A marine ranch environment monitoring method and system

Abstract

The present invention discloses a method and system for monitoring an ocean ranch environment, comprising: acquiring environmental monitoring data, the environmental monitoring data including remote sensing monitoring data, underwater drone data, and sensor data; performing feature extraction on the remote sensing monitoring data based on a preset deep residual network to obtain remote sensing features; performing feature extraction on the underwater drone data based on a preset Mask R-CNN network to obtain drone features; performing feature extraction on the sensor data based on a preset recurrent neural network to obtain sensor features; acquiring environmental condition data, adjusting feature weights of remote sensing features, drone features, and sensor features based on a preset conditional discriminant weight table, a preset benchmark weight, and the environmental condition data; performing weighted fusion of the remote sensing features, drone features, and sensor features based on the feature weights to generate fused features; constructing an environmental monitoring model based on a decision tree algorithm, inputting the fused features into the environmental monitoring model, and obtaining environmental monitoring results.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and in particular to a method and system for monitoring the environment of a marine ranch. Background Art

[0002] Marine ranch environmental monitoring technology is evolving towards intelligent, systematic, and multi-source integration. With the increasing intensity of marine resource development and the increasing frequency and magnitude of environmental changes, environmental monitoring is becoming increasingly important in ensuring the ecological security of marine ranches, improving aquaculture efficiency, and providing early warning of disaster risks. Scientific, real-time environmental monitoring not only helps to dynamically monitor key parameters such as water quality, temperature, salinity, and dissolved oxygen, but also provides data support for ecological assessments and management decisions.

[0003] Existing environmental monitoring methods for marine ranches primarily include remote sensing, ocean buoys, sensor networks, underwater robots, and acoustic equipment. Remote sensing technology is suitable for monitoring large areas of the ocean, but its spatial and temporal resolution is limited. Buoys and sensor networks offer continuous data collection capabilities, but their coverage is limited and deployment and maintenance costs are high. Underwater robots, while maneuverable and capable of diverse detection capabilities, are significantly limited by their endurance and real-time performance. Furthermore, existing monitoring methods lack consideration for multi-source data fusion, automatic identification and prediction of abnormal events, and stable operation in inclement weather, leaving them vulnerable. Summary of the Invention

[0004] The present invention provides a method and system for monitoring the environment of an ocean ranch, so as to improve the stability and accuracy of monitoring the environment of an ocean ranch.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method for monitoring the environment of a marine ranch, comprising:

[0006] Acquiring environmental monitoring data, including remote sensing monitoring data, underwater drone data, and sensor data;

[0007] Performing feature extraction on the remote sensing monitoring data based on a preset deep residual network to obtain remote sensing features;

[0008] Perform feature extraction on the underwater drone data based on a preset Mask R-CNN network to obtain drone features;

[0009] Performing feature extraction on the sensor data based on a preset recurrent neural network to obtain sensing features;

[0010] Acquiring environmental condition data, and adjusting feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data;

[0011] Performing weighted fusion on the remote sensing feature, the UAV feature, and the sensor feature based on the feature weight to generate a fusion feature;

[0012] An environmental monitoring model is constructed based on a decision tree algorithm, and the fusion features are input into the environmental monitoring model to obtain environmental monitoring results.

[0013] This method uses remote sensing data to cover the macroscopic ocean environment. UAVs capture local biological details, such as algae, and sensors provide continuous time-series parameters. This overcomes the limitations of a single data source and avoids insufficient response to local or dynamic environments. Different neural networks are then used to extract features for each type of data, dynamically adjusting feature weights based on environmental and weather conditions. This prevents inaccuracies in certain data measurements caused by these conditions, enhancing the ability of environmental monitoring to adapt to changing circumstances. The use of decision tree models for environmental monitoring improves the model's interpretability and decision-making efficiency, thereby increasing the stability and accuracy of marine ranch environmental monitoring.

[0014] Furthermore, the acquiring of environmental condition data and adjusting the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data include:

[0015] Acquiring the environmental condition data, wherein the environmental condition data includes weather data and environmental data;

[0016] Determining the importance scores of the remote sensing feature, the drone feature, and the sensor feature based on the weather data, the environmental data, and the conditional discriminant weight table; the conditional discriminant weight table includes a mapping relationship between the weather data and the importance score of each feature;

[0017] The feature weights of the remote sensing feature, the drone feature, and the sensor feature are adjusted based on the importance score and a preset reference weight.

[0018] The present invention improves the scientificity and rationality of feature weight adjustment by introducing environmental condition data. The conditional discriminant weight table is used to achieve dynamic adjustment of feature weights, thereby enhancing the flexibility of the system.

[0019] Furthermore, the adjusting the feature weights of the remote sensing feature, the drone feature, and the sensor feature based on the importance score and the preset reference weight includes:

[0020] Obtaining a reference weight for each feature, mapping each reference weight to a two-dimensional plane, constructing a reference triangle with the reference weights as vertices, and calculating the centroid of the reference triangle;

[0021] For each vertex of the reference triangle, calculate a unit vector whose center points to the vertex, and translate the vertex of the reference triangle along the direction of the unit vector according to the importance score to generate a deformed triangle;

[0022] Calculating a centroid offset vector of the deformed triangle, and mapping the centroid offset vector to a direction vector of each vertex to generate a regression factor for each feature;

[0023] Based on the regression factors of each feature, the benchmark weight of each feature is incrementally regressed to adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature.

[0024] The present invention intuitively reflects the balance or imbalance of the three types of feature importance scores through geometric offset, and directly drives weight adjustment based on the center of gravity offset without setting complex thresholds or empirical rules.

[0025] Furthermore, the remote sensing monitoring data is subjected to feature extraction based on a preset deep residual network to obtain remote sensing features, including:

[0026] The deep residual network includes a plurality of residual units, each of which is connected by a jump connection;

[0027] Performing feature extraction on the remote sensing monitoring data based on each residual unit to obtain a feature vector;

[0028] Each feature vector is fused based on skip connection to obtain remote sensing features.

[0029] This paper uses skip connections in residual networks to alleviate the vanishing gradient problem in deep networks, improving the depth and accuracy of feature extraction. Furthermore, by fusing multiple feature vectors, it enhances the expressiveness of remote sensing features and improves the accuracy of environmental monitoring.

[0030] Furthermore, the feature extraction of the underwater drone data based on the preset Mask R-CNN network to obtain drone features includes:

[0031] The Mask R-CNN network includes a backbone network and a region proposal network;

[0032] Perform feature extraction on the underwater drone data based on the backbone network to obtain a multi-scale feature map set;

[0033] Acquire a candidate region set based on the region proposal network and the multi-scale feature map set; the candidate region set includes a plurality of candidate regions and coordinate information of each candidate region;

[0034] Feature alignment processing and feature fusion processing are performed based on the candidate region set and the multi-scale feature map set to obtain drone features.

[0035] This paper improves the detection and recognition capabilities of underwater targets through the multi-scale feature extraction and region proposal mechanism of the Mask R-CNN network. The accuracy and robustness of drone features are enhanced through feature alignment and fusion processing.

[0036] Furthermore, the feature extraction of the sensor data based on the preset recurrent neural network to obtain the sensing features includes:

[0037] The recurrent neural network includes several layers of LSTM units;

[0038] Acquiring sensor data, including temperature data, salinity data, dissolved oxygen data, pH data, conductivity data, and turbidity data;

[0039] constructing a first learning projection matrix based on the temperature data, salinity data, and conductivity data;

[0040] Constructing a second learning projection matrix based on the dissolved oxygen data and the pH data;

[0041] constructing a third learning projection matrix based on the turbidity data;

[0042] Dynamically scaling the first learning projection matrix, the second learning projection matrix, and the third learning projection matrix to achieve cross-parameter interaction and generate a fusion vector;

[0043] Feature extraction is performed on the fusion vector based on the recurrent neural network to obtain sensing features.

[0044] This paper uses an LSTM network to effectively capture the time series features in sensor data, improving the timeliness and accuracy of feature extraction. It also enhances the ability to perceive environmental changes through multi-parameter sensor data input.

[0045] Furthermore, after constructing the environmental monitoring model based on the decision tree algorithm, inputting the fusion features into the environmental monitoring model, and obtaining the environmental monitoring results, the method further includes:

[0046] generating an error evaluation value and a credibility value based on the monitoring result, and generating a comprehensive evaluation score based on the error evaluation value and the credibility value;

[0047] The benchmark weight is incrementally adjusted according to the comprehensive evaluation score.

[0048] This invention improves the efficiency and interpretability of environmental monitoring decisions through a decision tree model. The introduction of error assessment and credibility analysis enables self-correction of monitoring results, enhancing the system's adaptability. Incremental adjustments to baseline weights ensure the system's stability and accuracy over long-term operation.

[0049] In a second aspect, the present invention provides a marine ranch environment monitoring system, comprising: a data acquisition module, a feature extraction module, a feature fusion module, and a monitoring module;

[0050] The data acquisition module is used to obtain environmental monitoring data, which includes remote sensing monitoring data, underwater drone data and sensor data;

[0051] The feature extraction module is used to extract features from the remote sensing monitoring data based on a preset deep residual network to obtain remote sensing features; to extract features from the underwater drone data based on a preset Mask R-CNN network to obtain drone features; and to extract features from the sensor data based on a preset recurrent neural network to obtain sensor features;

[0052] The feature fusion module is configured to obtain environmental condition data, adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data, and perform weighted fusion of the remote sensing feature, the UAV feature, and the sensor feature based on the feature weights to generate a fused feature;

[0053] The monitoring module is used to build an environmental monitoring model based on a decision tree algorithm, input the fusion features into the environmental monitoring model, and obtain environmental monitoring results.

[0054] Furthermore, the fusion module is used to obtain environmental condition data, and adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data, including:

[0055] Acquiring the environmental condition data, wherein the environmental condition data includes weather data and environmental data;

[0056] Determining the importance scores of the remote sensing feature, the drone feature, and the sensor feature based on the weather data, the environmental data, and the conditional discriminant weight table; the conditional discriminant weight table includes a mapping relationship between the weather data and the importance score of each feature;

[0057] The feature weights of the remote sensing feature, the drone feature, and the sensor feature are adjusted based on the importance score and a preset reference weight.

[0058] Furthermore, the fusion module is used to adjust the feature weights of the remote sensing feature, the drone feature, and the sensor feature based on the importance score and a preset reference weight, including:

[0059] Obtaining a reference weight for each feature, mapping each reference weight to a two-dimensional plane, constructing a reference triangle with the reference weights as vertices, and calculating the centroid of the reference triangle;

[0060] For each vertex of the reference triangle, calculate a unit vector whose center points to the vertex, and translate the vertex of the reference triangle along the direction of the unit vector according to the importance score to generate a deformed triangle;

[0061] Calculating a centroid offset vector of the deformed triangle, and mapping the centroid offset vector to a direction vector of each vertex to generate a regression factor for each feature;

[0062] Based on the regression factors of each feature, the benchmark weight of each feature is incrementally regressed to adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A schematic flow chart of a method for monitoring an ocean ranch environment provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0065] The terms "first," "second," and the like in the specification, claims, and drawings of this application are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0066] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0067] Example 1

[0068] See also Figure 1 , Figure 1 A schematic flow chart of a method for monitoring an ocean ranch environment provided by an embodiment of the present invention. The method includes steps 101 to 107, as follows:

[0069] Step 101: Acquire environmental monitoring data, including remote sensing monitoring data, underwater drone data, and sensor data;

[0070] In this embodiment, three types of data sources are used to collect marine ranch environmental information.

[0071] In this embodiment, remote sensing monitoring data, based on satellite or airborne remote sensing, can capture parameters such as sea surface temperature, ocean color (e.g., chlorophyll a concentration), sea surface salinity, and sea surface wind speed. Chlorophyll a concentration can reflect phytoplankton biomass and algal blooms. Remote sensing data has a wide coverage area and is beneficial for dynamic macro-environmental monitoring.

[0072] In this embodiment, macroscopic image data of the marine ranch is obtained through remote sensing monitoring data, thereby monitoring the environment of the marine ranch from a macroscopic perspective.

[0073] In this embodiment, underwater drones (AUVs / ROVs) are equipped with high-definition cameras, sonar, and other sensors to capture images and videos of the seafloor environment. Through autonomous cruising or pre-set routes, the drones can observe underwater targets such as fish distribution, corals / seaweed, and artificial structures (such as aquaculture facilities), providing high-resolution visual feature information. Continuous drone operation enables real-time monitoring of ecological dynamics and captures underwater drone data, including continuously generated image or video data.

[0074] In this example, multi-parameter environmental sensors are deployed on aquaculture cages or buoys to collect real-time water quality parameters such as temperature, salinity (conductivity), dissolved oxygen, pH, conductivity, and turbidity. These sensors continuously monitor fine-grained time-series data, supplementing the local environmental information from remote sensing and imaging data to generate sensor data.

[0075] Step 102: extracting features from the remote sensing monitoring data based on a preset deep residual network to obtain remote sensing features;

[0076] In this embodiment,

[0077] The remote sensing monitoring data is subjected to feature extraction based on a preset deep residual network to obtain remote sensing features, including:

[0078] The deep residual network includes a plurality of residual units, each of which is connected by a jump connection;

[0079] Performing feature extraction on the remote sensing monitoring data based on each residual unit to obtain a feature vector;

[0080] Evaluate the similarity between each pair of feature vectors, and construct a feature vector topology map based on the similarity;

[0081] The jump path of each feature is determined based on the feature vector topology map and the maximum weight path algorithm, and each feature vector is fused based on the jump path and jump connection to obtain remote sensing features.

[0082] In this embodiment, by obtaining remote sensing monitoring original image data as remote sensing monitoring data, the remote sensing monitoring original image data is input into a preset deep residual network, wherein the network includes multiple residual units (ResidualBlock), and each residual unit simultaneously learns the identity mapping and residual mapping through skip connection to extract low-level to high-level spatial features respectively.

[0083] In this embodiment, after passing through the residual units of each layer of the deep residual network in sequence, the output of the last layer of the network is the remote sensing feature that integrates multi-scale information.

[0084] In this embodiment, in the initial stage, the deep residual network focuses on extracting low-level, texture-level features, and usually adopts a basic residual block (Basic Block). This block consists of two 3×3 convolutional layers and skip connections. The number of input and output channels remains consistent, which helps to stabilize shallow feature learning.

[0085] In this embodiment, as the network deepens and demands higher feature expression capabilities and parameter efficiency, bottleneck blocks are used instead. Each bottleneck block consists of a 1×1 dimensionality reduction convolution, a 3×3 main convolution, and a 1×1 dimensionality increase convolution, ensuring the trainability of deep networks while controlling the number of parameters.

[0086] In this embodiment, the remote sensing monitoring original image data is output; in the residual unit feature extraction, I is sequentially input into the deep residual network, and the deep residual network includes a plurality of residual units (Residual Block), each residual unit connects the input through a skip connection (skip connection) The output of the consecutive convolutional layers within the unit Add together to learn the residual mapping, thereby extracting multi-level feature vectors { }.

[0087] In this embodiment, for any two eigenvectors, the similarity is calculated using cosine similarity to obtain a similarity matrix S:

[0088]

[0089] In this embodiment, an undirected graph is constructed , node set , corresponding to the eigenvector; where the edge connections and weights in the undirected graph represent (preset threshold), then The weights of the edges are taken as similarity values.

[0090] In this embodiment, define the "input feature" node With the Output Features node , and respectively with and Unconditional edge connection.

[0091] In this embodiment, in FIG. In the algorithm, we use the maximum weight path algorithm (such as modified Dijkstra or maximum flow path) to find the path from arrive The strongest connected path P∗;

[0092] In this embodiment, Get the optimal jump path List.

[0093] In this embodiment, based on the optimal jump path The list takes the corresponding vectors in order , and use a preset fusion function (such as weighted sum or small MLP) to fuse each jump to obtain the fusion result h. The fusion result h is combined with the original input Or add the features of the previous stage to achieve the "skip connection" effect and obtain the final remote sensing features:

[0094]

[0095] In this embodiment, the similarity graph can reflect the correlation between the features of each residual layer, so that the jump path focuses on the most relevant level, improving the efficiency and accuracy of feature fusion. At the same time, the path search based on the graph structure can skip redundant or noisy layers and only fuse highly correlated features, thereby improving the model's ability to capture details in complex scenarios.

[0096] Step 103: extracting features from the underwater drone data based on a preset Mask R-CNN network to obtain drone features;

[0097] In this embodiment, the feature extraction of the underwater drone data based on the preset Mask R-CNN network to obtain drone features includes:

[0098] The Mask R-CNN network includes a backbone network and a region proposal network;

[0099] Perform feature extraction on the underwater drone data based on the backbone network to obtain a multi-scale feature map set;

[0100] Acquire a candidate region set based on the region proposal network and the multi-scale feature map set; the candidate region set includes a plurality of candidate regions and coordinate information of each candidate region;

[0101] Feature alignment processing and feature fusion processing are performed based on the candidate region set and the multi-scale feature map set to obtain drone features.

[0102] In this embodiment, original image or video frame data collected by an underwater drone is obtained.

[0103] In this embodiment, the image frame is input into a preset Mask R-CNN backbone network, which includes an FPN (Feature Pyramid Network) for multi-scale feature extraction, and generates feature atlases {P2, P3, P4, P5} at different scales.

[0104] In this embodiment, on each feature map, the region proposal network (RPN) generates candidate target regions (RoIproposals); and performs RoIAlign operations on the candidate regions, mapping them to feature blocks of uniform size, and inputting them into the classification and bounding box regression branch and the mask branch respectively.

[0105] In this embodiment, the classification and bounding box regression branches are used to perform classification scoring and bounding box coordinate regression on each RoI feature block, and the category probability and bounding box adjustment amount of each target are output;

[0106] In this embodiment, the mask branch performs several convolution and deconvolution operations on each RoI feature block to generate a binary mask;

[0107] In this embodiment, the intermediate fully connected layer feature vector output by the classification branch, the features output by the border branch, and the convolution features output by the mask branch are cascaded or weightedly fused according to preset rules to obtain the final drone features.

[0108] Step 104: extracting features from the sensor data based on a preset recurrent neural network to obtain sensing features;

[0109] In this embodiment, the feature extraction of the sensor data based on the preset recurrent neural network to obtain the sensing features includes:

[0110] The recurrent neural network includes several layers of LSTM units;

[0111] Acquiring sensor data, including temperature data, salinity data, dissolved oxygen data, pH data, conductivity data, and turbidity data;

[0112] constructing a first learning projection matrix based on the temperature data, salinity data, and conductivity data;

[0113] Constructing a second learning projection matrix based on the dissolved oxygen data and the pH data;

[0114] constructing a third learning projection matrix based on the turbidity data;

[0115] Dynamically scaling the first learning projection matrix, the second learning projection matrix, and the third learning projection matrix to achieve cross-parameter interaction and generate a fusion vector;

[0116] Feature extraction is performed on the fusion vector based on the recurrent neural network to obtain sensing features.

[0117] In this embodiment, the sensor time series raw data sequence D = {d1, d2, ..., dT} is obtained as sensor data, where each It includes parameter vectors such as temperature, salinity, dissolved oxygen, pH, conductivity and turbidity collected at time t;

[0118] In this embodiment, D is input into the normalization module, each parameter dimension is normalized, and the normalized data sequence is output. .

[0119] In this embodiment, a first learning projection matrix is ​​constructed based on the temperature data, salinity data and conductivity data. : Construct a second learning projection matrix based on the dissolved oxygen data and pH data ; Construct a third learning projection matrix based on the turbidity data .

[0120] In this embodiment, for each projection Assign a learnable scaling factor :

[0121]

[0122] in, is the activation function.

[0123] In this embodiment, the first learning projection matrix, the second learning projection matrix and the third learning projection matrix are dynamically scaled to achieve cross-parameter interaction and generate a fusion vector :

[0124]

[0125] In this embodiment, the importance of various environmental parameters is adaptively adjusted through scaling, which is beneficial to eliminating data scale differences and enhancing coupled modeling capabilities.

[0126] In this embodiment, the vector sequence is fused at each moment The data is fed into a deep recurrent network, and multi-layer LSTM units (for example, 2 to 3 layers) are used to model the fused vector sequence, extract long-term dependency features, and finally obtain the sensor time series feature vector.

[0127] In this embodiment, the normalized time series data Input the data into a preset recurrent neural network (RNN) unit set step by step in chronological order. The RNN includes several layers of LSTM.

[0128] At the tth moment, the hidden state of the previous moment is With the current input The hidden state is updated according to the calculation function of the recurrent unit, and the current hidden state is output. ;

[0129] In this embodiment, after processing the end of the sequence After that, extract the hidden state at the last moment and (optionally) the set of intermediate hidden state vectors at each time step ;

[0130] In this embodiment, the hidden state and / or The vector obtained after pooling or attention weighting) is input into the fully connected layer, and after linear transformation and activation function processing, the fixed-dimensional sensor features are output.

[0131] Step 105: Acquire environmental condition data, and adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data;

[0132] In this embodiment, the acquiring of environmental condition data and adjusting the feature weights of the remote sensing feature, the drone feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data include:

[0133] Acquiring the environmental condition data, wherein the environmental condition data includes weather data and environmental data;

[0134] Determining the importance scores of the remote sensing feature, the drone feature, and the sensor feature based on the weather data, the environmental data, and the conditional discriminant weight table; the conditional discriminant weight table includes a mapping relationship between the weather data and the importance score of each feature;

[0135] The feature weights of the remote sensing feature, the drone feature, and the sensor feature are adjusted based on the importance score and a preset reference weight.

[0136] In this embodiment, a conditional discriminant weight table is established. This table maps weather data to the importance scores of each feature. Rows represent different environmental scenarios or weather conditions (e.g., "strong wind," "low visibility," "pollution warning," etc.), while columns correspond to different feature types (remote sensing features, underwater drone features, sensor features). The table contains the adjustment coefficients or target values ​​for each feature weight in that environment.

[0137] In this embodiment, the conditional discriminant weight table can define the baseline weight values ​​for each type of feature and the increase and decrease coefficients for different scenarios. For example, in a normal scenario, the remote sensing feature can be set at 0.4, the underwater drone feature at 0.4, and the sensor feature at 0.2. When encountering "strong winds", the rule table can specify that the drone weight be reduced by 0.2, the remote sensing weight be increased by 0.1, and the sensor weight be increased by 0.1.

[0138] In this embodiment, if the current environment meets multiple conditions simultaneously (e.g., strong winds and a pollution warning), the rule table can be configured with composite entries or applied sequentially by priority. For example, the highest-priority "strong wind" rule will be applied first, followed by the secondary adjustment caused by the "pollution warning."

[0139] Please refer to Table 1, which is a schematic table of conditional discriminant weights provided by an embodiment of the present invention.

[0140] Table 1

[0141]

[0142] In this embodiment, the importance score ranges from 0.0 to 1.0, with larger values ​​indicating a more reliable data source under those conditions. During actual fusion, the system first obtains current weather and environmental information, then uses a table lookup to obtain the corresponding scores. These scores are then used to calculate weight coefficients for feature fusion (e.g., weighted average) to adjust the contribution of each modal data. For example, rainy weather indicates reduced credibility of remote sensing data (a lower score), while drone data still has some reliability and sensor data has a higher credibility. Based on the weight ratios in the table, the system can reduce the influence of remote sensing features and increase the weight of sensor features, thereby improving the robustness of the fusion results. This conditional discriminant weight adjustment strategy allows for dynamic adaptation to changing marine environments and improves overall monitoring accuracy.

[0143] Step 106: performing weighted fusion on the remote sensing feature, the UAV feature, and the sensor feature based on the feature weight to generate a fusion feature;

[0144] In this embodiment, adjusting the feature weights of the remote sensing feature, the drone feature, and the sensor feature based on the importance score and the preset reference weight includes:

[0145] Obtaining a reference weight for each feature, mapping each reference weight to a two-dimensional plane, constructing a reference triangle with the reference weights as vertices, and calculating the centroid of the reference triangle;

[0146] For each vertex of the reference triangle, calculate a unit vector whose center points to the vertex, and translate the vertex of the reference triangle along the direction of the unit vector according to the importance score to generate a deformed triangle;

[0147] Calculating a centroid offset vector of the deformed triangle, and mapping the centroid offset vector to a direction vector of each vertex to generate a regression factor for each feature;

[0148] Based on the regression factors of each feature, the benchmark weight of each feature is incrementally regressed to adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature.

[0149] In this embodiment, the feature weights are updated according to the importance scores retrieved from the rule table.

[0150] In this embodiment, the triangle base points are first initialized, and a reference triangle is constructed on a two-dimensional plane with three preset reference weights as vertices. The center of gravity of the reference triangle is then calculated. For each vertex of the reference triangle, its external normal direction is calculated, that is, the unit vector from the center of gravity to the vertex. The normal is then translated along the direction of the unit vector according to the importance score. Specifically:

[0151] (1)

[0152] in, The vertices of the base triangle , is the importance score of the feature, is a unit vector, are the vertices of the deformed triangle, Represents remote sensing, Indicates drone, Indicates a sensor.

[0153] In this example, the three new vertices The deformed triangle is composed to calculate the center of gravity of the deformed triangle, and the center of gravity offset vector is calculated based on the center of gravity. Specifically:

[0154] (2)

[0155] (3)

[0156] in, represents the centroid of the deformed triangle, Represent the three vertices of the deformed triangle, represents the centroid of the base triangle, Indicates the center of gravity offset.

[0157] In this embodiment, The direction vector projected onto each vertex , get the regression factor of each feature:

[0158] (4)

[0159] in, is a regression factor that represents the component of the current triangle's "imbalance" in that direction.

[0160] In this embodiment, the benchmark weight is incrementally regressed based on the regression factor:

[0161] (5)

[0162] in, is the regression weight of each feature, is the baseline weight of each feature, is the regression coefficient.

[0163] In this embodiment, the regression weights of each feature are trimmed and normalized to obtain the final feature weights of each feature.

[0164] In this embodiment, the balance or imbalance of the importance scores of the three types of features is intuitively reflected through geometric offset, and the weight adjustment is directly driven based on the center of gravity offset, without setting complex thresholds or empirical rules.

[0165] Step 107: construct an environmental monitoring model based on a decision tree algorithm, input the fusion features into the environmental monitoring model, and obtain environmental monitoring results.

[0166] In this embodiment, remote sensing features, underwater drone features, and sensor features are weighted and fused to obtain a fused feature vector, which is used as the input of the environmental monitoring model.

[0167] In this embodiment, historical environmental monitoring data is used to train a decision tree model. The CART (Classification and Regression Tree) algorithm is selected. This algorithm uses the Gini index as a partitioning criterion and is suitable for classification and regression tasks.

[0168] In this embodiment, the environmental monitoring results include water quality grades (such as Class I to Class V), predicted values ​​of pollutant concentrations (such as ammonia nitrogen, phosphorus, heavy metals, etc.), abnormal event warnings (such as red tides, hypoxia, etc.) and ecological risk level assessments.

[0169] In this embodiment, after constructing the environmental monitoring model based on the decision tree algorithm, inputting the fusion features into the environmental monitoring model, and obtaining the environmental monitoring results, the following steps are further included:

[0170] generating an error evaluation value and a credibility value based on the monitoring result, and generating a comprehensive evaluation score based on the error evaluation value and the credibility value;

[0171] The benchmark weight is incrementally adjusted according to the comprehensive evaluation score.

[0172] In this embodiment, the error evaluation value e is calculated by comparing the model prediction results with the actual observation data. The error evaluation value e includes the mean absolute error (MAE) and the root mean square error (RMSE). Combined with the stability performance of the environmental detection model in multiple batch evaluations, the result reliability score r is calculated, such as the classification accuracy or regression value, to quantify the credibility of the detection result.

[0173] In this embodiment, the comprehensive evaluation score is specifically:

[0174] (3)

[0175] in, and It is a preset weight coefficient used to balance the impact of error and reliability on feedback.

[0176] In this embodiment, incrementally adjusting the benchmark weight according to the comprehensive evaluation score includes:

[0177] (4)

[0178] in, is the base weight, where For feature type The sensitivity coefficient of (remote sensing, drone, sensing) can be calibrated offline in the early stage.

[0179] In this example, remote sensing data covers the macroscopic ocean environment, drones capture local biological details such as algae, and sensors provide continuous time series parameters, overcoming the limitations of a single data source and avoiding insufficient response to local or dynamic environments. Different neural networks are then used to extract features for each type of data, dynamically adjusting feature weights based on environmental and weather conditions. This prevents inaccurate measurements of a particular type of data due to these conditions, enhancing the ability of environmental monitoring to adapt to changing circumstances. Using a decision tree model for environmental monitoring improves the model's interpretability and decision-making efficiency, thereby enhancing the stability and accuracy of marine ranch environmental monitoring.

[0180] In an embodiment of the present invention, a terminal device is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the above-mentioned marine ranch environment monitoring method is implemented.

[0181] In an embodiment of the present invention, a computer-readable storage medium is also provided, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned marine ranch environment monitoring method.

[0182] For example, a computer program may be divided into one or more modules, one or more of which are stored in a memory and executed by a processor to implement the present invention. One or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in a terminal device.

[0183] The terminal device may be a computing device such as a desktop computer, laptop, PDA, or cloud server. The terminal device may include, but is not limited to, a processor, memory, and display. Those skilled in the art will appreciate that the aforementioned components are merely examples of terminal devices and do not constitute a limitation of the terminal device. The terminal device may include more or fewer components, or a combination of certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, and the like.

[0184] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0185] The memory can be used to store computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store an operating system and at least one application required for a function (such as a sound playback function, a text conversion function, etc.); the data storage area can store data generated based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory can include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0186] If the module for monitoring the marine ranch environment is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. Computer-readable media can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. Those skilled in the art can understand and implement the present invention without expending any creative effort.

[0187] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for monitoring the environment of a marine ranch, characterized in that: include: Acquiring environmental monitoring data, including remote sensing monitoring data, underwater drone data, and sensor data; Performing feature extraction on the remote sensing monitoring data based on a preset deep residual network to obtain remote sensing features; Perform feature extraction on the underwater drone data based on a preset Mask R-CNN network to obtain drone features; Performing feature extraction on the sensor data based on a preset recurrent neural network to obtain sensing features; Acquiring environmental condition data, and adjusting feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data; Performing weighted fusion on the remote sensing feature, the UAV feature, and the sensor feature based on the feature weight to generate a fusion feature; Building an environmental monitoring model based on a decision tree algorithm, inputting the fusion features into the environmental monitoring model, and obtaining environmental monitoring results; The step of obtaining environmental condition data and adjusting the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset benchmark weight, and the environmental condition data includes: obtaining the environmental condition data, wherein the environmental condition data includes weather data and environmental data; determining importance scores of the remote sensing feature, the UAV feature, and the sensor feature based on the weather data, the environmental data, and the conditional discriminant weight table; wherein the conditional discriminant weight table includes a mapping relationship between the weather data and the importance score of each feature; and adjusting the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on the importance score and the preset benchmark weight. The adjusting the feature weights of the remote sensing feature, the UAV feature and the sensor feature based on the importance score and the preset benchmark weight includes: obtaining the benchmark weight of each feature, mapping the benchmark weight of each feature to a two-dimensional plane, constructing a benchmark triangle with the benchmark weight of each feature as a vertex, and calculating the center of gravity of the benchmark triangle; for each vertex of the benchmark triangle, calculating the unit vector of the center of gravity pointing to the vertex, and translating the vertex of the benchmark triangle along the direction of the unit vector according to the importance score to generate a deformed triangle; calculating the center of gravity offset vector of the deformed triangle, and mapping the center of gravity offset vector to the direction vector of each vertex to generate a regression factor of each feature; and performing incremental regression on the benchmark weight of each feature based on the regression factor of each feature to adjust the feature weights of the remote sensing feature, the UAV feature and the sensor feature.

2. A marine ranch environmental monitoring method according to claim 1, characterized in that: The remote sensing monitoring data is subjected to feature extraction based on a preset deep residual network to obtain remote sensing features, including: The deep residual network includes a plurality of residual units, each of which is connected by a jump connection; Performing feature extraction on the remote sensing monitoring data based on each residual unit to obtain a feature vector; Evaluate the similarity between each pair of feature vectors, and construct a feature vector topology map based on the similarity; Based on the feature vector topology map and the maximum weight path algorithm, the jump path of each feature vector is determined, and based on the jump path and jump connection, each feature vector is fused to obtain remote sensing features.

3. The marine ranch environmental monitoring method according to claim 1, wherein: The feature extraction of the underwater drone data based on the preset Mask R-CNN network to obtain drone features includes: The Mask R-CNN network includes a backbone network and a region proposal network; Perform feature extraction on the underwater drone data based on the backbone network to obtain a multi-scale feature map set; Acquire a candidate region set based on the region proposal network and the multi-scale feature map set; the candidate region set includes a plurality of candidate regions and coordinate information of each candidate region; Based on the coordinate information of each candidate region and the multi-scale feature map set, each candidate region is mapped into a feature block of a fixed size to obtain a feature block set; The feature block set is fused to obtain the drone features.

4. The marine ranch environmental monitoring method according to claim 1, wherein: The feature extraction of the sensor data based on the preset recurrent neural network to obtain the sensing features includes: The recurrent neural network includes several layers of LSTM units; Acquiring sensor data, including temperature data, salinity data, dissolved oxygen data, pH data, conductivity data, and turbidity data; constructing a first learning projection matrix based on the temperature data, salinity data, and conductivity data; Constructing a second learning projection matrix based on the dissolved oxygen data and the pH data; constructing a third learning projection matrix based on the turbidity data; Dynamically scaling the first learning projection matrix, the second learning projection matrix, and the third learning projection matrix to achieve cross-parameter interaction and generate a fusion vector; Feature extraction is performed on the fusion vector based on the recurrent neural network to obtain sensing features.

5. A method for monitoring marine ranch environment according to any one of claims 1 to 4, characterized in that: After constructing the environmental monitoring model based on the decision tree algorithm, inputting the fusion features into the environmental monitoring model, and obtaining the environmental monitoring results, the method further includes: generating an error evaluation value and a credibility value based on the monitoring result, and generating a comprehensive evaluation score based on the error evaluation value and the credibility value; The baseline weight of each feature is incrementally adjusted according to the comprehensive evaluation score.

6. A marine ranch environment monitoring system, used to implement a marine ranch environment monitoring method according to any one of claims 1 to 5, characterized in that: include: Data acquisition module, feature extraction module, feature fusion module and monitoring module; The data acquisition module is used to obtain environmental monitoring data, which includes remote sensing monitoring data, underwater drone data and sensor data; The feature extraction module is used to extract features from the remote sensing monitoring data based on a preset deep residual network to obtain remote sensing features; Performing feature extraction on the underwater drone data based on a preset Mask R-CNN network to obtain drone features; performing feature extraction on the sensor data based on a preset recurrent neural network to obtain sensor features; The feature fusion module is used to obtain environmental condition data and adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data; Performing weighted fusion on the remote sensing feature, the UAV feature, and the sensor feature based on the feature weight to generate a fusion feature; The monitoring module is used to build an environmental monitoring model based on a decision tree algorithm, input the fusion features into the environmental monitoring model, and obtain environmental monitoring results.

7. The marine ranch environmental monitoring system according to claim 6, characterized in that: The fusion module is used to obtain environmental condition data, and adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on a preset conditional discriminant weight table, a preset reference weight, and the environmental condition data, including: Acquiring the environmental condition data, wherein the environmental condition data includes weather data and environmental data; Determining the importance scores of the remote sensing feature, the drone feature, and the sensor feature based on the weather data, the environmental data, and the conditional discriminant weight table; the conditional discriminant weight table includes a mapping relationship between the weather data and the importance score of each feature; The feature weights of the remote sensing feature, the drone feature, and the sensor feature are adjusted based on the importance score and a preset reference weight.

8. The marine ranch environmental monitoring system according to claim 7, characterized in that: The fusion module is used to adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature based on the importance score and a preset reference weight, including: Obtaining a reference weight for each feature, mapping the reference weight for each feature to a two-dimensional plane, constructing a reference triangle with the reference weight for each feature as a vertex, and calculating the centroid of the reference triangle; For each vertex of the reference triangle, calculate a unit vector whose center of gravity points to the vertex, and translate the vertex of the reference triangle along the direction of the unit vector according to the importance score to generate a deformed triangle; Calculating a centroid offset vector of the deformed triangle and mapping the centroid offset vector to a direction vector of each vertex to generate a regression factor for each feature; Based on the regression factors of the respective features, incremental regression is performed on the benchmark weights of the respective features to adjust the feature weights of the remote sensing feature, the UAV feature, and the sensor feature.

Images