Offshore sea space intelligent monitoring system based on satellite remote sensing
By using a satellite remote sensing-based intelligent monitoring system, the system automatically identifies and dynamically adjusts monitoring strategies, thus solving the bottleneck of automated identification and dynamic response in the monitoring of marine activities in nearshore waters and achieving efficient and accurate supervision of marine activities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-04-07
AI Technical Summary
Existing technologies for monitoring marine activities in nearshore waters suffer from problems such as low efficiency of automated identification, reliance on human experience for identification accuracy, and a disconnect between monitoring strategies and real-time risks. These issues result in low regulatory efficiency and make it difficult to achieve automated, standardized extraction and dynamic response to marine activities.
An intelligent monitoring system based on satellite remote sensing is adopted, including modules for data preprocessing, feature extraction, matching and verification, cloud analysis, and strategy execution. It uses a deep learning neural network model to automatically identify the characteristics of marine activities and conducts multi-dimensional risk assessment through cloud analysis nodes to dynamically adjust monitoring strategies.
It enables rapid batch processing of massive amounts of remote sensing data, improves the efficiency and objectivity of feature extraction for marine activities, can intelligently allocate monitoring resources according to the target risk level, enhances the ability to continuously track and supervise violations, and improves the agility and accuracy of supervision.
Smart Images
Figure CN121811273A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of marine environmental remote sensing monitoring technology, specifically to an intelligent monitoring system for nearshore marine space based on satellite remote sensing. Background Technology
[0002] Nearshore marine spatial resource management heavily relies on effective monitoring of marine activities. Current monitoring primarily relies on satellite or aerial remote sensing, but bottlenecks exist in its implementation. Existing technologies typically employ manual visual interpretation or basic image processing methods to identify marine targets from remote sensing images. This approach is not only inefficient and unable to handle large-scale, high-frequency remote sensing data streams, but its accuracy also heavily depends on the personal experience of the interpreters, resulting in highly subjective and unstable results. It fails to automate and standardize the extraction of spatial location and type characteristics of marine activities, becoming a primary constraint on improving regulatory effectiveness.
[0003] The existing monitoring and response mechanisms are inadequate after suspected abnormal use of marine resources are detected. Most systems rely on pre-set, fixed monitoring schemes. This static management model cannot differentiate monitoring resources according to the risk level of the target, easily leading to both insufficient monitoring of high-risk targets and wasted resources on low-risk targets. The disconnect between monitoring strategies and real-time risks results in low overall regulatory efficiency and makes it difficult to promptly and effectively identify and track genuine violations.
[0004] In the field of nearshore marine use monitoring, there is an urgent need to overcome two major technological bottlenecks: automated identification and intelligent dynamic response. A technological solution is needed that can automatically and accurately extract marine use activity characteristics from remote sensing data, replacing manual methods; an adaptive mechanism needs to be established that can dynamically adjust monitoring strategies based on target risks, thereby achieving optimal allocation of monitoring resources and improving regulatory efficiency. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent monitoring system for nearshore marine space use based on satellite remote sensing, so as to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides an intelligent monitoring system for nearshore marine space use based on satellite remote sensing, the system comprising: The data preprocessing module is used to receive raw image data streams from remote sensing satellites and perform data cleaning and format standardization on the raw image data streams to generate standard format remote sensing data. The feature extraction module is used to input the standard format remote sensing data into the marine activity feature extraction model to identify and extract the spatial location information and activity type features of potential marine activities; The matching and verification module is used to match and verify the spatial location information and activity type characteristics with the list of authorized sea use activities pre-stored in the local database, and generate a verification anomaly report containing location deviation and type anomaly information when the matching and verification fails. The cloud communication module is used to upload the verification anomaly report and the corresponding standard format remote sensing data to the cloud analysis node; The cloud-based analytics node is used to initiate a multi-dimensional risk assessment process based on the verification anomaly report and generate a dynamic monitoring strategy for the sea use activities. The strategy execution module is used to adjust the monitoring frequency and monitoring parameters of the sea use activities according to the dynamic monitoring strategy, and to update the monitoring rule base of the local database.
[0007] Preferably, the data cleaning and format standardization processing of the original image data stream includes: detecting noisy pixels and transmission error data blocks in the original image data stream, smoothing the noisy pixels using a spatial filtering algorithm, reconstructing or marking and removing the transmission error data blocks, uniformly converting the processed image data to a preset geographic coordinate system and pixel resolution, performing radiometric calibration and atmospheric correction on the converted image data, and outputting the standard format remote sensing data.
[0008] Preferably, the step of inputting the standard format remote sensing data into the marine activity feature extraction model includes: loading a trained deep learning neural network model, inputting the standard format remote sensing data into the deep learning neural network model, extracting the low-level texture features of the image through the convolutional layer of the deep learning neural network model, reducing the dimensionality of the low-level texture features through the pooling layer, fusing the dimensionality-reduced features through the fully connected layer, outputting the spatial location coordinates and activity type probability distribution of the marine activity, and determining the final activity type feature from the activity type probability distribution according to a preset probability threshold.
[0009] Preferably, the process of matching and verifying the spatial location information and activity type features with the list of authorized sea use activities pre-stored in the local database includes: reading the list of authorized sea use activities from the local database, the list of authorized sea use activities containing a set of approved sea use activity boundary polygons and corresponding activity type codes; calculating the spatial inclusion relationship between the spatial location coordinates and each sea use activity boundary polygon; if a spatial inclusion relationship exists, comparing the consistency between the activity type features and the corresponding activity type codes; if the spatial location coordinates are not contained within any boundary polygon or the activity type features are inconsistent with the codes, the matching verification is determined to have failed.
[0010] Preferably, generating a verification anomaly report containing location deviation and type anomaly information when the matching verification fails includes: recording the timestamp and satellite sensor identifier of the matching verification failure, quantifying the distance deviation value between the spatial location coordinates and the nearest authorized boundary, recording the difference description between the activity type characteristics and the closest authorized type, and encapsulating the timestamp, sensor identifier, distance deviation value and difference description into a structured verification anomaly report.
[0011] Preferably, the process for initiating a multi-dimensional risk assessment based on the verification anomaly report includes: parsing the received verification anomaly report, extracting the distance deviation value and difference description, querying the historical violation case database, finding historical cases with similar distance deviation values and difference descriptions, calculating the risk level index of the current marine activity based on the handling records and subsequent development of the historical cases, assessing the potential impact of the marine activity on the marine ecosystem by combining real-time marine environmental data, and generating a risk assessment matrix by combining the risk level index and the potential impact.
[0012] Preferably, the generation of a dynamic monitoring strategy for the marine use activities includes: determining monitoring priorities based on a risk assessment matrix, setting revisit cycles and spatial resolution requirements for image acquisition based on the monitoring priorities, determining the image feature bands that need to be monitored in detail based on the difference description, and combining the revisit cycles, spatial resolution requirements, and feature band configurations to form a complete dynamic monitoring strategy configuration file.
[0013] Preferably, the step of adjusting the monitoring frequency and monitoring parameters of the marine use activities according to the dynamic monitoring strategy includes: receiving the dynamic monitoring strategy configuration file issued by the cloud analysis node, parsing the revisit period parameter in the configuration file, adjusting the imaging schedule of the satellite mission planning system, switching the imaging mode of the remote sensing satellite according to the spatial resolution requirements, and activating the corresponding sensor channel according to the characteristic band configuration.
[0014] Preferably, the process of updating the monitoring rule base of the local database includes: recording the correlation between the dynamic monitoring strategy configuration file and the verification anomaly report in the monitoring log; when the marine activity is continuously monitored for a preset time period, extracting all image feature change trajectories during the monitoring period; verifying the accuracy of the risk assessment matrix based on the feature change trajectories; optimizing and correcting the parameters in the risk assessment matrix based on the verification results; and updating the optimized parameters to the monitoring rule base of the local database.
[0015] Preferably, the construction process of the marine activity feature extraction model includes: collecting multi-temporal historical remote sensing image data, manually interpreting the historical remote sensing image data, marking the spatial location boundaries and activity types of marine activities to form a labeled dataset, dividing the labeled dataset into a training set, a validation set, and a test set according to a preset ratio, constructing an initial deep learning neural network model, the model including sequentially connected convolutional layers, pooling layers, and fully connected layers, using the training set to iteratively train the initial model, minimizing the loss function through backpropagation to optimize model parameters, using the validation set to verify the accuracy of the model during training, and adjusting the learning rate and iteration number hyperparameters, using the test set to evaluate the recognition accuracy and recall of the final model, and when both accuracy and recall reach a preset threshold, deploying the model to the feature extraction module.
[0016] Compared with the prior art, the beneficial effects of the present invention are: A specialized feature extraction model for marine activities is employed to process standard-format remote sensing data, enabling automatic and accurate identification of the spatial contours and classification of various marine activities, thus freeing manual labor from tedious image interpretation. This technical solution achieves rapid batch processing of massive amounts of remote sensing data, improving the efficiency and objectivity of feature extraction. It overcomes the shortcomings of traditional manual operations, such as slow speed, fatigue, and inconsistent standards, providing stable and reliable structured data input for subsequent matching and verification processes.
[0017] The cloud-based analysis node initiates a multi-dimensional risk assessment process based on verification anomaly reports. This process comprehensively considers multiple factors, including location deviations, type anomalies, and environmental sensitivity of marine activities, to quantitatively or qualitatively assess their potential risk levels. Monitoring strategies dynamically generated based on these assessment results enable the system to intelligently allocate monitoring resources according to the risk level of targets. High-risk targets receive higher frequency and better parameter monitoring, while low-risk targets receive appropriately reduced monitoring intensity, thereby improving the overall efficiency of satellite remote sensing data and computing resources.
[0018] The strategy execution module translates dynamic monitoring strategies into specific monitoring instructions and feeds them directly back to the monitoring system itself, enabling real-time, closed-loop updates of monitoring rules. This adaptive adjustment mechanism gives the entire monitoring system the ability to learn and evolve, continuously optimizing its monitoring behavior to cope with ever-changing marine usage conditions. The system's response has shifted from static and passive to dynamic and proactive, enhancing its ability to continuously track and monitor illegal or abnormal marine activities, and improving the agility and accuracy of overall supervision. Attached Figure Description
[0019] Figure 1This is a schematic diagram illustrating the working principle of the intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in this invention. Figure 2 A flowchart for data preprocessing; Figure 3 The flowchart for the feature extraction model; Figure 4 A comparison chart of core parameters for monitoring strategies with different priorities; Figure 5 Configuration diagram for monitoring parameters of risk level of marine activities in nearshore waters. Detailed Implementation
[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] Please see Figure 1 This invention provides an intelligent monitoring system for nearshore marine space use based on satellite remote sensing. The system includes: a data preprocessing module, a feature extraction module, a matching and verification module, a cloud communication module, a cloud analysis node, and a strategy execution module. The data preprocessing module receives raw image data streams from remote sensing satellites, performs data cleaning and format standardization on the raw image data streams, and generates standard format remote sensing data. The feature extraction module inputs the standard format remote sensing data into a marine activity feature extraction model to identify and extract the spatial location information and activity type features of potential marine activities. The matching and verification module matches and verifies the spatial location information and activity type features with a list of authorized marine activities pre-stored in a local database. When the matching and verification fails, a verification anomaly report containing location deviation and type anomaly information is generated. The cloud communication module uploads the verification anomaly report and the corresponding standard format remote sensing data to the cloud analysis node. The cloud analysis node initiates a multi-dimensional risk assessment process based on the verification anomaly report and generates a dynamic monitoring strategy for marine activities. The strategy execution module adjusts the monitoring frequency and monitoring parameters for marine activities according to the dynamic monitoring strategy and updates the monitoring rule base of the local database.
[0022] Example 1: See Figure 2In specific implementation, the data preprocessing module receives raw image data streams from remote sensing satellites. These raw image data streams contain multi-band image information. The data preprocessing module performs data cleaning operations on the raw image data streams. Data cleaning includes detecting noisy pixels and transmission error data blocks in the raw image data streams. Noisy pixels are points whose pixel values deviate abnormally from the surrounding area. Transmission error data blocks are generated due to data transmission interruptions or interference. The detection process uses a sliding window to scan the image data and calculates local statistical features to identify outliers. In some embodiments, a spatial filtering algorithm is applied to smooth noisy pixels. The spatial filtering algorithm uses a Gaussian convolution kernel for convolution operations. The mathematical expression of the Gaussian convolution kernel is:
[0023] in: This represents the weight value of the convolution kernel at a given location. and These are the relative coordinates within the convolution kernel. This is the standard deviation parameter that controls the smoothing level. The convolution operation iterates through each pixel of the image, weighting and summing the neighboring pixel values to suppress noise. It can be understood that the noise level of the image data is reduced after smoothing. The processing of transmitted erroneous data blocks includes data reconstruction or label removal. Data reconstruction uses interpolation algorithms to fill the erroneous region based on surrounding valid data, while label removal identifies erroneous data blocks as invalid and excludes them from further processing. Format standardization converts the cleaned image data to a preset geographic coordinate system and pixel resolution. The geographic coordinate system uses the WGS84 coordinate system, and the pixel resolution is adjusted to 10 meters. The conversion process involves resampling algorithms to maintain spatial consistency. Radiometric calibration converts image pixel values to surface reflectance or radiance, and atmospheric correction eliminates the effects of atmospheric scattering and absorption, outputting standard format remote sensing data. Optionally, radiometric calibration uses satellite calibration parameters to calculate absolute radiance, and atmospheric correction uses the dark pixel method to estimate aerosol optical thickness. In some embodiments, the data preprocessing module integrates multiple processing steps into a pipeline to ensure consistent data quality.
[0024] Example 2: See Figure 3In specific implementation, the feature extraction module inputs standard-format remote sensing data into the marine activity feature extraction model, loads a trained deep learning neural network model, and the marine activity feature extraction model receives standard-format remote sensing data as input. The model extracts the low-level texture features of the image through convolutional layers of the deep learning neural network model. These convolutional layers use multiple convolutional kernels to perform sliding window convolution operations to extract edge and texture information. In some embodiments, pooling layers reduce the dimensionality of the low-level texture features. These pooling layers employ max pooling, selecting the maximum value within a local region as the output to reduce the feature map size. Fully connected layers fuse the dimensionality-reduced features. These fully connected layers perform linear transformations and non-linear activations on the flattened feature vectors, outputting the spatial location coordinates of marine activities and the probability distribution of activity types. The probability distribution of activity types is obtained through... Function computation, The mathematical expression of the function is:
[0025] in: This represents the probability value of the i-th activity type. It is the raw score of the i-th node output by the fully connected layer. It is the total number of activity types. The function ensures that the sum of all probability values is 1. The final activity type features are determined from the activity type probability distribution based on a preset probability threshold of 0.7. When the probability value of an activity type exceeds the threshold, it is selected as the final feature. The process of building the marine activity feature extraction model involves collecting multi-temporal historical remote sensing image data, covering nearshore waters under different seasons and weather conditions. The historical remote sensing image data is manually interpreted by professionals, annotating the spatial boundaries and activity types of marine activities, including aquaculture, shipping, and tourism. A labeled dataset is then formed, containing image data and corresponding annotation information. This dataset is divided into training, validation, and test sets according to a preset ratio: 70% for training, 15% for validation, and 15% for test. An initial deep learning neural network model is constructed, consisting of sequentially connected convolutional layers, pooling layers, and fully connected layers. The convolutional layers use the ReLU activation function. The initial deep learning neural network model is iteratively trained using a training set. Iterative training optimizes model parameters by minimizing the loss function through backpropagation, employing cross-entropy loss. During iterative training, the cross-entropy loss function calculates the difference between the activity type probability distribution output by the deep learning neural network model and the actual activity types annotated by human interpretation in the training set. This loss value is propagated layer by layer through backpropagation, optimizing the weights and bias parameters in the model via gradient descent, thereby driving the model parameters to update in a direction that reduces the difference between predictions and annotations. The deep learning neural network model's accuracy is validated using a validation set, and the learning rate and iteration count hyperparameters are adjusted. The initial learning rate is 0.001. The recognition accuracy and recall of the final deep learning neural network model are evaluated using a test set. When both accuracy and recall reach preset thresholds, the deep learning neural network model is deployed to the feature extraction module. Optionally, data augmentation techniques are used during training to increase sample diversity. In some embodiments, the validation set is used to stop training early to prevent overfitting.
[0026] Example 3: In specific implementation, the matching verification module matches the spatial location information and activity type features output by the feature extraction module with the list of authorized sea use activities pre-stored in the local database. The matching verification module first reads the list of authorized sea use activities from the local database. This list contains a set of approved sea use activity boundary polygons and corresponding activity type codes. The list is stored in geospatial data format, and each sea use activity boundary polygon is defined by a series of vertex coordinates. The activity type codes use a standard classification system. The spatial inclusion relationship between the spatial location coordinates and each sea use activity boundary polygon is calculated. This spatial inclusion relationship is determined using the ray casting method or a point-in-polygon algorithm. The algorithm checks whether the spatial location coordinates are located inside the boundary polygon. If a spatial inclusion relationship exists, the consistency between the activity type features and the corresponding activity type codes is compared. The activity type features come from the type labels output by the feature extraction model, and the activity type codes are predefined strings or numerical codes. Consistency comparison uses string matching or a code mapping table. If the spatial location coordinates are not contained within any boundary polygon or the activity type features are inconsistent with the code, the matching verification is deemed to have failed. In some embodiments, spatial inclusion relationship calculations are accelerated using optimized algorithms, such as spatial index structures that reduce the number of polygon traversals.
[0027] When a match verification fails, a verification anomaly report is generated, containing information on position deviation and type anomaly. The report records the timestamp of the verification failure and the satellite sensor identifier, accurate to the millisecond level. The satellite sensor identifier uniquely identifies the satellite from which the data originated. The distance deviation between the spatial position coordinates and the nearest authorized boundary is quantified and calculated using the Euclidean distance formula, expressed as:
[0028] in: Indicates the distance deviation value. and These are the horizontal and vertical coordinates of the spatial location. and These are the x and y coordinates of the most recently authorized boundary point, determined by calculating the shortest distance from the spatial location coordinates to each edge of the boundary polygon. A description of the difference between the activity type characteristics and the closest authorized type is recorded, based on semantic differences in type encoding or predefined difference levels. Timestamps, sensor identifiers, distance deviation values, and difference descriptions are encapsulated into a structured verification anomaly report, stored in JSON or XML format. Optionally, elevation information can be incorporated into the distance deviation value calculation for 3D distance calculation. In some embodiments, the difference description includes type name differences and confidence level differences. It is understood that the verification anomaly report provides detailed anomaly data for cloud-based analytics.
[0029] Example 4: In specific implementation, the cloud analysis node initiates a multi-dimensional risk assessment process based on the verification anomaly report. It parses the received verification anomaly report, extracting the distance deviation value and difference description. The verification anomaly report is stored in a structured format, with the distance deviation value being numerical data and the difference description being text data. The system queries the historical violation case database, which contains historical case records. Each case record includes a case number, distance deviation value, difference description, handling record, and subsequent development. Historical cases with similar distance deviation values and difference descriptions are searched. Similarity matching uses a combination of Euclidean distance and text similarity algorithms. Based on the handling record and subsequent development of historical cases, the risk level index of the current sea use activity is calculated. The risk level index is obtained through weighted calculation, expressed by the formula:
[0030] in: Indicates the risk level index. It is the distance deviation value. It is a quantitative score describing the difference. and The weighting coefficients are determined through historical data regression analysis, and the difference description quantification score is obtained by mapping textual difference descriptions to predefined numerical levels. Combined with real-time marine environmental data, including water temperature, salinity, and chlorophyll concentration, the potential impact of marine activities on the marine ecosystem is assessed. The potential impact level is calculated based on an environmental sensitivity model. A risk assessment matrix is generated by integrating the risk level index and the potential impact level. This matrix is a two-dimensional table where rows represent risk level index levels, columns represent potential impact level levels, and cell content represents the overall risk level. In some embodiments, referring to Table 1, the historical violation case library uses a database index to accelerate queries. The risk assessment matrix provides a visual representation of risk. A dynamic monitoring strategy for marine activities is generated. Monitoring priorities are determined based on the risk assessment matrix, with high, medium, and low priorities. The revisit period and spatial resolution requirements for image acquisition are set based on the monitoring priorities, with higher priorities corresponding to shorter revisit periods and higher spatial resolution. Image feature bands requiring focused monitoring are determined based on the difference descriptions. The difference descriptions indicate abnormal activity types, and the feature band selection is based on the band response characteristics of multispectral images. The revisit period, spatial resolution requirements, and characteristic band configuration are combined to form a complete dynamic monitoring strategy configuration file, which is stored in JSON format. Optionally, the revisit period setting takes into account the satellite orbital period. In some embodiments, the characteristic band configuration includes visible and near-infrared bands. It is understood that the dynamic monitoring strategy configuration file guides the adjustment of monitoring parameters. Optionally, a monitoring priority mapping table is used for standardization settings.
[0031] Table 1: Historical Violation Case Matching Table
[0032] See Figure 4 In the monitoring strategy configuration phase of the dynamic monitoring strategy parameter analysis, the core parameters (average revisit period and average spatial resolution) of different priority monitoring strategies exhibit differentiated characteristics. Specifically, monitoring priorities are divided into three levels: high, medium, and low. High-priority strategies correspond to an average revisit period of 1.3 days and an average spatial resolution of 1.3 meters, reflecting the precise monitoring characteristics of "short cycle and high resolution." Medium-priority strategies correspond to an average revisit period of 4.0 days and an average spatial resolution of 7.7 meters, with parameter configurations between high and low priorities. Low-priority strategies correspond to an average revisit period of 8.3 days and an average spatial resolution of 17.7 meters, exhibiting the conventional monitoring characteristics of "long cycle and low resolution." The parameter distribution is consistent with the design logic of the dynamic monitoring strategy: higher priority means higher monitoring frequency and better spatial accuracy, to match the refined regulatory needs of high-risk marine activities.
[0033] Example 5: In specific implementation, the strategy execution module adjusts the monitoring frequency and parameters for marine activities according to the dynamic monitoring strategy. It receives the dynamic monitoring strategy configuration file from the cloud analysis node, which includes the revisit period, spatial resolution requirements, and characteristic band configuration. The module parses the revisit period parameter in the configuration file, which represents the time interval for the satellite to re-image a specific area in hours. It then adjusts the imaging schedule of the satellite mission planning system, which is a time-series plan for the satellite platform to execute observation tasks. Based on the spatial resolution requirements, which specify the ground pixel size of the image, the module switches the imaging mode of the remote sensing satellite. Imaging modes include panchromatic mode and multispectral mode, with different spatial resolutions corresponding to different imaging modes. According to the characteristic band configuration, which specifies the electromagnetic spectrum range to be activated, the module activates the corresponding sensor channels, which are hardware units that detect signals in specific bands. In some embodiments, the adjustment of the satellite mission planning system involves orbital parameter calculation and resource allocation.
[0034] The process of updating the monitoring rule base in the local database includes recording the correlation between the dynamic monitoring strategy configuration file and the verification anomaly report in the monitoring log. The monitoring log records the complete context information of each strategy adjustment in a time-series format, and the correlation is established through a unique task identifier. When marine activities are continuously monitored for a preset time period (set based on risk assessment results), all image feature change trajectories during the monitoring period are extracted. These trajectory data are obtained through continuous observation of the temporal changes in spectral and spatial characteristics of marine activities. The accuracy of the risk assessment matrix is verified based on these feature change trajectories. This verification process involves comparing the actual observed development with the predictions of the risk assessment matrix. Based on the verification results, the parameters in the risk assessment matrix are optimized and corrected. Parameter optimization uses the gradient descent method to adjust the weight coefficients. The iterative formula for the gradient descent method is:
[0035] in: This represents the optimized parameter values. These are the parameter values before optimization. It's the learning rate. It is a loss function Regarding parameters The gradient is calculated. The optimized parameters are then updated to the monitoring rule base in the local database, which stores all rules and parameters used for risk assessment and strategy generation. Optionally, time series analysis methods are used to extract feature change trajectories. In some embodiments, the parameter optimization process employs a batch update method to improve efficiency.
[0036] See Figure 5 In the dynamic monitoring strategy configuration for different risk levels, the coordinated adjustment of revisit period and spatial resolution reflects the risk-adaptive monitoring logic. Specifically, as the risk level progresses from "low risk" to "high risk," the revisit period (in hours) decreases in a stepwise manner (72 hours for low risk, 6 hours for high risk), while the spatial resolution (in meters) decreases linearly (30 meters for low risk, 4 meters for high risk). The core logic of this configuration is that high-risk marine activities require shorter revisit periods for high-frequency monitoring, while simultaneously improving the accuracy of activity feature identification with higher spatial resolution (smaller ground pixel size); low-risk activities, on the other hand, reduce resource consumption with longer revisit periods, and can meet monitoring needs with relatively lower spatial resolution. In terms of parameter configuration, the revisit period and spatial resolution for each risk level form a strict negative correlation; for example, a medium-risk level corresponds to a 24-hour revisit period and a 15-meter spatial resolution, achieving dynamic adaptation of monitoring resources to risk levels.
[0037] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0038] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A nearshore marine space intelligent monitoring system based on satellite remote sensing, characterized in that, The system includes: The data preprocessing module is used to receive raw image data streams from remote sensing satellites and perform data cleaning and format standardization on the raw image data streams to generate standard format remote sensing data. The feature extraction module is used to input the standard format remote sensing data into the marine activity feature extraction model to identify and extract the spatial location information and activity type features of potential marine activities; The matching and verification module is used to match and verify the spatial location information and activity type characteristics with the list of authorized sea use activities pre-stored in the local database, and generate a verification anomaly report containing location deviation and type anomaly information when the matching and verification fails. The cloud communication module is used to upload the verification anomaly report and the corresponding standard format remote sensing data to the cloud analysis node; The cloud-based analytics node is used to initiate a multi-dimensional risk assessment process based on the verification anomaly report and generate a dynamic monitoring strategy for the sea use activities. The strategy execution module is used to adjust the monitoring frequency and monitoring parameters of the sea use activities according to the dynamic monitoring strategy, and to update the monitoring rule base of the local database.
2. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 1, characterized in that, The data cleaning and format standardization process for the original image data stream includes: detecting noisy pixels and transmission error data blocks in the original image data stream; smoothing the noisy pixels using a spatial filtering algorithm; reconstructing or marking and removing the transmission error data blocks; uniformly converting the processed image data to a preset geographic coordinate system and pixel resolution; performing radiometric calibration and atmospheric correction on the converted image data; and outputting the standard format remote sensing data.
3. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 2, characterized in that, The step of inputting the standard format remote sensing data into the marine activity feature extraction model includes: loading a trained deep learning neural network model, inputting the standard format remote sensing data into the deep learning neural network model, extracting the low-level texture features of the image through the convolutional layer of the deep learning neural network model, reducing the dimensionality of the low-level texture features through the pooling layer, fusing the dimensionality-reduced features through the fully connected layer, outputting the spatial location coordinates and activity type probability distribution of the marine activity, and determining the final activity type feature from the activity type probability distribution according to a preset probability threshold.
4. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 3, characterized in that, The process of matching and verifying the spatial location information and activity type features with the list of authorized sea use activities pre-stored in the local database includes: reading the list of authorized sea use activities from the local database, the list of authorized sea use activities containing a set of approved sea use activity boundary polygons and corresponding activity type codes; calculating the spatial inclusion relationship between the spatial location coordinates and each sea use activity boundary polygon; if a spatial inclusion relationship exists, comparing the consistency between the activity type features and the corresponding activity type codes; if the spatial location coordinates are not contained within any boundary polygon or the activity type features are inconsistent with the codes, then determining that the matching verification has failed.
5. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 4, characterized in that, The process of generating a verification anomaly report containing location deviation and type anomaly information when a matching verification fails includes: recording the timestamp and satellite sensor identifier of the matching verification failure, quantifying the distance deviation value between the spatial location coordinates and the nearest authorized boundary, recording the difference description between the activity type characteristics and the closest authorized type, and encapsulating the timestamp, sensor identifier, distance deviation value and difference description into a structured verification anomaly report.
6. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 5, characterized in that, The process for initiating a multi-dimensional risk assessment based on the verification anomaly report includes: parsing the received verification anomaly report, extracting the distance deviation value and difference description, querying the historical violation case database, finding historical cases with similar distance deviation values and difference descriptions, calculating the risk level index of the current marine activity based on the handling records and subsequent development of the historical cases, assessing the potential impact of the marine activity on the marine ecosystem by combining real-time marine environmental data, and generating a risk assessment matrix by combining the risk level index and the potential impact.
7. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 6, characterized in that, The generation of a dynamic monitoring strategy for the marine use activities includes: determining monitoring priorities based on a risk assessment matrix; setting revisit cycles and spatial resolution requirements for image acquisition based on the monitoring priorities; determining the image feature bands that need to be monitored in particular based on the difference descriptions; and combining the revisit cycles, spatial resolution requirements, and feature band configurations to form a complete dynamic monitoring strategy configuration file.
8. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 7, characterized in that, The method for adjusting the monitoring frequency and parameters of the marine use activities according to the dynamic monitoring strategy includes: receiving the dynamic monitoring strategy configuration file issued by the cloud analysis node, parsing the revisit period parameter in the configuration file, adjusting the imaging schedule of the satellite mission planning system, switching the imaging mode of the remote sensing satellite according to the spatial resolution requirements, and activating the corresponding sensor channel according to the characteristic band configuration.
9. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 8, characterized in that, The process of updating the monitoring rule base of the local database includes: recording the correlation between the dynamic monitoring strategy configuration file and the verification anomaly report in the monitoring log; after the marine activity is continuously monitored for a preset time period, extracting all image feature change trajectories during the monitoring period; verifying the accuracy of the risk assessment matrix based on the feature change trajectories; optimizing and correcting the parameters in the risk assessment matrix based on the verification results; and updating the optimized parameters to the monitoring rule base of the local database.
10. The intelligent monitoring system for nearshore marine space use based on satellite remote sensing as described in claim 3, characterized in that, The construction process of the marine activity feature extraction model includes: collecting multi-temporal historical remote sensing image data, manually interpreting the historical remote sensing image data, marking the spatial location boundaries and activity types of marine activities to form a labeled dataset, dividing the labeled dataset into a training set, a validation set, and a test set according to a preset ratio, constructing an initial deep learning neural network model, the model including sequentially connected convolutional layers, pooling layers, and fully connected layers, iteratively training the initial model using the training set, minimizing the loss function through backpropagation to optimize model parameters, using the validation set to verify the accuracy of the model during training, and adjusting the learning rate and iteration number hyperparameters, using the test set to evaluate the recognition accuracy and recall of the final model, and deploying the model to the feature extraction module when both accuracy and recall reach a preset threshold.