Aquaculture monitoring method and system based on multi-modal sensing technology

Abstract

The application provides an aquaculture monitoring method and system based on a multi-modal sensing technology, relates to the technical field of aquaculture monitoring methods, and comprises the following steps: calculating a comprehensive water quality characteristic value of each grid node, multiplying and aggregating an adjacency matrix and a characteristic matrix, simultaneously processing the first characteristic matrix by using an expanded causal convolution to obtain a space-time characteristic matrix of each time step; calculating a comprehensive risk index by using a weighted fusion method, and classifying and warning the comprehensive risk index. High-resolution spatial modeling is realized by grid division, local differences of parameters such as dissolved oxygen, temperature and ammonia concentration in different areas of the pool bottom are accurately captured, the spatial correlation between grid nodes is dynamically constructed based on the adjacency matrix, the space-time dependence relationship is simultaneously modeled by combining the adjacency matrix multiplication aggregation and the expanded causal convolution, and errors caused by spatial discretization in traditional methods are avoided.

Description

Technical Field

[0001] This invention relates to the field of aquaculture monitoring methods, specifically to an aquaculture monitoring method and system based on multimodal sensing technology. Background Technology

[0002] Traditional aquaculture water quality monitoring methods often rely on single-parameter threshold warnings, such as those for excessive ammonia concentration, or data collected by fixed-location sensors. For example, excessive ammonia concentration can lead to the heterogeneous distribution of dissolved oxygen and temperature in the three-dimensional space of the pond, resulting in missed detections of local toxicity risks. Empirical weights are used to linearly superimpose multiple parameters, such as pH and free ammonia, ignoring the dynamic interaction effects between parameters, such as the exacerbation of ammonia toxicity by rising temperatures and the real-time impact of sudden pollution events. Traditional models struggle to integrate time-series trends, such as the accumulation of pollutants from continuous high temperatures and spatial diffusion patterns, such as the migration of pollutants from the bottom of the pond to the surface. They suffer from strong warning lag and rely on binary alarms, lacking a comprehensive risk assessment and graded early warning mechanism that integrates multi-dimensional parameters.

[0003] In the existing technology, document CN119534781A describes a multi-dimensional monitoring system that integrates acoustic, optical, and environmental parameter monitoring technologies. However, this method does not quantify water quality gradient changes in the depth, width, and length of the pond based on three-dimensional mesh partitioning and adjacency matrix construction to accurately locate pollution sources; it does not dynamically allocate parameter weights based on information entropy and toxicity factors to adapt to the nonlinear coupling effect of water quality parameters and improve the sensitivity of response to sudden pollution; it does not combine dilated causal convolution and Mahalanobis distance to simultaneously capture the long-term time-varying patterns and spatial diffusion paths of parameters to enhance risk prediction capabilities; and it does not classify low, medium, and high risks based on a comprehensive risk index, supplemented by dynamic threshold adjustment, to provide quantitative basis for precise drug administration, oxygenation, and other decisions, thereby reducing aquaculture loss rates. Therefore, there is an urgent need for aquaculture monitoring methods and systems based on multimodal sensing technology.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for monitoring aquaculture based on multimodal sensing technology, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The aquaculture monitoring method based on multimodal sensing technology includes the following steps: S1: Divide the aquaculture pond into grids along the depth, width and length directions at fixed intervals, mark each grid node, set the time window length and time step, and collect the dissolved oxygen concentration, pH value, temperature and ammonia concentration of each grid node at each time step; S2: At each time step, the free ammonia concentration is calculated using the dynamic dissociation equilibrium method based on the pH value and temperature of each grid node. All four types of parameters, namely dissolved oxygen concentration, pH value, temperature and free ammonia concentration, are normalized and mapped to a set interval. The interval is divided into multiple sub-intervals of equal length. The frequency ratio of each type of parameter in each sub-interval is counted, and the information entropy of each type of parameter is calculated. S3: Based on the information entropy result, set the toxicity factor of the parameter, calculate the weight factor of each parameter at each time step, and weight and sum the normalized free ammonia concentration, dissolved oxygen concentration, pH value and temperature of each grid node with the corresponding weight factor to obtain the comprehensive water quality characteristic value of each grid node at each time step, and construct the feature matrix with the comprehensive water quality characteristic values ​​of all grid nodes at one time step. S4: Set distance threshold rules, calculate the spatial relationship between grid nodes, construct the adjacency matrix, perform matrix multiplication and aggregation on the adjacency matrix and the feature matrix, activate with the ReLU function to obtain the first feature matrix, perform dilated causal convolution on the first feature matrix at all time steps to extract spatiotemporal features and obtain the spatiotemporal feature matrix; S5: The spatiotemporal feature matrix and the free ammonia concentration data are combined to form an ammonia concentration matrix. After standardization, the Mahalanobis distance method is used to calculate the preliminary toxicity index vector. The mean scalar of the preliminary toxicity vector and the free ammonia concentration matrix is ​​calculated. Based on the mean scalar, the weighted fusion method is used to calculate the comprehensive risk index. Based on the index, water quality risk classification and early warning are carried out, and the corresponding risk warning information is output.

[0007] Further, grid division and data collection are performed, with the following specific steps: The aquaculture pond is divided into grids along the depth, width, and length directions according to a preset fixed interval. Each grid node is marked, and a measurement point is set at each grid node. The time window length is set, and the dissolved oxygen concentration, pH value, temperature, and ammonia concentration of the measurement point are collected at preset time steps.

[0008] Further, the free ammonia concentration is calculated, and the specific steps are as follows: Calculate the percentage of free ammonia based on the pH value and temperature at each grid point in each time step: in, Indicates the first The first time step The dissociation constant of free ammonia in each grid cell; Indicates the first The first time step Temperature of each grid cell; Indicates the first The first time step The proportion of free ammonia to total ammonia nitrogen in each grid cell; Indicates the first The first time step pH of each grid cell; Indicates the time step index; The concentration of free ammonia was calculated using the dynamic dissociation equilibrium method. in, Indicates the first The first time step NH4 measured in each grid cell + concentration; Indicates the first The first time step The concentration of free ammonia in each grid cell.

[0009] Furthermore, the information entropy of each type of parameter is calculated, and the specific steps are as follows: All four parameters—dissolved oxygen concentration, pH value, temperature, and free ammonia concentration—were normalized and mapped to a set interval. The interval was then divided into multiple equal-length sub-intervals, and the frequency percentage of each parameter in each sub-interval was calculated. in, Indicates the total number of grid nodes; Indicates the length of the time window; Indicates the interval index number; Indicates the first The normalized first grid node of all grid nodes at the nth time step Class parameter value, , The time indicates the concentration of free ammonia. The time indicates the dissolved oxygen concentration. The time indicates the pH value. Time indicates temperature; Indicates the first The first time step The class parameter value falls into the first The probability of the frequency within a certain interval relative to the total frequency; Calculate the information entropy for each type of parameter: in, Indicates the first The first time step Information entropy of class parameter values; Represents the total number of intervals 。

[0010] Furthermore, the weighting factors of various parameters at each time step are calculated, and the specific steps are as follows: Based on the information entropy results, a toxicity factor is set, and the weighting factors of various parameters at each time step are calculated: in, Indicates the first The toxicity factor of class parameter values, and , , as well as ; Indicates the first The maximum information entropy of all categories of parameters at each time step; Indicates the first The first time step Weighting factors for class parameter values.

[0011] Furthermore, the comprehensive water quality characteristic value of each grid node at each time step is calculated, and the specific steps are as follows: The normalized free ammonia concentration, dissolved oxygen concentration, pH value, and temperature of each grid node at each time step are weighted and summed with weighting factors to obtain the comprehensive water quality characteristic value of each grid node at each time step: Indicates the normalized i-th The first time step The first grid node Class parameter value, , The time indicates the concentration of free ammonia. The time indicates the dissolved oxygen concentration. The time indicates the pH value. Time indicates temperature; Indicates the first The first time step The comprehensive water quality characteristic value of each grid node; This indicates the number of grid node indices.

[0012] Further, the spatiotemporal feature matrix for each time step is calculated, specifically through the following steps: At one time step, a feature matrix is ​​constructed using the comprehensive water quality feature values ​​of all grid nodes. Distance threshold rules are set, spatial relationships between grid nodes are calculated, and an adjacency matrix is ​​constructed. Specifically, if the distance between any two grid nodes is less than a preset distance and the difference in comprehensive water quality features between grid nodes at each time step does not exceed a preset difference, the value is set to 1; otherwise, it is set to 0. The adjacency matrix between all grid nodes at each time step is constructed sequentially. The adjacency matrix at each time step is multiplied and aggregated with the feature matrix at each time step, and then activated using the ReLU function to obtain the first feature matrix: in, Indicates the first The adjacency matrix at each time step; Indicates the first Feature matrices for each time step; Indicates the first The first feature matrix at each time step; Dilated causal convolution is performed on the first feature matrix at all time steps to extract spatiotemporal features, resulting in the spatiotemporal feature matrix: in, in, Indicates the first The contribution weight of each time step; Indicates the total number of sampling points; Indicates the first The spatiotemporal feature matrix at each time step.

[0013] Further, a preliminary toxicity vector is generated, the specific steps of which are as follows: The spatiotemporal feature matrix and free ammonia concentration data are combined to form an ammonia concentration matrix. This matrix is ​​then standardized and concatenated column-by-column. The mean scalar of each column of the concatenated matrix is ​​calculated. Based on the mean, the Mahalanobis distance vector of the concatenated matrix at each time step is calculated. The preliminary toxicity index vector is further calculated using the Mahalanobis distance method, based on the following formula: in, Indicates the first The Mahalanobis distance vector at each time step; Indicates the first [number] after splicing The matrix mean at each time step; Indicates the first [number] after splicing A matrix for each time step; Indicates the first The initial toxicity index vector at each time step; Indicates the scale parameter.

[0014] Furthermore, the comprehensive risk index is classified and warnings are issued. The specific steps are as follows: The mean scalar of the ammonia concentration matrix, consisting of the preliminary toxicity vector and the free ammonia concentration at each time step, is calculated separately. Based on the mean scalar, a weighted fusion method is used to calculate the comprehensive risk index, and the comprehensive risk index is used for classification and early warning. in, Indicates the first A comprehensive risk index for each time step; Indicates the first The mean scalar of the initial toxicity vector at each time step; Indicates the first The mean scalar of the ammonia concentration matrix at each time step; when A value less than 0.5 indicates that the aquaculture pond environment is normal. When the value is equal to 0.5 and less than 0.8, it indicates an abnormal environment in the aquaculture pond and a risk. When the value is equal to 0.8, the environment of the aquaculture pond is poor and an emergency warning is required.

[0015] The present invention also provides an aquaculture monitoring system based on multimodal sensing technology, the monitoring system being used to perform the above-described monitoring method, comprising: The data acquisition module is used to divide the aquaculture pond into grids along the depth, width and length directions at fixed intervals, mark each grid node, set the time window length and time step, and collect the dissolved oxygen concentration, pH value, temperature and ammonia concentration of each grid node at each time step; The parameter calculation module is used to calculate the free ammonia concentration at each time step based on the pH value and temperature of each grid node using the dynamic dissociation equilibrium method. It normalizes and maps all four types of parameters, namely dissolved oxygen concentration, pH value, temperature and free ammonia concentration, into a set interval, divides the interval into multiple sub-intervals of equal length, counts the frequency ratio of each type of parameter in each sub-interval, and calculates the information entropy of each type of parameter. The feature matrix acquisition module is used to set the toxicity factor based on the information entropy result, calculate the weight factor of various parameters at each time step, and perform weighted summation of the normalized free ammonia concentration, dissolved oxygen concentration, pH value and temperature of each grid node with the corresponding weight factor to obtain the comprehensive water quality feature value of each grid node at each time step, and construct the feature matrix with the comprehensive water quality feature values ​​of all grid nodes at one time step. The spatiotemporal matrix acquisition module is used to set distance threshold rules, calculate the spatial relationship between grid nodes, construct an adjacency matrix, aggregate the adjacency matrix and the feature matrix by matrix multiplication, activate it using the ReLU function to obtain the first feature matrix, perform dilated causal convolution on the first feature matrix at all time steps to extract spatiotemporal features and obtain the spatiotemporal feature matrix. The risk warning module is used to construct an ammonia concentration matrix by combining the spatiotemporal feature matrix with the free ammonia concentration data, perform standardization processing, calculate the preliminary toxicity index vector using the Mahalanobis distance method, calculate the mean scalar of the preliminary toxicity vector and the free ammonia concentration matrix, calculate the comprehensive risk index using the weighted fusion method based on the mean scalar, and perform water quality risk classification and warning based on the index, outputting the corresponding risk warning information.

[0016] Compared with the prior art, the beneficial effects of the present invention are: Based on three-dimensional mesh generation and adjacency matrix construction, the water quality gradient changes in the depth, width, and length of the pond are quantified to accurately locate pollution sources. By dynamically allocating parameter weights using information entropy and toxicity factors, the nonlinear coupling effect of water quality parameters is adaptively improved to enhance the sensitivity of response to sudden pollution. By combining dilated causal convolution and Mahalanobis distance, the long-term time-varying patterns and spatial diffusion paths of parameters are captured simultaneously, enhancing risk prediction capabilities. Based on a comprehensive risk index, low, medium, and high risk classifications are achieved, supplemented by dynamic threshold adjustments, providing quantitative basis for precise drug administration, oxygenation, and other decisions, thereby reducing aquaculture loss rates. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a graph showing the relationship between the percentage of free ammonia and pH value. Figure 3 This is a block diagram of the overall system structure of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example: Please see Figures 1-2 The present invention provides a technical solution: The aquaculture monitoring method based on multimodal sensing technology includes the following steps: S1: Divide the aquaculture pond into grids along the depth, width and length directions at fixed intervals, mark each grid node, set the time window length and time step, and collect the dissolved oxygen concentration, pH value, temperature and ammonia concentration of each grid node at each time step; The specific steps for grid division and data acquisition are as follows: The aquaculture pond is divided into grids along the depth, width, and length directions according to a preset fixed interval. Each grid node is marked, and a measurement point is set at each grid node. The time window length is set, and the dissolved oxygen concentration, pH value, temperature, and ammonia concentration of the measurement point are collected at preset time steps.

[0021] In the above process, by dividing the continuous aquaculture water body into high-resolution monitoring nodes by dividing the space into three dimensions of depth, width and length, the limitations of traditional single-point or surface monitoring are broken, and the parameter stratification and gradient differences of the water body in the vertical and horizontal directions can be fully captured. At the same time, by setting time windows and step sizes for periodic acquisition, the static spatial snapshot is extended into a dynamic time series. Dissolved oxygen concentration, pH value, temperature, and ammonia concentration are core physicochemical parameters affecting the aquatic environment and the survival of aquatic organisms. Dissolved oxygen is directly essential for the respiratory metabolism of aquatic organisms; excessively low concentrations can lead to suffocation and death. pH value directly affects the acid-base balance of the water and the activity of enzymes and ammonia toxicity within organisms. Temperature is a key physical factor regulating the metabolic rate, growth rate, and dissolved oxygen saturation of aquatic organisms. Ammonia concentration, especially non-ionized free ammonia, is highly toxic to aquatic animals, damaging gill tissue, disrupting osmotic balance, and inhibiting growth. Simultaneous collection of these four parameters allows for a systematic assessment of the overall suitability and potential risks of a water body from three interrelated dimensions: biological survival (dissolved oxygen), chemical environmental stability (pH and ammonia toxicity), and physical metabolic driving force (temperature).

[0022] S2: At each time step, the free ammonia concentration is calculated using the dynamic dissociation equilibrium method based on the pH value and temperature of each grid node. All four types of parameters, namely dissolved oxygen concentration, pH value, temperature and free ammonia concentration, are normalized and mapped to a set interval. The interval is divided into multiple sub-intervals of equal length. The frequency ratio of each type of parameter in each sub-interval is counted, and the information entropy of each type of parameter is calculated. The specific steps for calculating the free ammonia concentration are as follows: Calculate the percentage of free ammonia based on the pH value and temperature at each grid point in each time step: in, Indicates the first The first time step The dissociation constant of free ammonia in each grid cell; Indicates the first The first time step Temperature of each grid cell; Indicates the first The first time step The proportion of free ammonia to total ammonia nitrogen in each grid cell; Indicates the first The first time step pH of each grid cell; Indicates the time step index; In the above process, by introducing two key variables, temperature and pH, to dynamically calculate the dissociation constant, this process accurately quantifies the proportion of ammonia nitrogen in water as highly toxic non-ionic NH3; the dependent variable in the formula... Indicates the first The first time step The mass fraction of free ammonia in total ammonia nitrogen per grid cell. This parameter directly quantifies the proportion of free ammonia forms in the water that are highly toxic to aquatic organisms. Its technical advantage lies in transforming the measured total ammonia nitrogen concentration into an indicator that better reflects actual ecotoxicological risks, providing core data for subsequent accurate water quality toxicity assessment and early warning. The independent variable in the formula is temperature. and They influence the proportion of free ammonia through the chemical thermodynamics of the ammonia-ammonium equilibrium system; among which, temperature is determined by empirical formulas. Determine the dissociation constant Increased temperature will cause Decreasing the concentration of hydrogen ions in the solution affects the dissociation equilibrium of ammonia; pH value characterizes the concentration of hydrogen ions in the solution, and its magnitude directly determines the direction of the shift in chemical equilibrium, thus controlling the relative distribution of free ammonia and ionic ammonium; dependent variable With independent variable There is a positive correlation, meaning that an increase in pH value significantly increases the proportion of free ammonia; meanwhile, the dependent variable... pass With temperature It also shows a positive correlation, that is, as temperature increases, it decreases. This also leads to an increase in the proportion of free ammonia; 2729.92 is derived from the conversion between the standard enthalpy change and gas constant of the reaction, used to quantify the influence of temperature on the dissociation equilibrium constant; 273.15 is the conversion constant between the Celsius scale and the thermodynamic temperature scale (Kelvin), ensuring that the temperature unit conforms to the requirements of the physicochemical formula; 0.09018 is an empirical constant related to the standard entropy change of the reaction, used to correct the reference value of the formula; the selection of these three numbers is based on the dissociation reaction of ammonia in aqueous solution, i.e., NH4+. + NH3+ H + The thermodynamic principles and experimental data fitting.

[0023] In the above embodiments, 20 sets of data on pH value and corresponding free ammonia percentage are given to reflect the change of free ammonia percentage with pH value, as shown in Table 1: Table 1: Relationship between the percentage of free ammonia and pH value In Table 1 above, the fixed temperature is... The dissociation constant is In the case of pH, the proportion of free ammonia, the dependent variable, is positively correlated with the independent variable pH, that is, an increase in pH will significantly increase the proportion of free ammonia.

[0024] The concentration of free ammonia was calculated using the dynamic dissociation equilibrium method. in, Indicates the first The first time step NH4 measured in each grid cell + concentration; Indicates the first The first time step The concentration of free ammonia in each grid cell.

[0025] In the above process, the measured total ammonia nitrogen concentration The calculated percentage of free ammonia Multiplication dynamically separates and quantifies the free ammonia fraction with direct biotoxicity from total ammonia nitrogen. This step closely integrates chemical equilibrium theory with actual monitoring data, overcoming the potential bias that may arise from assessing toxicity solely based on total ammonia nitrogen concentration.

[0026] The specific steps for calculating the information entropy of each type of parameter are as follows: All four parameters—dissolved oxygen concentration, pH value, temperature, and free ammonia concentration—were normalized and mapped to a set interval. The interval was then divided into multiple equal-length sub-intervals, and the frequency percentage of each parameter in each sub-interval was calculated. in, Indicates the total number of grid nodes; Indicates the length of the time window; Indicates the interval index number; Indicates the first The normalized first grid node of all grid nodes at the nth time step Class parameter value, , The time indicates the concentration of free ammonia. The time indicates the dissolved oxygen concentration. The time indicates the pH value. Time indicates temperature; Indicates the first The first time step The class parameter value falls into the first The probability of the frequency within a certain interval relative to the total frequency; In the above process, the four normalized parameters are discretized and binned within a preset interval, and the frequency percentage of data points in each sub-interval is calculated. This process transforms the massive monitoring data from all grid nodes at each time step into a probability distribution sequence reflecting the spatial uniformity of various parameter values ​​at that moment; the independent variable in the formula is the set of normalized parameter values. and total frequency base Select As a core independent variable, it is used because the specific value of the parameter directly determines which preset sub-interval it will fall into, thus affecting the counting results within each interval; while the total frequency base... That is, the total number of grid nodes multiplied by the time window length represents the total number of all possible data points at that moment. Using this as the denominator for normalization ensures that the calculated value is accurate. A valid probability value is one between 0 and 1, and the sum of the probabilities across all intervals is 1. In the formula... In the calculation, the dependent variable The probability of a data point falling within a given interval is higher than the number of data points falling within that interval. Meanwhile, the dependent variable... With denominator Negative correlation means that, with a fixed count value, the more total data points there are, the smaller the calculated probability value, which reflects the normalization property of probability.

[0027] Calculate the information entropy for each type of parameter: in, Indicates the first The first time step Information entropy of class parameter values; Represents the total number of intervals 。

[0028] In the above process, the dependent variable Indicates the first The first time step The information entropy of parameter values, technically speaking, transforms the spatial distribution characteristics of parameters into a scalar index: a higher entropy value indicates a more uniform and unpredictable spatial distribution of parameter values; a lower entropy value indicates a more concentrated and ordered distribution of parameter values; the independent variable of the formula is the probability of each subinterval. ;choose As an independent variable, it is chosen because information entropy is essentially a function of a probability distribution; it directly measures the degree of disorder in that distribution. This represents the probability that a parameter value falls within a specific interval; the uniformity of its distribution is that of all... The similarity directly determines the magnitude of the entropy value; entropy is a comprehensive function of all these probability values; dependent variable With independent variable The relationship between them is non-monotonic, but is affected by the overall uniformity of the probability distribution; when the probability distribution is perfectly uniform, i.e., all... )hour, The probability distribution reaches its maximum value when it is extremely concentrated, i.e., a certain... When the value is close to 1 and the rest are close to 0, Therefore, the minimum value is obtained. It is positively correlated with the uniformity of the probability distribution.

[0029] S3: Based on the information entropy result, set the toxicity factor of the parameter, calculate the weight factor of each parameter at each time step, and weight and sum the normalized free ammonia concentration, dissolved oxygen concentration, pH value and temperature of each grid node with the corresponding weight factor to obtain the comprehensive water quality characteristic value of each grid node at each time step, and construct the feature matrix with the comprehensive water quality characteristic values ​​of all grid nodes at one time step. The specific steps for calculating the weighting factors of various parameters at each time step are as follows: Based on the information entropy results, a toxicity factor is set, and the weighting factors of various parameters at each time step are calculated: in, Indicates the first The toxicity factor of class parameter values, and , , as well as ; Indicates the first The maximum information entropy of all categories of parameters at each time step; Indicates the first The first time step Weighting factors for class parameter values.

[0030] The technical effect of the above process lies in realizing a dynamic weight allocation mechanism that integrates the inherent toxicity prior of parameters with the real-time spatial distribution anomaly. This is achieved by fixing the toxicity factor... The dynamic anomaly factor calculated based on information entropy In combination, this weighting formula can simultaneously reflect the inherent differences in the biological hazards of different water quality parameters, such as the highest toxicity of free ammonia, as well as the degree of non-uniformity or abnormal concentration of their spatial distribution at the current moment, i.e., the lower the entropy value, the more abnormal the distribution. Toxic factors The values ​​assigned are based on the priority ranking of the inherent toxicity and environmental sensitivity of various water quality parameters to aquaculture organisms. Free ammonia is given the highest weight due to its high toxicity and low safety threshold, and is directly associated with the risk of acute poisoning. Dissolved oxygen has a decisive impact on the survival of organisms, and hypoxia can lead to rapid and large-scale mortality, so its weight is second. pH plays an important regulatory role by affecting ammonia toxicity and the physiological state of organisms, but it usually has a wide tolerance range. Temperature, as a background physical factor, affects metabolism and dissolved oxygen, but its changes are usually slow and the organisms are relatively adaptable. This setting reflects the hierarchical weight allocation logic of "toxicity priority, survival critical, regulation auxiliary, and background reference". The independent variable in the formula is the toxicity factor. The ratio of the normalized information entropy to ,choose This is because different parameters have significantly different inherent toxicities to aquatic organisms and need to be used as a basis for weighting. This is because information entropy reflects the uniformity of the parameter space distribution, and its ratio to the maximum entropy is... This can be considered as "distribution anomaly": the lower the entropy value, the larger this value, indicating that the parameter is more anomalous in the current spatial state, and therefore should be given higher weight in the comprehensive evaluation; dependent variable With toxic factors They are positively correlated. Meanwhile, the dependent variable... and Positively correlated with entropy value Negative correlation: the more uneven the spatial distribution of the parameter at the current moment, i.e. the smaller the entropy, the higher its "abnormality" and the greater the dynamic weight obtained; the normalization operation of the denominator ensures that the sum of all weights is 1.

[0031] The specific steps for calculating the comprehensive water quality characteristic value for each grid node at each time step are as follows: The normalized free ammonia concentration, dissolved oxygen concentration, pH value, and temperature of each grid node at each time step are weighted and summed with weighting factors to obtain the comprehensive water quality characteristic value of each grid node at each time step: Indicates the normalized i-th The first time step The first grid node Class parameter value, , The time indicates the concentration of free ammonia. The time indicates the dissolved oxygen concentration. The time indicates the pH value. Time indicates temperature; Indicates the first The first time step The comprehensive water quality characteristic value of each grid node; This indicates the number of grid node indices.

[0032] The technical achievement in the above process lies in realizing the adaptive fusion of multimodal water quality parameters at spatiotemporal nodes, generating a feature scalar that can comprehensively reflect the local water quality status and global risk weights, by normalizing the four types of parameter values ​​of each grid node. With corresponding time step dynamic weights The weighted summation step aggregates discrete, multidimensional sensor data with different dimensions into a unified comprehensive water quality characteristic value. This takes into account both the inherent importance of different parameters to water quality and their anomalies in spatial distribution at the current moment, making... It not only represents the absolute level of water quality at that node, but also implies its relative risk level relative to the overall spatial state of the aquaculture pond.

[0033] S4: Set distance threshold rules, calculate the spatial relationship between grid nodes, construct the adjacency matrix, perform matrix multiplication and aggregation on the adjacency matrix and the feature matrix, activate with the ReLU function to obtain the first feature matrix, perform dilated causal convolution on the first feature matrix at all time steps to extract spatiotemporal features and obtain the spatiotemporal feature matrix; The specific steps for calculating the spatiotemporal feature matrix at each time step are as follows: At one time step, a feature matrix is ​​constructed using the comprehensive water quality feature values ​​of all grid nodes. Distance threshold rules are set, spatial relationships between grid nodes are calculated, and an adjacency matrix is ​​constructed. Specifically, if the distance between any two grid nodes is less than a preset distance and the difference in comprehensive water quality features between grid nodes at each time step does not exceed a preset difference, the value is set to 1; otherwise, it is set to 0. The adjacency matrix between all grid nodes at each time step is constructed sequentially. In the above process, by setting dual rules of distance threshold and feature difference threshold, this step not only identifies physically adjacent grid nodes in three-dimensional space, but also filters out nodes with similar water quality states (characterized by comprehensive feature values), thereby constructing an adjacency matrix that changes over time. This construction method effectively captures the potential spatial continuity and local homogeneity of water quality parameters in the water body; the determination of the preset distance is usually based on a comprehensive trade-off between water mixing characteristics, monitoring accuracy, and computational efficiency. Ideally, this distance can be based on experience; in practice, it is often set as a multiple of the sensor deployment spacing, such as 1.5 to 2 times the grid spacing, to ensure that each node can establish a connection with the physically adjacent point with the most direct hydraulic connection.

[0034] The adjacency matrix at each time step is multiplied and aggregated with the feature matrix at each time step, and then activated using the ReLU function to obtain the first feature matrix: in, Indicates the first The adjacency matrix at each time step; Indicates the first Feature matrices for each time step; Indicates the first The first feature matrix at each time step; In the above process, by using the adjacency matrix With characteristic matrix Multiplication occurs when the comprehensive water quality feature value of each node is linearly combined with the feature values ​​of all its "neighbor" nodes, defined according to distance and feature similarity rules, resulting in a new feature for each node. It not only includes its own information, but also incorporates the water quality status of its surrounding local area. Then, it is activated using the ReLU function, which on the one hand introduces non-linear expressive power, enabling the model to learn more complex spatial relationships; on the other hand, it filters out the negative information after aggregation, ensuring the effectiveness of the features.

[0035] Dilated causal convolution is performed on the first feature matrix at all time steps to extract spatiotemporal features, resulting in the spatiotemporal feature matrix: in, in, Indicates the first The contribution weight of each time step; Indicates the total number of sampling points; Indicates the first The spatiotemporal feature matrix at each time step.

[0036] In the above process, by introducing an attention mechanism into dilated causal convolution, adaptive weighted fusion of temporal features and effective extraction of long-term dependencies are achieved. This is achieved by using the first feature matrix at each time step... Input a multilayer perceptron and generate dynamic weights using the Sigmoid function. This process automatically adjusts the contribution of features within the historical window based on their importance at each moment, focusing on more informative or anomalous time points. Subsequently, causal convolution with a dilation factor is used to perform strafing sampling and aggregation of these weighted features along the temporal dimension. This expands the receptive field to capture longer-term temporal evolution patterns and periodic regularities while maintaining temporal causality—that is, utilizing only current and historical information. Finally, it outputs a feature matrix that integrates key spatiotemporal information. This provides a deeply processed spatiotemporal semantic representation for subsequent risk assessment.

[0037] S5: The spatiotemporal feature matrix and the free ammonia concentration data are combined to form an ammonia concentration matrix. After standardization, the Mahalanobis distance method is used to calculate the preliminary toxicity index vector. The mean scalar of the preliminary toxicity vector and the free ammonia concentration matrix is ​​calculated. Based on the mean scalar, the weighted fusion method is used to calculate the comprehensive risk index. Based on the index, water quality risk classification and early warning are carried out, and the corresponding risk warning information is output.

[0038] The steps to generate a preliminary toxicity vector are as follows: The spatiotemporal feature matrix and free ammonia concentration data are combined to form an ammonia concentration matrix. This matrix is ​​then standardized and concatenated column-by-column. The mean scalar of each column of the concatenated matrix is ​​calculated. Based on the mean, the Mahalanobis distance vector of the concatenated matrix at each time step is calculated. The preliminary toxicity index vector is further calculated using the Mahalanobis distance method, based on the following formula: in, Indicates the first The Mahalanobis distance vector at each time step; Indicates the first [number] after splicing The matrix mean at each time step; Indicates the first [number] after splicing A matrix for each time step; Indicates the first The initial toxicity index vector at each time step.

[0039] In the above process, the fused data of high-dimensional spatiotemporal features and free ammonia concentration is mapped to a toxicity index reflecting the degree to which the overall water quality deviates from the normal baseline using Mahalanobis distance. This is achieved by calculating the Mahalanobis distance between the fused feature matrix and its mean vector at each time step. This method not only considers the absolute deviation of each feature dimension, but also uses the covariance inverse matrix. By eliminating correlations between features and considering variability across dimensions, a "comprehensive outlier" in a multivariate statistical sense is obtained, which is then expressed using a formula. The distance is normalized into a toxicity index between 0 and 1, whose value increases monotonically with the increase of the Mahalanobis distance. This intuitively quantifies the comprehensive deviation of the current water quality status from the historical normal pattern, and realizes the effective condensation from multi-dimensional data to a single risk scalar. The independent variable of the formula is the square of the Mahalanobis distance. and scale parameters .choose It is the core independent variable because it is expressed through the formula. The calculation shows that it is itself a comprehensive multivariate statistical distance that integrates the mean deviation, variance, and correlation between features across various feature dimensions. It is the currently observed fusion feature matrix. Its mean vector represents the baseline state. It is the inverse of the covariance matrix, and the scale parameter The sensitivity of the exponential function to changes in distance is used to control its degree of sensitivity. Follow The rate of growth; in the formula In the middle, the dependent variable With independent variable A positive correlation exists: the larger the Mahalanobis distance, the greater the deviation between the current water quality state and the baseline state, and the higher the calculated toxicity index. The larger it is. At the same time, the dependent variable... With scale parameters Negative correlation: in the same Down, The larger the value, the slower the exponential function decays, and the more accurate the calculated value becomes. The smaller the value, the better. This function form ensures... The range of is [0,1), and it maps monotonically increasing. .

[0040] The specific steps for classifying and issuing early warnings based on the comprehensive risk index are as follows: The mean scalar of the ammonia concentration matrix, consisting of the preliminary toxicity vector and the free ammonia concentration at each time step, is calculated separately. Based on the mean scalar, a weighted fusion method is used to calculate the comprehensive risk index, and the comprehensive risk index is used for classification and early warning. in, Indicates the first A comprehensive risk index for each time step; Indicates the first The mean scalar of the initial toxicity vector at each time step; Indicates the first The mean scalar of the ammonia concentration matrix at each time step; In the above process, the mean of the preliminary toxicity vector and the mean of the free ammonia concentration are calculated first. The former represents the overall water quality anomaly degree after in-depth spatiotemporal feature extraction and multivariate statistical testing, while the latter directly reflects the absolute exposure level of the core toxic substances. Averaging these two values ​​aims to eliminate local fluctuations between spatial nodes and obtain a stable scalar that represents the global average risk status of the entire aquaculture pond at that moment. A weight ratio of 0.6 and 0.4 is used because the preliminary toxicity vector integrates multi-parameter spatiotemporal anomaly information, including free ammonia, making it more comprehensive and forward-looking in early warning. The mean of the free ammonia concentration serves as a verification and calibration benchmark with direct toxicological significance, ensuring that the assessment results do not deviate from the core biological toxicity facts. This weight allocation reflects the risk assessment logic of "overall anomaly early warning as the primary focus, core toxicity confirmation as a secondary consideration," improving early warning sensitivity while ensuring the toxicological reliability of the assessment results. when A value less than 0.5 indicates that the aquaculture pond environment is normal. When the value is equal to 0.5 and less than 0.8, it indicates an abnormal environment in the aquaculture pond and a risk. When the value is equal to 0.8, the environment of the aquaculture pond is poor and an emergency warning is required.

[0041] In the aforementioned process, by setting two key thresholds of 0.5 and 0.8, this grading mechanism divides risk assessment results into three progressive levels: "normal," "abnormal risk," and "severe requiring emergency warning," corresponding to different management response strategies such as routine monitoring, intervention, and emergency response. This not only provides aquaculture managers with clear and timely criteria for judging risk status, avoiding misjudgments or delays caused by ambiguous indicators, but also establishes a direct correlation between the severity of risk and the intensity of response measures through tiered warnings. This drives refined and tiered risk management, effectively improving the intelligence level of aquaculture environment monitoring and emergency response efficiency.

[0042] Please see Figure 3 The present invention also provides an aquaculture monitoring system based on multimodal sensing technology, the monitoring system being used to perform the above-described monitoring method, comprising: The data acquisition module is used to divide the aquaculture pond into grids along the depth, width and length directions at fixed intervals, mark each grid node, set the time window length and time step, and collect the dissolved oxygen concentration, pH value, temperature and ammonia concentration of each grid node at each time step; The parameter calculation module is used to calculate the free ammonia concentration at each time step based on the pH value and temperature of each grid node using the dynamic dissociation equilibrium method. It normalizes and maps all four types of parameters, namely dissolved oxygen concentration, pH value, temperature and free ammonia concentration, into a set interval, divides the interval into multiple sub-intervals of equal length, counts the frequency ratio of each type of parameter in each sub-interval, and calculates the information entropy of each type of parameter. The feature matrix acquisition module is used to set the toxicity factor based on the information entropy result, calculate the weight factor of various parameters at each time step, and perform weighted summation of the normalized free ammonia concentration, dissolved oxygen concentration, pH value and temperature of each grid node with the corresponding weight factor to obtain the comprehensive water quality feature value of each grid node at each time step, and construct the feature matrix with the comprehensive water quality feature values ​​of all grid nodes at one time step. The spatiotemporal matrix acquisition module is used to set distance threshold rules, calculate the spatial relationship between grid nodes, construct an adjacency matrix, aggregate the adjacency matrix and the feature matrix by matrix multiplication, activate it using the ReLU function to obtain the first feature matrix, perform dilated causal convolution on the first feature matrix at all time steps to extract spatiotemporal features and obtain the spatiotemporal feature matrix. The risk warning module is used to construct an ammonia concentration matrix by combining the spatiotemporal feature matrix with the free ammonia concentration data, perform standardization processing, calculate the preliminary toxicity index vector using the Mahalanobis distance method, calculate the mean scalar of the preliminary toxicity vector and the free ammonia concentration matrix, calculate the comprehensive risk index using the weighted fusion method based on the mean scalar, and perform water quality risk classification and warning based on the index, outputting the corresponding risk warning information.

[0043] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0044] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0045] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0046] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring aquaculture based on multimodal sensing technology, characterized by the following steps: include: S1: Divide the aquaculture pond into grids along the depth, width and length directions at fixed intervals, mark each grid node, set the time window length and time step, and collect the dissolved oxygen concentration, pH value, temperature and ammonia concentration of each grid node at each time step; S2: At each time step, the free ammonia concentration is calculated using the dynamic dissociation equilibrium method based on the pH value and temperature of each grid node. All four types of parameters, namely dissolved oxygen concentration, pH value, temperature and free ammonia concentration, are normalized and mapped to a set interval. The interval is divided into multiple sub-intervals of equal length. The frequency ratio of each type of parameter in each sub-interval is counted, and the information entropy of each type of parameter is calculated. S3: Based on the information entropy result, set the toxicity factor of the parameter, calculate the weight factor of each parameter at each time step, and weight and sum the normalized free ammonia concentration, dissolved oxygen concentration, pH value and temperature of each grid node with the corresponding weight factor to obtain the comprehensive water quality characteristic value of each grid node at each time step, and construct the feature matrix with the comprehensive water quality characteristic values ​​of all grid nodes at one time step. S4: Set distance threshold rules, calculate the spatial relationship between grid nodes, construct the adjacency matrix, perform matrix multiplication and aggregation on the adjacency matrix and the feature matrix, activate with the ReLU function to obtain the first feature matrix, perform dilated causal convolution on the first feature matrix at all time steps to extract spatiotemporal features and obtain the spatiotemporal feature matrix; S5: Construct an ammonia concentration matrix by combining the spatiotemporal feature matrix with the free ammonia concentration data, perform standardization processing, calculate the preliminary toxicity index vector using the Mahalanobis distance method, and calculate the mean scalar of the preliminary toxicity vector and the free ammonia concentration matrix. Based on the mean scalar, a weighted fusion method is used to calculate the comprehensive risk index, and water quality risk classification and early warning are carried out based on the index, outputting the corresponding risk warning information.

2. The aquaculture monitoring method based on multimodal sensing technology according to claim 1, characterized in that, The specific steps for grid division and data acquisition are as follows: The aquaculture pond is divided into grids along the depth, width, and length directions according to a preset fixed interval. Each grid node is marked, and a measurement point is set at each grid node. The time window length is set, and the dissolved oxygen concentration, pH value, temperature, and ammonia concentration of the measurement point are collected at preset time steps.

3. The aquaculture monitoring method based on multimodal sensing technology according to claim 2, characterized in that, The specific steps for calculating the free ammonia concentration are as follows: Calculate the percentage of free ammonia based on the pH value and temperature at each grid point in each time step: in, Indicates the first The first time step The dissociation constant of free ammonia in each grid cell; Indicates the first The first time step Temperature of each grid cell; Indicates the first The first time step The proportion of free ammonia to total ammonia nitrogen in each grid cell; Indicates the first The first time step pH of each grid cell; Indicates the time step index; The concentration of free ammonia was calculated using the dynamic dissociation equilibrium method. in, Indicates the first The first time step NH4 measured in each grid cell + concentration; Indicates the first The first time step The concentration of free ammonia in each grid cell.

4. The aquaculture monitoring method based on multimodal sensing technology according to claim 3, characterized in that, The specific steps for calculating the information entropy of each type of parameter are as follows: All four parameters—dissolved oxygen concentration, pH value, temperature, and free ammonia concentration—were normalized and mapped to a set interval. The interval was then divided into multiple equal-length sub-intervals, and the frequency percentage of each parameter in each sub-interval was calculated. in, Indicates the total number of grid nodes; Indicates the length of the time window; Indicates the interval index number; Indicates the first The normalized first grid node of all grid nodes at the nth time step Class parameter value, , The time indicates the concentration of free ammonia. The time indicates the dissolved oxygen concentration. The time indicates the pH value. Time indicates temperature; Indicates the first The first time step The class parameter value falls into the first The probability of the frequency within a certain interval relative to the total frequency; Calculate the information entropy for each type of parameter: in, Indicates the first The first time step Information entropy of class parameter values; Represents the total number of intervals 。 5. The aquaculture monitoring method based on multimodal sensing technology according to claim 4, characterized in that, The specific steps for calculating the weighting factors of various parameters at each time step are as follows: Based on the information entropy results, a toxicity factor is set, and the weighting factors of various parameters at each time step are calculated: in, Indicates the first The toxicity factor of class parameter values, and , , as well as ; Indicates the first The maximum information entropy of all categories of parameters at each time step; Indicates the first The first time step Weighting factors for class parameter values.

6. The aquaculture monitoring method based on multimodal sensing technology according to claim 5, characterized in that, The specific steps for calculating the comprehensive water quality characteristic value for each grid node at each time step are as follows: The normalized free ammonia concentration, dissolved oxygen concentration, pH value, and temperature of each grid node at each time step are weighted and summed with weighting factors to obtain the comprehensive water quality characteristic value of each grid node at each time step: Indicates the normalized i-th The first time step The first grid node Class parameter value, , The time indicates the concentration of free ammonia. The time indicates the dissolved oxygen concentration. The time indicates the pH value. Time indicates temperature; Indicates the first The first time step The comprehensive water quality characteristic value of each grid node; This indicates the number of grid node indices.

7. The aquaculture monitoring method based on multimodal sensing technology according to claim 6, characterized in that, The specific steps for calculating the spatiotemporal feature matrix at each time step are as follows: At one time step, a feature matrix is ​​constructed using the comprehensive water quality feature values ​​of all grid nodes. Distance threshold rules are set, spatial relationships between grid nodes are calculated, and an adjacency matrix is ​​constructed. Specifically, if the distance between any two grid nodes is less than a preset distance and the difference in comprehensive water quality features between grid nodes at each time step does not exceed a preset difference, the value is set to 1; otherwise, it is set to 0. The adjacency matrix between all grid nodes at each time step is constructed sequentially. The adjacency matrix at each time step is multiplied and aggregated with the feature matrix at each time step, and then activated using the ReLU function to obtain the first feature matrix: in, Indicates the first The adjacency matrix at each time step; Indicates the first Feature matrices for each time step; Indicates the first The first feature matrix at each time step; Dilated causal convolution is performed on the first feature matrix at all time steps to extract spatiotemporal features, resulting in the spatiotemporal feature matrix: in, in, Indicates the first The contribution weight of each time step; Indicates the total number of sampling points; Indicates the first The spatiotemporal feature matrix at each time step.

8. The aquaculture monitoring method based on multimodal sensing technology according to claim 7, characterized in that, The steps to generate a preliminary toxicity vector are as follows: The spatiotemporal feature matrix and free ammonia concentration data are combined to form an ammonia concentration matrix. This matrix is ​​then standardized and concatenated column-by-column. The mean scalar of each column of the concatenated matrix is ​​calculated. Based on the mean, the Mahalanobis distance vector of the concatenated matrix at each time step is calculated. The preliminary toxicity index vector is further calculated using the Mahalanobis distance method, based on the following formula: in, Indicates the first The Mahalanobis distance vector at each time step; Indicates the first [number] after splicing The matrix mean at each time step; Indicates the first [number] after splicing A matrix for each time step; Indicates the first The initial toxicity index vector at each time step; Indicates the scale parameter.

9. The aquaculture monitoring method based on multimodal sensing technology according to claim 1, characterized in that, The specific steps for classifying and issuing early warnings based on the comprehensive risk index are as follows: The mean scalar of the ammonia concentration matrix, consisting of the preliminary toxicity vector and the free ammonia concentration at each time step, is calculated separately. Based on the mean scalar, a weighted fusion method is used to calculate the comprehensive risk index, and the comprehensive risk index is used for classification and early warning. in, Indicates the first A comprehensive risk index for each time step; Indicates the first The mean scalar of the initial toxicity vector at each time step; Indicates the first The mean scalar of the ammonia concentration matrix at each time step; when A value less than 0.5 indicates that the aquaculture pond environment is normal. When the value is equal to 0.5 and less than 0.8, it indicates an abnormal environment in the aquaculture pond and a risk. When the value is equal to 0.8, the environment of the aquaculture pond is poor and an emergency warning is required.

10. An aquaculture monitoring system based on multimodal sensing technology, characterized in that: The monitoring system is used to perform the monitoring method according to any one of claims 1-9, including: The data acquisition module is used to divide the aquaculture pond into grids along the depth, width and length directions at fixed intervals, mark each grid node, set the time window length and time step, and collect the dissolved oxygen concentration, pH value, temperature and ammonia concentration of each grid node at each time step; The parameter calculation module is used to calculate the free ammonia concentration at each time step based on the pH value and temperature of each grid node using the dynamic dissociation equilibrium method. It normalizes and maps all four types of parameters, namely dissolved oxygen concentration, pH value, temperature and free ammonia concentration, into a set interval, divides the interval into multiple sub-intervals of equal length, counts the frequency ratio of each type of parameter in each sub-interval, and calculates the information entropy of each type of parameter. The feature matrix acquisition module is used to set the toxicity factor based on the information entropy result, calculate the weight factor of various parameters at each time step, and perform weighted summation of the normalized free ammonia concentration, dissolved oxygen concentration, pH value and temperature of each grid node with the corresponding weight factor to obtain the comprehensive water quality feature value of each grid node at each time step, and construct the feature matrix with the comprehensive water quality feature values ​​of all grid nodes at one time step. The spatiotemporal matrix acquisition module is used to set distance threshold rules, calculate the spatial relationship between grid nodes, construct an adjacency matrix, aggregate the adjacency matrix and the feature matrix by matrix multiplication, activate it using the ReLU function to obtain the first feature matrix, perform dilated causal convolution on the first feature matrix at all time steps to extract spatiotemporal features and obtain the spatiotemporal feature matrix. The risk warning module is used to construct an ammonia concentration matrix by combining the spatiotemporal feature matrix with the free ammonia concentration data, perform standardization processing, calculate the preliminary toxicity index vector using the Mahalanobis distance method, calculate the mean scalar of the preliminary toxicity vector and the free ammonia concentration matrix, calculate the comprehensive risk index using the weighted fusion method based on the mean scalar, and perform water quality risk classification and warning based on the index, outputting the corresponding risk warning information.

Images