warning and decision-making. S5. Construct a graph neural network using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges, and output carbon sink predictions and key influencing factors through the graph neural network. It should be noted that starting from the overall ecosystem level, the study comprehensively considers multiple biological and environmental factors and their interrelationships, achieving a multi-dimensional and in-depth assessment of the seaweed ecosystem. It can accurately predict the trend of carbon sink changes and identify key influencing factors, providing a basis for formulating targeted ecological protection and restoration measures. S6. When the predicted carbon sink value falls below a dynamic threshold, a brown tide warning signal is generated. Reinforcement learning is then used to optimize the drone's patrol path and automatically increase the sampling frequency of water quality sensors in the abnormal target sea area. It should be noted that timely dynamic responses based on real-time monitoring data and ecosystem assessment results have achieved the coordinated optimization of monitoring, early warning and decision-making, and improved the efficiency and accuracy of responding to abnormal ecological events in seaweed fields. By reasonably adjusting the allocation of monitoring resources and strengthening monitoring of abnormal areas, it provides guarantees for the implementation of effective ecological intervention measures, helps to reduce ecological disaster losses and maintain the stability and health of the seaweed field ecosystem.

[0024] This embodiment implements the entire process from data collection, fusion, in-situ verification, anomaly detection, ecological assessment to dynamic early warning, realizing the coordinated use of multi-source data and comprehensive monitoring of the ecosystem, providing an overall framework and foundation for the implementation of subsequent specific steps, and improving the accuracy, efficiency and intelligence level of seaweed field monitoring.

[0025] Furthermore, as a refinement and expansion of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process in this embodiment, another multi-fusion seaweed field ecosystem observation method is provided, which includes the following steps: S1. Determine the target sea area, obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data using a hyperspectral imager mounted on an unmanned aerial vehicle, and obtain three-dimensional biomass point cloud data using a ship-borne lidar. The specific steps of step S1 are as follows: S11. Obtaining a chlorophyll a concentration distribution map of the target sea area by using a satellite remote sensing sensor according to a preset first spatial resolution to generate a chlorophyll a spatial dataset; For example, the satellite remote sensing sensor is a Sentinel-2 MSI sensor; the preset first spatial resolution is 10m; the chlorophyll a spatial dataset is in GeoTIFF format; S12. Collecting spectral data of the seaweed canopy using a hyperspectral imager carried by the drone according to a preset second spatial resolution, a preset flight altitude, a set number of bands, and a set spectral range; For example, the second spatial resolution is 0.5 m, the preset flight altitude is 50 m, the number of bands is greater than or equal to 200, and the seaweed canopy spectral data is a hyperspectral cube in the 400-1000 nm band; S13. The ship-borne laser radar scans the three-dimensional structure of the seaweed field using a preset wavelength, preset point cloud density, and preset vertical accuracy, generating three-dimensional biomass point cloud data in a point cloud format representing the vertical distribution of biomass; S2. Align the chlorophyll a concentration data of the target sea area with the algae canopy spectral data using a U-Net network to obtain two-dimensional raster data. Then, fuse the two-dimensional raster data with the three-dimensional biomass point cloud data using a 3D convolutional neural network to generate a three-dimensional biomass model. The specific steps of step S2 are as follows: S21. Pre-build a U-Net network model and train it using a training set constructed using a historical chlorophyll a spatial dataset collected by satellite remote sensing sensors and a historical algae canopy spectral dataset collected by a hyperspectral imager carried by a drone. Specifically, a U-Net network model with a skip connection structure is pre-built, and the loss function is set to the cross entropy with a spectral weight factor:

[0026]

[0027] in, is the loss function value, which is used to measure the difference between the model prediction result and the true value; N is the total number of samples, that is, the number of pixels involved in the calculation; is the weight factor, reflecting the importance of the i-th pixel; is the maximum NDVI value of all pixels and is used to normalize the weight; is the true label, indicating the category of the i-th pixel (such as seaweed / non-seaweed); is the probability that the i-th pixel belongs to the seaweed category predicted by the model; is the normalized vegetation index of the i-th pixel, and the calculation formula is:

[0028] in, is the reflectivity in the near-infrared band, is the reflectivity of red light band; The training set is constructed by combining the chlorophyll a spatial dataset collected by satellite remote sensing sensors and the algae canopy spectral data collected by the hyperspectral imager carried by the UAV. The training is performed using the Adam optimizer until the loss function converges or the maximum number of iterations is met. S22. Spatially align the chlorophyll a spatial dataset collected in real time by satellite remote sensing sensors with the algae canopy spectral data collected in real time by a hyperspectral imager carried by a drone using the trained U-Net network model to generate registered two-dimensional raster data. S23. Construct a 3D convolutional neural network model by sequentially connecting a 3D convolutional layer, a cross-modal attention layer, and a trilinear upsampling layer. Train the model using historically aligned 2D raster data and historically rasterized 3D biomass point cloud data. Specifically, a 3D convolutional neural network model is constructed, which includes: 3D convolution layer, with 4 input channels (corresponding to the RGB color + intensity values ​​of the lidar point cloud), 64 output channels, and a convolution kernel size of 3×3×3, used to extract the spatial structural features of the point cloud data; The cross-modal attention layer receives the output features of the 3D convolution layer and the spectral features of the hyperspectral data (270 bands), dynamically fuses multi-source information through the attention mechanism, and outputs a dimension of 256; Three linear upsampling layers with a magnification of 2 are used to restore the spatial resolution and generate a fused feature map aligned with the 2D raster data; The loss function of the 3D convolutional neural network model is defined as the weighted mean square error:

[0029] in, is the voxel-level weight, which is determined by the density of the lidar point cloud and hyperspectral NDVI values Jointly determine: ; It is the ground-based biomass value, obtained through eDNA sampling and underwater robot verification; During training, the AdamW optimizer is used to perform training of set batches according to the learning rate, and training of preset epochs is performed in each batch; For example, the batch size is 8 and the training round is 200 epochs; S24. After rasterizing the real-time 3D biomass point cloud data, the real-time registered 2D raster data is input into the trained 3D convolutional neural network model, and the 3D biomass model stored in voxels of a set size is output; Specifically, the rasterization processing formula is as follows:

[0030] in, is the voxel resolution (spatial discretization granularity); is the mean value of the point cloud in the x direction, The mean value of the point cloud in the y direction, used for local coordinate system alignment; , is a height weighted function that suppresses seabed noise ( is the vertical coordinate of the point); is the indicator function, which determines whether the point falls into the target voxel; For example, the size voxel is set to 0.5m×0.5m×0.2m voxel; Each voxel contains: Chlorophyll concentration (Chl, unit: μg / L); Biomass dry weight (Biomass, unit: g / m³); S3. Deploy water quality sensors in the target sea area to monitor water quality parameters in real time. Use an underwater robot equipped with a microscopic camera to collect benthic video streams. Deploy an eDNA sampler to collect eDNA metagenomic sequencing data. Generate an in situ validation dataset based on the water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. Use the in situ validation dataset to perform in situ validation and calibration of the three-dimensional biomass model. The specific steps of step S3 are as follows: S31. Deploy a water quality sensor array in the target sea area to collect water quality parameters in real time and obtain time series data of water quality parameters, including dissolved oxygen, pH value, nutrients, turbidity, and chlorophyll a; the nutrients include 、 、 ; For example, the time series data of water quality parameters adopt minute-level data; S32. Scan benthic organisms in the middle layer of the target sea area using an underwater robot equipped with a microscopic camera along a spiral path, and output a video stream and key frame images; Specifically, the middle layer of the target sea area is 2-5m; S33. Deploy eDNA samplers at a set density in the surface layer of the target sea area, collect water samples, perform on-site metagenomic sequencing, and output species composition data; Specifically, the surface layer of the target sea area is 0-2m; Exemplarily, species composition data are in FASTQ format; S34. Compare the three-dimensional biomass point cloud data retrieved by the lidar with the species composition data observed by the eDNA sampler. If the deviation is greater than a set amplitude, calculate a calibration coefficient and use the calibration coefficient to calibrate the three-dimensional biomass point cloud data. For example, the setting range is 15%, and the calibration coefficient is calculated by the following formula:

[0031] in, is the calibration coefficient, is the biomass value inverted by lidar, is an on-site observation value (such as an eDNA estimate or an underwater robot measurement value), is the relative error term; S35. Correcting the chlorophyll a spatial dataset collected by satellite remote sensing using water quality parameters collected by the water quality sensor array;

[0032] in, is the calibrated chlorophyll a concentration; Chlorophyll a concentration retrieved from satellite remote sensing; is the dissolved oxygen concentration, is the turbidity of water; S4. The edge node uses a lightweight LSTM model to detect anomalies in water quality parameters and uses an improved YOLOv5 model to identify benthic species in real time. The specific steps of step S4 are as follows: S41. The water quality sensor is used as the first edge node, and a lightweight LSTM model is deployed on the first edge node. The water quality parameter time series data is input into the lightweight LSTM model for water quality anomaly detection, and the anomaly probability value is output; Specifically, the lightweight LSTM model architecture includes: The input layer is 7-dimensional water quality parameters (dissolved oxygen, pH, 、 、 , turbidity, chlorophyll a), time window length T = 10; The hidden layer is a single LSTM layer with 8 units, followed by a temporal attention layer with a compression rate of 0.5:

[0033] in, is the temporal attention weight; is a learnable weight matrix; is the attention context vector, which is used to evaluate the importance of each time step; It is a hyperbolic tangent activation function that introduces nonlinearity to capture complex temporal patterns; is an exponential function that maps attention scores to probability distributions; The output layer is a 2D fully connected layer (anomaly probability + parameter importance score), with a parameter size of ≤1KB; The training uses the Focal loss function:

[0034] in, , an adjustment factor to control the weight of difficult and easy samples; Is the probability model prediction probability, for positive samples , negative samples ; S42. The underwater robot is used as the second edge node, and an improved YOLOv5 model is deployed on the second edge node. The video stream of the benthic image is input into the improved YOLOv5 model for benthic species identification; Specifically, the original YOLOv5 model has been improved as follows: First is the optimization of Backbone: Replace the C3 module with the Ghost module to reduce the number of parameters; It should be noted that by replacing the C3 module, while maintaining the feature expression capability, the computational complexity is reduced to adapt to edge devices; Add eDNA prior guidance layer, weight matrix as follows:

[0035]

[0036] Where C is the number of backbone network feature map channels (e.g., 512), and S is the number of species categories identified by eDNA sequencing (e.g., seaweed, fish, invertebrates). is the feature vector obtained from the global average pooling result of the i-th channel; Indicates the species whose eDNA was sequenced Mapped to a d-dimensional vector; Indicates the similarity between the calculated feature and the species; It should be noted that by adding an eDNA prior guidance layer, the eDNA species abundance information was injected into the backbone network to enhance the characteristic response to key species (e.g., toxic algae); Then the dynamic category weight is introduced to improve the detection head, as follows:

[0037] in, is the dynamic class weight, which is used to adjust the loss contribution of the detection head to species c; is a hyperparameter that controls the strength of the eDNA prior information; is the relative abundance of species c in eDNA sequencing (e.g., a certain toxic algae accounts for 0.8); is the maximum eDNA abundance value among all species categories and is used for normalization; It should be noted that the detection head is improved through dynamic category weighting to enable the model to focus on ecologically key species (such as invasive species marked by eDNA), thereby improving the accuracy of small target detection; Finally, SIoU loss is used instead of CIoU, angle loss as follows:

[0038] in, is the prediction box rotation angle, is the real frame rotation angle, Overall, the penalty angle deviation is →4 π When , Λ→1, it is the maximum penalty; It should be noted that compared to CIoU, SIoU explicitly models angle differences, which accelerates convergence and improves positioning accuracy; S5. Using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges, a graph neural network is constructed. The graph neural network outputs carbon sink prediction values ​​and key influencing factors. The specific steps of step S5 are as follows: S51. Construct algae node features, animal node features, and environmental node features of the graph neural network based on the calibrated 3D biomass model, benthic organism identification results, and water quality parameters; Specifically, algae node characteristics include three-dimensional biomass mean and Shannon diversity index; The three-dimensional biomass mean was calculated using a three-dimensional biomass model to reflect the spatial distribution density of algae; Shannon Diversity Index:

[0039]

[0040] Where S represents the number of algal species identified by eDNA sequencing. represents the relative abundance of species i, is the number of sequencing reads; Animal node features include species number, average body length, and animal entropy; The number of species is the number of animal groups (e.g., fish, invertebrates) detected by the improved YOLOv5 model; Average body length was measured using a video tracking algorithm; The kinetic entropy is calculated as follows:

[0041] in, is the velocity at time t (e.g. estimated by the SLAM algorithm), is the velocity probability distribution (e.g. obtained by kernel density estimation); Environmental node characteristics are the Z-score standardized values ​​of water quality parameters; S52. Establish edge relationships in graph neural networks based on the ecological relationships between node features; Specifically, the edge relations include algae-animal edge weights and environment-organism edge weights; Algae-animal edge weight:

[0042] in, The number of co-occurrences of species i (algae) and j (animals) (e.g., statistics from association rule mining); is a normalization term to avoid high-frequency species dominating the weight; Environment-Organism Edge Weight:

[0043] in Environmental parameters (such as DO) and biomass Pearson correlation coefficient; S53. Constructing the graph convolutional message passing layer of the graph neural network:

[0044] in, is the normalized degree; is the trainable weight matrix of the lth layer (initialized to N(0,0.1)); σ is the ReLU activation function (introducing nonlinearity, σ(x)=max(0,x)); S54. Construct carbon sink prediction branch and impact factor branch for graph neural network; Specifically, the carbon sink prediction branch:

[0045] in, It is the aggregation characteristic of algae nodes; It is the aggregation feature of the environment node; is the weight of the fully connected layer (d=64 is the hidden layer dimension); ∈[0,1], is the predicted value of normalized carbon flux; Impact Factor Branch:

[0046] in, is the characteristic map of the kth environmental parameter (such as the convolution response of pH value); Z is the normalization constant; It should be noted that the impact factor branch Quantify the gradient contribution of environmental parameters to carbon sink predictions and identify key drivers (e.g., the effect of DO on respiration); S6. When the predicted carbon sink value falls below the dynamic threshold, a brown tide warning signal is generated. The patrol path of the drone is then optimized based on reinforcement learning, and the sampling frequency of the water quality sensor in the abnormal target sea area is automatically increased. The specific steps of step S6 are as follows: S61. Calculate the moving average and standard deviation of the carbonization prediction value within the sampling period, and set the dynamic threshold value based on the moving average and standard deviation; S62. Compare the real-time carbonization prediction value with the dynamic threshold value, and generate a brown tide warning signal when the real-time carbon sink prediction value is less than the dynamic threshold value; S63. Construct a state space consisting of the drone's position, remaining battery power, and the carbon sink gradient tensor output by the graph neural network, and an action space consisting of the drone's heading angle increment and velocity increment. Then, construct a loss function and perform reinforcement learning on the drone's patrol path. Specifically, the loss function is as follows:

[0047] Where IoU(⋅) is the intersection-over-union ratio of the scanned area to the high-risk area (e.g., when IoU=0.7, 70% of the risk area is covered); ΔE t : Energy consumption increment (estimated by the motor model P=Fv, where F is thrust and v is speed); λ1=0.7,λ2=0.3 are weight coefficients (determined by multi-objective optimization); S64. For the identified abnormal target sea area, increase the sampling frequency of the water quality sensor according to the preset amplitude.

[0048] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0049] like Figure 2 As shown, the following is an embodiment of the multi-fusion seaweed field ecosystem observation system provided by the embodiment of the present disclosure. This system and the multi-fusion seaweed field ecosystem observation method of the above-mentioned embodiments belong to the same inventive concept. For details not fully described in the embodiment of the multi-fusion seaweed field ecosystem observation system, please refer to the embodiment of the above-mentioned multi-fusion seaweed field ecosystem observation method.

[0050] The system includes: The multi-scale data acquisition module is used to determine the target sea area, obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data through a hyperspectral imager mounted on a drone, and obtain three-dimensional biomass point cloud data through a ship-borne lidar; The multimodal data registration module is used to align the chlorophyll a concentration data of the target sea area with the spectral data of the seaweed canopy through the U-Net network to obtain two-dimensional raster data. The two-dimensional raster data is then fused with the three-dimensional biomass point cloud data through a 3D convolutional neural network to generate a three-dimensional biomass model. A dynamic in situ sampling network construction module is used to deploy water quality sensors in the target sea area to monitor water quality parameters in real time. An underwater robot equipped with a microscopic camera collects benthic video streams and deploys eDNA samplers to collect eDNA metagenomic sequencing data. An in situ validation dataset is generated based on water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. This in situ validation dataset is then used to perform in situ validation and calibration of the three-dimensional biomass model. An adaptive anomaly detection module, which uses a lightweight LSTM model at the edge node to detect anomalies in water quality parameters and identifies benthic species in real time using a modified YOLOv5 model; A multi-dimensional ecological assessment module is used to construct a graph neural network using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges. The graph neural network then outputs carbon sink predictions and key influencing factors. The dynamic early warning decision module is used to generate a brown tide warning signal when the carbon sink prediction value falls below the dynamic threshold. It then optimizes the drone patrol path based on reinforcement learning and automatically increases the sampling frequency of water quality sensors in abnormal target sea areas.

[0051] This embodiment achieves the coordinated use of multi-source data and comprehensive monitoring of the ecosystem through the interactive collaboration of a multi-scale data acquisition module, a multimodal data registration module, a dynamic in-situ sampling network construction module, an adaptive anomaly detection module, a multi-dimensional ecological assessment module, and a dynamic early warning decision-making module. It provides an overall framework and foundation for the implementation of subsequent specific steps, and improves the accuracy, efficiency, and intelligence level of seaweed field monitoring.

[0052] The multi-fusion seaweed field ecosystem observation method provided in the embodiment of the present application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiment of the present invention does not constitute a limitation on the electronic device, and the electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. In the embodiment of the present invention, the electronic device includes but is not limited to a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or required herein.

[0053] The electronic device may include a processor, an external memory interface, an internal memory, a universal serial bus (USB) interface, a charging management module, a power management module, a battery, a wireless communication module, an audio module, a speaker, a microphone, a sensor module, a button, a camera, a display, and a SIM card interface, etc.

[0054] It is understood that the structures illustrated in the embodiments of the present application do not constitute specific limitations on the electronic device. In other embodiments of the present application, the electronic device may include more or fewer components than shown, or combine or separate certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0055] A processor may include one or more processing units, such as a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0056] The processor can be the nerve center and command center of the electronic device. The controller can generate operation control signals based on the instruction opcode and timing signal to complete the control of instruction fetching and execution.

[0057] The processor may also include a memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or is reusing. If the processor needs to use the instruction or data again, it can directly call it from the memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0058] The above-mentioned electronic equipment realizes the determination of the target sea area of ​​the multi-fusion seaweed field ecosystem observation method of the present application, obtains chlorophyll a concentration data through satellite remote sensing, obtains seaweed canopy spectral data through a drone equipped with a hyperspectral imager, and obtains three-dimensional biomass point cloud data through a ship-borne lidar; aligns the chlorophyll a concentration data of the target sea area with the seaweed canopy spectral data through a U-Net network to obtain two-dimensional raster data, and then fuses the two-dimensional raster data with the three-dimensional biomass point cloud data through a 3D convolutional neural network to generate a three-dimensional biomass model; deploys water quality sensors in the target sea area to monitor water quality parameters in real time, collects benthic video streams through an underwater robot equipped with a microscope camera, deploys eDNA samplers to collect eDNA metagenome sequencing data, generates an in situ verification data set based on water quality parameters, benthic video streams, and eDNA metagenome sequencing data, and uses The three-dimensional biomass model was verified and calibrated in situ using an in-situ verification dataset. A lightweight LSTM model was used at the edge node to detect anomalies in water quality parameters, and benthic species were identified in real time using an improved YOLOv5 model. A graph neural network was constructed using the calibrated three-dimensional biomass model, benthic identification results, and water quality parameters as nodes, and the ecological relationships between species as edges. The graph neural network then outputted carbon sink prediction values ​​and key influencing factors. When the carbon sink prediction value was lower than the dynamic threshold, a brown tide warning signal was generated. The drone patrol path was then optimized based on reinforcement learning, and the sampling frequency of water quality sensors in abnormal target sea areas was automatically increased. This technical solution achieved the coordinated use of multi-source data and comprehensive monitoring of the ecosystem, providing an overall framework and foundation for the implementation of subsequent specific steps, and improving the accuracy, efficiency, and intelligence level of seaweed field monitoring.

[0059] The storage medium provided in this application stores a program product that can realize a multi-fusion seaweed field ecosystem observation method.

[0060] The multi-fusion seaweed field ecosystem observation method includes: determining the target sea area, obtaining chlorophyll a concentration data through satellite remote sensing, obtaining seaweed canopy spectral data through drone-mounted hyperspectral imagers, and obtaining three-dimensional biomass point cloud data through ship-mounted lidar; aligning the chlorophyll a concentration data of the target sea area with the seaweed canopy spectral data through the U-Net network to obtain two-dimensional raster data, and then fusing the two-dimensional raster data with the three-dimensional biomass point cloud data through a 3D convolutional neural network to generate a three-dimensional biomass model; deploying water quality sensors in the target sea area to monitor water quality parameters in real time, collecting benthic video streams through underwater robots equipped with microscopic cameras, and deploying eDNA samplers to collect eDNA metagenomic sequencing data. , eDNA metagenomic sequencing data is used to generate an in situ verification data set, and the in situ verification data set is used to perform in situ verification and calibration of the three-dimensional biomass model; the edge node uses the calibrated three-dimensional biomass model and water quality parameters to execute a lightweight LSTM model to detect anomalies in water quality parameters, and uses the improved YOLOv5 model to identify benthic species in real time; a graph neural network is constructed with the calibrated three-dimensional biomass model, benthic identification results and water quality parameters as nodes, and the ecological relationship between species as edges, and the graph neural network outputs the carbon sink prediction value and key influencing factors; when the carbon sink prediction value is lower than the dynamic threshold, a brown tide warning signal is generated, and then the drone patrol path is optimized based on reinforcement learning, and the sampling frequency of water quality sensors in the abnormal target sea area is automatically increased.

[0061] In some possible embodiments, the multi-fusion seaweed field ecosystem observation method disclosed herein can be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps of various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.

[0062] The storage medium of the present disclosure can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0063] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-integrated seaweed field ecosystem observation method, characterized in that: The steps include: S1. Identify the target sea area and obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data using a hyperspectral imager mounted on an unmanned aerial vehicle, and obtain three-dimensional biomass point cloud data using a ship-borne lidar. S2. Align the chlorophyll a concentration data of the target sea area with the algae canopy spectral data using a U-Net network to generate 2D raster data. This 2D raster data is then fused with the 3D biomass point cloud data using a 3D convolutional neural network to generate a 3D biomass model. S3. Deploy water quality sensors in the target sea area to monitor water quality parameters in real time. Use an underwater robot equipped with a microscopic camera to collect benthic video streams. Deploy an eDNA sampler to collect eDNA metagenomic sequencing data. Generate an in situ validation dataset based on the water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. Use this in situ validation dataset to perform in situ validation and calibration of the 3D biomass model. S4. Edge nodes use a lightweight LSTM model to detect anomalies in water quality parameters and identify benthic species in real time using an improved YOLOv5 model. S5. Construct a graph neural network using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges, and output carbon sink predictions and key influencing factors through the graph neural network. S6. When the predicted carbon sink value falls below the dynamic threshold, a brown tide warning signal is generated. Reinforcement learning is then used to optimize the drone patrol path and automatically increase the sampling frequency of water quality sensors in abnormal target sea areas.

2. The multi-fusion seaweed field ecosystem observation method according to claim 1, characterized in that: The specific steps of step S1 are as follows: S11. Obtaining a chlorophyll a concentration distribution map of the target sea area by using a satellite remote sensing sensor according to a preset first spatial resolution to generate a chlorophyll a spatial dataset; S12. Collecting spectral data of the seaweed canopy using a hyperspectral imager carried by the drone according to a preset second spatial resolution, a preset flight altitude, a set number of bands, and a set spectral range; S13. The shipborne lidar uses a preset wavelength, preset point cloud density, and preset vertical accuracy to scan the three-dimensional structure of the seaweed field, generating three-dimensional biomass point cloud data in a point cloud format that represents the vertical distribution of biomass.

3. The multi-fusion seaweed field ecosystem observation method according to claim 2, characterized in that: The specific steps of step S2 are as follows: S21. Pre-build a U-Net network model and train it using a training set constructed using a historical chlorophyll a spatial dataset collected by satellite remote sensing sensors and a historical algae canopy spectral dataset collected by a hyperspectral imager carried by a drone. S22. Spatially align the chlorophyll a spatial dataset collected in real time by satellite remote sensing sensors with the algae canopy spectral data collected in real time by a hyperspectral imager carried by a drone using the trained U-Net network model to generate registered two-dimensional raster data. S23. Construct a 3D convolutional neural network model by sequentially connecting a 3D convolutional layer, a cross-modal attention layer, and a trilinear upsampling layer. Train the model using historically aligned 2D raster data and historically rasterized 3D biomass point cloud data. S24. After rasterizing the real-time three-dimensional biomass point cloud data, the real-time registered two-dimensional raster data is input into the trained 3D convolutional neural network model, and the three-dimensional biomass model stored in voxels of a set size is output.

4. The multi-fusion seaweed field ecosystem observation method according to claim 3, characterized in that: The specific steps of step S3 are as follows: S31. Deploy a water quality sensor array in the target sea area to collect water quality parameters in real time and obtain time series data of water quality parameters, including dissolved oxygen, pH value, nutrients, turbidity, and chlorophyll a; S32. Scan benthic organisms in the middle layer of the target sea area using an underwater robot equipped with a microscopic camera along a spiral path, and output a video stream and key frame images; S33. Deploy eDNA samplers at a set density in the surface layer of the target sea area, collect water samples, perform on-site metagenomic sequencing, and output species composition data; S34. Compare the three-dimensional biomass point cloud data retrieved by the lidar with the species composition data observed by the eDNA sampler. If the deviation is greater than a set amplitude, calculate a calibration coefficient and use the calibration coefficient to calibrate the three-dimensional biomass point cloud data. S35. Correct the chlorophyll a spatial dataset collected by satellite remote sensing using water quality parameters collected by a water quality sensor array.

5. The multi-fusion seaweed field ecosystem observation method according to claim 4, characterized in that: The specific steps of step S4 are as follows: S41. The water quality sensor is used as the first edge node, and a lightweight LSTM model is deployed on the first edge node. The water quality parameter time series data is input into the lightweight LSTM model for water quality anomaly detection, and the anomaly probability value is output; S42. Use the underwater robot as the second edge node, deploy an improved YOLOv5 model on the second edge node, and input the video stream of the benthic organism image into the improved YOLOv5 model to identify the benthic organism species.

6. The multi-fusion seaweed field ecosystem observation method according to claim 5, characterized in that: The specific steps of step S5 are as follows: S51. Construct algae node features, animal node features, and environmental node features of the graph neural network based on the calibrated 3D biomass model, benthic organism identification results, and water quality parameters; S52. Establish edge relationships in graph neural networks based on the ecological relationships between node features; S53. Constructing the graph convolutional message passing layer of the graph neural network: in, is the normalized degree; is the trainable weight matrix of layer l; σ is the ReLU activation function; S54. Construct carbon sink prediction branch and impact factor branch for graph neural network.

7. The multi-fusion seaweed field ecosystem observation method according to claim 6, characterized in that: The specific steps of step S6 are as follows: S61. Calculate the moving average and standard deviation of the carbonization prediction value within the sampling period, and set the dynamic threshold value based on the moving average and standard deviation; S62. Compare the real-time carbonization prediction value with the dynamic threshold value, and generate a brown tide warning signal when the real-time carbon sink prediction value is less than the dynamic threshold value; S63. Construct a state space consisting of the drone's position, remaining battery power, and the carbon sink gradient tensor output by the graph neural network, and an action space consisting of the drone's heading angle increment and velocity increment. Then, construct a loss function and perform reinforcement learning on the drone's patrol path. S64. For the identified abnormal target sea area, increase the sampling frequency of the water quality sensor according to the preset amplitude.

8. A multi-integrated seaweed field ecosystem observation system, characterized in that: include: The multi-scale data acquisition module is used to determine the target sea area, obtain chlorophyll a concentration data through satellite remote sensing, obtain seaweed canopy spectral data through a hyperspectral imager mounted on a drone, and obtain three-dimensional biomass point cloud data through a ship-borne lidar; The multimodal data registration module is used to align the chlorophyll a concentration data of the target sea area with the spectral data of the seaweed canopy through the U-Net network to obtain two-dimensional raster data. The two-dimensional raster data is then fused with the three-dimensional biomass point cloud data through a 3D convolutional neural network to generate a three-dimensional biomass model. A dynamic in situ sampling network construction module is used to deploy water quality sensors in the target sea area to monitor water quality parameters in real time. An underwater robot equipped with a microscopic camera collects benthic video streams and deploys eDNA samplers to collect eDNA metagenomic sequencing data. An in situ validation dataset is generated based on water quality parameters, benthic video streams, and eDNA metagenomic sequencing data. This in situ validation dataset is then used to perform in situ validation and calibration of the three-dimensional biomass model. An adaptive anomaly detection module, which uses a lightweight LSTM model at the edge node to detect anomalies in water quality parameters and identifies benthic species in real time using a modified YOLOv5 model; A multi-dimensional ecological assessment module is used to construct a graph neural network using the calibrated three-dimensional biomass model, benthic organism identification results, and water quality parameters as nodes and interspecies ecological relationships as edges. The graph neural network then outputs carbon sink predictions and key influencing factors. The dynamic early warning decision module is used to generate a brown tide warning signal when the carbon sink prediction value falls below the dynamic threshold. It then optimizes the drone patrol path based on reinforcement learning and automatically increases the sampling frequency of water quality sensors in abnormal target sea areas.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the multi-fusion seaweed field ecosystem observation method as claimed in any one of claims 1 to 7 are implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-fusion seaweed field ecosystem observation method according to any one of claims 1 to 7 are implemented.

Images