Sea area plankton density prediction method based on environmental factors and acoustics

Abstract

The invention discloses a sea area plankton density prediction method based on environmental factors and acoustics, and belongs to the technical field of marine organism prediction.The sea area plankton density prediction method comprises the steps that multi-source data are obtained from a target sea area, and the multi-source data comprise acoustic echo data and environmental factor data; extracting a noise level from the acoustic echo data; judging whether the noise level exceeds a preset threshold value or not, and if yes, performing denoising processing on the acoustic echo data to obtain a pure acoustic signal feature set; normalizing the features in the pure sound wave signal feature set and the environmental factor data; and obtaining an estimation result of the plankton distribution density through a multiple regression equation according to the normalized pure acoustic signal characteristics and the environmental factor data. The sea area plankton density prediction method based on environmental factors and acoustics solves the problem of inaccurate prediction caused by complex marine environment in the aspect of plankton density prediction at present.

Description

Technical Field

[0001] This invention relates to the field of marine organism prediction technology, and in particular to a method for predicting marine planktonic density based on environmental factors and acoustics. Background Technology

[0002] The health and stability of marine ecosystems are crucial for the sustainable development of the global environment and human society. As the foundation of the marine food chain, plankton plays an irreplaceable role in maintaining ecological balance and the sustainability of fishery resources. However, in recent years, climate change, environmental pollution, and human activities such as the construction of marine facilities have profoundly impacted the marine environment, leading to significant changes in the distribution and density of plankton. This not only threatens the stability of ecosystems but may also pose potential risks to the safety of marine engineering projects.

[0003] Traditional methods for monitoring phytoplankton density are often limited to measuring environmental factors such as chlorophyll a concentration, turbidity, and dissolved oxygen. These methods struggle to address the complex environmental changes across large marine areas and cannot meet the demands of real-time dynamic monitoring. While some methods utilize sound wave reflection for phytoplankton monitoring, these only reflect the distribution of phytoplankton communities. The signals are affected by various interference factors in the marine environment, making it difficult to directly convert them into reliable density data. Therefore, current phytoplankton density predictions suffer from inaccuracies due to the complexity of the marine environment. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides a method for predicting marine plankton density based on environmental factors and acoustics, in order to solve the aforementioned problems.

[0005] The technical solution adopted by this invention to solve its technical problem is: a method for predicting marine plankton density based on environmental factors and acoustics, comprising the following steps: S1: Acquire multi-source data from the target sea area, the multi-source data including acoustic echo data and environmental factor data; S2: Extract noise levels from acoustic echo data; S3: Determine whether the noise level exceeds a preset threshold. If it does, perform noise reduction processing on the acoustic echo data to obtain a clean acoustic signal feature set. S4: Normalize the feature and environmental factor data in the feature set of pure acoustic wave signals; S5: Based on the characteristics of the normalized pure acoustic signal and environmental factor data, the estimated distribution density of plankton is obtained through a multiple regression equation.

[0006] Preferably, in step S2, an acoustic echo data with the same timestamp is classified based on environmental factor data using a classification algorithm to obtain acoustic echo data with different labels; then, the corresponding noise level is extracted based on the acoustic echo data with the same label. In step S3, it is determined whether the noise level of acoustic echo data with different labels exceeds the corresponding preset threshold, which is determined according to the category of acoustic echo data.

[0007] Optionally, in step S2, the environmental factor data includes chlorophyll a concentration, turbidity, and dissolved oxygen. In step S2, chlorophyll a concentration, turbidity, and dissolved oxygen at the same time point are spliced ​​together to obtain spliced ​​data, and the spliced ​​data is used to form an environmental factor set. The environmental factor set is initially classified using KMeans clustering from the Scikit-learn library to obtain preliminary clustering results. Based on the preliminary clustering results, corresponding pseudo-labels are generated, and each environmental factor data is assigned a corresponding pseudo-label to obtain environmental factor data with pseudo-labels. XGBoost is then trained using the environmental factor data with pseudo-labels to obtain the trained XGBoost. The steps for supervised classification using trained XGBoost include: using the environmental factor data as input to trained XGBoost, and outputting the classified environmental factor data and corresponding labels.

[0008] Specifically, in step S2, based on the classified environmental factor data, acoustic echo data with the same timestamp is obtained and assigned a label of the same category as the classified environmental factor data.

[0009] It is worth noting that in step S2, acoustic echo data of different labels are combined into different acoustic feature sets, and the noise level is determined by calculating the standard deviation of acoustic echo data in different acoustic feature sets.

[0010] Preferably, in step S3, a preset threshold corresponding to the label of the acoustic feature set is obtained from a preset threshold library; Based on a preset threshold, it is determined whether the noise level exceeds the threshold, and a determination result is obtained; If the judgment result shows an excess, a preset denoising method is used to process the sound wave feature set to obtain a clean sound wave signal feature set.

[0011] Optionally, in step S4, a normalization method is used to adjust the characteristics, chlorophyll a content, turbidity, and dissolved oxygen amplitude of the pure sound wave signal feature set, respectively, to obtain the normalized pure sound wave signal feature set, chlorophyll a content set, turbidity set, and dissolved oxygen set, wherein the data amplitudes in the normalized pure sound wave signal feature set, chlorophyll a content set, turbidity set, and dissolved oxygen set are all within the preset interval [0,1].

[0012] It is worth noting that in step S5, a prediction model is established using the linear regression algorithm from the sklearn library. This model uses the normalized set of pure acoustic signal features, chlorophyll a content set, turbidity set, and dissolved oxygen set as input variables, and the plankton distribution density collected at the same time point as the output variable. The relationship between the input and output variables is fitted using the least squares method to obtain the regression coefficient matrix β. The least squares method is achieved by solving the formula... To determine the regression coefficients, X represents the input variable matrix, and y represents the output variable vector. Represents the covariance matrix, matrix Each element in the matrix is ​​the inner product of the feature column corresponding to the input variable matrix and the target vector y; After normalizing the newly collected acoustic echo data, chlorophyll a content concentration, oxygen demand data, and dissolved oxygen data, the data were input into the prediction model to obtain the estimated results of phytoplankton distribution density.

[0013] Specifically, after step S5, step S6 is also included, which includes: for the estimated result of the plankton distribution density, if the estimated result of the plankton distribution density is higher than a preset density threshold, an alarm is triggered.

[0014] The beneficial effects of this invention are as follows: The method for predicting marine plankton density based on environmental factors and acoustics acquires acoustic echo data and environmental factor data from the target sea area. Then, based on the environmental factor data, a classification algorithm is used to classify the acoustic echo data, extracting the noise level and determining whether it exceeds a preset threshold specific to the category. If it does, a filtering method is applied to remove noise, obtaining a clean acoustic signal feature set. Subsequently, the features of the clean feature set and the environmental factor data are normalized, and a multiple regression equation is constructed to obtain the estimated plankton distribution density, thus solving the problems of data noise interference and multivariate correlation analysis. The technical effect of this solution is that it achieves high-precision and real-time plankton density estimation, improving the efficiency and accuracy of marine ecological monitoring. Attached Figure Description

[0015] Figure 1This is a flowchart of a marine planktonic density prediction method based on environmental factors and acoustics in one embodiment of the present invention. Figure 2 This is a flowchart illustrating the steps of step S2 in one embodiment of the present invention; Figure 3 This is a flowchart of a marine plankton density prediction method based on environmental factors and acoustics, as described in another embodiment of the present invention. Detailed Implementation

[0016] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0017] like Figure 1-3 As shown, a method for predicting marine planktonic density based on environmental factors and acoustics includes the following steps: S1: Acquire multi-source data from the target sea area, the multi-source data including acoustic echo data and environmental factor data; S2: Extract noise levels from acoustic echo data; S3: Determine whether the noise level exceeds a preset threshold. If it does, perform noise reduction processing on the acoustic echo data to obtain a clean acoustic signal feature set. S4: Normalize the feature and environmental factor data in the feature set of pure acoustic wave signals; S5: Based on the characteristics of the normalized pure acoustic signal and environmental factor data, the estimated distribution density of plankton is obtained through a multiple regression equation.

[0018] This solution addresses the ecological risks that may be caused by abnormally high phytoplankton density in the marine environment, such as blockage of nuclear power plant cold source intakes. The method integrates these business scenarios into a logically related core problem: how to accurately estimate phytoplankton distribution density through real-time multi-source data analysis and provide timely warnings of potential safety risks to nuclear power plant cold source intakes.

[0019] The proposed method for predicting marine plankton density based on environmental factors and acoustics acquires acoustic echo data and environmental factor data (chlorophyll a concentration, turbidity, and dissolved oxygen) from the target sea area. Then, based on the environmental factor data, a classification algorithm is used to classify the acoustic echo data, extracting noise levels and determining whether they exceed category-specific preset thresholds. If they do, filtering methods are applied to remove noise, resulting in a clean acoustic signal feature set. Subsequently, the features of the clean feature set and the environmental factor data are normalized, and a multiple regression equation is constructed to obtain the estimated plankton distribution density, thus solving the problems of data noise interference and multivariate correlation analysis. The technical advantage of this scheme is that it achieves high-precision and real-time plankton density estimation, improving the monitoring efficiency of marine plankton at nuclear power plant cold source intakes.

[0020] In this embodiment, a portable split-beam scientific fish finder, either a Simrad EK80 or EY60, was used to collect acoustic echo data of the sea area. The acoustic equipment was fixed to the port side midway using a mounting device. The acoustic equipment was calibrated according to specifications before the survey began. Characteristic values ​​such as chlorophyll a concentration, turbidity, and dissolved oxygen were obtained using real-time online monitoring equipment such as a multi-parameter water quality analyzer. In this scheme, chlorophyll a content is an important indicator parameter of phytoplankton biomass in seawater; turbidity can reflect the total amount of suspended particulate matter in the water to a certain extent; dissolved oxygen (DO) is fundamental to the survival of marine life, and its concentration directly affects the ecological capacity and biological productivity of the water body.

[0021] It is worth noting that in step S2, a classification algorithm is used to classify acoustic echo data with the same timestamp based on environmental factor data to obtain acoustic echo data with different labels; then, the corresponding noise level is extracted based on the acoustic echo data with the same label. In step S3, it is determined whether the noise level of acoustic echo data with different labels exceeds the corresponding preset threshold, which is determined according to the category of acoustic echo data.

[0022] Preferably, in step S2, the environmental factor data includes chlorophyll a concentration, turbidity, and dissolved oxygen. In step S2, chlorophyll a concentration, turbidity, and dissolved oxygen at the same time point are spliced ​​together to obtain spliced ​​data, and an environmental factor set is obtained by assembling the spliced ​​data. The environmental factor set is initially classified using KMeans clustering from the Scikit-learn library to obtain preliminary clustering results. Based on the preliminary clustering results, corresponding pseudo-labels are generated, and each environmental factor data is assigned a corresponding pseudo-label to obtain environmental factor data with pseudo-labels. XGBoost is then trained using the environmental factor data with pseudo-labels to obtain the trained XGBoost. The steps for supervised classification using trained XGBoost include: using the environmental factor data as input to trained XGBoost, and outputting the classified environmental factor data and corresponding labels.

[0023] In this embodiment, KMeans is an unsupervised learning algorithm that iteratively optimizes the allocation of data points into K clusters, where the center of each cluster is the mean of all points in that cluster. Specifically, when processing environmental factor data including chlorophyll a concentration, turbidity, and dissolved oxygen, the data are first standardized to eliminate dimensional differences. Then, chlorophyll a concentration, turbidity, and dissolved oxygen at the same time point are concatenated. A suitable K value is then selected, for example, by observing the change curve of the sum of squares within the cluster using the elbow method. When the curve tends to flatten, K=3 is selected, indicating that the data is divided into three environmental quality categories. Next, the KMeans clustering algorithm initializes the cluster centers, calculates the Euclidean distance from each data point to the center, assigns the point to the nearest cluster, and updates the center position. This process is repeated until the center no longer moves significantly, thus obtaining the preliminary clustering results. For example, in marine monitoring operations, a sample at a certain timestamp contains chlorophyll a concentration of 0.05 mg / L, turbidity of 2.5 mg / L, and dissolved oxygen of 6.0 mg / L. The resulting spliced ​​data is 0.05-2.5-6.0. After processing with KMeans, these samples are divided into three clusters: high pollution, moderate pollution, and low pollution. Pseudo-labels such as 0, 1, and 2 are assigned according to the cluster assignment, thus forming spliced ​​data with pseudo-labels. Then, the pseudo-labels from the spliced ​​data with pseudo-labels are assigned to the three environmental factor data in the spliced ​​data, resulting in environmental factor data without labels.

[0024] XGBoost is an ensemble learning framework based on gradient boosting decision trees. It improves prediction accuracy by constructing multiple decision trees and combining them in a weighted manner. The training process involves minimizing loss functions such as cross-entropy and using regularization to prevent overfitting.

[0025] Specifically, the environmental factor data with pseudo-labels is first divided into training and validation sets, for example, 80% is used for training. The input features are environmental factor data such as chlorophyll a concentration, turbidity and dissolved oxygen, and the target is pseudo-labels. Then, parameters are set such as learning rate of 0.1 and maximum depth of 5. XGBoost starts from a simple tree, calculates the residuals and builds the next tree to fit the residuals. Iterates multiple times until the validation set performance is optimal, thus obtaining the trained XGBoost model.

[0026] When using trained XGBoost for supervised classification, newly acquired environmental factor data (chlorophyll a concentration, turbidity, and dissolved oxygen) are used as input to XGBoost, and the output is the classified environmental factor data and its corresponding label. Specifically, the entire environmental factor data is input, and the trained XGBoost traverses each tree to calculate the score and sum them to obtain the classification label for each sample. For example, highly polluted samples are labeled as category 0, thereby achieving refined data classification. This method improves the robustness of classification because XGBoost can handle non-linear relationships.

[0027] Optionally, in step S2, based on the classified environmental factor data, acoustic echo data with the same timestamp is obtained and assigned a label of the same category as the classified environmental factor data.

[0028] In this embodiment, timestamps are first matched. By acquiring acoustic echo data with timestamps consistent with the classified environmental factor data, it is ensured that the acquired acoustic echo data was collected at the same time as the environmental factor data. In this case, the label category of the environmental factor data is the label category of the acoustic signal feature. For example, in marine ecological monitoring, a set of newly collected environmental factor data containing chlorophyll a concentration, turbidity, and dissolved oxygen is input. The trained XGBoost output label is 1, indicating moderate pollution. This label is then extended to the associated acoustic echo data, indicating that the acoustic echo data collected at the same timestamp belongs to the data obtained under moderate pollution conditions.

[0029] Specifically, in step S2, acoustic echo data of different labels are combined into different acoustic feature sets, and the noise level is determined by calculating the standard deviation of the acoustic echo data in the different acoustic feature sets.

[0030] Here, standard deviation is a statistical indicator used to measure the dispersion of data points relative to the mean. The calculation process involves first obtaining the average value of all acoustic echo data within the acoustic feature set, then calculating the sum of squares of the differences between each acoustic echo data and the average value, and finally taking the square root and dividing it by the total number of samples in the acoustic feature set minus one, thereby obtaining the noise level used to quantify the noise.

[0031] It is worth noting that in step S3, a preset threshold corresponding to the label of the acoustic feature set is obtained from the preset threshold library; Based on a preset threshold, it is determined whether the noise level exceeds the threshold, and a determination result is obtained; If the judgment result shows an excess, a preset denoising method is used to process the sound wave feature set to obtain a clean sound wave signal feature set.

[0032] In this embodiment, the preset threshold library is a database that stores noise thresholds under various labels. For example, in a marine ecological monitoring system, this library may contain thresholds such as 3.0 for high pollution labels, 2.5 for medium pollution labels, and 1.5 for low pollution labels. These thresholds are derived statistically from historical acoustic data and are used to assess the degree of signal interference. Specifically, the acquisition process involves first obtaining the labels of the acoustic feature set, for example, an acoustic feature set is labeled as high pollution, and then querying the preset threshold library for matching entries. The corresponding values ​​are extracted using a database retrieval function. This process ensures the specificity of the preset threshold library.

[0033] In one possible implementation, a preset threshold is used to determine whether the noise level exceeds the limit, resulting in a judgment. The core of this judgment is comparing the previously calculated noise level with the preset threshold extracted from a preset threshold library. If the noise level exceeds the threshold, it is considered to exceed the limit. If the judgment result indicates an exceedance, a preset denoising method is used to process the acoustic feature set to obtain a clean acoustic signal feature set. The preset denoising method is the `medfilt` function from the SciPy library, a median filtering technique. Its principle is to replace each data point with the median value within the window using a sliding window, thereby smoothing out abnormal fluctuations.

[0034] Specifically, in the processing, for a set of acoustic features labeled as highly polluted, such as -45dB, -50dB, -60dB and some noise points, first define a window size of 3, and then apply a mean-value filter to the sequence in the acoustic feature set point by point. For example, the mean value of -45dB, -50dB, and -60dB in the window is -50dB, and this value is used to replace the center point. This method is particularly useful in marine environmental monitoring operations because it can preserve the edge features of the signal without introducing too much distortion.

[0035] This denoising process can be extended to multi-label scenarios. For example, for a set of low-pollution acoustic features, if the noise level does not exceed the preset threshold corresponding to low pollution, it can be skipped directly. However, for a set of acoustic features that exceeds the preset threshold, the spectral data can be further integrated by combining the medfilt function to obtain a set of clean acoustic signal features.

[0036] Preferably, in step S4, a normalization method is used to adjust the characteristics, chlorophyll a concentration, turbidity, and dissolved oxygen amplitude of the pure sound wave signal feature set, respectively, to obtain the normalized pure sound wave signal feature set, chlorophyll a concentration set, turbidity set, and dissolved oxygen set, respectively. The data amplitudes in the normalized pure sound wave signal feature set, chlorophyll a concentration set, turbidity set, and dissolved oxygen set are all within the preset interval [0,1].

[0037] In this embodiment, normalization is a data preprocessing technique. Its principle is to map data of different dimensions or ranges to a unified interval. It is usually achieved through the minimum-maximum normalization formula, that is, subtract the minimum value from each data point and divide by the difference between the maximum value and the minimum value, so that all values ​​fall between [0,1].

[0038] Specifically, in marine ecological monitoring systems, this method is first applied to the characteristics of a pure acoustic signal feature set. For example, a pure acoustic signal feature set contains characteristic acoustic echo data, the amplitude of which may range from -80dB to -20dB. The system will identify the minimum value of -80dB and the maximum value of -20dB, and then calculate each feature value. For example, a value of -50dB will be converted to (-50-(-80)) / (-20-(-80))=30 / 60=0.5. This process ensures that the relative relationship of the features remains unchanged.

[0039] For chlorophyll a concentration, turbidity, and dissolved oxygen, this normalization is also performed independently for each of these parameters. For example, for chlorophyll a concentration, in a marine pollution monitoring mission, the system collected a series of chlorophyll a concentration values, such as 0.001 mg / L, 0.01 mg / L, 0.02 mg / L, 0.04 mg / L, and 0.05 mg / L. First, the minimum value of 0.001 mg / L and the maximum value of 0.05 mg / L were determined. Then, for 0.01 mg / L, the normalization was calculated as (0.01-0.001) / (0.05-0.001)≈0.018 / 0.049≈0.367. This normalized value falls within the range [0,1], thus forming a set of chlorophyll a concentration values. For example, in marine fisheries resource assessment, for turbidity sequences such as 20 mg / L, 50 mg / L, 100 mg / L, 150 mg / L, and 200 mg / L, the minimum value of 20 mg / L and the maximum value of 200 mg / L are first identified. Then, the formula (50-20) / (200-20) = 30 / 180 ≈ 0.167 is applied to 50 mg / L to obtain the normalized turbidity set. Similarly, in marine biodiversity monitoring, for data points such as 4 mg / L, 6 mg / L, 7 mg / L, 8 mg / L, and 10 mg / L, the system calculates the minimum value of 4 mg / L and the maximum value of 10 mg / L. The 6 mg / L value is then normalized to (6-4) / (10-4) = 2 / 6 ≈ 0.333, producing the dissolved oxygen set.

[0040] Optionally, in step S5, a prediction model is established using the linear regression algorithm from the sklearn library. The prediction model uses the normalized set of pure acoustic signal features, chlorophyll a concentration set, turbidity set, and dissolved oxygen set as input variables, and the plankton distribution density collected at the same time point as the output variable. The relationship between the input and output variables is fitted using the least squares method to obtain the regression coefficient matrix β. The least squares method is obtained by solving the formula... To determine the regression coefficients, X represents the input variable matrix, and y represents the output variable vector. Represents the covariance matrix, matrix Each element in the matrix is ​​the inner product of the feature column corresponding to the input variable matrix and the target vector y; After normalizing the newly collected acoustic echo data, chlorophyll a concentration, oxygen demand data, and dissolved oxygen data, the data were input into the prediction model to obtain the estimated results of phytoplankton distribution density.

[0041] Linear regression is a supervised learning method used to capture the linear relationship between input features and output targets. In the field of marine ecological monitoring, this helps to predict the distribution density of plankton from multi-source data.

[0042] Specifically, when constructing the model, acoustic echo data, chlorophyll a concentration, turbidity, and dissolved oxygen at the same time stamp are extracted from the normalized pure acoustic signal feature set, chlorophyll a concentration set, turbidity set, and dissolved oxygen set as input variables. These variables represent various indicators of the marine environment. Through these inputs, the model learns their correlation with plankton density, thereby providing a reliable predictive basis in actual seabed resource assessment operations.

[0043] Using the distribution density of plankton collected at the same time point as the output variable means that during the training phase, a labeled dataset needs to be prepared, such as the density values ​​of plankton samples collected from historical ocean surveys at the same time point as the input variable. These values ​​are obtained using a CX33 biological microscope, and the unit of plankton density is cells / m³. The output variable corresponds to the input variable, forming training sample pairs.

[0044] In one possible implementation, the relationship between the input and output variables is fitted using the least squares method. This is an optimization technique aimed at minimizing the squared error between the predicted and actual values. Although matrix operations are involved in calculating the regression coefficient matrix β, the sklearn library automatically handles these details in practice. Specifically, in marine pollution assessment operations, the input matrix X consists of normalized multi-column features. For example, a row of data might include acoustic echo data of 0.4, chlorophyll a concentration of 0.3, turbidity of 0.5, and dissolved oxygen of 0.6. The corresponding output y is a plankton density of 15,000. The prediction model adjusts the coefficients through multiple iterations to make the fitted line as close as possible to the data points, which improves the sensitivity to ecological changes in operational practice.

[0045] After extracting the regression coefficient matrix β from the prediction model, the coefficients in the regression coefficient matrix β represent the influence weight of each input variable on the output density. Among them, acoustic echo data, chlorophyll a concentration, turbidity and dissolved oxygen all have corresponding coefficients in the regression coefficient matrix β.

[0046] In one possible implementation, a key step is to normalize the newly input acoustic echo data, chlorophyll a concentration, oxygen demand data, and dissolved oxygen data to ensure they are on the same scale as the training data. Finally, the estimated plankton distribution density is calculated based on the acoustic echo data, chlorophyll a concentration, turbidity, and dissolved oxygen, as well as their coefficients in the regression coefficient matrix β.

[0047] Specifically, after step S5, step S6 is also included, which includes: for the estimated result of the plankton distribution density, if the estimated result of the plankton distribution density is higher than a preset density threshold, an alarm is triggered.

[0048] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A method for predicting marine planktonic density based on environmental factors and acoustics, characterized in that, Includes the following steps: S1: Acquire multi-source data from the target sea area, the multi-source data including acoustic echo data and environmental factor data; S2: Extract noise levels from acoustic echo data; S3: Determine whether the noise level exceeds a preset threshold. If it does, perform noise reduction processing on the acoustic echo data to obtain a clean acoustic signal feature set. S4: Normalize the feature and environmental factor data in the feature set of pure acoustic wave signals; S5: Based on the characteristics of the normalized pure acoustic signal and environmental factor data, the estimated distribution density of plankton is obtained through a multiple regression equation.

2. The method for predicting marine planktonic density based on environmental factors and acoustics according to claim 1, characterized in that: In step S2, an acoustic echo data with the same timestamp is classified based on environmental factor data to obtain acoustic echo data with different labels; then, the corresponding noise level is extracted based on the acoustic echo data with the same label. In step S3, it is determined whether the noise level of acoustic echo data with different labels exceeds the corresponding preset threshold, and the preset threshold is determined according to the category of acoustic echo data.

3. The method for predicting marine planktonic density based on environmental factors and acoustics according to claim 2, characterized in that: In step S2, the environmental factor data includes chlorophyll a concentration, turbidity, and dissolved oxygen. In step S2, chlorophyll a concentration, turbidity, and dissolved oxygen at the same time point are spliced ​​together to obtain spliced ​​data, and the spliced ​​data is used to form an environmental factor set. The environmental factor set is initially classified using KMeans clustering from the Scikit-learn library to obtain preliminary clustering results. Based on the preliminary clustering results, corresponding pseudo-labels are generated, and each environmental factor data is assigned a corresponding pseudo-label to obtain environmental factor data with pseudo-labels. XGBoost is then trained using the environmental factor data with pseudo-labels to obtain the trained XGBoost. The steps for supervised classification using trained XGBoost include: using the environmental factor data as input to trained XGBoost, and outputting the classified environmental factor data and corresponding labels.

4. The method for predicting marine planktonic density based on environmental factors and acoustics according to claim 3, characterized in that: In step S2, based on the classified environmental factor data, acoustic echo data with the same timestamp is obtained and assigned a label of the same category as the classified environmental factor data.

5. The method for predicting marine plankton density based on environmental factors and acoustics according to claim 2, characterized in that: In step S2, acoustic echo data from different labels are combined into different acoustic feature sets, and the noise level is determined by calculating the standard deviation of the acoustic echo data in the different acoustic feature sets.

6. The method for predicting marine plankton density based on environmental factors and acoustics according to claim 5, characterized in that: In step S3, a preset threshold corresponding to the label of the acoustic feature set is obtained from the preset threshold library; Based on a preset threshold, it is determined whether the noise level exceeds the threshold, and a determination result is obtained; If the judgment result shows an excess, a preset denoising method is used to process the sound wave feature set to obtain a clean sound wave signal feature set.

7. The method for predicting marine planktonic density based on environmental factors and acoustics according to claim 3, characterized in that: In step S4, a normalization method is used to adjust the characteristics, chlorophyll a concentration, turbidity, and dissolved oxygen amplitude of the pure sound wave signal feature set, respectively, to obtain the normalized pure sound wave signal feature set, chlorophyll a concentration set, turbidity set, and dissolved oxygen set. The data amplitudes in the normalized pure sound wave signal feature set, chlorophyll a concentration set, turbidity set, and dissolved oxygen set are all within the preset interval [0,1].

8. The method for predicting marine plankton density based on environmental factors and acoustics according to claim 7, characterized in that: In step S5, a prediction model is established using the linear regression algorithm from the sklearn library. This model uses the normalized set of pure acoustic signal features, chlorophyll a concentration, turbidity, and dissolved oxygen as input variables, and the plankton distribution density collected at the same time point as the output variable. The relationship between the input and output variables is fitted using the least squares method to obtain the regression coefficient matrix β. The least squares method is achieved by solving the formula... To determine the regression coefficients, X represents the input variable matrix, and y represents the output variable vector. Represents the covariance matrix, matrix Each element in the matrix is ​​the inner product of the feature column corresponding to the input variable matrix and the target vector y; After normalizing the newly collected acoustic echo data, chlorophyll a concentration, oxygen demand data, and dissolved oxygen data, the data were input into the prediction model to obtain the estimated results of phytoplankton distribution density.

9. The method for predicting marine planktonic density based on environmental factors and acoustics according to claim 1, characterized in that: The process includes step S6 after step S5, which includes: triggering an alarm if the estimated density of the plankton distribution is higher than a preset density threshold.

Images