Track fusion method based on open source data and AIS data

Abstract

The invention discloses a track fusion method based on open source data and AIS data, and relates to the technical field of Internet open source data processing and situation fusion. An acquired AIS original message is analyzed to generate a track of a ship target, open source text data acquired through the Internet also includes track information of the ship target, entity extraction is performed by adopting a deep neural network algorithm, the open source text data is processed and recognized, and the track information of the ship target is acquired. Information such as an IMO number, an MMSI number and a ship name of a ship in the data and information such as occurrence time and occurrence place of a related event are identified, ship information and event information are integrated to form a trace point of the ship, and the mark, time and space information of the ship correspond to records in an AIS original message, so that the information of the ship is obtained. And inserting the ship trace point obtained from the open source text data into the track of the ship in the AIS, and further smoothing the track through a Kalman filtering algorithm to form track information after ship target fusion, thereby realizing track fusion based on the open source data and the AIS data.

Description

Technical Field

[0001] This invention relates to the field of Internet open source data processing and situation fusion technology, and discloses a trajectory fusion method based on open source data and AIS data. Background Technology

[0002] This approach integrates ship target description data from multiple sources, fusing open-source text data obtained via the internet (such as data from news websites, social media, and official announcements) with AIS data collected by equipment. This process compensates for blind spots and false alarms caused by relying on a single source. While AIS data can provide real-time information on a ship's reported position, it is susceptible to issues like ship identification number camouflage or AIS transmission equipment shutdown, leading to missing tracks. In such cases, obtaining the ship's trajectory from open-source text data effectively supplements the missing tracks caused by AIS data. By fusing open-source and AIS data, the completeness and accuracy of ship target tracks are improved, enabling earlier detection of anomalies, enhanced law enforcement and compliance monitoring, and improved maritime situational awareness and security control capabilities. Summary of the Invention

[0003] The technical problem to be solved by this invention is to overcome the difficulties in the above-mentioned background technology and provide a trajectory fusion method based on open source data and AIS data.

[0004] The technical problem to be solved by this invention is achieved by the following technical solution: A trajectory fusion method based on open-source data and AIS data includes the following steps: (1) Extract ship information and event information from open-source text data using deep neural network algorithms; (2) Based on the location of the event, obtain the latitude and longitude information of the event location through a reverse geocoding library; (3) A serialized record is formed based on the ship information and the latitude and longitude information of the location of the event. Ship target track data within a set time range before and after the event time is obtained from AIS data. Through spatiotemporal alignment and association, the serialized record and the ship target track data formed by AIS data are fused to form fused track data.

[0005] Furthermore, step (1) specifically involves: (101) Locate the data source address of the ship-related information and use data acquisition tools to capture the original HTML, JSON or XML format text data; (102) Extract plain text corpus from the original HTML, JSON or XML format text data, clean and segment it to form open source text data; (103) Design corresponding event templates for maritime incident information, including event trigger words, event time and event location; (104) Entity recognition is performed using a deep neural network algorithm with a BiLSTM-CRF network structure. The elements included in the event template are extracted from the open-source text data, and spatiotemporal alignment is performed to convert the local time into standard time and clarify the ambiguous time expression into an accurate time. Among them, entities include ships and locations. (105) Using relation extraction technology, identify entities and relationships between entities in the text, locate ship identifiers in open-source text data, and extract the ship's IMO number, MMSI number and ship name; (106) The ships and the relationships between them, the time and location of the events are organized and stored in a “trinity” structure to form ship information and event information.

[0006] Furthermore, step (2) specifically involves: (201) Design the KD-Tree data structure to form an offline reverse geocoding library; (202) Based on the location of the event stored in the event information, perform a place name semantic query through the reverse geocoding library to obtain the latitude and longitude information of the location of the event.

[0007] Furthermore, step (3) specifically involves: (301) The event occurrence time and latitude and longitude information of the ship target extracted from multiple open source text information are spliced ​​together according to the time sequence and spatial continuity, and deduplication and outlier filtering are performed to form a serialized record; (302) Based on the time of the event, extract the ship target track records within a set time range before and after the time from the AIS data, perform data cleaning, outlier removal and format standardization processing to form ship target track data that can be compared with the serialized records formed by open source text in the same dimension. (303) Using ship identification and ship target track data, the serialized records formed by open source text and the ship target track data formed by AIS data are matched and corresponded to form the same continuous curve; (304) Using the Kalman filter algorithm, track abrupt changes, jitter and outliers are removed on the continuous curve, so that the final track achieves the optimal balance between accuracy and smoothness, forming fused track data.

[0008] The advantages of this invention compared to the prior art are: This invention features clear logic, strong formal knowledge expression capabilities, and ease of application integration and expansion. Attached Figure Description

[0009] Figure 1 This is a flowchart of the trajectory fusion method according to an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of the BiLSTM-CRF network structure used for deep neural network computation according to an embodiment of the present invention. Detailed Implementation

[0011] Below, in conjunction with Figures 1 to 2 The present invention will be further described below.

[0012] A trajectory fusion method based on open-source data and AIS data, such as Figure 1 As shown, it includes the following steps: (1) By employing a deep neural network algorithm with a BiLSTM-CRF network structure, ship information and event information are extracted from open-source text data. Specifically: (101) Locate the data source address of the ship-related information and use data acquisition tools to capture the original HTML, JSON or XML format text data; (102) Extract plain text corpus from the original HTML, JSON or XML format text data, clean and segment it to form open source text data; (103) Design corresponding event templates for maritime incident information, including event trigger words, event time and event location; (104) Entity recognition is performed using a deep neural network algorithm with a BiLSTM-CRF network structure. The elements included in the event template are extracted from the open-source text data, and spatiotemporal alignment is performed to convert the local time into standard time and clarify the ambiguous time expression into an accurate time. Among them, entities include ships and locations. (105) Using relation extraction technology, identify entities and relationships between entities in the text, locate ship identifiers in open-source text data, and extract the ship's IMO number, MMSI number and ship name; (106) The ships and the relationships between them, the time and location of the events are organized and stored in a “trinity” structure to form ship information and event information.

[0013] like Figure 2 As shown, the overall model architecture consists of an input layer, a BERT embedding layer, an LSTM encoding layer (including forward LSTM and backward LSTM), and a CRF layer. The input layer first segments the original sentence (e.g., "US Reagan...") into a token sequence based on words or sub-words. The BERT embedding layer uses a pre-trained BERT model to generate a context-aware vector for each token in the sequence, thereby capturing contextual information, resolving the polysemous word problem, and forming a vector sequence. The LSTM encoding layer extracts the serialized temporal features from the vector sequence output by the BERT embedding layer, concatenates the forward and backward states, and obtains the final text representation encoding. The CRF layer decodes the representation encoding and determines the optimal annotation information for words or subwords based on the score. The scoring function is defined as follows: ;

[0014] in, To output the total number of label sequences; Input the observation sequence (such as oranges, time series data); To output the label sequence; The feature function describes the location. Observation Below, tags Is it valid? For the first tag in the sequence The label for each location; For a given first A sequence of labels The conditional probability.

[0015] Based on the scores, the best option is selected, and the Viterbi algorithm is used to solve for the optimal label sequence. This allows the extraction of ship information and the time and location of events from the open-source text, such as when a ship and another ship entered a certain location at a certain time.

[0016] (2) Based on the location of the event, obtain the latitude and longitude information of the event location through a reverse geocoding library; Design a KD-Tree data structure to form an offline reverse geocoding library; using the geographic information of the event extracted in step (1), and based on the Java-based offline reverse geocoding library, utilize the KD-Tree data structure to achieve efficient place name semantic query. For example, descriptions such as "Yokohsuka" and "near Yokosuka Base" appearing in the text need to be accurate to calculable latitude and longitude coordinates.

[0017] (3) A serialized record is formed based on the ship information and the latitude and longitude information of the event location. Ship target track data within a set time range before and after the event time is obtained from AIS data. Through spatiotemporal alignment and correlation, the serialized record and the ship target track data formed by AIS data are fused to form fused track data. Specifically: (301) The event occurrence time and latitude and longitude information of the ship target extracted from multiple open source text information are spliced ​​together according to the time sequence and spatial continuity, and deduplication and outlier filtering are performed to form a serialized record; (302) Based on the time of the event, extract the ship target track records within a set time range before and after the time from the AIS data, perform data cleaning, outlier removal and format standardization processing to form ship target track data that can be compared with the serialized records formed by open source text in the same dimension. (303) Using ship identification and ship target track data, the serialized records formed by open source text and the ship target track data formed by AIS data are matched and corresponded to form the same continuous curve; (304) Using the Kalman filter algorithm, track abrupt changes, jitter and outliers are removed on the continuous curve, so that the final track achieves the optimal balance between accuracy and smoothness, forming fused track data.

[0018] The Kalman filter algorithm is an effective recursive filter that can fuse a series of noisy data and prediction data based on physical laws. Ship trajectory data is characterized by well-defined physical laws, high inertia, and slow turning. Ship target tracks obtained from AIS data often exhibit positioning errors or multipath jitter. The Kalman filter algorithm is used to smooth the prediction and measurement fusion, removing abrupt changes, jitter, and outliers to form smooth trajectory data.

[0019] First, define the state-space model and the state vector. This indicates the current status of the ship target, such as its position. ,speed The ship's target status can be described as follows: ; Assuming the ship target moves at a constant speed, then Where F is the state transition matrix, This refers to the state of the ship target at the previous moment. This is the noise data from the previous time step. Position observations were obtained using AIS data. ,in Noise during the process and measuring noise The data follows a Gaussian distribution with covariances Q and R. The initial state can be set using the first AIS data point. Subsequent data are added for prediction and updating. During state prediction, the previous posterior state is projected onto the current time step, system dynamics and process noise are calculated, and the covariance matrix is ​​updated. The predicted data is combined with the actual AIS data using "optimal weighting" to generate a smoother estimate of the current state.

Claims

1. A trajectory fusion method based on open-source data and AIS data, characterized in that, Includes the following steps: (1) Extract ship information and event information from open-source text data using deep neural network algorithms; (2) Based on the location of the event, obtain the latitude and longitude information of the event location through a reverse geocoding library; (3) A serialized record is formed based on the ship information and the latitude and longitude information of the location of the event. Ship target track data within a set time range before and after the event time is obtained from AIS data. Through spatiotemporal alignment and association, the serialized record and the ship target track data formed by AIS data are fused to form fused track data.

2. The trajectory fusion method based on open-source data and AIS data according to claim 1, characterized in that, Step (1) is as follows: (101) Locate the data source address of the ship-related information and use data acquisition tools to capture the original HTML, JSON or XML format text data; (102) Extract plain text corpus from the original HTML, JSON or XML format text data, clean and segment it to form open source text data; (103) Design corresponding event templates for maritime incident information, including event trigger words, event time and event location; (104) Entity recognition is performed using a deep neural network algorithm with a BiLSTM-CRF network structure. The elements included in the event template are extracted from the open-source text data, and spatiotemporal alignment is performed to convert the local time into standard time and clarify the ambiguous time expression into an accurate time. Among them, entities include ships and locations. (105) Using relation extraction technology, identify entities and relationships between entities in the text, locate ship identifiers in open-source text data, and extract the ship's IMO number, MMSI number and ship name; (106) The ships and the relationships between them, the time and location of the events are organized and stored in a “trinity” structure to form ship information and event information.

3. The trajectory fusion method based on open-source data and AIS data according to claim 1, characterized in that, Step (2) is as follows: (201) Design the KD-Tree data structure to form an offline reverse geocoding library; (202) Based on the location of the event stored in the event information, perform a place name semantic query through the reverse geocoding library to obtain the latitude and longitude information of the location of the event.

4. The trajectory fusion method based on open-source data and AIS data according to claim 1, characterized in that, Step (3) is as follows: (301) The event occurrence time and latitude and longitude information of the ship target extracted from multiple open source text information are spliced ​​together according to the time sequence and spatial continuity, and deduplication and outlier filtering are performed to form a serialized record; (302) Based on the time of the event, extract the ship target track records within a set time range before and after the time from the AIS data, perform data cleaning, outlier removal and format standardization processing to form ship target track data that can be compared with the serialized records formed by open source text in the same dimension. (303) Using ship identification and ship target track data, the serialized records formed by open source text and the ship target track data formed by AIS data are matched and corresponded to form the same continuous curve; (304) Using the Kalman filter algorithm, track abrupt changes, jitter and outliers are removed on the continuous curve, so that the final track achieves the optimal balance between accuracy and smoothness, forming fused track data.

Images