A marine data monitoring method based on a wireless communication system

Abstract

The application relates to the technical field of marine data monitoring, and discloses a marine data monitoring method based on a wireless communication system, which comprises the following steps: S1, data acquisition, obtaining marine monitoring data of a target; S2, data preprocessing, performing format conversion, deduplication, time alignment and data interpolation processing on the data obtained in the data acquisition; S3, data fusion, based on a time interleaving algorithm and a space clustering algorithm, matching and fusing the processed data; S4, target tracking, under the condition of target data loss, adopting a Kalman filtering method to predict a target trajectory; and S5, monitoring output, generating a fusion data set and outputting dynamic information of the target. Through the combination of deep learning target identification, an LSTM network state prediction and efficient wireless communication technology, the application realizes more accurate marine target monitoring, stable real-time data transmission and efficient target state tracking.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine data monitoring, in particular to a marine data monitoring method based on a wireless communication system. BACKGROUND

[0002] Currently, marine data monitoring mainly relies on radar, AIS (Automatic Identification System) and satellite remote sensing technology. These technologies have good performance in target detection, basic trajectory tracking and communication coverage. Radar monitoring can continuously track targets in bad weather, AIS system can provide the identity and position information of ships, and satellite remote sensing can cover a wide range of sea areas, providing support for long-distance target monitoring. In addition, some monitoring systems also introduce filtering algorithms for target state estimation, combined with certain rule algorithms, to improve the efficiency of data processing. Overall, existing technologies can already meet some marine monitoring needs and achieve basic target identification and state tracking in specific environments.

[0003] Although existing technologies have made some progress, there are still some deficiencies in the face of more complex marine environments. First, the recognition accuracy of traditional target recognition methods decreases significantly in complex environments, especially when the sea conditions change dramatically or the target behavior is abnormal, the misjudgment rate is high, which affects situation awareness. Second, conventional target trajectory prediction methods are unstable when dealing with non-uniform speed and highly maneuverable targets, with large trajectory deviation, leading to tracking errors. Third, the transmission stability of wireless communication systems in marine environments is insufficient, and data is easily affected by signal attenuation or interference, leading to delays and losses of monitoring information. Finally, the monitoring system still has a high degree of dependence on human labor, and the level of intelligence is limited, with short boards in unexpected event handling and autonomous decision-making. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a marine data monitoring method based on a wireless communication system, which solves the problems of low target recognition accuracy, inaccurate target state prediction and unstable data transmission in the prior art.

[0005] To achieve the above purpose, the present application is implemented by the following technical scheme: a marine data monitoring method based on a wireless communication system, comprising the following steps:

[0006] S1, data acquisition, obtaining marine monitoring data of the target;

[0007] S2, data preprocessing, performing format conversion, deduplication, time alignment and data interpolation processing on the data obtained by the data acquisition;

[0008] S3, data fusion, based on time interleaving algorithm and space clustering algorithm, matching and fusing the processed data;

[0009] S4, target tracking, in the case of missing target data, using Kalman filtering method to predict target trajectory;

[0010] S5, monitoring output, generating fusion data set, outputting the dynamic information of the target.

[0011] Preferably, the data collection includes:

[0012] Receiving ship automatic identification system data through the AIS data receiving module, and parsing the static information and dynamic information of the ship;

[0013] Obtaining radar detection data of the marine target through the radar detection data receiving module, and extracting the position information and motion parameters of the target;

[0014] Obtaining mobile terminal device data through the mobile communication terminal acquisition positioning module, and parsing the position information and unique identification of the terminal;

[0015] Using the data storage module to cache the collected data to ensure data integrity.

[0016] Preferably, the data preprocessing includes:

[0017] Using time window method to remove duplicate processing of AIS data, radar data and mobile communication terminal data;

[0018] Using the geodetic coordinate conversion model to unify the geographic coordinate standardization processing of the position information of different data sources;

[0019] Using linear interpolation method to complete the missing data and ensure the continuity of the data time series;

[0020] Using filtering algorithm to denoise the original data to reduce the influence of data outliers.

[0021] Preferably, the data fusion includes:

[0022] Using time interleaving algorithm to align the time stamps of different data sources, and eliminating data with time error exceeding threshold value;

[0023] Using spatial clustering algorithm to match the target position information of different data sources, and setting matching threshold according to spatial distance;

[0024] Using target association method to match AIS data, radar data and mobile communication terminal data, and establishing multi-source fusion target data set;

[0025] Using data consistency detection method to optimize the fusion data to improve the reliability of the data.

[0026] Preferably, the target tracking comprises:

[0027] The Kalman filtering method is adopted to estimate the target state, and the target position at the next time is predicted based on the state transition equation;

[0028] In the case of AIS data missing, radar data and mobile communication terminal data are used to predict the target trajectory, and the state estimation parameters are dynamically adjusted;

[0029] In the case of radar data missing, AIS data and mobile communication terminal data are used to predict the target trajectory, and the trajectory deviation is compensated;

[0030] In the case of mobile communication terminal data missing, AIS data and radar data are used to predict the target trajectory, and the target speed change is corrected;

[0031] The data confidence evaluation method is adopted to judge the accuracy of the tracking target data, and the tracking parameters are adjusted.

[0032] Preferably, the monitoring output comprises:

[0033] According to the time synchronization information of the target, the fusion data set is generated, and the historical data of the target is stored;

[0034] The output of the fused target dynamic information includes the position, speed, heading and time stamp of the target;

[0035] According to the source information of the target data, the data source category is identified, and the data source weight is adjusted;

[0036] The data storage and management module is adopted to store the fusion data in a classified manner, and provide a data calling interface;

[0037] The data visualization module is adopted to convert the fusion data into graphical display, and supports real-time monitoring and query.

[0038] Preferably, the time interleaving algorithm comprises:

[0039] The time difference of different data sources is calculated, and the time matching window is set to ensure the time synchronization of the data;

[0040] The interpolation calculation method is adopted to correct the time stamp of different data sources, and the time alignment processing is carried out;

[0041] The data with time error exceeding the set threshold is eliminated, and the effective data is retained for subsequent fusion;

[0042] The time synchronization mechanism is adopted to adjust the time alignment of different data sources, and ensure the time consistency of the data.

[0043] Preferably, the spatial clustering algorithm comprises:

[0044] The spatial distance between AIS data, radar data and mobile communication terminal data is calculated, and a distance matching threshold is set;

[0045] A target clustering method is used to spatially cluster targets from different data sources, and data belonging to the same target is merged;

[0046] A data association analysis method is used to compare target trajectories and ensure the stability of target matching;

[0047] A trajectory matching method is used to associate and analyze the historical trajectory and real-time trajectory of the target, and optimize the matching degree of the target position.

[0048] Preferably, the Kalman filtering method comprises:

[0049] A state transition equation is used to predict the next time state of the target, and the target motion parameters are calculated;

[0050] The target state is corrected according to the observation equation, and the Kalman gain matrix is updated to adjust the target prediction result;

[0051] A target state estimation algorithm is used to optimize the target trajectory and improve the accuracy of target positioning;

[0052] An error correction method is used to correct the trajectory prediction error of the target and optimize the motion trajectory of the target.

[0053] Preferably, the source information of the target data comprises:

[0054] When the target only depends on AIS data, the data source is identified as AIS, and the trajectory information of the AIS target is stored;

[0055] When the target only depends on radar data, the data source is identified as radar, and the detection data of the radar target is stored;

[0056] When the target only depends on mobile communication terminal data, the data source is identified as mobile communication terminal, and the position information of the mobile terminal is stored;

[0057] When the target fuses multiple data sources, the data source is identified as fusion data, and the weight of the fusion data is calculated according to the confidence of the data source;

[0058] A data classification management method is used to classify and store target data from different data sources, and provide a multi-level data management interface.

[0059] The present application provides a marine data monitoring method based on a wireless communication system. The following advantages are provided:

[0060] 1、The application improves the target classification accuracy by combining deep learning and multi-dimensional target feature analysis, solving the problem that traditional classification technology cannot accurately respond to complex environmental changes. Compared with the simple rule classification in the prior art, the application can more flexibly adapt to different scenes.

[0061] 2、The LSTM network is used to accurately predict the target motion, which not only improves the estimation accuracy of the target state, but also can correct the tracking results in real time. In this way, the system responds faster to dynamically changing targets, and is much more stable than traditional methods.

[0062] 3、With the help of wireless communication technology, the application ensures efficient and real-time data transmission, reduces delay and packet loss. Compared with the shortcomings of the existing monitoring system, the optimization of the application makes real-time feedback more stable, and improves the reliability of offshore monitoring.

[0063] 4、The target recognition and decision support mechanism of the application realizes intelligent offshore monitoring. This not only reduces the burden of manual intervention, but also improves the overall monitoring efficiency. Compared with traditional labor-intensive operations, this level of automation makes the system respond more quickly. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The method flowchart of the application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the application will be described below in conjunction with the drawings in the specification of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0066] Please refer to the accompanying Figure 1 The embodiment of the application provides an offshore data monitoring method based on a wireless communication system, which includes the following steps:

[0067] S1, data acquisition, obtaining offshore monitoring data of the target;

[0068] Generally, the monitoring of offshore targets depends on multiple data sources, and the measurement method, time synchronization mechanism and data format of each data source are different. Therefore, the embodiment provides a multi-source data acquisition method to ensure that the monitoring information of offshore targets is comprehensive, stable, and can effectively support the operation of the subsequent data processing module.

[0069] In this embodiment, data collection mainly includes AIS data reception, radar detection data reception, mobile communication terminal data collection and data storage buffer mechanism, which ensures that information from different data sources can be stored according to unified standards, and provides a basis for subsequent fusion processing.

[0070] AIS (Automatic Identification System) is an automatic broadcast system based on VHF (Very High Frequency) communication. Ships periodically send their status information, including static information and dynamic information.

[0071] AIS static data is mainly used to identify the identity of the ship, including:

[0072] Ship unique identification code ( ): Corresponds to MMSI (Maritime Mobile Service Identity).

[0073] Ship name ( ): The registered name of the ship.

[0074] Ship type ( ): Such as cargo ship, tanker, fishing boat, etc.

[0075] Size information ( ): Length and width of the ship.

[0076] Draft ( ): The maximum draft value of the ship.

[0077] AIS static data is usually manually input by the ship and stored in the shipborne AIS terminal device, and generally does not change frequently.

[0078] AIS dynamic data is automatically updated by shipborne GPS or navigation control system, including:

[0079] Current position ( ): Latitude and longitude coordinates in WGS-84 coordinate system.

[0080] Speed ( ): Unit is knots (knots), i.e. 1 nautical mile / hour.

[0081] Heading ( ): 0-360 degrees, indicating the angle of motion relative to the north direction.

[0082] Send timestamp ( ): Records the UTC time of sending the current data.

[0083] AIS data is usually broadcasted at a frequency of every 2-10 seconds, the faster the ship speed, the higher the update frequency.

[0084] Radar (Radio Detection and Ranging) uses the principle of electromagnetic wave ranging, through receiving the echo signal reflected by the target, to calculate its distance, direction and motion state.

[0085] The basic measurement parameters of radar include:

[0086] Target distance ( ): the distance from the radar to the target, the calculation formula is:

[0087] ;

[0088] Where: is the propagation speed of electromagnetic wave in air; is the time interval between radar signal transmission and reception (unit: seconds).

[0089] When calculating the target latitude and longitude ( , ), coordinate conversion needs to be performed in combination with the radar site location ( , ):

[0090] ;

[0091] ;

[0092] Where: , is the latitude and longitude coordinates of the radar station; is the target distance (unit: meters); is the target azimuth angle (unit: radian); is the target latitude detected by the radar, in degrees (°); is the target longitude detected by the radar.

[0093] Some high-end maritime radars have the function of estimating the size of the target. The radar can calculate the size of the target according to the echo intensity of the target:

[0094] ;

[0095] Where: is the equivalent radar cross section (RCS) of the target; is the received echo power; is the radar transmission power; is the system calibration coefficient.

[0096] In general, RCS values ​​can be used for target classification, such as distinguishing between large ships, small vessels, or marine buoys.

[0097] Mobile communication terminal data is used for positioning based on wireless signaling. Generally, positioning methods include multi-base station triangulation and wireless fingerprint matching.

[0098] The location of a target device can be calculated by measuring the signal transmission time from the target device to different base stations.

[0099] ;

[0100] in: Target to base station Distance (unit: meters); The base station signal transmission time (unit: seconds); This refers to the speed of signal propagation.

[0101] The latitude and longitude of the target can be calculated using the least squares method, as shown in the following formula:

[0102] ;

[0103] in: The latitude of the mobile terminal target is expressed in degrees (°). Longitude of the target mobile terminal, in degrees (°); The coordinates of the base station; The distance from the target to each base station.

[0104] In areas with sparse base stations, a signal fingerprint database matching method can be used to estimate the target location using historical signal strength data.

[0105] To ensure time synchronization between different data sources, this embodiment employs a short-term caching mechanism, within a set time window. Internal storage is used to align data over time during subsequent data fusion.

[0106] ;

[0107] in: The latest AIS data timestamp; This is the latest AIS data timestamp, in seconds (s). This is the latest radar data timestamp, in seconds (s). This is the latest timestamp of mobile communication terminal data, in seconds (s).

[0108] Under normal circumstances, The time range can be set to 10-30 seconds to ensure that the system can integrate the most recent data.

[0109] Through different measurement principles and calculation methods, the accuracy and timeliness of the monitoring data are ensured. At the same time, through the data caching mechanism, the time alignment of multi-source data is ensured, providing a solid data foundation for subsequent data fusion, tracking analysis and other steps.

[0110] S2, data preprocessing, format conversion, de-duplication, time alignment and data interpolation processing are performed on the data obtained by data acquisition;

[0111] Using unprocessed data for fusion may lead to target matching errors, trajectory breaks or abnormal predictions. Therefore, before entering data fusion, a series of data preprocessing operations are needed to unify the data format, remove redundant data, adjust the timestamp synchronization, fill in the missing information and optimize the data quality.

[0112] In this embodiment, data preprocessing mainly involves format conversion, data de-duplication, time alignment, data interpolation completion and abnormal data processing, to ensure that each data source is in the same time reference system and spatial coordinate system for fusion calculation, thereby improving the monitoring accuracy.

[0113] The coordinate system and time standard used by different data sources may be different. In order to ensure the uniformity of subsequent calculations, format conversion is needed.

[0114] Generally, AIS data uses WGS-84 coordinate system, i.e. latitude and longitude coordinates, while radar data usually uses polar coordinate system, and mobile communication data may use projection coordinate system (such as UTM coordinate system). In order to ensure the consistency of calculation, all data in this embodiment are converted to Earth-Centered, Earth-Fixed (ECEF) coordinate system.

[0115] In one possible implementation, the formula for converting WGS-84 latitude and longitude coordinates to ECEF coordinates is as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] Wherein: , , is the three-dimensional coordinate in ECEF coordinate system, with unit of meter (m); , is the latitude and longitude of the target, with unit of degree (°); is the altitude of the target, with unit of meter (m); is the long semi-axis of the earth, which is 6378137 meters; is the first eccentricity of the Earth; is the radius of curvature of the prime vertical circle.

[0120] For radar data, it is necessary to convert to local Cartesian coordinates first, and then to ECEF coordinate system due to the polar coordinate representation. In this embodiment, the formula for radar polar coordinate conversion is as follows:

[0121] ;

[0122] ;

[0123] wherein: , : the two-dimensional coordinates of the target detected by the radar in the radar station coordinate system, in meters (m); : the distance from the target to the radar station, in meters (m); : the azimuth angle of the target measured by the radar, in radians (rad).

[0124] Since AIS, radar and mobile communication terminal may collect information of the same target multiple times in a short time, data deduplication is required. In this embodiment, time window method and spatial threshold method are used to filter duplicate data.

[0125] As an option, within a set time window , check whether there is information of the same target in multiple data sources:

[0126] ;

[0127] wherein: , : the time stamp of the target data, in seconds (s); : the time window threshold, in seconds (s), usually 1-10s.

[0128] If the time interval meets the above conditions, further calculate the spatial distance between the targets:

[0129] ;

[0130] wherein: : the spatial distance between the target data points and , in meters (m); , , and , , : the ECEF coordinates of the target, in meters (m).

[0131] The data timestamps of different data sources may not be consistent and need to be synchronized. The present embodiment uses linear interpolation method for time alignment to estimate the target position when data is missing.

[0132] ;

[0133] ;

[0134] Wherein: , is the interpolated target longitude and latitude, in degrees (°); , is the target longitude and latitude of the previous time point, in degrees (°); , is the target longitude and latitude of the next time point, in degrees (°); is the current time point to be calculated, in seconds (s); , is the timestamp of the adjacent data point, in seconds (s).

[0135] In order to remove measurement errors and sudden data, the present embodiment uses sliding window mean filtering method to smooth the data:

[0136] ;

[0137] ;

[0138] Wherein: , is the filtered target longitude and latitude, in degrees (°); , is the target longitude and latitude of the previous time point, in degrees (°); is the target longitude and latitude of the next time point, in degrees (°); is the size of the sliding window, usually 3-5.

[0139] For numerical anomaly data such as speed and heading, the present embodiment uses median filtering to remove outliers, to ensure the continuity and stability of the data.

[0140] Through the format conversion, data deduplication, time alignment, interpolation completion and abnormal data processing, the stability and consistency of the data input are ensured, and high quality data support is provided for subsequent data fusion.

[0141] S3, data fusion, based on time interleaving algorithm and spatial clustering algorithm, matching and fusion of processed data;

[0142] The core objective of data fusion is to integrate observation information from multiple data sources to make target state estimation more accurate and improve the stability of trajectory prediction. Since different data sources have different observation characteristics, error models, and data update frequencies, a single data source may contain missing or inaccurate information. Therefore, data fusion methods are needed to reasonably weight the observation results from different data sources and combine them with state estimation algorithms to obtain a more accurate and stable target motion state. In this embodiment, weighted fusion, extended Kalman filtering (EKF), and joint probabilistic data association (JPDA) are employed to improve the reliability and stability of data fusion.

[0143] In this embodiment, weighted fusion is first performed to calculate the contribution of different data sources to the target state estimation. Generally, for a given target... The observations from the data sources are respectively , , ..., The formula for weighted fusion is:

[0144] ;

[0145] in: The fused target state estimate is in meters (m). For the first The measurement values ​​of the target from each data source are in meters (m). For the first The weight coefficients of each data source satisfy the normalization constraint:

[0146] ;

[0147] In this embodiment, the weighting coefficients are calculated based on the measurement error variance of the data source. The calculation method is as follows:

[0148] ;

[0149] in: For the first The measurement error variance of each data source, in m. 2 This indicates the stability of the observations from the data source.

[0150] In one possible implementation, if the error characteristics of some data sources change dynamically over time, recursive least squares (RLS) or adaptive Kalman filtering methods are used to dynamically adjust the weights to ensure the accuracy of data fusion.

[0151] In this embodiment, to improve the accuracy of target state estimation, an extended Kalman filter (EKF) is used for state prediction and updating. The target's state vector is defined as:

[0152] ;

[0153] where: is the two-dimensional position coordinate of the target at time in meters (m); is the velocity of the target in and directions in meters per second (m / s).

[0154] State prediction equation:

[0155] ;

[0156] where: is the state vector of the target at time ; is the state transition matrix, describing the target motion model:

[0157] ;

[0158] where: is the time step in seconds (s); is the process noise, following a Gaussian distribution with mean 0 and covariance matrix , representing the influence of system noise.

[0159] Measurement equation:

[0160] ;

[0161] where: is the measurement vector, containing the target position observed by the sensor; is the measurement noise, following a Gaussian distribution with mean 0 and covariance matrix , representing the influence of measurement error; is the measurement matrix, defined as: ;

[0162] Kalman gain calculation:

[0163] ;

[0164] where: is the Kalman gain matrix; is the predicted covariance matrix; is the measurement noise covariance matrix; is the weighting matrix, used to calculate the gain, considering the measurement error and the prior estimation error; is the observation matrix, representing the relationship between the system state and the measurement space.

[0165] State update formula:

[0166] ;

[0167] ;

[0168] wherein: is an identity matrix, ensuring the correctness of matrix operations; is the actual measurement value, representing the observed quantity at time ; is the updated error covariance matrix, representing the uncertainty of the system state estimation error at time ; is the prior error covariance matrix, representing the covariance of the system state estimation error before time .

[0169] In a multi-target environment, there may be multiple observations corresponding to different targets, so this embodiment adopts the Joint Probabilistic Data Association (JPDA) method to calculate the association probability between each observation and the target.

[0170] In this embodiment, Mahalanobis Distance is used for target association:

[0171] ;

[0172] wherein: is the Mahalanobis Distance, measuring the matching degree of the observation and the predicted target; is the th observation data, with units of meters (m); is the target state estimation value, with units of meters (m); is the residual covariance matrix, defined as: ;

[0173] If the Mahalanobis Distance satisfies:

[0174] ;

[0175] wherein: is the set association threshold, then it is considered that the observation matches the target.

[0176] This embodiment provides a complete data fusion method, covering weighted fusion, Extended Kalman Filter (EKF), and Joint Probabilistic Data Association (JPDA), which improves the target matching accuracy by reasonably allocating data source weights, estimating target states, and combining association methods, making the data fusion result more reliable, and laying a solid foundation for subsequent target trajectory prediction and situation analysis.

[0177] S4. Target tracking: In the case of missing target data, the Kalman filter method is used to predict the target trajectory.

[0178] S4 combines motion models, filtering algorithms, smoothing methods, and deep learning techniques in its trajectory prediction process. These methods allow it to consider the target's motion characteristics and environmental factors, thus providing a better estimate of the target's future position and state.

[0179] In the first step of trajectory prediction, the target's motion model is used to describe the change of the target's state over time, using the "state vector definition" disclosed in S3.

[0180] In one possible implementation, the state transition of the target can be represented by a linear motion model, using the "state prediction equation" disclosed in S3.

[0181] In certain situations, if the target's motion possesses higher dynamic characteristics, the model can be extended to an acceleration model, for example:

[0182] ;

[0183] in: For the goal at all times The x-axis position, in meters (m); For the goal at all times The y-axis position, in meters (m); For the goal at all times The x-axis velocity, in meters per second (m / s); For the goal at all times The y-axis velocity, in meters per second (m / s); The acceleration component of the target is expressed in meters per second squared.

[0184] For acceleration models, the state transition matrix F needs to be adjusted accordingly to include acceleration information.

[0185] To improve prediction accuracy, filtering and smoothing techniques are employed to reduce errors. Generally, short-term observations may be affected by noise, causing trajectory fluctuations. Therefore, it is necessary to smooth the target's historical trajectory.

[0186] In this embodiment, a moving average filter is used for trajectory smoothing, and its formula is as follows:

[0187] ;

[0188] in: The target state vector is smoothed, with units of meters (m) and meters per second (m / s). To adjust the sliding window size; This represents the target state at N past time points.

[0189] Moving average filtering can effectively reduce the impact of sudden data changes on target state estimation by calculating the mean of historical data, thereby improving the stability of prediction.

[0190] To further improve the reliability of the prediction, this embodiment employs Bayesian estimation for joint state estimation. The core idea of ​​Bayesian estimation is to progressively optimize the target state estimation by updating the posterior probability distribution. In the state prediction process, the formula for Bayesian estimation is:

[0191] ;

[0192] in: Indicates given past observation data The goal is to be achieved at all times. The posterior probability of the state; The state transition probability of the target is usually given by the motion model; For the goal at all times The posterior probability distribution of the state.

[0193] This method improves the accuracy of target state estimation by recursively updating the posterior probability and integrating historical information with current observation data.

[0194] For complex nonlinear target state estimation, this embodiment uses particle filtering (PF) to handle nonlinear and non-Gaussian processes. Particle filtering represents the target state as a set of particles and assigns a weight to each particle to represent its contribution to the target state.

[0195] The state update process of particle filtering can be represented as:

[0196] ;

[0197] ;

[0198] in: For the first Each particle at time The state; For the first The weight of each particle; The observation likelihood function represents the degree of matching between the particle state and the observed value.

[0199] Particle filtering improves prediction accuracy by eliminating low-weight particles through a resampling process and concentrating particle distribution in high-weight regions.

[0200] In some embodiments, the present embodiment also employs a long short-term memory network (LSTM) to further optimize trajectory prediction. LSTM can capture long-term dependencies in time series through its special gating structure, effectively improving the accuracy of prediction.

[0201] The state update formula of LSTM is as follows:

[0202] ;

[0203] Wherein: are the activation values of the forget gate, input gate and output gate, respectively, ranging from 0 to 1, used to control the selection and update of information flow; is the weight matrix of LSTM; is the bias term of LSTM; is the memory state of the LSTM unit; is the hidden state output of LSTM; is the candidate memory unit, representing the influence of the current input information on the memory state; is the output gate (Output Gate), which determines how much of the current memory state will affect the output; is the hyperbolic tangent activation function, used to generate outputs ranging from -1 to 1, commonly used to handle information smoothing; is the hidden state of the previous time step, as part of the current time step input, providing context information from the previous time step to the current time step; is the input at the current time step, usually the observation data or input features at the current time step.

[0204] Through this method, LSTM can learn the complex patterns in the target trajectory, thus providing more accurate state prediction.

[0205] Step S4 improves the accuracy of target trajectory prediction by combining motion models, filtering methods, particle filtering, Bayesian estimation and deep learning techniques. The parameter definitions involved in each formula are disclosed in detail to ensure that those skilled in the art can accurately understand the meaning and function of each formula, and thus provide reliable technical support for more complex trajectory prediction and dynamic tracking.

[0206] S5, monitoring the output, generating a fusion data set, and outputting the dynamic information of the target;

[0207] S5 By analyzing the target state information and trajectory prediction results obtained from the previous steps, the type of target can be identified, providing necessary decision support for the system. Target recognition and classification is not only a simple distinction of targets, but also an accurate judgment made through in-depth analysis of the target's motion pattern, appearance characteristics, and other related information. Therefore, the implementation of this step requires detailed analysis based on the multi-dimensional characteristics of the target (including dynamic and static characteristics), and the use of appropriate classification algorithms to complete the target recognition and classification.

[0208] In the process of target recognition and classification, representative features need to be extracted from the target state information. The target state information usually includes the target's position, velocity, acceleration, and other dynamic characteristics. However, relying solely on these dynamic characteristics may not be sufficient to fully identify the target. Therefore, in this embodiment, the target's motion pattern, appearance characteristics, and other information are further introduced to classify the target based on more comprehensive features.

[0209] This embodiment uses classification algorithms such as multi-layer perception (MLP) and support vector machine (SVM) to implement target classification. By utilizing the predicted target state information from the previous steps, these classification models are trained based on historical data to accurately classify target types.

[0210] In this method, the target state vector is input as a feature, and the classification model outputs the class label of each target after training. The input of the target state vector and the output of the classification label are optimized through a loss function, which is generally a cross-entropy loss function with the form:

[0211] ;

[0212] where: is the state vector of the target at time , with units of meters (m) and meters per second (m / s); is the classification label of the target, indicating the target type; is the total number of target types, indicating the number of target classes; is the predicted probability that the target belongs to class , which is usually the probability value output by the classification model.

[0213] This loss function optimizes the difference between the target output and the actual label, updating the parameters (such as weights and biases) of the classification model during training. This process continuously improves the classification accuracy and enables the classification model to make correct classification judgments when facing unknown targets.

[0214] Once the targets are classified, the classification results will provide key information for subsequent decision-making and action planning. In some embodiments, the invention also incorporates the motion trajectory features of the targets to further improve the accuracy of target recognition. For example, if a target exhibits a similar motion pattern to a known target, the system will confirm it by combining these trajectory information, thereby improving the accuracy of classification.

[0215] After the targets are successfully classified, the system will take different decision measures according to different target categories. For example, the system may choose different trajectory tracking strategies according to the target type, or execute corresponding countermeasures for hostile targets, while for friendly targets, appropriate trajectory maintenance strategies are selected.

[0216] Classification results can also further affect the estimation of target state. If the system discovers that the type of the target has changed through target recognition, the classification result will immediately drive the re-estimation of the target state. For example, the dynamic characteristics of the target such as speed, acceleration, etc. are corrected to ensure the accuracy and consistency of the prediction.

[0217] To improve the performance of target recognition, convolutional neural networks (CNN) based on deep learning are also used in some embodiments to process the image data or radar echo data of the target, further improving the classification of the target. In this implementation, CNN can extract important feature information from more complex target patterns by automatically learning the spatial and temporal features in the input data, thereby improving the recognition accuracy.

[0218] S5 extracts multi-dimensional features by combining the results of the aforementioned trajectory prediction and data fusion, and the classification model can effectively distinguish different types of targets. The formulas and parameter definitions involved in each step have been disclosed in detail to ensure that those skilled in the art can fully understand their functions and significance. These details not only provide reliable technical support for the target recognition and classification process, but also lay a solid foundation for the subsequent decision-making process.

[0219] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring data at sea based on a wireless communication system, characterized by, The method comprises the following steps: S1, data acquisition, obtaining offshore monitoring data of the target; S2, data preprocessing, performing format conversion, deduplication, time alignment and data interpolation processing on the data obtained in the data acquisition; S3, data fusion, based on time interleaving algorithm and spatial clustering algorithm, matching and fusing the processed data; S4, target tracking, under the condition of target data loss, using Kalman filtering method to predict target trajectory; S5, monitoring output, generating a fusion data set and outputting the dynamic information of the target; The data acquisition comprises: receiving ship automatic identification system data through an AIS data receiving module, and analyzing the static information and dynamic information of the ship; acquiring radar detection data of the offshore target through a radar detection data receiving module, and extracting the position information and motion parameters of the target; acquiring data of the mobile terminal device through a mobile communication terminal acquisition and positioning module, and analyzing the position information and unique identifier of the terminal; using a data storage module to cache the collected data to ensure data integrity; The data fusion comprises: using a time interleaving algorithm to align the timestamps of different data sources and eliminate data with time errors exceeding a threshold value; using a spatial clustering algorithm to match the target position information of different data sources and setting a matching threshold value based on spatial distance; using a target association method to match AIS data, radar data and mobile communication terminal data, and establishing a multi-source fusion target data set; using a data consistency detection method to optimize the fusion data to improve the reliability of the data; The target tracking comprises: using a Kalman filtering method to estimate the target state and predicting the target position at the next time based on the state transition equation; under the condition of AIS data loss, using radar data and mobile communication terminal data to predict the target trajectory and dynamically adjusting the state estimation parameters; under the condition of radar data loss, using AIS data and mobile communication terminal data to predict the target trajectory and compensating for the trajectory deviation; under the condition of mobile communication terminal data loss, using AIS data and radar data to predict the target trajectory and correcting the target speed change; using a data confidence evaluation method to judge the data accuracy of the tracked target and adjusting the tracking parameters; The spatial clustering algorithm comprises: calculating the spatial distance between AIS data, radar data and mobile communication terminal data and setting a distance matching threshold value; using a target clustering method to spatially cluster targets from different data sources and merging data belonging to the same target; using a data association analysis method to compare target trajectories and ensure the stability of target matching; using a trajectory matching method to perform correlation analysis on the historical trajectory and real-time trajectory of the target and optimizing the matching degree of the target position.

2. The offshore data monitoring method based on a wireless communication system according to claim 1, characterized in that, The data preprocessing comprises: using a time window method to perform deduplication processing on AIS data, radar data and mobile communication terminal data; using a geodetic coordinate conversion model to perform unified geographic coordinate standardization processing on the position information of different data sources; The missing data is completed by using linear interpolation method, and the continuity of the time series data is ensured. The original data is denoised by using filtering algorithm to reduce the influence of data outliers.

3. The method according to claim 1, wherein, The monitoring output includes: According to the time synchronization information of the target, a fusion data set is generated, and the historical data of the target is stored; The output of the fused target dynamic information includes the position, speed, heading and timestamp of the target; According to the source information of the target data, the data source category is identified, and the data source weight is adjusted; The fusion data is stored by using data storage and management module, and data calling interface is provided; The fusion data is converted into graphical display by using data visualization module, and real-time monitoring query is supported.

4. The method according to claim 1, wherein, The time interleaving algorithm includes: Calculate the time difference of different data sources, and set the time matching window to ensure the time synchronization of data; The time stamp of different data sources is corrected by using interpolation calculation method, and time alignment processing is carried out; The data with time error exceeding the set threshold is rejected, and the effective data is retained for subsequent fusion; The time synchronization mechanism is adopted to adjust the time alignment of different data sources, and the time consistency of data is ensured.

5. The method according to claim 1, wherein, The Kalman filtering method includes: The next time state of the target is predicted by using state transition equation, and the target motion parameters are calculated; The target state is corrected according to the observation equation, and the Kalman gain matrix is updated to adjust the target prediction result; The target state estimation algorithm is used to optimize the target trajectory and improve the accuracy of target positioning; The error correction method is used to correct the trajectory prediction error of the target and optimize the motion trajectory of the target.

6. The offshore data monitoring method based on a wireless communication system according to claim 1, wherein, The source information of the target data includes: When the target only depends on AIS data, identify the data source as AIS, and store the trajectory information of AIS target; When the target only depends on radar data, identify the data source as radar, and store the detection data of radar target; When the target only depends on mobile communication terminal data, identify the data source as mobile communication terminal, and store the position information of mobile terminal; When the target fuses multiple data sources, identify the data source as fusion data, and calculate the weight of fusion data according to the confidence of data source; The data classification management method is used to store the target data of different data sources, and multi-level data management interface is provided.

Images