International ocean observation data fusion and achievement mutual recognition system suitable for joint scientific investigation

Abstract

This application relates to the field of international marine scientific research collaboration and mutual recognition technology. The invention discloses an international marine observation data fusion and results mutual recognition system suitable for joint scientific expeditions, including a multi-source data standardization access module, a polar-adaptive intelligent fusion module, an international results mutual recognition standard and verification module, a hierarchical sharing and mutual recognition process management module, a collaborative iteration and operation and maintenance management module, a blockchain full-process traceability system, and an international joint collaborative operation and maintenance mechanism. This system addresses the heterogeneity of international data through multi-source data standardization preprocessing, utilizes intelligent fusion technology to achieve accurate data fusion under complex polar / out-of-sea conditions, establishes a two-way verification mutual recognition system, and employs blockchain to ensure full-process traceability. The system supports hierarchical sharing and dynamic iterative optimization, and is operated and maintained collaboratively by international parties, ensuring long-term adaptation to both parties' standard updates and equipment iterations, effectively improving the efficiency of joint scientific research data sharing and the credibility of results mutual recognition.

Description

Technical Field

[0001] This invention relates to the integrated application of artificial intelligence basic models, neural operator learning, privacy computing, blockchain cross-chain interoperability and digital twin technology in the field of international marine scientific research collaboration. Specifically, it involves a marine observation data fusion and results mutual recognition system suitable for international joint scientific research. Background Technology

[0002] International joint marine scientific expeditions are an important form of cooperation in polar and open-sea marine research. However, existing technologies face many bottlenecks in the data processing and mutual recognition of results between the two sides, making it difficult to meet the needs of efficient joint scientific expeditions.

[0003] 1. Significant differences exist in the equipment, data standards, and formats used for ocean observation by international parties, creating a serious data heterogeneity barrier. Traditional data processing often employs simple splicing methods, resulting in data utilization rates of less than 40%. The fused data exhibits poor consistency and cannot effectively integrate multi-source data.

[0004] 2. There is a lack of a unified and operable mutual recognition standard system for international scientific research results. The mutual recognition process mainly relies on offline document transmission and multi-department coordination, which takes up 70% of the original cycle. The cycle for a single mutual recognition of results can be as long as 15 to 20 days, which is a pain point in the industry.

[0005] 3. The general data fusion model is not adapted to the complex working conditions such as low temperature, strong current, and sea ice interference in the polar / outer sea. Its ability to identify abnormal data and its system fault tolerance are weak. The fusion accuracy on the polar test dataset is only about 80%, which cannot meet the high-precision requirements of joint scientific expeditions.

[0006] 4. The access, fusion, results generation, mutual recognition and review, and application feedback of marine observation data are fragmented and have not formed a closed-loop collaborative management system. Data processing and results output cannot be synchronized with the progress of scientific research field operations, resulting in low efficiency of results transformation.

[0007] 5. The lack of a collaborative operation and maintenance mechanism involving both international parties makes it difficult for the system to adapt to updates in data standards and iterations in observation equipment. Furthermore, there is no reliable means of tracing the entire process of mutual recognition of results, and key step records can be tampered with, resulting in insufficient credibility and authority of joint results.

[0008] To address the aforementioned issues, there is currently no technical solution that can simultaneously resolve multiple problems such as data heterogeneity, inefficient mutual recognition of results, poor model adaptability, insufficient process collaboration, and lack of long-term operation and maintenance. There is an urgent need to develop a marine observation data fusion and results mutual recognition system specifically suitable for international joint scientific expeditions. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention proposes a marine observation data fusion and results mutual recognition system suitable for international joint scientific expeditions, comprising:

[0010] The multi-source data standardization access module is used to preprocess multi-source international ocean observation data, solve the problem of heterogeneity of ocean observation data from both sides, and realize unified access and standardized processing of multi-dimensional data;

[0011] The polar-adaptive intelligent fusion module is connected to the multi-source data standardization access module to receive pre-processed data, achieve accurate fusion of multi-dimensional data, and adapt and optimize for complex polar / ocean conditions to generate standardized results.

[0012] An international achievement mutual recognition standard and verification module is connected to the polar-adapted intelligent fusion module to establish an international scientific research achievement mutual recognition system and to perform two-way verification of the standardized achievements.

[0013] The hierarchical sharing and mutual recognition process management module is connected to the international achievement mutual recognition standard and verification module. It is used to manage the open sharing of verified achievements according to the hierarchical sharing mechanism and record the entire process information.

[0014] The collaborative iteration and operation and maintenance management module is connected to the above modules respectively. It is used to update and optimize the data standards, fusion model parameters and mutual recognition index system based on the feedback data of scientific research applications, so as to achieve long-term system adaptation.

[0015] The blockchain full-process traceability system, integrated with the hierarchical sharing and mutual recognition process management module, is used to record the information of the entire process of the results in an immutable manner.

[0016] An international joint operation and maintenance mechanism, with the participation of both international parties, ensures that the system dynamically adapts to updates in data standards and iterations in observation equipment.

[0017] Furthermore, the multi-source data standardization access module includes:

[0018] The International Marine Observation Data Two-Way Mapping Standard Library is used to integrate the indicator definitions, format specifications, accuracy requirements, and unit conversion rules of core data types in hydrology, meteorology, ecology, and geology from both international parties. It covers a variety of core data types and reserves interfaces for equipment and standard extensions.

[0019] The multi-source data automated adaptation and processing unit is connected to the international marine observation data bidirectional mapping standard library and is used to perform standardized processing operations such as format conversion, unit unification, index calibration, and initial screening of abnormal data.

[0020] A standardized data storage unit, connected to the multi-source data automated adaptation and processing unit, is used to generate standardized data files compatible with international dual standards and label metadata, and store them in an international distributed encrypted database, supporting bidirectional data synchronization and fast retrieval.

[0021] Furthermore, the initial screening of abnormal data in the multi-source data automated adaptation processing unit adopts a two-level screening mechanism:

[0022] The first level of screening involves setting reasonable physical range thresholds for observation indicators based on common sense in marine physics and historical scientific expedition experience, and eliminating invalid data that clearly exceeds the threshold.

[0023] The second level of screening uses the isolated forest algorithm, which combines data point values, spatiotemporal context features, and consistency of multi-parameter physical relationships to identify and classify outliers: invalid data caused by random errors are directly removed, while suspicious data that may indicate rare marine phenomena are marked, retained, and sent for manual review.

[0024] Furthermore, the polar-adaptive intelligent fusion module includes:

[0025] The fusion model architecture design unit adopts a hierarchical fusion and dynamic weighting architecture. The bottom layer is a data-level fusion based on the improved KNN algorithm, the middle layer is a feature-level fusion based on a one-dimensional convolutional neural network, and the top layer is a decision-level fusion based on an optimized random forest algorithm.

[0026] A polar operating condition adaptation and optimization unit, connected to the fusion model architecture design unit, is used to embed a polar marine environment interference factor compensation algorithm and a dynamic equipment error correction mechanism, and to optimize the initial screening of abnormal data, reducing fusion errors caused by environmental interference and equipment errors; and

[0027] The fusion result quality assessment unit is connected to the fusion model architecture design unit. It is used to establish a fusion data quality assessment index system, automatically generate quality reports, and trigger a re-fusion process for unqualified data.

[0028] Furthermore, the improved KNN algorithm uses weighted Euclidean distance instead of standard Euclidean distance, combines time, space, and observation feature weight factors to calculate similarity, and adopts a dynamic adjustment mechanism instead of a fixed K value, dynamically adjusting the K value according to the local density index of the data batch; the optimized random forest algorithm introduces a weighted voting mechanism in the training phase, assigning higher weights to decision trees that perform well on historical polar data, and supports feature importance analysis.

[0029] Furthermore, the international achievement mutual recognition standards and verification module includes:

[0030] The mutual recognition standard system construction unit is used to jointly develop a mutual recognition indicator system covering two types of outputs: data products and scientific research reports, and to clarify the accuracy threshold, expression standards, and credibility assessment dimensions.

[0031] The two-way verification mechanism design unit is connected to the mutual recognition standard system construction unit. It adopts a combination of automatic verification and manual review. The automatic verification stage adopts a parallel dual-channel verification architecture, which synchronously calls the independent standardized evaluation algorithms of both parties to verify the scientific research results. When the results conflict, a three-level arbitration mechanism is triggered. It also integrates a dynamic threshold adjustment module for polar data.

[0032] A manual review unit, connected to the bidirectional verification mechanism design unit, is used for joint review of results that fail automatic verification or are highly sensitive; and

[0033] The mutual recognition result generation unit is connected to the two-way verification mechanism design unit and the manual review unit, respectively. It is used to generate international bilingual mutual recognition certificates for verified results, push modification suggestions for unqualified results, and support resubmission for verification.

[0034] Furthermore, the hierarchical sharing and mutual recognition process management module includes:

[0035] The hierarchical sharing mechanism unit establishes a three-tier sharing system of open, restricted, and core based on the sensitivity level of the results, and configures access control policies for unverified access, joint authorization by two responsible persons, and dual key verification respectively;

[0036] A standardized mutual recognition process unit, connected to the hierarchical sharing mechanism unit, is used to build a fully online process and supports automatic reminders for process nodes.

[0037] The blockchain end-to-end traceability system unit is connected to the standardized mutual recognition process unit. It adopts a permissioned consortium blockchain architecture, stores end-to-end traceability metadata on the chain, and achieves immutability and two-way traceability of operation records through smart contract encoding of mutual recognition process rules.

[0038] Furthermore, the collaborative iteration and operation and maintenance management module includes:

[0039] The international joint iteration mechanism unit is used to update standards and models through quarterly technical collaboration meetings and to establish a rapid response channel for issues to jointly resolve disputes.

[0040] The model and system optimization unit is connected to the international joint iterative mechanism unit. It is used to perform incremental learning and parameter optimization on the abnormal data identification algorithm and environmental interference compensation mechanism based on scientific research application feedback data. It also adopts a modular design and reserves expansion interfaces.

[0041] The operation and maintenance monitoring and support unit is connected to the model and system optimization unit to build a system operation status monitoring platform, monitor the operation of data transmission, fusion processing, and mutual recognition processes in real time, realize automatic fault alarm, self-healing and backup plan triggering, and perform data backup and security testing regularly.

[0042] Furthermore, the monitoring platform of the operation and maintenance monitoring and protection unit adopts a baseline adaptive threshold strategy to set alarm rules, triggers alarms in layers according to the severity of the anomaly, and has a preset fault self-healing strategy library to perform fault detection and automatic response.

[0043] Furthermore, information on the steps of submission, verification, review, modification, and mutual recognition of results is recorded immutably. The international joint operation and maintenance mechanism is jointly participated in by both international parties, ensuring that the system dynamically adapts to updates in data standards and iterations in observation equipment.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0045] It pioneered an international standard for bidirectional mapping of multi-dimensional marine data and a fusion architecture of hierarchical fusion and dynamic weighting, which fundamentally solved the problem of data heterogeneity. The data fusion accuracy reached over 92%, and the data utilization rate was improved by 45%, far exceeding the technical effect of traditional simple splicing methods.

[0046] Establish a dedicated international scientific research achievement mutual recognition standard system and a fully online mutual recognition process. Adopt a two-way verification mechanism with automatic verification as the main method and manual review as a supplement. The mutual recognition cycle of achievements has been shortened from 15-20 days to within 72 hours, solving the industry pain point of difficulty in mutual recognition of achievements. Moreover, blockchain traceability ensures that the mutual recognition results are authoritative and credible.

[0047] The fusion model incorporates a polar environment interference compensation algorithm and a dynamic equipment error correction mechanism, and optimizes the abnormal data accurate identification module. It can adapt to complex polar / oceanic conditions such as low temperature, strong current, and sea ice interference. The system's robustness far exceeds that of general fusion models, and it fully meets the technical requirements of international offshore / polar joint scientific expeditions.

[0048] It achieves deep integration of the entire process of data access, fusion processing, results generation, mutual recognition and review, and application feedback. Data processing and results output can be synchronized with the progress of scientific expeditions in real time, providing immediate support for on-site decision-making and improving the efficiency of results transformation by 60%.

[0049] An international joint iteration mechanism has been established. The system adopts a modular design and reserves expansion interfaces, which can dynamically adapt to the updates of data standards of both parties, the iteration of observation equipment, and the expansion of scientific research scenarios. At the same time, the full-process traceability of blockchain and distributed encrypted storage ensure the security and credibility of the system, ultimately forming a replicable and scalable technical support paradigm for international marine scientific research cooperation. Attached Figure Description

[0050] Figure 1 A system architecture diagram for the fusion of international marine observation data;

[0051] Figure 2 A flowchart for the fusion of international marine observation data;

[0052] Figure 3 A flowchart illustrating the process of mutual recognition of results;

[0053] Figure 4 This is a diagram of the architecture of the international data mapping standard library. Detailed Implementation

[0054] To gain a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings.

[0055] This invention proposes a system for the fusion and mutual recognition of marine observation data suitable for international joint scientific expeditions, such as... Figure 1 As shown, it includes the following five core modules:

[0056] (a) Multi-source data standardization access module

[0057] The core solution addresses the heterogeneity of international marine observation data, enabling unified access and preprocessing of multi-dimensional data.

[0058] 1. Construct an international standard library for two-way mapping of marine observation data.

[0059] It integrates the indicator definitions, format specifications, accuracy requirements, and unit conversion rules of core data types from both parties, including hydrology, meteorology, ecology, and geology. It covers more than 20 core data types such as CTD data, ocean current data, and plankton monitoring data, and reserves interfaces for equipment and standard expansion.

[0060] Specific examples of the 20 core data types are as follows: CTD (Conductivity, Temperature, Depth) profile data, shipborne ADCP (Advanced Dip-Conversion Program) current data, long-term observation data of moored moorings, underway XBT (Xaponic Bipolar Transmission) temperature profile data, Argo buoy profile data, drifting buoy trajectory data, sea ice thickness radar observation data, sea ice concentration satellite remote sensing data, chlorophyll concentration fluorometer data, nutrient automatic analyzer data, dissolved oxygen sensor data, pH sensor data, turbidimeter data, wave height and wave period data, sea surface height satellite altimetry data, sea surface temperature satellite radiometer data, sea surface salinity satellite microwave radiometer data, seabed topography multibeam bathymetry data, sediment trap flux data, and plankton web sample analysis data. The standard mapping table for each data type includes: field naming conventions (international English, Chinese, and English), unit conversion formulas, accuracy level conventions, quality control mark definitions, and missing value coding rules.

[0061] Automated Adaptation and Processing of Multi-Source Data

[0062] During data access, the system automatically identifies the data source and data type based on a pre-set international marine observation data bidirectional mapping standard library, and performs standardization processing:

[0063] (1) Format conversion: The system has a built-in parser (supporting CSV, XML, NetCDF and other formats), which identifies the original format based on the file metadata and converts it into a unified JSON structure within the system.

[0064] (2) Unit Consistency: Based on the unit conversion rules in the standard library, the system automatically identifies the original data unit (such as pressure unit "mmHg" or "hPa") and converts it to international units or units agreed upon by both parties through linear conversion formulas, such as pressure hPa = pressure mmHg × 1.3332239. For conversions that depend on environmental parameters such as depth and latitude, the system dynamically calls the correction factor table to ensure accuracy.

[0065] (3) Index Calibration: To address the differences in international equipment accuracy, such as the CTD data accuracy of ±0.002°C vs. in some countries, the data is uniformly mapped to a common benchmark accuracy using a predefined linear transformation function in the standard library, and higher weights are assigned to high-precision data. For index definitions or threshold conflicts, such as chlorophyll concentration standards, a three-level rule is applied: international standards are preferred; if no international standard exists, the jointly agreed standard is used; disputed indicators are automatically marked and manual arbitration is triggered.

[0066] (4) Initial screening of abnormal data: This is carried out simultaneously in the standardized process, using a two-level screening mechanism that combines thresholding and machine learning. The system first pre-sets reasonable physical range thresholds for various observation indicators in the mapping standard library based on common sense in marine physics and historical scientific expedition experience, such as the seawater temperature range [-2°C, 15°C] and salinity range [28 PSU, 35 PSU] in the Bering Strait. After data access, the system automatically compares the value of each data point with the threshold range of the corresponding indicator. Data that clearly exceeds the threshold range is judged as invalid data and directly removed, with the reason for removal recorded in the system log. Such data is usually caused by hard errors such as equipment power failure or sensor malfunction. Data filtered through thresholding is further refined using an unsupervised learning algorithm. In addition to the data point values, the algorithm input features also include contextual features, such as the difference rate with adjacent measurement points, the trend of change over time, and the consistency of the physical relationship with related parameters. The algorithm learns and models the distribution pattern of normal data to identify abnormal points in the feature space that significantly deviate from the distribution of the main data. For identified anomalies, the system classifies and processes them according to their characteristic patterns: if the data is determined to be invalid due to random errors such as momentary equipment interference or data transmission errors, it is directly removed; if the characteristic pattern may correspond to rare marine phenomena or special environmental interference, it is judged as suspicious data, marked as "suspicious" and retained in a dedicated database. At the same time, a suggested review list containing data IDs and descriptions of anomalies is generated and pushed to international data quality control experts for manual review. The experts will then make the final decision on whether to include the data in the analysis or remove it.

[0067] 3. Standardized data storage

[0068] Generates standardized data files compatible with international dual standards, annotates metadata such as acquisition time, equipment model, observation location, and processing flow, and stores them in an international distributed encrypted database, supporting bidirectional data synchronization and rapid retrieval.

[0069] In a preferred embodiment, a cognitive semantic standardization access module is embedded in the multi-source data standardization access module. This module is equipped with a multimodal large language model pre-trained on millions of marine science and technology documents and hundreds of thousands of sets of international historical observation data, as well as a vector knowledge graph-enhanced retrieval engine built based on international marine observation standards. This cognitive semantic standardization access module performs semantic-level intelligent parsing of multi-source heterogeneous international marine observation data, automatically identifying key observation elements in scanned copies of handwritten observation logs, photographs of paper records, and multilingual emails. For undefined fields generated by new observation equipment, it automatically analyzes semantic features and retrieves similar standard fields in the knowledge graph, generating mapping suggestions and confidence scores. For fields with ambiguity, it infers the physical meaning by combining contextual information such as observation location depth, seasonal characteristics, and equipment type, achieving dynamic standard mapping and autonomous discovery of emerging data types, thus overcoming the bottleneck of traditional rule engines in handling unstructured and semantically ambiguous data.

[0070] (ii) Polar-adaptive intelligent fusion module

[0071] The core functionality enables precise fusion of multi-dimensional data to adapt to complex working conditions.

[0072] 1. Fusion Model Architecture Design

[0073] A layered fusion architecture combined with dynamic weighting is adopted, with data-level fusion at the bottom layer. An improved KNN algorithm is used to align multi-source data and remove redundancy. The specific implementation process is as follows: First, data preprocessing and feature vector construction are performed. International multi-source data processed by the standardized access module are uniformly indexed according to observation time and geographic coordinates. Each data point is constructed as a feature vector containing the observation value and its spatiotemporal context information (timestamp, latitude, longitude, and depth). In the similarity calculation stage, weighted Euclidean distance is used instead of standard Euclidean distance. Its calculation formula comprehensively considers the weighting factors (denoted as wt, ws, and wv) of time, space, and observation features. The weight values ​​are set based on historical scientific expedition data analysis and domain expert experience, ensuring that data points with closer timestamps and more adjacent geographic locations receive higher similarity weights.

[0074] SpatialDist(i,j) is a function for calculating the spherical distance between two points to ensure the accuracy of geographic distance calculation. This method effectively overcomes the nearest neighbor matching bias caused by inconsistent spatiotemporal scales. For neighborhood selection, a dynamic adjustment mechanism replaces the fixed K value. The algorithm dynamically adjusts the K value based on the local density index of the current batch of data being processed: automatically increasing the K value in sparse data regions to capture a sufficient number of nearest neighbors for effective alignment; and decreasing the K value in dense data regions to avoid introducing irrelevant or redundant information. This mechanism ensures the robustness of the algorithm under different data distributions. After completing the nearest neighbor search, data alignment and redundancy removal are performed. For the target data point to be aligned, the algorithm finds its K nearest neighbors using the above method and aligns the data by comparing the feature values ​​of the target point and its nearest neighbors. Subsequently, based on the statistical distribution of the feature values ​​of the nearest neighbors, if the difference between the target point and its nearest neighbors exceeds a preset redundancy threshold, the point is determined to be redundant data and removed. Finally, a clean dataset with significantly reduced redundancy after alignment is output for subsequent feature-level fusion.

[0075] Through the above improvements, this algorithm achieves accurate spatiotemporal alignment of international heterogeneous ocean observation data on the test dataset.

[0076] The middle layer employs feature-level fusion, extracting core features from various dimensions of data through convolutional neural networks. This network structure is specifically designed for profile data arranged along depth or time dimensions, effectively capturing local continuous change patterns. The core architecture of the network consists of alternating stacked one-dimensional convolutional layers and pooling layers. The convolutional layers use multiple one-dimensional convolutional kernels to slide along the sequence direction, identifying the correlation features between adjacent sampling points through a local perception mechanism. For example, a convolutional kernel of size 3 can continuously perceive the temperature and salinity change trends of three adjacent sampling points in the depth direction, thereby effectively identifying the feature patterns of key ocean phenomena such as thermoclines and salinity fronts. A ReLU activation function is applied after each convolutional operation to enhance nonlinear expressive power. The pooling layers use max pooling to reduce the dimensionality of features, improving the model's translation invariance and computational efficiency while preserving salient features. Through multiple layers of convolutional-pooling operations, the network achieves a step-by-step extraction from shallow local features to high-level abstract features. Shallow convolutional layers primarily capture basic patterns, such as temperature gradients and salinity fluctuations, while deeper convolutional layers fuse these local features to form more representative higher-order features, such as complete thermocline structures and multi-parameter coupling relationships. The final output feature map gathers the core spatial patterns of data from various dimensions, providing high-quality input for decision-level fusion. Compared to traditional manual feature extraction methods, this 1D-CNN structure can adaptively learn complex nonlinear relationships in marine data, significantly improving the accuracy of feature representation and the overall performance of the fusion model. The top layer is for decision-level fusion, using a random forest algorithm to achieve multi-feature comprehensive decision output. This algorithm effectively improves the model's generalization ability and robustness by constructing multiple decision trees and integrating their prediction results.

[0077] The specific technical implementation includes the following steps:

[0078] First, the input features are standardized and preprocessed. The input to the random forest is a multi-dimensional feature vector extracted by the feature-level fusion module, covering abstract representations of various types of data such as hydrology, meteorology, and ecology, including trends in temperature, salinity, and depth profiles, and spatiotemporal patterns of ocean currents. To ensure the comparability of features with different dimensions, the input features need to be standardized. Randomness is introduced during the decision tree construction process to enhance model diversity. The Bootstrap sampling method is used to generate multiple subsets of the training data, each of which is used to independently train a decision tree. When splitting a decision tree node, some features are randomly selected as candidate splitting features, and the split point is evaluated using the Gini coefficient or information gain to avoid overfitting and improve model stability. Multi-tree collaborative decision-making is achieved through a voting mechanism. Each decision tree independently predicts the input features, outputting class labels in classification tasks and continuous values, such as optimized estimates of temperature and salinity, in regression tasks.

[0079] The system synthesizes the predictions of all decision trees through majority voting or averaging to form the final decision. The algorithm has been adapted and optimized to address the unique characteristics of polar conditions. A weighted voting mechanism is introduced during the training phase, assigning higher weights to decision trees that perform well on historical polar data, thereby improving the model's tolerance to noise and distribution shifts in polar data. Testing has verified that this random forest algorithm reduces the mean squared error by approximately 18% in fusion decision-making and supports feature importance analysis, significantly improving the accuracy and interpretability of the decisions.

[0080] In a preferred embodiment, an algorithm for compensating for interference factors in the polar marine environment (low temperature, strong current, signal interference) is embedded to establish a dynamic correction mechanism for equipment errors. Environmental interference compensation mainly targets factors such as low polar temperatures, strong currents, and unstable signal transmission. In low-temperature environments, sensor measurements are prone to drift. The system establishes a temperature-error correction function based on the equipment's factory calibration data and historical low-temperature measurement records. Taking a CTD sensor as an example, when the water temperature is below -2°C, the system automatically calls a piecewise correction algorithm to compensate for the original measurement value. The correction formula combines the difference between the ambient temperature and the sensor's reference calibration temperature, and achieves nonlinear error fitting through a combination of quadratic and linear terms. For ADCP velocity data fluctuations caused by strong currents, dynamic smoothing based on Kalman filtering is used, combined with flow profile continuity constraints, to suppress invalid fluctuations and retain the true flow velocity trend. For signal loss caused by satellite transmission packet loss or sudden interference, the system uses time series prediction methods to reconstruct the missing segments and verifies the rationality of the reconstruction results through correlation with data from neighboring stations. Dynamic correction of equipment errors addresses system errors caused by internationally sourced equipment. The system incorporates a built-in equipment error feature library, integrating calibration error statistics for various sensors from past scientific expeditions. When new data is received, the system compares the current equipment output with the baseline values ​​of similar equipment in the feature library under the same environment in real time to identify any stable deviations. If a systematic error is identified, an adaptive weighted least squares algorithm is activated to dynamically update the correction parameters within a sliding time window, achieving online calibration of the equipment output. After each batch of data processing, the correction model compares the correction results with high-precision baseline data. If the residual exceeds the tolerance, the error feature library update process is automatically triggered, forming a closed-loop optimization. The anomaly identification module is optimized based on the above compensation and correction. This module integrates environmentally compensated data and uses the isolated forest algorithm combined with multi-dimensional features such as the spatiotemporal continuity and physical constraint consistency of data points for anomaly detection, reducing the probability of misjudgment due to environmental interference. For identified abnormal data, the system analyzes its causes using the equipment error model, classifies and marks it, and pushes it to the manual review interface for further expert judgment. Through the aforementioned compensation algorithm and dynamic correction model, the system reduces errors caused by environmental interference by approximately 40% in polar testing, achieving an equipment system error correction accuracy of over 95%. The optimized abnormal data identification module, combined with international data quality assessment standards, achieves precise quality control of polar marine observation data through a multi-level collaborative mechanism. This module first performs rapid initial screening of incoming data based on preset physical parameter thresholds, directly marking data significantly exceeding reasonable ranges as hard errors and removing them. For data within the threshold range, the module employs an isolated forest algorithm combined with local outlier detection for refined identification. The algorithm's input features encompass data numerical values, spatiotemporal continuity, and consistency of multi-parameter physical relationships.To address the spatiotemporal heterogeneity of polar data, the system dynamically adjusts anomaly detection thresholds: tolerance is relaxed in sparse data areas to avoid mistakenly deleting rare data; detection sensitivity is enhanced in dense data areas. The module also collaborates with environmental interference compensation algorithms and equipment error correction models to analyze the causes of identified anomalies. If the anomaly pattern closely matches known interference, an automatic repair process is triggered; if the anomaly cannot be explained by existing models and exhibits spatiotemporal clustering, it is retained as suspicious data and pushed for manual review to avoid misjudging potential marine phenomena. The system retrains the detection model quarterly based on newly added scientific research data and optimizes algorithm weights using the results of manual review.

[0081] 2. Quality assessment of fusion results

[0082] like Figure 2 As shown, a data fusion quality assessment index system is established, automatically generating quality reports. Unqualified data triggers a re-fusion process to ensure the reliability of the fused data. The system first assesses data consistency, using correlation coefficients and root mean square error to quantify the statistical correlation between international source data and fused data. For example, for temperature, salinity, and depth profile data, the system automatically calculates the correlation between the fused results and the original observation data on the vertical profile and spatiotemporal grid (target > 0.95), analyzes the deviation distribution characteristics, and identifies systematic offsets caused by sensor calibration differences. Integrity assessment uses a spatiotemporal gridding method to statistically analyze the ratio of the fused data coverage to the total amount of original data (target > 98%), marks uncovered grid cells, and verifies the rationality of missing areas in conjunction with scientific expedition plans. Accuracy assessment uses high-precision benchmark data as a reference, calculates the RMSE and MAE of the fused data in a specific verification area, and dynamically adjusts the fusion model parameters. Addressing the spatiotemporal heterogeneity of ocean data, the system analyzes spatial correlation using spatiotemporal variability functions and combines sliding windows to detect abrupt changes in time series, assessing the temporal smoothness and spatial continuity of the fused results. The physical rationality assessment is embedded in a physical rule base, automatically verifying whether the data conforms to constraints such as temperature-salinity relationship and geostrophic equilibrium, and marking data points that violate physical laws. The system automatically generates an assessment report containing scores for each indicator and visual analysis. A threshold trigger mechanism is set: if a core indicator falls below the threshold, re-fusion is automatically triggered; when secondary indicators are abnormal, an alert is generated and pushed to manual review. The assessment results are used to optimize the fusion model, for example, automatically adjusting the weight allocation in the weighted fusion algorithm when systematic biases are identified.

[0083] In one specific embodiment, the consistency metric uses the Pearson correlation coefficient, requiring a correlation coefficient greater than 0.95 between the fused result and the original observation data; the integrity metric uses spatiotemporal grid coverage, calculated using a grid of 0.1° × 0.1° × 1 day, requiring a coverage rate greater than 98%; the accuracy metric uses root mean square error (RMSE), requiring RMSE ≤ 0.05°C for temperature data, RMSE ≤ 0.02 PSU for salinity data, and RMSE ≤ 0.1 m / s for flow velocity data. The automatically generated quality report includes: scores for each metric, detailed explanations of failed items, a heatmap of spatial error distribution, and time-series comparison curves. The re-fusion process is triggered when any core metric falls below a threshold. During re-fusion, the fusion model parameters are automatically adjusted, prioritizing the data source with higher weights.

[0084] In a preferred embodiment, the polar-adaptive intelligent fusion module is embedded with a physical constraint neural operator fusion module and a generative anomaly detection module.

[0085] The physical constraint neural operator fusion module is connected to the cognitive semantic standardization access module. It is configured with Fourier neural operators embedded with Navier-Stokes equation residual constraints, first law of thermodynamics constraints, and seawater state equation constraints, a graph attention network, and a physical information neural network. This module is used to learn a mapping operator from sparse, irregular observations to a high-resolution fused field in an infinite-dimensional function space. It captures the coupling relationship between global patterns and local features in the frequency domain through Fourier transform, supporting super-resolution reconstruction. Physical regularization constraints ensure that the fusion result satisfies mass, momentum, and energy conservation. The graph attention network automatically learns the information transfer weights between heterogeneous observation stations, replacing manual weighting rules, achieving continuous spatial learning guided by physical laws, and avoiding false structures caused by physical inconsistencies in observation blind spots.

[0086] The generative anomaly detection module is connected to the cognitive semantic standardization access module and is equipped with a probabilistic manifold learning engine based on a diffusion model and a contrastive learning protection mechanism. This module uses a diffusion model to learn the probability distribution of historical normal data and jointly determines anomalies by combining reconstruction error and path likelihood. For data points determined to be anomalous, it further performs similarity calculations by comparing them with known rare marine phenomenon feature templates. When the similarity exceeds a threshold, it is determined to be a potential scientific discovery rather than an equipment malfunction and is marked as a priority review category, thus solving the problem of traditional discriminative methods mistakenly deleting real marine extreme events.

[0087] (III) International Standards for Mutual Recognition of Achievements and Verification Module

[0088] The core is to establish a mutual recognition system for research results, and achieve accurate two-way mutual recognition.

[0089] 1. Establishment of a mutual recognition standards system

[0090] A joint international scientific research achievement mutual recognition index system will be developed, covering two types of achievements: data products (fusion datasets, statistical analysis results) and research reports (observational conclusions, trend analysis). The system will clarify the accuracy threshold (data product error ≤ 5%), expression standards, and credibility assessment dimensions (data source, processing procedure, verification method).

[0091] 2. Design of Two-Way Verification Mechanism

[0092] The system employs a combined approach of automatic verification and manual review. The specific technical process of the automatic verification phase is as follows: The system uses a parallel dual-channel verification architecture, simultaneously calling independent standardized evaluation algorithms from both international parties to verify the same scientific research results. The Chinese standard algorithm channel loads the Chinese scientific research data quality specifications to quantitatively evaluate the consistency, accuracy, and standardization of the results. The national standard algorithm channel, based on the other party's marine observation standards, uses the other party's unique verification logic to evaluate the same indicators. The two algorithms operate in isolated environments using Docker containers to ensure the evaluation process is independent and reliable. After verification, the system automatically aligns the output results of both algorithms and generates a comparison matrix. If both algorithms agree on the same indicator, a verification pass conclusion is directly generated; if the results conflict, the system automatically triggers a three-level arbitration mechanism: priority is given to re-verification using internationally accepted standards; if international standards do not explicitly specify, a decision-making model based on weighted voting is initiated; indicators that cannot be consistently agreed upon are marked as high-dispute points and pushed to the manual review queue. Considering the special characteristics of polar data, the system integrates a dynamic threshold adjustment module. For example, regarding salinity data from the Arctic Ocean, the system automatically relaxes the accuracy tolerance (from ±0.02 PSU to ±0.05 PSU) based on the statistical distribution characteristics of historical joint scientific expedition data, avoiding misjudgments caused by environmental interference. Finally, the system integrates the conclusions of both algorithms to generate a joint verification report, marking the consistency score of both algorithms, and automatically locating the failure points for results that fail verification, providing targeted modification suggestions. All verification data and arbitration records are uploaded to the blockchain traceability system in real time to ensure the process is auditable. This mechanism reduced the misjudgment rate to below 3% in testing, significantly improving the scientific rigor and fairness of the mutually recognized results.

[0093] 3. During the manual review stage, the international joint audit team reviews the results that fail the automatic verification or are highly sensitive, and forms a review opinion.

[0094] 4. Generation of mutual recognition results

[0095] Validated results automatically generate international bilingual mutual recognition certificates, indicating the mutual recognition number, reviewer, and verification conclusion; unqualified results are pushed with modification suggestions and support resubmission for verification, ensuring the accuracy and authority of the mutual recognition results.

[0096] In a preferred embodiment, the international achievement mutual recognition standard and verification module embeds a zero-knowledge verifiable mutual recognition module. This module is connected to the physical constraint neural operator fusion module and is configured with a proof generation and verification circuit based on the zk-SNARKs protocol, as well as a secure multi-party computation arbitration protocol based on the GMW scheme. This zero-knowledge verifiable mutual recognition module encodes the data fusion algorithm into an arithmetic circuit, generating a zero-knowledge proof of the processing correctness. Without disclosing the original observation data and model parameters, it proves to the verifier that the output result was indeed correctly calculated by the agreed algorithm. When the verification results of both parties conflict, the secure multi-party computation arbitration protocol is initiated, jointly calculating the consistency score under cryptographic guarantees. Neither party can spy on the other's input, achieving bidirectional verifiable mutual recognition and fair arbitration under privacy protection.

[0097] In a specific embodiment, the international bilingual mutual recognition certificate includes the following information: certificate number, achievement name, achievement type, submitting organization, submission date, verification conclusion, measured value of accuracy index, mutual recognition validity period, digital signature, and QR code. Suggestions for revising unqualified achievements include: specific failed indicators, reference standard clauses, suggested revision directions, and the allowed timeframe for a second submission.

[0098] (iv) Hierarchical sharing and mutual recognition process management module

[0099] like Figure 3 As shown, the core achievement is precise sharing of results and collaborative management throughout the entire process.

[0100] 1. Tiered sharing mechanism

[0101] A three-tiered sharing system is established based on the sensitivity level of the results. Publicly available results are free to all collaborating institutions, restricted results require joint authorization from both project leaders, and core results are only open to the core research team and require dual key verification.

[0102] The specific technical details are as follows: Open-level sharing is for mutually recognized and verified basic marine environmental data and publicly available scientific expedition reports. This type of data allows anonymous access to researchers from all collaborating institutions without identity verification. The system embeds standard digital watermarks in the data to identify its source, primarily for non-classified scenarios such as scientific paper citations and public science education. Restricted-level sharing is applicable to raw observation datasets containing specific cruise details, high-precision fused data products, and preliminary analysis conclusions. Access control employs an international dual-responsibility joint approval mechanism. Applicants must submit a usage description online, and after approval by both project leaders, the system generates a temporary access token valid for 7–30 days. Data transmission is protected using AES-256 encryption to ensure it is used exclusively by members of the collaborating project team for in-depth analysis and model validation, and other scientific research activities. Core-level sharing covers key waterway hydrogeological information, sensitive species distribution data, and unpublished materials with military or commercial value. This level implements a dual-key verification mechanism. Access requires joint decryption by the core teams of both international parties, each holding a key. Each operation records the device MAC address, access time, and other operation fingerprints, which are then stored on the blockchain. Data storage employs a hybrid encryption method, using RSA-2048 asymmetric encryption for the transmission key, combined with AES-256 encryption for the data body, ensuring that it is used exclusively by high-level personnel of both parties for waterway safety planning and major scientific research decisions. This system achieves automatic policy execution through dynamic permission tags and smart contracts. When the confidentiality level of the results changes, the system automatically updates access permissions and notifies relevant parties. Tests show that this access control mechanism achieves an accuracy rate of 99.8%, with no instances of unauthorized access.

[0103] In a preferred embodiment, a hierarchical knowledge-sharing module is embedded, which is connected to a zero-knowledge verifiable mutual recognition module. This module is configured with a differential privacy-preserving federated learning framework, a CKKS homomorphic encryption computation unit, and a knowledge distillation engine. This module is used to construct a three-tiered knowledge-sharing architecture: public-level (model distillation sharing), restricted-level (federated gradient sharing), and core-level (homomorphic encryption computation). The data ontology remains within each entity's territory. The paradigm shift from data sharing to knowledge sharing is achieved through exchanging student models, noisy gradient updates, or encrypted state computation, complying with cross-border data flow regulations.

[0104] In one specific embodiment, the dual-key verification mechanism is clearly defined: The Shamir secret sharing scheme is adopted, splitting the decryption key into two parts. One country holds one key fragment, and the other country holds the other, preventing either party from decrypting independently. When accessing core-level data, authorized representatives from both parties must be online simultaneously within an agreed-upon time window (set to 15 minutes) to perform secure calculations on the key fragments using a hardware security module, reconstructing the complete decryption key. Each decryption operation generates an independent audit log, recording the operation time, operator identity, accessed data range, and operation purpose description. The logs are uploaded to the blockchain for evidence storage in real time.

[0105] 2. Standardized mutual recognition process

[0106] The system establishes online processes for submission, automatic verification, manual review, public announcement of results, and mutual recognition and archiving. It supports international parties in submitting results online, viewing review progress, and obtaining mutual recognition certificates. The system also provides automatic reminders for process milestones, thereby improving collaboration efficiency.

[0107] 3. End-to-end traceability system

[0108] A tamper-proof traceability system built on blockchain technology records the entire process of research results, from generation, submission, verification to mutual recognition. It includes modification records, review comments, and verification data, supporting two-way traceability and ensuring transparency and verifiability in the mutual recognition process. The system uses a permissioned consortium blockchain as its underlying architecture. Participating nodes must be authorized and certified by both international parties before access, ensuring the privacy and controllability of data interaction. Node deployment covers core international scientific research institutions; for example, one country's node is located at the National Marine Technology Center, while another country's node is deployed at its corresponding marine research institute. Both parties act as equal nodes, jointly maintaining the distributed ledger to achieve co-governance and sharing. On-chain storage primarily consists of lightweight traceability metadata, rather than raw scientific research data, to improve system efficiency. Each record contains key fields: Achievement ID serves as a unique identifier linking to the specific scientific research achievement; an operation timestamp precisely records the time the operation occurred; an operation type description such as "Submit Application," "Automatic Verification Passed," "Our Review," or "Recipient Confirmation"; the operation entity identifies the executing agency or automated module; a state hash verifies the integrity of the current achievement file; a digital signature enables identity authentication and non-repudiation; and a preceding transaction hash points to the previous operation, forming a chain-like association. Smart contracts encode the mutual recognition process rules and are automatically triggered at key nodes. For example, when an achievement is submitted, a "Create" record is generated and the state hash is initialized and uploaded to the chain; upon verification or approval, the contract verifies the operation permissions and packages the result with the new state hash onto the chain; upon completion of mutual recognition, a "Mutual Recognition Certificate" record is automatically generated, and the certificate hash is permanently fixed on the chain. Any attempt to tamper with the record will result in the state hash verification failing and being rejected by network nodes, thus ensuring the immutability of the entire process. The traceability and verification mechanism allows authorized participants to independently verify the complete mutual recognition trajectory of any achievement by querying on-chain records. The system provides a visual query interface, allowing users to retrieve traceability information by achievement ID, time range, and other criteria. Tests show that the system achieves 100% tamper-proof operation records, and the traceability query response time is less than 3 seconds, significantly improving the transparency and credibility of the mutual recognition process.

[0109] In a preferred embodiment, the blockchain end-to-end traceability system embeds a cross-chain mutual trust traceability module, integrated into the aforementioned modules. This module is configured with a heterogeneous chain interoperability architecture based on the Cosmos IBC protocol, a decentralized identity system conforming to the W3C DID standard, and a verifiable credential mechanism. This module enables interoperability between our FISCO BCOS chain and the Hyperledger Fabric chain via a cross-chain relay chain, supporting asset locking, state verification, and message passing between heterogeneous chains. It assigns decentralized identifiers to participating institutions and personnel, achieving fine-grained access control through verifiable credentials. Smart contracts automatically execute mutual recognition processes, and key functions set multi-signature thresholds, ensuring tamper-proof recording of information throughout the entire process, cross-border trusted verification, and future expansion to multiple country nodes.

[0110] (v) Collaborative Iteration and Operation and Maintenance Management Module

[0111] 1. International collaborative iteration mechanism

[0112] International technical collaboration meetings are held quarterly, such as Figure 4 As shown, the data mapping standard library was updated, model parameters were integrated, and the mutual recognition indicator system was improved to adapt to the needs of newly added observation equipment and scientific research; a rapid response channel for problems was established to jointly investigate and resolve disputes in system operation and mutual recognition of results.

[0113] 2. Model and System Optimization

[0114] The fusion model is continuously trained based on scientific expedition feedback data, optimizing the anomaly data identification algorithm and environmental interference compensation mechanism. The modular design reserves expansion interfaces to support rapid integration of new data types, output formats, and mutual recognition scenarios. The optimization process is continuously carried out by collecting and utilizing feedback data generated during the scientific expedition application phase. The system collects two types of key data in real time: first, the final judgment results of manual review by quality inspection experts on the "anomaly data" and "suspicious data" automatically marked by the system, confirming anomalies, misjudgments, or special phenomena; second, the accuracy fluctuations and error distribution characteristics of the fusion model under different environmental conditions. After cleaning and labeling, this feedback data constitutes a training set for algorithm optimization. For example, samples manually confirmed as misjudged are marked as "negative samples," and samples of special marine phenomena are marked as "protected samples." The optimization of the anomaly data identification algorithm mainly includes dynamic threshold adjustment and incremental model learning. Dynamic threshold adjustment is based on statistical analysis of misjudged cases, automatically correcting the detection threshold. For example, when the actual reasonable fluctuation range of a physical parameter in a specific sea area or season exceeds the original preset threshold, the system will automatically relax the threshold based on statistical distribution to reduce false alarms. For data patterns that frequently lead to misjudgments, such as short-term sensor drift, protective labels are added to the feature space to reduce the sensitivity of the algorithm. Incremental model learning involves retraining detection models such as Isolation Forest every quarter based on newly added feedback data, optimizing their decision boundaries, and assigning higher weights to "negative samples" during training to improve the model's ability to distinguish easily confused patterns. The optimization of the environmental interference compensation mechanism focuses on parameter correction and pattern library updates. By analyzing the correlation between environmental parameters and fusion errors in the feedback data, the parameters of the compensation algorithm are corrected. For example, if the low-temperature drift of the CTD sensor in the -2°C to -5°C range is found to be higher than the model's original prediction, linear regression is used to refit the coefficients in the drift compensation formula. Simultaneously, newly identified environmental interference patterns are added to the interference pattern library to expand the scope of the compensation algorithm. The optimized algorithm and model must undergo closed-loop verification before deployment. First, backtesting was conducted on historical data, and then small-scale testing was carried out on newly added scientific expedition data to ensure that its generalization performance was improved without introducing any negative impact.

[0115] 3. Operation and Maintenance Monitoring and Support

[0116] A system operation status monitoring platform is established to monitor the operation of data transmission, fusion processing, and mutual recognition processes in real time. Automatic alarms are triggered and backup plans are activated in case of faults. Regular data backups and security checks are performed to ensure system data security and stable operation. The transmission layer monitoring focuses on the status of international bilateral data synchronization channels, collecting data transmission success rates, transmission delays, and data packet integrity verification values ​​in real time, enabling independent monitoring of communication link quality. Processing layer metrics dynamically track the backlog of fusion task queues, CPU and memory resource utilization, and the execution efficiency of core algorithm modules, such as processing 10,000 records in a KNN data alignment task within ≤30 seconds. Business layer monitoring statistically analyzes the timeliness, node pass rate, and frequency of process timeouts at each node of the mutual recognition process. The platform automatically collects server resource status, application logs, and database performance metrics by deploying event tracking agents, aggregating the collected data to a time-series database (such as InfluxDB) for centralized storage every 10 seconds. Business flow data obtains node status in real time through the process engine interface and is associated with resource monitoring data for storage, forming an end-to-end traceable data link. Regarding the alarm mechanism, the platform adopts a baseline adaptive threshold strategy based on dynamically adjusted historical data. For example, during weekday morning peak hours, the transmission latency threshold is automatically relaxed to 8 seconds to adapt to network load fluctuations. Alarm rules are triggered in layers according to the severity of the anomaly: minor anomalies are pushed to the operations and maintenance (O&M) desk for processing; serious faults simultaneously send SMS notifications to the technical leaders of both parties and automatically activate backup transmission links. The platform provides a visual monitoring interface, dynamically displaying core indicator trend charts and system topology link status diagrams on a large dashboard, highlighting abnormal nodes. It supports drilling down from abnormal indicators to query specific error logs, stack information, and related data samples, assisting O&M personnel in quickly locating the root cause of problems. To improve system reliability, the platform has a pre-set fault self-healing strategy library. When an anomaly is detected in the converged service, the container instance restart process is automatically triggered; if the restart fails, it switches to a degraded operation mode. At the same time, the blockchain nodes are monitored in real time to verify the data synchronization consistency between the two nodes, and the consensus algorithm repair process is automatically triggered when a data fork is detected.

[0117] In a preferred embodiment, an autonomously evolving digital twin module is embedded, which is connected to each of the aforementioned modules. This module is configured with a real-time digital twin of the marine observation system, a multi-agent deep reinforcement learning decision engine, and a human-in-the-loop continuous learning mechanism. This module is used to construct real-time digital mirrors of the international joint scientific expedition fleet, buoy arrays, and underwater mooring networks. The twin maintains second-level synchronization with the physical entities, predicts the status of observation blind spots in real time, and optimizes voyage planning. Four types of agents are established: a data quality control agent, a standard negotiation agent, a model optimization agent, and a compliance review agent. These agents learn collaborative strategies through multi-agent deep reinforcement learning algorithms and autonomously handle routine operation and maintenance decisions. The system learns online from manual review results and expert opinions, incrementally updating the strategy network after each voyage, enabling continuous evolution of the system as scientific expedition experience accumulates.

[0118] The five modules work together. After being preprocessed by the standardized access module, the multi-source data from international ocean observations are fed into the intelligent fusion module for precise fusion and generate standardized results. The results are submitted to the mutual recognition and verification module for two-way verification. After passing the verification, the data is opened according to the hierarchical sharing mechanism, and the information throughout the process is recorded in the traceability system. The scientific research application feedback-driven collaborative iteration module updates the standards and models to achieve long-term system adaptation.

[0119] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An international marine observation data fusion and results mutual recognition system suitable for joint scientific expeditions, characterized in that: include: The multi-source data standardization access module is used to preprocess multi-source data from international ocean observations, enabling unified access and standardized processing of multi-dimensional data. The polar-adaptive intelligent fusion module, connected to the multi-source data standardization access module, receives pre-processed data, achieves multi-dimensional data fusion, and performs adaptation and optimization for complex polar or offshore conditions, generating standardized results. The polar-adaptive intelligent fusion module includes a fusion model architecture design unit and a polar condition adaptation and optimization unit. The fusion model architecture design unit adopts a hierarchical fusion and dynamic weighting architecture. The bottom layer is a data-level fusion based on the improved KNN algorithm, the middle layer is a feature-level fusion based on a one-dimensional convolutional neural network, and the top layer is a decision-level fusion based on an optimized random forest algorithm. The polar working condition adaptation and optimization unit is connected to the fusion model architecture design unit. It is used to embed the polar marine environment interference factor compensation algorithm and the equipment error dynamic correction mechanism, and to optimize the initial screening of abnormal data in the multi-source data automatic adaptation and processing unit, so as to reduce the fusion error caused by environmental interference and equipment error. And a fusion result quality assessment unit, connected to the fusion model architecture design unit, used to establish a fusion data quality assessment index system, automatically generate quality reports, and trigger a re-fusion process for unqualified data; An international achievement mutual recognition standard and verification module is connected to the polar-adapted intelligent fusion module to establish an international scientific research achievement mutual recognition system and to perform two-way verification of the standardized achievements. The hierarchical sharing and mutual recognition process management module is connected to the international achievement mutual recognition standard and verification module. It is used to manage the open sharing of verified achievements according to the hierarchical sharing mechanism and record the entire process information. The collaborative iteration and operation and maintenance management module is connected to the above modules respectively, and is used to update and optimize the data standards, fusion model parameters and mutual recognition indicator system based on the feedback data of scientific research applications; The blockchain full-process traceability system, integrated with the hierarchical sharing and mutual recognition process management module, is used to record the information of the entire process of the results in an immutable manner. An international joint operation and maintenance mechanism, with the participation of both international parties, ensures that the system dynamically adapts to updates in data standards and iterations in observation equipment.

2. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 1, characterized in that, The multi-source data standardization access module includes: The International Marine Observation Data Two-Way Mapping Standard Library is used to integrate the indicator definitions, format specifications, accuracy requirements, and unit conversion rules of core data types in hydrology, meteorology, ecology, and geology from both international parties. It includes multiple core data types and reserves interfaces for equipment and standard extensions. The multi-source data automated adaptation and processing unit is connected to the international marine observation data bidirectional mapping standard library and is used to perform standardized processing operations such as format conversion, unit unification, index calibration, and initial screening of abnormal data. A standardized data storage unit, connected to the multi-source data automated adaptation and processing unit, is used to generate standardized data files compatible with international dual standards and label metadata, and store them in an international distributed encrypted database, supporting bidirectional data synchronization and fast retrieval.

3. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 2, characterized in that, The initial screening of abnormal data in the multi-source data automated adaptation and processing unit adopts a two-level screening mechanism: The first level of screening involves setting reasonable physical range thresholds for observation indicators based on common sense in marine physics and historical scientific expedition experience, and eliminating invalid data that exceeds the threshold. The second level of screening uses the isolated forest algorithm, which combines data point values, spatiotemporal context features, and consistency of multi-parameter physical relationships to identify and classify outliers: invalid data caused by random errors are directly removed, while suspicious data that may indicate rare marine phenomena are marked, retained, and sent for manual review.

4. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 1, characterized in that, The improved KNN algorithm uses weighted Euclidean distance and combines time, space, and observation feature weight factors to calculate similarity. It also employs a dynamic adjustment mechanism to dynamically adjust the K value based on the local density index of the data batch. The optimized random forest algorithm introduces a weighted voting mechanism during the training phase, assigning higher weights to decision trees that perform well on historical polar data, and supports feature importance analysis.

5. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 1, characterized in that, The international achievement mutual recognition standards and verification module includes: The mutual recognition standard system construction unit is used to jointly develop a mutual recognition indicator system covering two types of outputs: data products and scientific research reports, and to clarify the accuracy threshold, expression standards, and credibility assessment dimensions. The two-way verification mechanism design unit is connected to the mutual recognition standard system construction unit. It adopts a combination of automatic verification and manual review. The automatic verification stage adopts a parallel dual-channel verification architecture, which synchronously calls the independent standardized evaluation algorithms of both parties to verify the scientific research results. When the results conflict, a three-level arbitration mechanism is triggered. It also integrates a dynamic threshold adjustment module for polar data. The manual review unit, connected to the two-way verification mechanism design unit, is used to jointly review results that fail automatic verification or are highly sensitive; and the mutual recognition result generation unit, connected to the two-way verification mechanism design unit and the manual review unit respectively, is used to generate international bilingual mutual recognition certificates for verified results, push modification suggestions for unqualified results, and support secondary submission for verification.

6. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 1, characterized in that, The hierarchical sharing and mutual recognition process management module includes: The hierarchical sharing mechanism unit establishes a three-tier sharing system of open, restricted, and core based on the sensitivity level of the results, and configures access control policies for unverified access, joint authorization by two responsible persons, and dual key verification respectively; A standardized mutual recognition process unit, connected to the hierarchical sharing mechanism unit, is used to build a fully online process and supports automatic reminders for process nodes. The blockchain end-to-end traceability system unit is connected to the standardized mutual recognition process unit. It adopts a permissioned consortium blockchain architecture, stores end-to-end traceability metadata on the chain, and achieves immutability and two-way traceability of operation records through smart contract encoding of mutual recognition process rules.

7. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 1, characterized in that, The collaborative iteration and operation and maintenance management module includes: The international joint iteration mechanism unit is used to update standards and models through quarterly technical collaboration meetings and to establish a rapid response channel for issues to jointly resolve disputes. The model and system optimization unit is connected to the international joint iterative mechanism unit. It is used to perform incremental learning and parameter optimization on the abnormal data identification algorithm and environmental interference compensation mechanism based on scientific research application feedback data. It also adopts a modular design and reserves expansion interfaces. The operation and maintenance monitoring and support unit is connected to the model and system optimization unit to build a system operation status monitoring platform, monitor the operation of data transmission, fusion processing, and mutual recognition processes in real time, realize automatic fault alarm, self-healing and backup plan triggering, and perform data backup and security testing regularly.

8. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions as described in claim 7, characterized in that, The monitoring platform of the operation and maintenance monitoring and protection unit adopts a baseline adaptive threshold strategy to set alarm rules, triggers alarms in layers according to the severity of the anomaly, and has a preset fault self-healing strategy library to perform fault detection and automatic response.

9. The international marine observation data fusion and results mutual recognition system applicable to joint scientific expeditions according to any one of claims 1-8, characterized in that, Information on the submission, verification, review, modification, and mutual recognition of results is recorded immutably. The international joint operation and maintenance mechanism is jointly participated in by both parties to ensure that the system dynamically adapts to updates in data standards and iterations in observation equipment.

Images