Deep learning-based offshore inorganic nitrogen remote sensing monitoring method

Abstract

The invention relates to the technical field of offshore inorganic nitrogen concentration testing, and discloses an offshore inorganic nitrogen remote sensing monitoring method based on deep learning, which comprises the following steps: acquiring first geographic space information and first remote sensing image data of a target sea area; extracting the wave band reflectivity of the target wave band from the first remote sensing image data; according to a predetermined inorganic nitrogen concentration correlation factor, the first geographic space information and the extracted wave band reflectivity are transformed, and wave band features and geographic features are generated respectively; the generated wave band features and geographic features are imported into an offshore inorganic nitrogen remote sensing monitoring model, and the offshore inorganic nitrogen remote sensing monitoring model outputs spatial distribution of the inorganic nitrogen concentration of the target sea area; the offshore inorganic nitrogen remote sensing monitoring model is constructed based on a bare eagle search algorithm and a Bayesian algorithm on the basis of the convolutional neural network. Therefore, a novel offshore inorganic nitrogen remote sensing monitoring mode based on deep learning is provided, and the test workload is reduced while the test accuracy is ensured.

Description

Technical Field

[0001] This invention relates to the field of nearshore marine water quality research technology, and in particular to a remote sensing monitoring method for nearshore inorganic nitrogen based on deep learning. Background Technology

[0002] Dissolved inorganic nitrogen (DIN) is a key indicator of eutrophication levels in water bodies. Excessive DIN concentrations have a significant destructive impact on fisheries resource conservation, coastal ecological security, and regional sustainable development. Monitoring DIN concentrations can provide data and decision support for nearshore water quality management. Therefore, accurate monitoring of DIN concentrations is of great importance for the rational utilization of marine resources and ecological protection.

[0003] Currently, the traditional methods in the field of water quality monitoring are mainly field sampling and laboratory analysis. However, these methods have problems such as being time-consuming, costly, and unable to accurately reflect the spatiotemporal changes in water quality. Summary of the Invention

[0004] The purpose of this invention is to address the problems in existing monitoring of dissolved inorganic nitrogen in nearshore waters, such as poor adaptability of remote sensing indices, low model robustness, and insufficient inversion accuracy due to the complex optical composition and strong spatiotemporal heterogeneity of water bodies. This invention proposes a remote sensing monitoring method and model for nearshore inorganic nitrogen based on deep learning.

[0005] In a first aspect, embodiments of the present invention provide a deep learning-based remote sensing monitoring method for nearshore inorganic nitrogen, used in a nearshore inorganic nitrogen remote sensing monitoring model. The method includes: acquiring first geospatial information and first remote sensing image data of a target sea area; extracting the band reflectance of a target band from the first remote sensing image data; transforming the first geospatial information and the band reflectance extracted from the first remote sensing image data according to a predetermined inorganic nitrogen concentration correlation factor to generate band features and geographic features respectively; importing the generated band features and geographic features into the nearshore inorganic nitrogen remote sensing monitoring model, which outputs the spatial distribution of inorganic nitrogen concentration in the target sea area; wherein the nearshore inorganic nitrogen remote sensing monitoring model is constructed based on a convolutional neural network; the nearshore inorganic nitrogen remote sensing monitoring model uses a vulture search algorithm to simulate the three-stage predation behavior of vultures, locking multiple potential areas in the high-dimensional search space of the convolutional neural network hyperparameters to find the starting point for local optimization; and using a Bayesian algorithm to perform local adjustments based on the results of the vulture search algorithm.

[0006] In some embodiments, the method further includes: obtaining measured values ​​of band reflectance and inorganic nitrogen concentration at actual sampling points in the experimental area; combining the band reflectance and geospatial information of the actual sampling points to obtain candidate correlation factors; analyzing the correlation between the candidate correlation factors and the measured values ​​of inorganic nitrogen concentration, and determining the inorganic nitrogen concentration correlation factor from the candidate correlation factors based on the correlation.

[0007] In some embodiments, obtaining the measured values ​​of band reflectance and inorganic nitrogen concentration at the measured sampling points in the experimental area includes: delineating the experimental area, determining the geospatial information of the measured sampling points in the experimental area, wherein the geospatial information includes the longitude, latitude, and seawater depth of the sampling points; and collecting the inorganic nitrogen concentration at the measured sampling points as the measured value of inorganic nitrogen concentration.

[0008] In some embodiments, obtaining the measured values ​​of band reflectance and inorganic nitrogen concentration at the actual sampling points in the experimental area includes: obtaining the acquisition time information of the measured inorganic nitrogen concentration; and obtaining second remote sensing image data that matches the sampling point location information and acquisition time information from the remote sensing dataset.

[0009] In some embodiments, obtaining the measured values ​​of band reflectance and inorganic nitrogen concentration at the measured sampling points in the experimental area includes: extracting the band reflectances B1, B2, B3, B4, B5, B6, and B7 of the measured sampling points from the acquired second remote sensing image data.

[0010] In some embodiments, the inorganic nitrogen concentration correlation factor includes one or more of the following: water depth, longitude, latitude, B1, B4, B1\B2, B1\B3, B1\B5, B3\B4, B1-B2, B1-B3, B1-B6, B1-B7, B2-B4, B3-B4, B4-B5, B4-B6, and B1*B4.

[0011] In some embodiments, the training sample generation step of the nearshore inorganic nitrogen remote sensing monitoring model includes: obtaining the measured values ​​of inorganic nitrogen concentration at actual sampling points; generating training feature data corresponding to the correlation factors of inorganic nitrogen concentration based on the band reflectance B1, B2, B3, B4, B5, B6, and B7 of the actual sampling points and based on the second geospatial information, wherein the training feature data includes training band features and training geographic features; and binding the training feature data of the actual sampling points and the measured values ​​of inorganic nitrogen concentration as training samples, wherein the second geospatial information is the geospatial information of the actual sampling points.

[0012] In some embodiments, the process of simulating the three-stage predation behavior of vultures using a vulture search algorithm, and identifying multiple potential regions in the high-dimensional search space of the convolutional neural network hyperparameters to pinpoint the starting point for local optimization, includes: The search space selection stage: Based on prior information such as the population's optimal position, the population's average position, and the individual's current position, the vulture selects the optimal search space, described as follows: ;in, For the first The updated location of the vulture; The parameter represents the control position change, and its value ranges from [1.5, 2]. A random number between (0, 1); The location of the optimal individual bald eagle; The average position of all individual bald eagles; For the first The current location of the vulture.

[0013] Searching for prey in the search space: Within the selected search space, the vulture flies in a spiral pattern to search for prey, accelerating the search process and seeking the optimal dive position. Its expression is: ;in, and The polar coordinates represent the position of the vulture in its spiral flight, with values ​​ranging from (-1, 1). Indicates the first The current location of the vulture.

[0014] Dive-and-capture prey phase: The vulture uses the optimal dive position as its starting point, accelerating towards its target prey. The position update formula is: ;in, and The polar coordinates of the vulture at this stage; A random number between (0, 1); and These represent the movement intensity of the vulture towards the optimal position and the center position, respectively, with values ​​of [1, 2].

[0015] Secondly, embodiments of the present invention provide a nearshore inorganic nitrogen remote sensing monitoring model based on deep learning, applicable to the method described in any of the first aspects. The inorganic nitrogen remote sensing monitoring model includes: a convolutional neural network adjusting hyperparameters based on a vulture search algorithm and a Bayesian algorithm; wherein the convolutional neural network is trained using training samples to obtain the nearshore inorganic nitrogen remote sensing monitoring model; wherein the vulture search algorithm is used to simulate the three-stage predation behavior of vultures, locking multiple potential regions in the high-dimensional search space of the convolutional neural network hyperparameters to pinpoint the starting point for local optimization; and the Bayesian algorithm is used to further refine the model based on the results of the vulture search algorithm. The process involves local adjustments; specifically, the training sample generation step of the nearshore inorganic nitrogen remote sensing monitoring model includes: acquiring measured inorganic nitrogen concentration values ​​at actual sampling points; generating training feature data corresponding to inorganic nitrogen concentration correlation factors based on the band reflectances B1, B2, B3, B4, B5, B6, and B7 of the actual sampling points, and based on second geospatial information. The training feature data includes training band features and training geographic features; and binding the training feature data and measured inorganic nitrogen concentration values ​​of the actual sampling points as training samples, wherein the second geospatial information is the geospatial information of the actual sampling points.

[0016] In some embodiments, the step of determining the inorganic nitrogen concentration correlation factor includes: obtaining the band reflectance and measured inorganic nitrogen concentration values ​​of the measured sampling points in the experimental area; combining the band reflectance and geospatial information of the measured sampling points to obtain candidate correlation factors; analyzing the correlation between the candidate correlation factors and the measured inorganic nitrogen concentration values, and determining the inorganic nitrogen concentration correlation factor from the candidate correlation factors based on the correlation.

[0017] The deep learning-based remote sensing monitoring method and model for nearshore inorganic nitrogen proposed in this application effectively solves the problem of poor adaptability of remote sensing indices caused by the complexity of optical components in nearshore waters through a multi-source feature fusion mechanism, significantly improving the feature characterization capability of DIN concentration inversion under complex environments. By utilizing a CNN-BES-BO hybrid optimization model and a collaborative mechanism of global search (BES) and local optimization (BO), high-precision dynamic monitoring of DIN concentration is achieved, providing a stable foundation for accurate testing of DIN concentration in nearshore waters, while reducing the workload of testing while ensuring test accuracy. Attached Figure Description

[0018] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0019] Figure 1This is a flowchart of an embodiment of the deep learning-based nearshore inorganic nitrogen remote sensing monitoring method according to the present invention.

[0020] Figure 2 This is a schematic diagram of an embodiment of the deep learning-based nearshore inorganic nitrogen remote sensing monitoring model according to the present invention.

[0021] Figure 3 This is a schematic diagram of an application scenario of the deep learning-based nearshore inorganic nitrogen remote sensing monitoring method according to the present invention.

[0022] Figure 4 This is a schematic diagram of the basic structure of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0023] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0024] Please refer to Figure 1 This illustrates the flowchart of an embodiment of a deep learning-based remote sensing monitoring method for nearshore inorganic nitrogen according to the present invention. Figure 1 The deep learning-based nearshore inorganic nitrogen remote sensing monitoring method shown includes the following steps:

[0025] Step 101: Obtain the first geospatial information and the first remote sensing image data of the target sea area.

[0026] Step 102: Extract the band reflectance of the target band from the first remote sensing image data.

[0027] Step 103: According to the predetermined inorganic nitrogen concentration correlation factor, transform the first geospatial information and the band reflectance extracted from the first remote sensing image data to generate band features and geographic features respectively.

[0028] Step 104: Import the generated band features and geographic features into the nearshore inorganic nitrogen remote sensing monitoring model. The nearshore inorganic nitrogen remote sensing monitoring model outputs the spatial distribution of inorganic nitrogen concentration in the target sea area.

[0029] Among them, the nearshore inorganic nitrogen remote sensing monitoring model is constructed based on a convolutional neural network; the nearshore inorganic nitrogen remote sensing monitoring model uses the vulture search algorithm to simulate the three-stage behavior of vultures preying on each other, and locks down multiple potential regions in the high-dimensional search space of the convolutional neural network hyperparameters, so as to find the starting point for local optimization; the Bayesian algorithm is used to make local adjustments based on the results of the vulture search algorithm.

[0030] The deep learning-based remote sensing monitoring method for nearshore inorganic nitrogen proposed in this embodiment effectively solves the problem of poor adaptability of remote sensing indices caused by the complexity of optical components in nearshore waters through a multi-source feature fusion mechanism, significantly improving the feature characterization capability of DIN concentration inversion under complex environments. By utilizing a CNN-BES-BO hybrid optimization model and a collaborative mechanism of global search (BES) and local optimization (BO), high-precision dynamic monitoring of DIN concentration is achieved, providing a stable foundation for DIN testing in nearshore waters, and reducing the workload of testing while ensuring testing accuracy.

[0031] It should be noted that remote sensing monitoring of DIN concentration in nearshore waters still faces many challenges due to the complex optical composition of water and human activities. The optical characteristics of water vary significantly across different sea areas, influenced by factors such as suspended solids, chlorophyll, and colored dissolved organic matter. This results in significant differences in the spectral characteristics of DIN across different regions, increasing the difficulty of remote sensing retrieval. The method provided in this embodiment is adaptable to the complex and variable nearshore environment and has good applicability in different seasons and water types. It can achieve accurate, rapid, and wide-range monitoring of DIN concentration in nearshore waters, demonstrating greater adaptability and robustness.

[0032] In some embodiments, the method further includes: obtaining measured values ​​of band reflectance and inorganic nitrogen concentration at actual sampling points in the experimental area; combining the band reflectance and geospatial information of the actual sampling points to obtain candidate correlation factors; analyzing the correlation between the candidate correlation factors and the measured values ​​of inorganic nitrogen concentration, and determining the inorganic nitrogen concentration correlation factor from the candidate correlation factors based on the correlation.

[0033] In some embodiments, obtaining the measured values ​​of band reflectance and inorganic nitrogen concentration at the measured sampling points in the experimental area includes: delineating the experimental area, determining the geospatial information of the measured sampling points in the experimental area, wherein the geospatial information includes the longitude, latitude, and seawater depth of the sampling points; and collecting the inorganic nitrogen concentration at the measured sampling points as the measured value of inorganic nitrogen concentration.

[0034] In some embodiments, obtaining the measured values ​​of band reflectance and inorganic nitrogen concentration at the actual sampling points in the experimental area includes: obtaining the acquisition time information of the measured inorganic nitrogen concentration; and obtaining second remote sensing image data that matches the sampling point location information and acquisition time information from the remote sensing dataset.

[0035] In some embodiments, obtaining the measured values ​​of band reflectance and inorganic nitrogen concentration at the measured sampling points in the experimental area includes: extracting the band reflectances B1, B2, B3, B4, B5, B6, and B7 of the measured sampling points from the acquired second remote sensing image data.

[0036] In some embodiments, the inorganic nitrogen concentration correlation factor includes one or more of the following: water depth, longitude, latitude, B1, B4, B1\B2, B1\B3, B1\B5, B3\B4, B1-B2, B1-B3, B1-B6, B1-B7, B2-B4, B3-B4, B4-B5, B4-B6, and B1*B4.

[0037] In some embodiments, the training sample generation step of the nearshore inorganic nitrogen remote sensing monitoring model includes: obtaining the measured values ​​of inorganic nitrogen concentration at actual sampling points; generating training feature data corresponding to the correlation factors of inorganic nitrogen concentration based on the band reflectance B1, B2, B3, B4, B5, B6, and B7 of the actual sampling points and based on the second geospatial information, wherein the training feature data includes training band features and training geographic features; and binding the training feature data of the actual sampling points and the measured values ​​of inorganic nitrogen concentration as training samples.

[0038] In some embodiments, the process of simulating the three-stage predation behavior of vultures using a vulture search algorithm, and identifying multiple potential regions in the high-dimensional search space of the convolutional neural network hyperparameters to pinpoint the starting point for local optimization, includes: The search space selection stage: Based on prior information such as the population's optimal position, the population's average position, and the individual's current position, the vulture selects the optimal search space, described as follows: ;in, For the first The updated location of the vulture; The parameter represents the control position change, and its value ranges from [1.5, 2]. A random number between (0, 1); The location of the optimal individual bald eagle; The average position of all individual bald eagles; For the first The current location of the vulture.

[0039] Searching for prey in the search space: Within the selected search space, the vulture flies in a spiral pattern to search for prey, accelerating the search process and seeking the optimal dive position. Its expression is: ;in, and The polar coordinates represent the position of the vulture in its spiral flight, with values ​​ranging from (-1, 1). Indicates the first The current location of the vulture.

[0040] Dive-and-capture prey phase: The vulture uses the optimal dive position as its starting point, accelerating towards its target prey. The position update formula is: ;in, and The polar coordinates of the vulture at this stage; A random number between (0, 1); and These represent the movement intensity of the vulture towards the optimal position and the center position, respectively, with values ​​of [1, 2].

[0041] Further reference Figure 2 This invention provides an embodiment of a deep learning-based remote sensing monitoring model for nearshore inorganic nitrogen, which is similar to... Figure 1 Corresponding to the method embodiments shown, this model can be specifically applied to various electronic devices.

[0042] like Figure 2 As shown, the model includes a convolutional neural network with hyperparameters adjusted based on the vulture search algorithm and the Bayesian algorithm.

[0043] The convolutional neural network is trained using training samples to obtain a nearshore inorganic nitrogen remote sensing monitoring model.

[0044] Specifically, the vulture search algorithm is used to simulate the three-stage behavior of vultures hunting, and multiple potential regions are identified in the high-dimensional search space of the hyperparameters of the convolutional neural network, so as to find the starting point for local optimization; the Bayesian algorithm is used to make local adjustments based on the results of the vulture search algorithm.

[0045] The training sample generation step of the nearshore inorganic nitrogen remote sensing monitoring model includes: obtaining the measured values ​​of inorganic nitrogen concentration at actual sampling points; generating training feature data corresponding to the correlation factors of inorganic nitrogen concentration based on the band reflectance B1, B2, B3, B4, B5, B6, and B7 of the actual sampling points and based on the second geospatial information, wherein the training feature data includes training band features and training geographic features; and binding the training feature data of the actual sampling points and the measured values ​​of inorganic nitrogen concentration as training samples.

[0046] The nearshore inorganic nitrogen remote sensing monitoring model proposed in this embodiment effectively solves the problem of poor adaptability of remote sensing indices caused by the complexity of optical components in nearshore waters through a multi-source feature fusion mechanism, and significantly improves the feature characterization capability of DIN concentration inversion under complex environments. By using the CNN-BES-BO hybrid optimization model, a high-precision dynamic monitoring of DIN concentration is achieved through a collaborative mechanism of global search (BES) and local optimization (BO), providing a stable foundation for accurate prediction of DIN concentration in nearshore waters.

[0047] This application proposes a method for preprocessing the collected data.

[0048] The experimental area was delineated, and the geospatial information of the actual sampling points in the experimental area was determined.

[0049] The inorganic nitrogen concentration at the actual sampling points was collected and used as the measured value of inorganic nitrogen concentration.

[0050] Outliers in the measured inorganic nitrogen concentrations at the sampling points were removed. As an example, the quartile method was used to remove outliers from the measured values ​​at the sampling points.

[0051] Outliers are observations that deviate from the majority of other data points. Outliers can arise from various causes, such as measurement errors, sensor calibration deviations, data entry errors, or real extreme hydrological events. If left unaddressed, these outliers can severely interfere with the model's training and prediction processes, leading to decreased model generalization ability and even erroneous conclusions. This invention employs the quartile method to identify and process outliers. Effectively processed training data can significantly improve data accuracy, facilitating the construction of more accurate and reliable nearshore inorganic nitrogen remote sensing monitoring models.

[0052] This application proposes a method for processing remote sensing data.

[0053] For remote sensing image data, outlier removal is also performed. First, sample points with remote sensing image quality problems are excluded; this mainly involves removing low-quality sample points whose corresponding remote sensing images are covered by clouds, have astigmatism, or are affected by solar flares.

[0054] This ensures the quality and reliability of the extracted spectral reflectance data.

[0055] Obtain the sampling point location information and the collection time information of the measured inorganic nitrogen concentration at the actual sampling points.

[0056] Based on the sampling point location information and acquisition time information, remote sensing image data of the experimental area is obtained. This step includes time matching and location matching to align the remote sensing image data with the measured values ​​of inorganic nitrogen concentration.

[0057] Time matching. The time difference between the acquisition time of remote sensing image data and the collection time of measured inorganic nitrogen concentration is within a predetermined time range. For example, the collection time of measured inorganic nitrogen concentration is within two days before and after the acquisition time of remote sensing image data.

[0058] Location matching. A location is considered to be matched when the location information of the sampling point is within the geographic coverage area of ​​the corresponding remote sensing image pixel.

[0059] Therefore, remote sensing image data that are time- and location-matched with the measured values ​​of inorganic nitrogen concentration can be identified.

[0060] From the acquired remote sensing image data, the band reflectances B1, B2, B3, B4, B5, B6, and B7 of the measured sampling points are extracted.

[0061] The meanings of spectral reflectance B1, B2, B3, B4, B5, B6, and B7 are explained below as examples.

[0062] B1 is the red light band.

[0063] B2 is the near-infrared band.

[0064] B3 is the blue light band.

[0065] B4 is the green light band.

[0066] B5 is the shortwave infrared band.

[0067] B6 is the shortwave infrared band.

[0068] B7 is the shortwave infrared band.

[0069] This application provides a correlation analysis method.

[0070] Obtain the band reflectances B1, B2, B3, B4, B5, B6, and B7 of the measured sampling points, and obtain the geospatial information of the measured sampling points.

[0071] By combining the band reflectance and geospatial information of the measured sampling points, candidate correlation factors (Bi, Bi-Bj, Bi+Bj, Bi / Bj, Bi*Bj; i=1,2,3,4,5,6,7; j=1,2,3,4,5,6,7) are obtained.

[0072] Correlation analysis is used to identify highly correlated factors, which are then used as correlation factors for inorganic nitrogen concentration. Optionally, Spearman correlation coefficient (Spearman) analysis can be used to identify factors highly correlated with inorganic nitrogen concentration (water depth, longitude, latitude, B1, B4, B1 / B2, B1 / B3, B1 / B5, B3 / B4, B1-B2, B1-B3, B1-B6, B1-B7, B2-B4, B3-B4, B4-B5, B4-B6, B1*B4).

[0073] In Bi-Bj, "-" represents a minus sign; in Bi+Bj, "+" represents a plus sign; in Bi / Bj, " / " represents a division sign; and in Bi*Bj, "*" represents a multiplication sign.

[0074] This application proposes the model structure and training samples for a nearshore inorganic nitrogen remote sensing monitoring model.

[0075] The training samples include: measured values ​​of inorganic nitrogen concentration and features corresponding to the selected inorganic nitrogen concentration correlation factors.

[0076] A deep learning-based remote sensing monitoring model for nearshore inorganic nitrogen can be trained using a model with a CNN-BES-BO structure.

[0077] The CNN-BES-BO model is built on a CNN network and uses the Bald Eagle Search (BES) algorithm to simulate the three-stage behavior of bald eagles hunting. It quickly identifies multiple potential regions in the high-dimensional search space of CNN hyperparameters, finding the starting point for local optimization. The Bayesian algorithm (Bo) is used to make fine local adjustments based on the results of BES, further improving the accuracy of hyperparameters.

[0078] We simulated the three stages of bald eagle hunting behavior (selecting the search space, searching for prey, and swooping down to catch it) to optimize CNN hyperparameters.

[0079] These three stages together constitute the core mechanism of the Bald Eagle Search (BES) algorithm, such as Figure 3 As shown, the specific process is as follows: Step 1, Search Space Selection: Based on prior information such as the population's optimal position, the population's average position, and the individual's current position, the vulture selects the optimal search space, described as follows: ;in, For the first The updated location of the vulture; The parameter represents the control position change, and its value ranges from [1.5, 2]. A random number between (0, 1); The location of the optimal individual bald eagle; The average position of all individual bald eagles; For the first The current location of the vulture.

[0080] The second step, the prey search phase: Within the selected search area, the vulture flies in a spiral pattern to search for prey, accelerating the search process and seeking the optimal dive position. Its expression is: ;in, and The polar coordinates represent the position of the vulture in its spiral flight, with values ​​ranging from (-1, 1). Indicates the first The current location of the vulture.

[0081] The third step, the dive-and-capture phase: The vulture uses the optimal dive position as its starting point, accelerating towards its target prey. The position update formula is: ;in, and The polar coordinates of the vulture at this stage; A random number between (0, 1); and These represent the movement intensity of the vulture towards the optimal position and the center position, respectively, with values ​​of [1, 2].

[0082] Next, the Bayesian algorithm (BO) was used to make local fine-tuning based on the results of BES, further improving the accuracy of hyperparameters, giving full play to the advantages of the two algorithms, and improving the overall performance of the CNN model.

[0083] In Convolutional Neural Networks (CNNs), the settings of hyperparameters have a crucial impact on model performance and training efficiency. Hyperparameters include input image pixels, number of convolutional layers, kernel size, stride, pooling layer parameters, and learning rate.

[0084] Optionally, the learning rate and the number of convolutional kernels can be optimized using the Bald Eagle Search algorithm, thereby improving the speed of establishing a nearshore inorganic nitrogen remote sensing monitoring model and increasing the accuracy of the established model.

[0085] Please refer to Figure 3 The specific process may include: First step, data preprocessing: read water quality data, process missing values, divide the training set and test set, and normalize the data.

[0086] The second step is to set the parameters of the CNN-BES-BO model: the CNN network contains 3 convolutional layers and 2 fully connected layers, and set the population size and maximum number of iterations for the vulture search algorithm.

[0087] The third step is to set the fitness function: use the mean squared error (MSE) between the predicted and actual values ​​as the fitness function. The smaller the MSE value, the higher the accuracy of the prediction.

[0088] The fourth step is to optimize the hyperparameters of the CNN using the Bald Eagle Search algorithm and then perform fine-tuning using Bayesian optimization: the optimal parameters obtained by the Bald Eagle Search algorithm are used as the starting point for Bayesian optimization, and further optimization is performed to obtain the final hyperparameters.

[0089] Step 5, Inorganic nitrogen concentration prediction: Build a CNN model using optimized hyperparameters, train it on the training set, then make predictions on the test set, and calculate performance metrics and visualize the results.

[0090] This can improve the training speed and effectiveness of nearshore inorganic nitrogen remote sensing monitoring models.

[0091] This application proposes a prediction process based on a nearshore inorganic nitrogen remote sensing monitoring model.

[0092] Acquire geospatial information and first remote sensing image data of the target sea area.

[0093] Extract the band reflectance value of the target band from the first remote sensing image data.

[0094] Based on a predetermined inorganic nitrogen concentration correlation factor, spatial geographic information and extracted band reflectance values ​​are transformed to generate band features and geographic features.

[0095] The generated band features and geographic features are imported into the nearshore inorganic nitrogen remote sensing monitoring model, and the spatial distribution of inorganic nitrogen concentration in the target sea area is output by the nearshore inorganic nitrogen remote sensing monitoring model.

[0096] The following is for reference. Figure 4 This diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of the present invention. The terminal devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0097] like Figure 4 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage device 408 into a random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device 400. The processing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0098] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0099] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a storage device 408, or installed from a ROM 402. When the computer program is executed by the processing device 401, it performs the functions defined in the methods of the embodiments of the present invention.

[0100] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0101] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in sequential order. In certain circumstances, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0102] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A remote sensing monitoring method for nearshore inorganic nitrogen based on deep learning, characterized in that, The method for a nearshore inorganic nitrogen remote sensing monitoring model includes: Acquire the first geospatial information and first remote sensing image data of the target sea area; Extract the band reflectance of the target band from the first remote sensing image data; Based on a predetermined inorganic nitrogen concentration correlation factor, the first geospatial information and the band reflectance extracted from the first remote sensing image data are transformed to generate band features and geographic features respectively. The generated band features and geographic features are imported into the nearshore inorganic nitrogen remote sensing monitoring model, which outputs the spatial distribution of inorganic nitrogen concentration in the target sea area. Among them, the nearshore inorganic nitrogen remote sensing monitoring model is constructed based on a convolutional neural network; the nearshore inorganic nitrogen remote sensing monitoring model uses the vulture search algorithm to simulate the three-stage behavior of vultures preying on each other, and locks down multiple potential regions in the high-dimensional search space of the convolutional neural network hyperparameters, so as to find the starting point for local optimization; the Bayesian algorithm is used to make local adjustments based on the results of the vulture search algorithm.

2. The method according to claim 1, characterized in that, Before acquiring the first geospatial information and first remote sensing image data of the target sea area, the method further includes: Obtain the measured values ​​of band reflectance and inorganic nitrogen concentration at the actual sampling points in the experimental area; Candidate correlation factors are obtained by combining the measured band reflectance and geospatial information of the sampling points. The correlation between candidate correlation factors and measured values ​​of inorganic nitrogen concentration was analyzed, and the correlation factors of inorganic nitrogen concentration were determined from the candidate correlation factors based on the correlation.

3. The method according to claim 2, characterized in that, The acquisition of the measured values ​​of band reflectance and inorganic nitrogen concentration at the actual sampling points in the experimental area includes: The experimental area was delineated, and the geospatial information of the actual sampling points in the experimental area was determined. The geospatial information included the longitude, latitude, and seawater depth of the sampling points. The inorganic nitrogen concentration at the actual sampling points was collected and used as the measured value of inorganic nitrogen concentration.

4. The method according to claim 3, characterized in that, The acquisition of the measured values ​​of band reflectance and inorganic nitrogen concentration at the actual sampling points in the experimental area includes: Obtain the acquisition time information of the measured values ​​of inorganic nitrogen concentration; Obtain second remote sensing image data from the remote sensing dataset that matches the sampling point location information and acquisition time information.

5. The method according to claim 4, characterized in that, The acquisition of the measured values ​​of band reflectance and inorganic nitrogen concentration at the actual sampling points in the experimental area includes: From the acquired second remote sensing image data, the band reflectances B1, B2, B3, B4, B5, B6, and B7 of the measured sampling points are extracted.

6. The method according to claim 5, characterized in that, The inorganic nitrogen concentration correlation factors include one or more of the following: water depth, longitude, latitude, B1, B4, B1\B2, B1\B3, B1\B5, B3\B4, B1-B2, B1-B3, B1-B6, B1-B7, B2-B4, B3-B4, B4-B5, B4-B6, and B1*B4.

7. The method according to claim 6, characterized in that, The training sample generation steps of the nearshore inorganic nitrogen remote sensing monitoring model include: Based on the measured band reflectances B1, B2, B3, B4, B5, B6, and B7 of the sampling points, and based on the second geospatial information, training feature data corresponding to the inorganic nitrogen concentration correlation factor is generated. The training feature data includes training band features and training geographic features, wherein the second geospatial information is the geospatial information of the measured sampling points. The training feature data of the actual sampling points are bound to the actual inorganic nitrogen concentration values ​​and used as training samples.

8. The method according to claim 1, characterized in that, The method utilizes a vulture search algorithm to simulate the three-stage predation behavior of vultures, identifying multiple potential regions in the high-dimensional search space of convolutional neural network hyperparameters to pinpoint starting points for local optimization, including: Search Space Selection Phase: Based on prior information such as the population's optimal location, the population's average location, and the individual's current location, the vulture selects the optimal search space, described as follows: ； in For the first The updated location of the vulture; The parameter represents the control position change, and its value ranges from [1.5, 2]. A random number between (0, 1); The location of the optimal individual bald eagle; The average position of all individual bald eagles; For the first The current location of the vulture; Searching for prey in the search space: Within the selected search space, the vulture flies in a spiral pattern to search for prey, accelerating the search process and seeking the optimal dive position. Its expression is: ； in and The polar coordinates represent the position of the vulture in its spiral flight, with values ​​ranging from (-1, 1). Indicates the first The current location of the vulture; Dive-and-capture prey phase: The vulture uses the optimal dive position as its starting point, accelerating towards its target prey. The position update formula is: ； in and The polar coordinates of the vulture at this stage; A random number between (0, 1); and These represent the movement intensity of the vulture towards the optimal position and the center position, respectively, with values ​​of [1, 2].

Images