An intelligent processing system for marine observation data
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FOURTH INSTITUTE OF OCEANOGRAPHY MINISTRY OF NATURAL RESOURCES (CHINA ASEAN COUNTRIES JOINT RESEAR
- Filing Date
- 2025-09-26
- Publication Date
- 2026-07-03
Smart Images
Figure CN121213682B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of marine observation technology, specifically to an intelligent system for processing marine observation data. Background Technology
[0002] Ocean eddies, as an important dynamic phenomenon that is ubiquitous in the ocean, have a crucial impact on air-sea interaction, ocean material transport, ecological environment evolution, and climate system regulation. Their monitoring and analysis is one of the core topics in the field of ocean observation. Traditional ocean eddy identification often relies on manual interpretation or semi-automatic methods based on physical parameter thresholds (such as using the closed contour line characteristics of sea level height anomalies (SLA)). However, due to the complexity of remote sensing image data (such as noise interference and heterogeneity of multi-source data) and the dynamic change characteristics of the eddies themselves (such as irregular shape and large scale differences), there are problems such as low identification efficiency, insufficient accuracy, and strong subjectivity.
[0003] Application CN115293662A discloses a method and system for intelligent computing of marine observation data that integrates parallel and distributed approaches. This method relates to the field of intelligent computing of marine observation time-series data streams. It acquires marine observation data streams from each channel in real time and stores them in a distributed cluster. The data streams undergo preprocessing, including out-of-order processing, deduplication, and missing data removal. Based on the preprocessed marine observation data streams, a supercomputing MPI parallel training model is used to train a multi-channel online learning model, obtaining the latest intelligent computing model for each channel. Using the Flink distributed stream processing system, the latest intelligent computing model for each channel is selected for real-time inference and prediction of the continuously flowing marine observation data. This invention is suitable for multi-channel, multi-task application scenarios, effectively supporting online learning and inference tasks for streaming data and the management of high-throughput sensor data, enabling rapid iterative upgrades of multi-channel computing models and real-time data inference.
[0004] For ocean observation, remote sensing is generally used to identify eddies on the sea surface. However, in actual observation and processing, human intervention is required to identify eddies and then assess their status in conjunction with related observation data. This traditional method cannot effectively identify eddies or densely populated areas, and therefore cannot achieve good automatic observation results. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides an intelligent marine observation data processing system that solves the problems of not being able to effectively confirm eddies or dense areas.
[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent processing system for marine observation data, comprising:
[0007] The image data acquisition end acquires remote sensing images associated with the observation area;
[0008] The image feature analysis unit performs feature verification on the acquired remote sensing images, determines the gradient features associated with image points, completes the gradient pixel calibration process, and then locks the gradient contour line. The specific method is as follows:
[0009] The Sobel algorithm is used to identify the vertical and longitudinal gradients associated with different image points within the remote sensing image. Based on the identified vertical and longitudinal gradients, the composite gradient associated with the corresponding image points is then determined. ;
[0010] The comprehensive gradient associated with different image points in the remote sensing image is compared with the preset value Y1: if the comprehensive gradient is ≥ Y1, the corresponding image point is recorded as the gradient point; otherwise, no marking is made. Y1 is the preset value.
[0011] Connect consecutive gradient points in the remote sensing image to confirm the outline of the associated points, and identify several gradient outline lines in the remote sensing image.
[0012] The texture vortex calibration end identifies the feature points associated with the corresponding gradient contours based on the curvature characteristics of different gradient contours. Then, cluster analysis is performed on the identified feature points to locate the texture vortex. The specific method is as follows:
[0013] The confirmed gradient contour line is confirmed by arc. Starting from one end of the gradient contour line, the arc is confirmed from the other end. The moving segments between connected points are confirmed, and the arc is confirmed in sequence. Multiple moving segments with the same arc are integrated to confirm a single arc segment. Then the same processing method is used to confirm the subsequent arc segments, so that a single gradient contour line is divided into multiple arc segments with different arc characteristics.
[0014] Based on the arc features of each arc segment, identify several arc points associated with the arc segment, and then lock the feature points from the remote sensing image. The straight-line distance between the feature points and the several arc points is the same. The feature points associated with the arc segment are confirmed in sequence.
[0015] Based on several different feature points identified by different gradient contours, a set of feature points is randomly selected as the center point, and a search circle about the center point is generated with a preset radius R1 as the search range. The number of feature points inside the search circle is denoted as Q. Then, other feature points are successively used as center points, and the number of feature points Q in the corresponding search circle is simultaneously confirmed. The maximum value is selected from the confirmed Q, and the center point associated with the maximum value is denoted as the selected point. The selected point and the feature points associated with the corresponding search circle are all denoted as vortex points. The arc segments associated with several vortex points are confirmed, and the texture vortex associated with this vortex point is generated.
[0016] Then, the same processing method is used to confirm the vortexes for other feature points that are not labeled as vortex points. Several texture vortices existing in the remote sensing image are labeled in sequence, and the remote sensing image with the texture vortexes labeled is transmitted to the feature region clustering end.
[0017] In the feature region clustering stage, the remote sensing image with completed texture vortex calibration is further processed. Based on the positional characteristics of different texture vortices, dense regions are identified. The specific method is as follows:
[0018] Identify the selected points marked within each texture vortex, confirm the two selected points with the closest straight-line distance from the remote sensing image, connect the two selected points, confirm the connecting line, and then confirm the selected point with the closest perpendicular distance to the connecting line. Generate an undetermined triangle by connecting the connecting line and the current selected point, and record the area parameter M1 of the current undetermined triangle.
[0019] Based on the confirmed undetermined triangle, randomly select other surrounding points to generate a quadrilateral and record the area M2 of the corresponding quadrilateral. If M1 and M2 satisfy: M2 < 2M1, then retain the currently confirmed quadrilateral; otherwise, reconstruct the quadrilateral. If there is no quadrilateral that satisfies M2 < 2M1, then terminate the current dense confirmation process.
[0020] Based on the confirmed quadrilateral, randomly select other selected points around it to generate a pentagon, and record the area of the pentagon as M3. Then identify whether M3 satisfies: M3 < 3M1. If it does, keep the currently confirmed pentagon. If it does not, continue to confirm other pentagons. If there is no pentagon that satisfies M2 < 3M1, record the selected points associated with the corresponding quadrilateral as dense points and terminate the current dense confirmation process.
[0021] Similarly, based on the corresponding polygons confirmed in the previous stage, the newly added polygons associated in the next stage are confirmed, and the dense points are confirmed in turn. When all selected points in the remote sensing image have completed the dense confirmation process, several dense points associated with each dense confirmation process are regarded as similar dense points, and multiple texture vortices associated with similar dense points are recorded as the same dense region.
[0022] Preferred options also include:
[0023] The data classification and display terminal identifies other observation data associated with the corresponding dense regions based on several marked dense regions, and bundles the other observation data with multiple texture vortices associated with the corresponding single dense region to generate a bundled data package belonging to the corresponding dense region. The terminal then displays the different bundled data packages associated with different dense regions.
[0024] This invention provides an intelligent processing system for ocean observation data. Compared with existing technologies, it has the following advantages:
[0025] This invention uses the Sobel algorithm to accurately extract gradient features from remote sensing images. By comparing the gradient with a preset threshold, the gradient contour is locked, enabling rapid capture of key features around vortices. This significantly improves the efficiency and accuracy of initial vortex identification. Furthermore, through arc feature analysis and clustering algorithms, the gradient contour is transformed into specific texture vortex calibration. In particular, through the analysis of feature point search circles, the center position of the vortex is accurately locked, solving the problems of blurred vortex boundaries and difficulty in center positioning in traditional methods. This upgrades vortex identification from contour perception to refined point calibration.
[0026] By employing a polygonal area progressive verification method, the distribution density of multiple vortices is quantitatively analyzed, enabling rapid identification of densely clustered vortex areas. This provides an intuitive basis for understanding the spatial distribution patterns and correlations of vortices, overcoming the limitations of single-vortex analysis. By integrating dense areas with related observation data through bundled data packages, the data is visualized and categorized, facilitating rapid review and analysis by researchers. Furthermore, it provides clear and accurate reference data for decision-making in marine environmental monitoring and disaster early warning, effectively enhancing the application value and decision support capabilities of marine observation data. Attached Figure Description
[0027] Figure 1 This is a schematic diagram of the principle framework of the present invention. Detailed Implementation
[0028] 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.
[0029] First Embodiment
[0030] Please see Figure 1 This application provides an intelligent processing system for marine observation data, including an image data acquisition end, an image feature analysis end, a texture vortex calibration end, a feature region clustering end, and a data classification and display end, wherein the image data acquisition end, the image feature analysis end, the texture vortex calibration end, the feature region clustering end, and the data classification and display end are electrically connected from the output node to the input node in sequence;
[0031] Among them, the image data acquisition end acquires remote sensing images associated with the observation area and transmits the acquired remote sensing images to the image feature analysis end. When acquiring images, there are access permissions, and the corresponding remote sensing images are updated regularly. The acquired remote sensing images are all the most recently updated images, and the remote sensing images are only for the relevant images of the corresponding observation area at sea.
[0032] Among them, the image feature analysis end performs feature verification on the acquired remote sensing image. Based on the gradient features associated between image points, the gradient pixels in the remote sensing image are sequentially calibrated, and the gradient contour line is locked based on the calibrated gradient pixels. The gradient contour line is the outer contour line formed by connecting several consecutive gradient pixels.
[0033] The gradient contour line is determined as follows:
[0034] The Sobel algorithm is used to identify the vertical and longitudinal gradients associated with different image points within the remote sensing image. Based on these identified gradients, the comprehensive gradient associated with each image point is determined. Both the vertical and longitudinal gradients are obtained by convolving and summing the gradients of the corresponding pixel's surrounding pixels and their associated weight factors. Since the methods for identifying pixel vertical and longitudinal gradients are quite common, they will not be elaborated upon here. ;
[0035] The comprehensive gradient associated with different image points in the remote sensing image is compared with the preset value Y1: if the comprehensive gradient is ≥ Y1, the corresponding image point is recorded as the gradient point; otherwise, no marking is made. Y1 is the preset value, and its specific value is determined by the operator based on experience.
[0036] Connect consecutive gradient points in the remote sensing image to confirm the outline of the associated points, and identify several gradient outline lines in the remote sensing image.
[0037] When gradient contour lines generally exist, it means that there is a vortex above the corresponding sea surface. Each vortex has several surrounding arcs of gradient contours, which allows for quick and effective identification of gradient contour lines.
[0038] The texture vortex calibration end performs feature verification on the remote sensing image after gradient contour line calibration. Based on the curvature characteristics of different gradient contour lines, it identifies the feature points associated with the corresponding gradient contour lines. Then, it performs cluster analysis on the identified feature points to lock the texture vortex. The specific locking method is as follows:
[0039] The confirmed gradient contour line is confirmed by arc. Starting from one end of the gradient contour line, the arc is confirmed from the other end. The moving segments between connected points are confirmed, and the arc is confirmed in sequence. Multiple moving segments with the same arc are integrated to confirm a single arc segment. Then the same processing method is used to confirm the subsequent arc segments, so that a single gradient contour line is divided into multiple arc segments with different arc characteristics.
[0040] Based on the arc features of each arc segment, identify several arc points associated with the arc segment, and then lock the feature points from the remote sensing image. The straight-line distance between the feature points and the several arc points is the same. The feature points associated with the arc segment are confirmed in sequence.
[0041] Based on several different feature points identified by different gradient contours, a set of feature points is randomly selected as the center point, and a search circle about the center point is generated with a preset radius R1 as the search range. The number of feature points inside the search circle is denoted as Q. Then, other feature points are successively used as center points, and the number of feature points Q in the corresponding search circle is simultaneously confirmed. From the confirmed Q, the maximum value is selected, and the center point associated with the maximum value is denoted as the selected point. The selected point and the feature points associated with the corresponding search circle are all denoted as vortex points. The arc segments associated with several vortex points are confirmed, and the texture vortex associated with this vortex point is generated.
[0042] Then, the same processing method is used to confirm the vortexes for other feature points that are not labeled as vortex points. Several texture vortices existing in the remote sensing image are labeled in sequence, and the remote sensing image with the texture vortexes labeled is transmitted to the feature region clustering end.
[0043] Specifically, the several arc segments associated with the texture vortex all have different arc centers, and the center point is the confirmed feature point. When a corresponding vortex is generated on the corresponding sea surface, the associated feature points should all be located at the center of the corresponding vortex. Therefore, by using the method of finding the same point to confirm the same point, the feature points associated with the corresponding texture vortex can be quickly confirmed. In the specific confirmation process, the associated texture vortices in the remote sensing image are confirmed in sequence, which facilitates subsequent feature verification.
[0044] Second Embodiment
[0045] The first embodiment mainly focuses on the process of confirming texture vortices on the sea surface. This embodiment mainly focuses on the clustering process of several texture vortices. From the center position of several associated texture vortices, feature clustering analysis is performed to identify whether multiple texture vortices are located in the same location area and clustering processing is performed.
[0046] In the feature region clustering stage, the remote sensing images with completed texture vortex calibration are reprocessed. Based on the positional characteristics of different texture vortices, dense regions are identified. The specific identification method is as follows:
[0047] Identify the selected points marked within each texture vortex, confirm the two selected points with the closest straight-line distance from the remote sensing image, connect the two selected points, confirm the connecting line, and then confirm the selected point with the closest perpendicular distance to the connecting line. Generate an undetermined triangle by connecting the connecting line and the current selected point, and record the area parameter M1 of the current undetermined triangle.
[0048] Based on the confirmed undetermined triangle, randomly select other surrounding points to generate a quadrilateral and record the area M2 of the corresponding quadrilateral. If M1 and M2 satisfy: M2 < 2M1, then retain the currently confirmed quadrilateral; otherwise, reconstruct the quadrilateral. If there is no quadrilateral that satisfies M2 < 2M1, then terminate the current dense confirmation process.
[0049] Based on the confirmed quadrilateral, randomly select other selected points around it to generate a pentagon, and record the area of the pentagon as M3. Then identify whether M3 satisfies: M3 < 3M1. If it does, keep the currently confirmed pentagon. If it does not, continue to confirm other pentagons. If there is no pentagon that satisfies M2 < 3M1, record the selected points associated with the corresponding quadrilateral as dense points and terminate the current dense confirmation process.
[0050] Similarly, based on the corresponding polygons confirmed in the previous stage, the newly added polygons associated in the next stage are confirmed, and the dense points are confirmed in turn. When all selected points in the remote sensing image have completed the dense confirmation process, several dense points associated with each dense confirmation process are taken as similar dense points, and multiple texture vortices associated with similar dense points are recorded as the same dense region.
[0051] Specifically, several texture vortices are identified within the remote sensing image. Ten vortices are identified, namely A1, A2, ..., A10. After confirmation, A1 and A2 form a line, which, combined with A3, generates a triangle, meaning A3 is closest to the line. In the subsequent confirmation process, A4 and A6 are identified as dense points. Therefore, the currently identified dense points include {A1, A2, A3, A4, A6}. These dense points belong to a dense region. If other dense points exist, the corresponding confirmation process needs to be executed. First, the two closest points are identified and a line is generated, then the triangle is constructed, and then the dense region is confirmed again, completing the confirmation process for the subsequent dense regions.
[0052] The data classification and display terminal identifies other observation data associated with the corresponding dense areas based on several marked dense areas, and bundles the other observation data with multiple texture vortices associated with the corresponding single dense area to generate a bundled data package belonging to the corresponding dense area. The terminal then displays the different bundled data packages associated with different dense areas.
[0053] This section facilitates the subsequent data classification and verification process, achieving optimal data classification and processing results, and displays the data in categories for review by relevant personnel, enabling timely data organization and observation, and prompting appropriate action.
[0054] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.
[0055] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. An intelligent processing system for marine observation data, characterized in that, include: The image data acquisition end acquires remote sensing images associated with the observation area; The image feature analysis end performs feature verification on the acquired remote sensing images, determines the gradient features associated between image points and completes the calibration process of gradient pixels, and then locks the gradient contour line. The texture vortex calibration end identifies the feature points associated with the corresponding gradient contours based on the curvature characteristics of different gradient contours. Then, cluster analysis is performed on the identified feature points to locate the texture vortex. The specific method is as follows: The confirmed gradient contour line is confirmed by arc. Starting from one end of the gradient contour line, the arc is confirmed from the other end. The moving segments between connected points are confirmed, and the arc is confirmed in sequence. Multiple moving segments with the same arc are integrated to confirm a single arc segment. Then the same processing method is used to confirm the subsequent arc segments, so that a single gradient contour line is divided into multiple arc segments with different arc characteristics. Based on the arc features of each arc segment, identify several arc points associated with the arc segment, and then lock the feature points from the remote sensing image. The straight-line distance between the feature points and the several arc points is the same. The feature points associated with the arc segment are confirmed in sequence. Based on several different feature points identified by different gradient contours, a set of feature points is randomly selected as the center point, and a search circle about the center point is generated with a preset radius R1 as the search range. The number of feature points inside the search circle is denoted as Q. Then, other feature points are successively used as center points, and the number of feature points Q in the corresponding search circle is simultaneously confirmed. The maximum value is selected from the confirmed Q, and the center point associated with the maximum value is denoted as the selected point. The selected point and the feature points associated with the corresponding search circle are all denoted as vortex points. The arc segments associated with several vortex points are confirmed, and the texture vortex associated with this vortex point is generated. Then, the same processing method is used to confirm the vortexes for other feature points that are not labeled as vortex points. Several texture vortices existing in the remote sensing image are labeled in sequence, and the remote sensing image with the texture vortexes labeled is transmitted to the feature region clustering end. In the feature region clustering stage, the remote sensing images that have completed texture vortex calibration are reprocessed to identify dense regions based on the positional characteristics of different texture vortices.
2. The intelligent processing system for ocean observation data according to claim 1, wherein, The image feature analysis terminal uses the following method to lock the gradient contour line: The Sobel algorithm is used to confirm the vertical gradient and the vertical gradient associated with different image points in the remote sensing image, and the comprehensive gradient associated with the corresponding image points is locked according to the confirmed vertical gradient and vertical gradient, and ; The comprehensive gradient associated with different image points in the remote sensing image is compared with the preset value Y1: if the comprehensive gradient is ≥ Y1, the corresponding image point is recorded as the gradient point, where Y1 is the preset value. Connect consecutive gradient points within the remote sensing image to confirm the outlines of the associated points, and identify several gradient outlines within the remote sensing image.
3. The intelligent oceanographic data processing system of claim 2, wherein, If the combined gradient is less than Y1, no labeling is performed.
4. The intelligent ocean observation data processing system of claim 1, wherein, The specific method for identifying dense regions in the feature region clustering endpoint is as follows: Identify the selected points marked within each texture vortex, confirm the two selected points with the closest straight-line distance from the remote sensing image, connect the two selected points, confirm the connecting line, and then confirm the selected point with the closest perpendicular distance to the connecting line. Generate an undetermined triangle by connecting the connecting line and the current selected point, and record the area parameter M1 of the current undetermined triangle. Based on the confirmed undetermined triangle, randomly select other surrounding points to generate a quadrilateral and record the area M2 of the corresponding quadrilateral. If M1 and M2 satisfy: M2 < 2M1, then retain the currently confirmed quadrilateral; otherwise, reconstruct the quadrilateral. If there is no quadrilateral that satisfies M2 < 2M1, then terminate the current dense confirmation process. Based on the confirmed quadrilateral, randomly select other selected points around it to generate a pentagon, and record the area of the pentagon as M3. Then identify whether M3 satisfies: M3 < 3M1. If it does, keep the currently confirmed pentagon. If it does not, continue to confirm other pentagons. If there is no pentagon that satisfies M2 < 3M1, record the selected points associated with the corresponding quadrilateral as dense points and terminate the current dense confirmation process. Similarly, based on the corresponding polygons confirmed in the previous stage, the newly added polygons associated in the next stage are confirmed, and the dense points are confirmed in turn. When all selected points in the remote sensing image have completed the dense confirmation process, several dense points associated with each dense confirmation process are regarded as similar dense points, and multiple texture vortices associated with similar dense points are recorded as the same dense region.
5. The intelligent oceanographic data processing system of claim 1, wherein, Also includes: The data classification and display terminal identifies other observation data associated with the corresponding dense regions based on several marked dense regions, and bundles the other observation data with multiple texture vortices associated with the corresponding single dense region to generate a bundled data package belonging to the corresponding dense region. The terminal then displays the different bundled data packages associated with different dense regions.