Knowledge and data fusion driven sea surface target type identification method and device

Abstract

The invention provides a knowledge and data fusion driven sea surface target type identification method and device. The method comprises the following steps: acquiring AIS data of a to-be-identified target; based on the AIS data, using a grid method to extract context information of a to-be-identified target; performing feature extraction on the trajectory information corresponding to the AIS data by using a pre-trained Transform network to obtain first feature information; performing feature extraction on the context information by using a pre-trained CNN network to obtain second feature information; performing target identification based on the first feature information and the second feature information to obtain an initial identification result; establishing an identification rule by using the decision tree, and performing feature extraction and classification on the track information and the context information based on the identification rule to obtain a rule identification result; and performing decision-level fusion on the initial recognition result and the rule recognition result based on a Dempster combination rule to obtain a final recognition result, thereby improving the accuracy of sea surface target recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of target recognition, and in particular to a method and device for recognizing sea surface target types driven by knowledge and data fusion. Background Art

[0002] In recent years, with the deepening of globalization and the booming development of international trade, maritime transportation has become an indispensable part of global economic activities. According to statistics, more than 90% of the world's trade goods rely on maritime transportation for circulation, which makes the safety and efficiency of maritime transportation particularly important. As an important means to ensure navigation safety and optimize maritime transportation management, the Automatic Identification System (AIS) effectively improves the monitoring level and security capabilities of maritime traffic by exchanging ship dynamic information in real time. How to accurately identify the type of sea surface targets, especially when the targets may be deliberately disguised, has become a technical problem that needs to be solved in the current field of maritime security.

[0003] In response to the challenge of sea surface target recognition, researchers have carried out a lot of research work and made some progress. Among them, trajectory-based target recognition method is a common technical means at present. This method collects and analyzes the navigation trajectory data of ships, and uses data-driven methods to explore the differences in motion characteristics between different target types, thereby realizing the classification and recognition of targets. In addition, in order to improve the accuracy of recognition, some studies have also tried to incorporate sea surface situational information into the recognition model, such as marine meteorology, channel layout, traffic flow, etc., to enrich the contextual information of target recognition. The introduction of these situational information has, to a certain extent, enhanced the model's adaptability to complex sea surface environments and improved the accuracy and robustness of recognition.

[0004] Although the target recognition method based on the fusion of trajectory and contextual information has improved the accuracy of sea surface target recognition to a certain extent, there are still many shortcomings. First, pure data-driven methods are often difficult to effectively distinguish when dealing with different types of targets with similar motion characteristics, resulting in a decrease in recognition accuracy. Secondly, when the target deliberately disguises its navigation trajectory or sends false information, it is difficult for existing methods to accurately identify its true identity, thereby increasing the difficulty of maritime security. In addition, although the introduction of contextual information provides more contextual information for the recognition model, most of the existing methods rely on neural networks for data processing and feature extraction, lacking effective mining and utilization of deep-level knowledge in contextual information, which limits the further improvement of recognition performance. Therefore, it is an urgent need for current research to develop a sea surface target recognition technology that can comprehensively consider trajectory information and sea surface contextual information and effectively deal with target camouflage situations. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a method and device for identifying sea surface target types driven by knowledge and data fusion.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a method for identifying sea surface target types driven by knowledge and data fusion, comprising:

[0008] Obtain AIS data of the target to be identified;

[0009] Based on AIS data, the grid method is used to extract the context information of the target to be identified; the context information includes: water depth information, offshore distance and traffic density;

[0010] The pre-trained Transformer network is used to extract features of the trajectory information corresponding to the AIS data to obtain the first feature information;

[0011] The pre-trained CNN network is used to extract features from the context information to obtain the second feature information;

[0012] Performing target recognition based on the first feature information and the second feature information to obtain an initial recognition result;

[0013] The decision tree is used to establish recognition rules, and the trajectory information and context information are extracted and classified based on the recognition rules to obtain the rule recognition results;

[0014] The initial recognition result and the rule recognition result are fused at the decision level based on the Dempster combination rule to obtain the final recognition result.

[0015] Optionally, based on AIS data, the water depth information of the target to be identified is extracted by using a grid method:

[0016] Determine the latitude and longitude coordinates of the target to be identified based on AIS data;

[0017] Calculate the distance between the current longitude and latitude coordinates and all coordinates in the original water depth database, and select the water depth value corresponding to the coordinate in the original water depth database closest to the current longitude and latitude coordinates as the water depth value corresponding to the current longitude and latitude coordinates;

[0018] Traverse all longitude and latitude coordinates to generate water depth information of the target to be identified divided into grids.

[0019] Optionally, based on AIS data, the grid method is used to extract the offshore distance of the target to be identified:

[0020] Determine the ocean grid where the AIS data of the target to be identified is located, and traverse to obtain the number of longitude grids and latitudinal grids between the land grids closest to the ocean grid where the AIS data is located;

[0021] Based on the number of meridional grids and the number of latitudinal grids, the offshore grid distance is obtained using Manhattan distance;

[0022] The offshore grid distance is multiplied by the preset grid size to obtain the offshore distance; among them, grids with water depth greater than 0 are all ocean grids, and grids with water depth less than or equal to 0 are all land grids.

[0023] Optionally, a pre-trained Transformer network is used to extract features of the trajectory information corresponding to the AIS data to obtain first feature information, including:

[0024] Using a sine function or a cosine function to positionally encode points in the trajectory information to obtain encoded trajectory information;

[0025] The pre-trained Transformer network is used to extract features of the encoded trajectory information to obtain the first feature information; when the index of the point in the trajectory information is an even number, the sine function is used for position encoding; when the index of the point in the trajectory information is an odd number, the cosine function is used for position encoding.

[0026] Optionally, performing target recognition based on the first feature information and the second feature information to obtain an initial recognition result includes:

[0027] The attention mechanism is used to perform weighted fusion processing on the first feature information and the second feature information to obtain a fusion feature;

[0028] The fused features are classified by a pre-trained classifier to obtain the initial recognition result.

[0029] Optionally, a decision tree is used to establish recognition rules, and feature extraction and classification based on the recognition rules are performed on the trajectory information and context information to obtain rule recognition results, including:

[0030] Obtain trajectory features corresponding to trajectory information;

[0031] Obtaining context features corresponding to the context information;

[0032] The decision tree and information gain method are used to classify the trajectory features and context features to obtain the rule recognition results.

[0033] Optionally, the trajectory features include:

[0034] Average speed, average heading, average speed change, average heading change, maximum speed change, maximum heading change and maximum speed.

[0035] Optionally, the context features include: maximum water depth, average water depth, minimum water depth, maximum distance from shore, average distance from shore and minimum distance from shore.

[0036] In a second aspect, the present invention provides a sea surface target type recognition device driven by knowledge and data fusion, the sea surface target type recognition device driven by knowledge and data fusion includes: an acquisition unit, an extraction unit, a classification unit and a fusion unit;

[0037] The acquisition unit is used to: acquire AIS data of the target to be identified;

[0038] The extraction unit is used to: extract the context information of the target to be identified based on the AIS data by using a grid method; the context information includes: water depth information, offshore distance and traffic density;

[0039] The extraction unit is also used to: use a pre-trained Transformer network to extract features from the trajectory information corresponding to the AIS data to obtain first feature information;

[0040] The extraction unit is also used to: extract features from the context information using a pre-trained CNN network to obtain second feature information;

[0041] The classification unit is used to: perform target recognition based on the first feature information and the second feature information to obtain an initial recognition result;

[0042] The classification unit is also used to: establish recognition rules using decision trees, extract and classify features of trajectory information and context information based on recognition rules, and obtain rule recognition results;

[0043] The fusion unit is used to perform decision-level fusion of the initial recognition result and the rule recognition result based on the Dempster combination rule to obtain the final recognition result.

[0044] In a third aspect, the present invention provides a sea surface target type identification device driven by knowledge and data fusion, comprising: a processor, a storage medium and a bus, the storage medium storing machine-readable instructions executable by the processor, when the sea surface target type identification device driven by knowledge and data fusion is running, the processor and the storage medium communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the sea surface target type identification method driven by knowledge and data fusion as described in the first aspect above.

[0045] The present invention provides a method and device for identifying sea surface target types driven by the fusion of knowledge and data. Among them, the method for identifying sea surface target types driven by the fusion of knowledge and data includes: obtaining AIS data of the target to be identified; based on the AIS data, using the grid method to extract the context information of the target to be identified; the context information includes: water depth information, distance from the shore, and traffic density; using a pre-trained Transformer network to extract feature information from the trajectory information corresponding to the AIS data to obtain first feature information; using a pre-trained CNN network to extract feature information from the context information to obtain second feature information; performing target identification based on the first feature information and the second feature information to obtain an initial identification result; using a decision tree to establish an identification rule, performing feature extraction and classification based on the identification rule on the trajectory information and the context information to obtain a rule identification result; fusing the initial identification result and the rule identification result at the decision level based on the Dempster combination rule to obtain a final identification result. In the present invention, the first feature information and the second feature information respectively extracted by the pre-trained Transformer network and the pre-trained CNN network are used together for target identification, that is, an identification result based on data is generated, and at the same time, the problem of inaccurate feature extraction when using a single grid for target identification is avoided; secondly, based on the initial identification result, a decision tree is used to establish an identification rule, and feature extraction and classification based on the identification rule are performed on the trajectory information and the context information to obtain a rule identification result, that is, an identification result of the target based on knowledge; finally, the initial identification result and the rule identification result are jointly used to determine the final identification result, realizing the effective combination of the deep knowledge in the corresponding time-series data and the corresponding information of the trajectory and the context, and improving the accuracy of sea surface target identification in the case of target camouflage.

[0046] The following will further elaborate on the present invention in conjunction with the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic flow chart of a method for identifying sea surface target types driven by the fusion of knowledge and data provided by an embodiment of the present invention;

[0048] Figure 2 Exemplarily shows a traffic density map of 6 common ship types;

[0049] Figure 3 Exemplarily shows a flow processing block diagram of a method for identifying sea surface target types driven by the fusion of knowledge and data;

[0050] Figure 4 Exemplarily shows a final identification result diagram when a fishing boat is camouflaged as a cargo ship;

[0051] Figure 5A schematic diagram of the structure of a sea surface target type recognition device driven by knowledge and data fusion provided by an embodiment of the present invention;

[0052] Figure 6 A schematic diagram of the structure of a sea surface target type recognition device driven by knowledge and data fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0054] In order to improve the accuracy of sea surface target recognition under target camouflage conditions, an embodiment of the present invention provides a sea surface target type recognition method driven by knowledge and data fusion. Figure 1 The present invention provides a flow chart of a method for identifying sea surface target types driven by knowledge and data fusion. Figure 1 As shown, the method includes:

[0055] S101: Acquire AIS data of a target to be identified.

[0056] It should be noted that AIS data is data collected using the AIS system. AIS (Automatic Identification System) is an automatic tracking system installed on ships (targets to be identified) for exchanging electronic data with neighboring ships, AIS shore stations, satellites and other equipment for ship traffic control identification and positioning, playing a huge role in the field of marine traffic safety management. Compared with SAR images, the acquisition of SAR images is limited by conditions such as satellite orbits and weather, while AIS data can continuously provide ship identification information on a global scale. In addition, since AIS data mainly contains basic information about ships (position, speed, heading, ship name and type, etc.), the data structure is simple, so the storage requirements of AIS data are low and the transmission efficiency is high.

[0057] However, since the AIS system relies on radio signal propagation, signal interference or data transmission errors may cause noise in the AIS data. For example, the positioning data of some ships may be affected by weather, equipment problems or signal interference, resulting in inaccurate or abnormal position data. In addition, some ships may tamper with, forge or turn off AIS information, including location, speed, heading, ship name type and other data, in order to carry out some illegal activities. Therefore, judging the target type by the ship name type in the AIS data alone may be misleading or misidentified.

[0058] S102: Based on the AIS data, a grid method is used to extract context information of the target to be identified.

[0059] Contextual information includes: water depth information, distance from shore and traffic density.

[0060] Optionally, based on AIS data, the water depth information of the target to be identified is extracted by using a grid method:

[0061] Determine the latitude and longitude coordinates of the target to be identified based on AIS data;

[0062] Calculate the distance between the current longitude and latitude coordinates and all coordinates in the original water depth database, and select the water depth value corresponding to the coordinate in the original water depth database closest to the current longitude and latitude coordinates as the water depth value corresponding to the current longitude and latitude coordinates;

[0063] Traverse all longitude and latitude coordinates to generate water depth information of the target to be identified divided into grids.

[0064] It should be noted that the original water depth database is a water depth data table with uniform spatial distribution obtained by preprocessing with the grid method to address the problem of uniform spatial distribution of water depth data in the preset research area.

[0065] It can be understood that, in the embodiment of the present invention, by adopting the grid method to acquire context information, the amount of calculation can be reduced and the processing speed of target recognition can be improved.

[0066] Optionally, based on AIS data, the grid method is used to extract the offshore distance of the target to be identified:

[0067] Determine the ocean grid where the AIS data of the target to be identified is located, and traverse to obtain the number of longitude grids and latitudinal grids between the land grids closest to the ocean grid where the AIS data is located;

[0068] Based on the number of longitude grids and latitudinal grids, the offshore grid distance is obtained using Manhattan distance;

[0069] The offshore grid distance is multiplied by the preset grid size to obtain the offshore distance; grids with water depth greater than 0 are all ocean grids, and grids with water depth less than or equal to 0 are all land grids.

[0070] Furthermore, based on AIS data, the grid method is used to extract the traffic density of the target to be identified as follows:

[0071] Determine the grid where the AIS data of the target to be identified is located, and determine the number of AIS data of all types of ships corresponding to the grid. Since the AIS data of the same marine mobile communication service identification code MMSI (that is, the same ship) is only recorded once, the number of all ships of the type corresponding to the target to be identified in the grid where the target to be identified is located can be obtained according to the AIS data, which is used as the traffic density.

[0072] Figure 2 Traffic density diagrams for six common ship types are shown as examples. Figure 2 As shown, the vessel types include: sailboats + yachts, passenger ships, fishing boats, cargo ships, oil tankers and tugboats. Figure 2 The orange area represents land, the white area represents ocean, and the blue area represents traffic density. The darker the blue area, the greater the traffic density of ships in the area. During the navigation process of a ship, each track point will correspond to the traffic density value of each type of ship. This value can be used to assist in identifying the type of ship. Specifically, from Figure 2 It can be seen that different types of ships often appear in different areas. Sailboats and yachts are mostly found in offshore areas, passenger ship tracks are mostly found on fixed routes between two lands, fishing boats are mostly found near fishing grounds in offshore areas, and only cargo ships and tankers appear in the open ocean areas far from land. Tugboats usually appear in areas far from land and are mainly responsible for rescuing stranded ships, but they will not appear in the open ocean areas like cargo ships and tankers.

[0073] S103: Using a pre-trained Transformer network, extract features of the trajectory information corresponding to the AIS data to obtain first feature information.

[0074] Optionally, S103 may specifically include:

[0075] Using a sine function or a cosine function to positionally encode points in the trajectory information to obtain encoded trajectory information;

[0076] The pre-trained Transformer network is used to extract features of the encoded trajectory information to obtain the first feature information; when the index of the point in the trajectory information is an even number, the sine function is used for position encoding; when the index of the point in the trajectory information is an odd number, the cosine function is used for position encoding.

[0077] It should be noted that the generation process of the pre-trained Transformer network includes:

[0078] Get the initial Transformer network and trajectory sample information;

[0079] Input the trajectory sample information into the initial Transformer network, and continuously train the initial Transformer network in the direction of decreasing the first loss function corresponding to the initial Transformer network;

[0080] The initial Transformer network corresponding to when the loss value of the first loss function is less than the preset first loss threshold is used as the pre-trained Transformer network, wherein the network structures of the initial Transformer network and the pre-trained Transformer network are exactly the same.

[0081] S104: Use the pre-trained CNN network to extract features of the context information to obtain second feature information.

[0082] It should be noted that the generation process of the pre-trained CNN network is similar to that of the pre-trained Transformer network, including:

[0083] Get the initial CNN network and context sample information;

[0084] Input the context sample information into the initial CNN network, and continuously train the initial CNN network in the direction in which the second loss function corresponding to the initial CNN network decreases;

[0085] The initial CNN network corresponding to the time when the loss value of the second loss function is less than the preset second loss threshold is used as the pre-trained CNN network, wherein the network structures of the initial CNN network and the pre-trained CNN network are exactly the same.

[0086] S105. Perform target recognition based on the first feature information and the second feature information to obtain an initial recognition result.

[0087] Optionally, S105 may specifically include:

[0088] The attention mechanism is used to perform weighted fusion processing on the first feature information and the second feature information to obtain a fusion feature;

[0089] The fused features are classified by a pre-trained classifier to obtain the initial recognition result.

[0090] Optionally, the pre-trained classifier may employ a pre-trained fully connected layer.

[0091] Furthermore, in this embodiment, the weighted fusion coefficient of the first feature information and the second feature information has a sum value of 1, and the specific weighted fusion coefficient can be determined according to prior information or actual application scenarios, or adjusted during the training process of the pre-trained CNN network and the pre-trained Transformer network, which is not limited in this embodiment of the present invention.

[0092] S106 , using the decision tree to establish recognition rules, extracting and classifying features of the trajectory information and context information based on the recognition rules, and obtaining rule recognition results.

[0093] Optionally, S106 may specifically include:

[0094] Obtain trajectory features corresponding to trajectory information;

[0095] Obtaining context features corresponding to the context information;

[0096] The decision tree and information gain method are used to classify the trajectory features and context features to obtain the rule recognition results.

[0097] Alternatively, a decision tree is a supervised learning algorithm for classification tasks that can automatically discover data classification rules through training data and present these rules in a visual tree structure for easy understanding and interpretation. Information gain is used to measure the degree to which information uncertainty is reduced after a feature splits the data.

[0098] In this embodiment, the decision tree is specifically a pre-trained visual tree structure using information gain, trajectory sample features, and context sample features. The visual tree structure can be used to visually display the pre-trained recognition rules.

[0099] Optionally, the trajectory features include:

[0100] Average speed, average heading, average speed change, average heading change, maximum speed change, maximum heading change and maximum speed.

[0101] Optionally, the context features include: maximum water depth, average water depth, minimum water depth, maximum distance from shore, average distance from shore and minimum distance from shore.

[0102] S107, performing decision-level fusion on the initial recognition result and the rule recognition result based on the Dempster combination rule to obtain the final recognition result.

[0103] Specifically, when the target recognition type given by the initial recognition result and the rule recognition result is the same, the final recognition result is directly determined. Only when the recognition results given by the initial recognition result and the rule recognition result are inconsistent, the Dempster combination rule is effective. Specifically, the commutative law and associative law of the Dempster combination rule can be used to fuse the initial recognition result and the rule recognition result at the decision level to obtain the final recognition result.

[0104] The embodiment of the present invention provides a method for identifying sea surface target types driven by knowledge and data fusion, which uses the first feature information and the second feature information corresponding to the extracted information of the pre-trained Transformer network and the pre-trained CNN network to jointly perform target identification, that is, generate a data-based identification result, and at the same time avoid the problem of inaccurate feature extraction when using a single grid for target identification; secondly, based on the initial recognition result, a recognition rule is established using a decision tree, and feature extraction and classification based on the recognition rule are performed on the trajectory information and the context information to obtain a rule recognition result, that is, generate a knowledge-based target recognition result; finally, the initial recognition result and the rule recognition result are used to jointly determine the final recognition result, which realizes the effective combination of the time series data corresponding to the trajectory and the context and the deep knowledge in the corresponding information, and improves the accuracy of sea surface target recognition under the condition of target camouflage.

[0105] In order to generally illustrate the execution process of a method for identifying sea surface target types driven by knowledge and data fusion provided by an embodiment of the present invention, Figure 3 The following is an exemplary flowchart of a method for identifying sea surface target types driven by knowledge and data fusion. Figure 3 As shown, firstly, the context information is extracted and the feature extraction of the context information is performed based on the pre-trained CNN network (CNN network) to obtain the second feature information. The feature extraction of the trajectory information corresponding to the AIS data is performed using the pre-trained Transformer network (Transformer network) to obtain the first feature information, and the first feature information and the second feature information are feature fused based on the attention weighted fusion method, and the target type is identified to obtain the initial recognition result. Then, based on the decision tree, the context features obtained from the context information and the trajectory features obtained from the trajectory information are used to perform target recognition processing based on the recognition rule to obtain the rule recognition result. Finally, the initial recognition result and the rule recognition result are fused at the decision level based on the Dempster combination rule to obtain the final recognition result, and the result is output.

[0106] In order to verify the effectiveness of the sea surface target type recognition method driven by knowledge and data fusion provided by the embodiment of the present invention, a simulation experiment was also conducted in the embodiment of the present invention.

[0107] In this embodiment, six most common ship types are selected for six-category classification. The selected ship types include: Cargo, Fishing, Sailing, Passenger, Pleasure and Tanker. 500 trajectories are selected for each target, and 100 data points are selected for each trajectory.

[0108] Since type recognition is a multi-classification problem, accuracy (acc), precision (Precision), recall (Recall) and F1 score (F1-Score) are used to evaluate the experimental results.

[0109] The pre-trained Transformer network processing process is recorded as module 1, the pre-trained CNN network processing process is recorded as module 2, and the decision tree processing process is recorded as module 3. Only the first four types of ships are identified, and the final identification results are shown in Table 1 below.

[0110] Table 1 Comparison of test results

[0111]

[0112]

[0113] It can be seen from Table 1 that the sea surface target type recognition method driven by knowledge and data fusion proposed in the present invention significantly improves the recognition accuracy. The effect of the present invention is further demonstrated below through an example of a fishing boat disguised as a cargo ship to evade detection and engage in illegal fishing. The final results are shown in Table 2 below.

[0114] Table 2 Comparison of test results

[0115]

[0116] From the above experiments, it can be seen that when the number of categories is large and rough recognition is required, the fusion recognition effect of deep neural network modules 1 and 2 is better, while when the number of categories is small and more detailed recognition is required, the effect of extracting recognition rules using decision trees and then performing recognition is better. The sea surface target type recognition method driven by knowledge and data fusion proposed in the present invention can combine the advantages of both. Further, on this basis, Figure 4 The final recognition result diagram when the fishing boat is disguised as a cargo ship is shown as an example. Figure 4 As shown in the figure, the horizontal axis represents the predicted category based on the method of the present invention, and the vertical axis represents the actual category. Among them, the actual number of cargo ships is 101, and the actual number of fishing boats is 99. After identification based on the method of the present invention, 91 and 96 represent the number of cargo ships and fishing boats that are accurately identified, respectively, 10 represents 10 cargo ships that are mistakenly identified as fishing boats, and 3 represents 3 fishing boats that are mistakenly identified as cargo ships.

[0117] From the above analysis, it can be seen that the probability of misidentifying a fishing boat as a cargo ship based on the method of the present invention is smaller than the probability of misidentifying a cargo ship as a fishing boat. This is mainly because in this embodiment, the decision tree uses the maximum speed and the average speed as the first and second division nodes based on information gain, and some cargo ships are mistaken for fishing boats because they do not reach the speed that should be achieved during normal navigation for some reasons. This is consistent with the identification rules obtained by the present invention, which also reflects the practical significance of the present invention.

[0118] The method provided in the embodiment of the present invention can be applied to an electronic device. Specifically, the electronic device can be: a desktop computer, a portable computer, an intelligent mobile terminal, a server, etc., which is not limited in the embodiment of the present invention.

[0119] Based on the same inventive concept, an embodiment of the present invention also provides a sea surface target type identification device driven by knowledge and data fusion. Figure 5 A schematic diagram of the structure of a sea surface target type recognition device driven by knowledge and data fusion provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, the sea surface target type recognition device includes: an acquisition unit 501, an extraction unit 502, a classification unit 503 and a fusion unit 504;

[0120] The acquisition unit 501 is used to: acquire AIS data of the target to be identified;

[0121] The extraction unit 502 is used to extract the context information of the target to be identified based on the AIS data by using a grid method; the context information includes: water depth information, offshore distance and traffic density;

[0122] The extraction unit 502 is further used to: use a pre-trained Transformer network to extract features from the trajectory information corresponding to the AIS data to obtain first feature information;

[0123] The extraction unit 502 is further used to: extract features from the context information using a pre-trained CNN network to obtain second feature information;

[0124] The classification unit 503 is used to: perform target recognition based on the first feature information and the second feature information to obtain an initial recognition result;

[0125] The classification unit 503 is further used to: establish a recognition rule using a decision tree, extract and classify features of the trajectory information and context information based on the recognition rule, and obtain a rule recognition result;

[0126] The fusion unit 504 is used to perform decision-level fusion on the initial recognition result and the rule recognition result based on the Dempster combination rule to obtain the final recognition result.

[0127] Figure 6A schematic diagram of the structure of a sea surface target type recognition device driven by knowledge and data fusion provided by an embodiment of the present invention includes: a processor 610, a storage medium 620 and a bus 630, wherein the storage medium 620 stores machine-readable instructions executable by the processor 610, and when the sea surface target type recognition device driven by knowledge and data fusion is running, the processor 610 communicates with the storage medium 620 via the bus 630, and the processor 610 executes the machine-readable instructions to execute the steps of the above method embodiment. The specific implementation method and technical effect are similar and will not be repeated here.

[0128] The storage medium may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage. Optionally, the storage medium may also be at least one storage device located away from the aforementioned processor.

[0129] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0130] It should be noted that the terms "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention.

[0131] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0132] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other changes of the above disclosed embodiments by viewing the drawings and the disclosed content. In the description of the present invention, the term "comprising" does not exclude other components or steps, "one" or "an" does not exclude multiple situations, and the meaning of "multiple" is two or more, unless otherwise clearly and specifically limited. In addition, certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0133] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A method for identifying sea surface target types driven by knowledge and data fusion, characterized in that: include: Obtain AIS data of the target to be identified; Based on the AIS data, a grid method is used to extract context information of the target to be identified; The context information includes: water depth information, distance from shore and traffic density; Using a pre-trained Transformer network to extract features from the trajectory information corresponding to the AIS data to obtain first feature information; Using a pre-trained CNN network to extract features from the context information to obtain second feature information; Performing target recognition based on the first feature information and the second feature information to obtain an initial recognition result; Establishing recognition rules by using a decision tree, extracting and classifying features of the trajectory information and the context information based on the recognition rules, and obtaining rule recognition results; The initial recognition result and the rule recognition result are fused at the decision level based on the Dempster combination rule to obtain the final recognition result.

2. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 1 is characterized in that: Based on the AIS data, the water depth information of the target to be identified is extracted using the grid method as follows: Determine the latitude and longitude coordinates of the target to be identified according to the AIS data; Calculate the distance between the current longitude and latitude coordinates and all coordinates in the original water depth database, and select the water depth value corresponding to the coordinate in the original water depth database that is closest to the current longitude and latitude coordinates as the water depth value corresponding to the current longitude and latitude coordinates; All the latitude and longitude coordinates are traversed to generate the water depth information of the target to be identified divided into grids.

3. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 1 is characterized in that: Based on the AIS data, the method of extracting the offshore distance of the target to be identified by using the grid method is as follows: Determine the ocean grid where the AIS data of the target to be identified is located, and traverse to obtain the number of longitude grids and the number of latitudinal grids between the land grids closest to the ocean grid where the AIS data is located; Based on the number of longitudinal grids and the number of latitudinal grids, the offshore grid distance is obtained using the Manhattan distance; The offshore grid distance is multiplied by a preset grid size to obtain the offshore distance; wherein, grids with a water depth greater than 0 are all ocean grids, and grids with a water depth less than or equal to 0 are all the land grids.

4. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 1 is characterized in that: The using of the pre-trained Transformer network to extract features from the trajectory information corresponding to the AIS data to obtain first feature information includes: Using a sine function or a cosine function to perform position encoding on points in the trajectory information to obtain encoded trajectory information; The pre-trained Transformer network is used to perform feature extraction on the encoded trajectory information to obtain the first feature information; wherein, when the index of the point in the trajectory information is an even number, the sine function is used for position encoding; when the index of the point in the trajectory information is an odd number, the cosine function is used for position encoding.

5. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 1 is characterized in that: The performing target recognition based on the first feature information and the second feature information to obtain an initial recognition result includes: Using an attention mechanism to perform weighted fusion processing on the first feature information and the second feature information to obtain a fusion feature; The fusion features are classified by a pre-trained classifier to obtain the initial recognition result.

6. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 1 is characterized in that: The method of establishing a recognition rule by using a decision tree, extracting and classifying the trajectory information and the context information based on the recognition rule, and obtaining a rule recognition result includes: Acquire trajectory features corresponding to the trajectory information; Acquire context features corresponding to the context information; The trajectory features and the context features are classified and processed by using a decision tree and information gain method to obtain the rule recognition result.

7. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 6 is characterized in that: The trajectory features include: Average speed, average heading, average speed change, average heading change, maximum speed change, maximum heading change and maximum speed.

8. The method for identifying sea surface target types driven by knowledge and data fusion according to claim 6 is characterized in that: The context features include: maximum water depth, average water depth, minimum water depth, maximum offshore distance, average offshore distance and minimum offshore distance.

9. A sea surface target type recognition device driven by knowledge and data fusion, characterized in that: The sea surface target type recognition device driven by knowledge and data fusion includes: an acquisition unit, an extraction unit, a classification unit and a fusion unit; The acquisition unit is used to: acquire AIS data of the target to be identified; The extraction unit is used to extract context information of the target to be identified based on the AIS data using a grid method; the context information includes water depth information, offshore distance and traffic density; The extraction unit is further used to: use a pre-trained Transformer network to extract features from the trajectory information corresponding to the AIS data to obtain first feature information; The extraction unit is also used to: use a pre-trained CNN network to extract features from the context information to obtain second feature information; The classification unit is used to: perform target recognition based on the first feature information and the second feature information to obtain an initial recognition result; The classification unit is further used to: establish a recognition rule using a decision tree, extract and classify the trajectory information and the context information based on the recognition rule, and obtain a rule recognition result; The fusion unit is used to perform decision-level fusion on the initial recognition result and the rule recognition result based on the Dempster combination rule to obtain a final recognition result.

10. A sea surface target type recognition device driven by knowledge and data fusion, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the sea surface target type identification device driven by knowledge and data fusion is running, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to perform the steps of the sea surface target type identification method driven by knowledge and data fusion as described in any one of claims 1 to 8.

Images