A Multi-Sensor Fusion Identification Method for Marine Vessels

Abstract

This invention discloses a multi-sensor fusion identification method for maritime vessels, belonging to the field of multi-sensor information fusion and target recognition technology. The method includes: constructing an identification framework and a causal graph model of maritime vessel type-sensor-marine environment; generating counterfactual facts based on interference quantifiers; calculating evidence confidence and counterfactual weights; calculating information conflict between the causal space and the evidence space; adaptive processing driven by conflict type; and outputting the maritime vessel type identification result. This invention, by fully integrating the causal model into the D-S evidence theory framework, achieves counterfactual generation, constructs a measure of information conflict between the causal space and the evidence space, and enables causal-evidence collaborative fusion, which helps solve long-standing problems such as sea state interference and difficulty in distinguishing the root causes of information conflicts.

Description

Technical Field

[0001] This invention belongs to the field of multi-sensor information fusion and target recognition technology, specifically involving a multi-sensor fusion recognition method for maritime vessels based on counterfactual generation and information conflict resolution. It is applicable to the fusion recognition of maritime vessel types using multi-source heterogeneous sensors in scenarios such as maritime traffic management, port monitoring, and maritime safety. Background Technology

[0002] Several studies have applied DS evidence theory to the field of multi-sensor information fusion for maritime vessels. Patent CN202410013591.X discloses a "method for multi-source data fusion of inland waterway vessel navigation," which uses DS evidence theory to fuse measurements from AIS, radar, and video at time k associated with the same target. Patent CN202210594119.0 discloses a "method for risk evolution analysis of ship pilotage process based on FRAM-DBN," which maps key risk factors and their coupling relationships in operational scenarios to a Bayesian (BN) network topology model. It uses DS evidence theory and parameter adaptive algorithms to determine the prior and conditional probabilities of BN nodes, supporting the construction of an analytical model for the risk evolution of ship pilotage processes. Patent CN201310201242.2 discloses a "real-time monitoring method for vessels underway based on AIS and VTS information fusion," which constructs trust functions for each focal element in the vessel identification framework based on the Kalman prediction algorithm and performs evidence synthesis within the DS evidence theory identification framework. While the aforementioned patents have achieved certain results in their respective fields, they have not solved the following technical problems: First, these methods struggle to distinguish between information conflicts caused by different factors, such as changes in the target itself (e.g., ship turning, speed changes), sensor malfunctions, and marine environmental interference (e.g., sea fog, wave clutter), which can easily lead to misjudgments. Second, sensors are often affected by marine environmental factors such as sea fog, rain, snow, wave clutter, diurnal light variations, and electromagnetic interference, resulting in a decline in the quality of observational data. Traditional methods rely solely on current sensor data and lack the ability to causally infer "what should be seen," making them susceptible to being misled by changes in the marine environment. Third, traditional DS evidence theory only deals with the correlation between evidence and fails to uncover the causal mechanisms behind the observational data, resulting in limited accuracy in complex dynamic environments and a lack of interpretability in the identification results.

[0003] To address the aforementioned issues, this invention proposes a multi-sensor fusion identification method for maritime vessels based on counterfactual generation and information conflict resolution. By fully incorporating the causal hierarchy model into the DS evidence theory framework, it achieves counterfactual generation and virtual-evidence collaborative fusion, and constructs a measurement mechanism for the degree of information conflict between the causal space and the evidence space. This effectively solves long-standing technical challenges such as small-sample vessel identification, sea state interference countermeasures, and conflict root cause differentiation. Summary of the Invention

[0004] In view of this, the present invention provides a multi-sensor fusion identification method for marine vessels, which solves the problems of data sparsity and sea state interference by generating counterfactual facts, and refines the root causes of conflicts by resolving information conflicts between causal space and evidence space, thereby improving the accuracy, robustness and interpretability of vessel type identification under complex sea conditions.

[0005] The technical solution adopted in this invention is as follows:

[0006] A multi-sensor fusion identification method for marine vessels includes the following steps:

[0007] Step 1: Construct a causal graph model of ship type, sensor, and marine environment;

[0008] Step 2: Generate counterfactual evidence based on the interference calculation.

[0009] Step 3: Calculate the confidence level of true evidence and the weight of counterfactual evidence;

[0010] Step 4: Calculate the causal spatial conflict degree and the real evidence spatial conflict degree;

[0011] Step 5: Construct a conflict type identification table and perform adaptive collaborative fusion based on the conflict type.

[0012] Furthermore, the specific method of step 1 is as follows:

[0013] Step 1-1: Construct the node set V of the cause-effect graph, including the ship type variable T, sensor variable S, and marine environmental variable E;

[0014] Steps 1-2: Construct the directed edge set L of the causal graph, where the directed edges include: edges from ship type variables to sensor observation variables, and edges from marine environmental variables to sensor observation variables.

[0015] Steps 1-3: Construct the edge weight set W of the directed edges, where W contains elements... This represents the causal influence strength of directed edge b. The value range is [0,1];

[0016] Steps 1-4: Construct a causal graph model G=(V,L,W) of the marine vessel-sensor-marine environment.

[0017] Furthermore, the specific method for step 2 is as follows:

[0018] Step 2-1, let Θ be the identification frame, containing all mutually exclusive and complete basic propositions, i.e., all possible ship types. Introduce the interference quantifier do(·) for sensor S. i Set ship type and marine environment Calculate the basic probability distribution of counterfactual evidence :

[0019]

[0020] in, This indicates that the maritime vessel type T is forcibly set as the specified vessel type. , This indicates that the marine environment E is the set marine environment. A is a proposition. Θ, To enable sensor S to operate under intervention conditions and set marine environmental conditions. i The observation results support the proposition. The probability of;

[0021] Step 2-2, for each ship type Repeat step 2-1 to obtain counterfactual evidence under all the assumed conditions.

[0022] Furthermore, the specific method for step 3 is as follows:

[0023] Step 3-1, calculate the degree of mutual support C among the true evidence. real :

[0024]

[0025]

[0026] Where n is the number of sensors. and Sensors and The observation results represent the basic probability allocation for the corresponding propositions;

[0027] Step 3-2, calculate the confidence level α of the true evidence:

[0028]

[0029]

[0030] in, , Under normal historical conditions The mean and standard deviation, For adjustment coefficients, The value ranges from 2 to 3, where λ is the sensitivity coefficient. This applies to scenarios involving civilian vessel traffic management and port monitoring. The value ranges from 1 to 2, for maritime safety monitoring scenarios. The value is between 2 and 3, for security-sensitive scenarios. The value ranges from 3 to 5; the value of α is (0,1), where α=1 indicates complete trust in the evidence and α=0 indicates complete distrust of the evidence.

[0031] Step 3-3, calculate sensor S i causal path strength :

[0032]

[0033] in, For all sensors S i The set of causal paths ending at p, where p represents a causal path consisting of directed edges b. Let b be the weight of edge b. ;

[0034] Steps 3-4: Calculate sensor S i Counterfactual evidence weight :

[0035]

[0036] in, For sensor S j The causal path strength, where n is the number of sensors.

[0037] Furthermore, step 4 is specifically implemented as follows:

[0038] Step 4-1: Based on the causal graph model from Step 1 and the counterfactual evidence from Step 2, calculate the sensor S. i In a given marine environment Expected observation allocation :

[0039]

[0040] in, The ship types are derived from historical statistics. The prior probability, Assuming the ship type is The basic probability distribution of counterfactual evidence obtained at that time;

[0041] Step 4-2, calculate sensor S i True observational basic probability allocation Basic probability allocation of counterfactual evidence The difference between them, i.e. the degree of causal spatial conflict :

[0042]

[0043] Step 4-3: Calculate the spatial conflict of true evidence among all sensors. The average value is then used as the global real space conflict index. :

[0044]

[0045]

[0046] Where A and B are propositions, and n is the number of sensors.

[0047] Furthermore, step 5 is specifically implemented as follows:

[0048] Step 5-1: Calculate the causal space conflict threshold and the evidence space conflict threshold, and construct an identification table for information conflict types. The specific method is as follows:

[0049] Step 5-1-1: Calculate the causal spatial conflict threshold based on historical normal data. :

[0050]

[0051] in, and All sensors under normal historical conditions The population mean and population standard deviation, This refers to the corresponding adjustment coefficient;

[0052] Step 5-1-2: Calculate the spatial conflict threshold of true evidence based on historical normal data. :

[0053]

[0054] in, and These represent conflicts in the global real evidence space under normal historical conditions. The mean and standard deviation, This refers to the corresponding adjustment coefficient;

[0055] Step 5-1-3: Construct an identification table for information conflict types between counterfactual evidence and true evidence, and provide the characteristics, diagnostic conclusions and handling strategies for each conflict type;

[0056] Among them, the characteristic of type I is: the presence of sensor S i , making ,and The diagnosis is: environmental mutation; the treatment strategy is: update the cause-effect graph structure and adjust the state of environmental nodes.

[0057] Type II is characterized by the presence of sensor S.i , making ,and The diagnosis was: sensor failure or local interference. The treatment strategy was: use a discount factor to reduce the weight of evidence from the suspected sensor.

[0058] Type III is characterized by the presence of sensor S. i , making ,and The diagnosis was: the actual state of the ship has changed, and the handling strategy is: retain the conflict information and trigger a reassessment of the ship's state.

[0059] Type IV is characterized by the following: for all sensors, it satisfies ,and The diagnosis conclusion is: normal situation. The treatment strategy is: no additional treatment is required, and routine collaborative fusion is performed directly.

[0060] Step 5-2, define Dempster combination rules and discount operations:

[0061] For the two basic probability assignments and Dempster combination rules Defined as:

[0062]

[0063]

[0064] The discount operation γ·m is defined as:

[0065]

[0066] Step 5-3: Perform adaptive collaborative fusion based on conflict type:

[0067] Step 5-3-1, for type I:

[0068] Identify all sensor nodes affected by the environment;

[0069] Using historical window data, the weights of the "Environment → Sensor" edge are updated via linear regression. ;

[0070] Update the causal graph model and recalculate the expected observation assignments using step 4-1. ;

[0071] Computational fusion results :

[0072]

[0073]

[0074]

[0075] Where α is the degree of trust in the true evidence, and its value ranges from (0,1). α=1 indicates that the evidence is completely trusted, and α=0 indicates that the evidence is completely distrusted. For sensor S n The basic probability distribution of true observations; For sensor S n The basic probability assignment weights of counterfactual evidence;

[0076] Step 5-3-2, for Type II:

[0077] The sensor with the highest average conflict rate with other sensors is located;

[0078] Calculate the discount factor ;

[0079] Correcting the basic probability assignment of the sensor: ;

[0080] Computational fusion results :

[0081]

[0082] Step 5-3-3, for Type III:

[0083] Mark the state change points and reinitialize the tracking filter;

[0084] Based on the re-evaluated tracking filter, the results of true evidence fusion and discounted counterfactual evidence fusion are recalculated. :

[0085]

[0086]

[0087] Step 5-3-4, for type IV:

[0088] Without re-estimating, we directly perform collaborative fusion to obtain the fusion result. :

[0089]

[0090] .

[0091] The present invention has the following beneficial effects:

[0092] 1. This invention generates counterfactuals of "what should be observed" by interfering with the quantizer do(·). When evidence is missing or interfered with, it can provide data support for subsequent conflict resolution and improved identification accuracy based on the causal graph of "what should be observed if there is no interference".

[0093] 2. This invention identifies conflict types and processing strategies by constructing information conflict results between counterfactual facts and evidence, thereby realizing the identification of evidence conflicts and updating the fusion strategy, and improving the fusion accuracy.

[0094] 3. This invention corrects deviations by providing "expected observations" through counterfactual evidence, which can improve the accuracy of identification under conditions of marine environmental interference or changes in the motion state of ships at sea.

[0095] 4. By constructing an identification table for information conflict types between counterfacts and evidence, this invention can explain the reasons for conflicts between evidence, make the identification process transparent, and improve the credibility of the fusion method.

[0096] 5. This invention does not rely on special sensors, can be adapted to different types of sensors, and has wide applicability. Attached Figure Description

[0097] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0098] Figure 2 This is a schematic diagram of the marine vessel type-sensor-marine environment causal graph model in an embodiment of the present invention.

[0099] Figure 3 This is a schematic diagram of the process for generating counterfactuals based on interference calculations in an embodiment of the present invention.

[0100] Figure 4 This is a flowchart of the conflict type-driven adaptive processing in an embodiment of the present invention. Detailed Implementation

[0101] The present invention will now be described in further detail with reference to the accompanying drawings.

[0102] A multi-sensor fusion identification method for marine vessels includes the following steps:

[0103] Step 1: Construct a causal graph model of ship type, sensor, and marine environment; the specific method is as follows:

[0104] Step 1-1: Construct the node set V of the cause-effect graph, including the ship type variable T, sensor variable S, and marine environmental variable E;

[0105] Steps 1-2: Construct the directed edge set L of the causal graph, where the directed edges include: edges from ship type variables to sensor observation variables, and edges from marine environmental variables to sensor observation variables.

[0106] Steps 1-3: Construct the edge weight set W of the directed edges, where W contains elements... This represents the causal influence strength of directed edge b. The value range is [0,1];

[0107] Steps 1-4: Construct a causal graph model G=(V,L,W) of the marine vessel-sensor-marine environment.

[0108] Step 2: Generate counterfactual evidence based on the interference calculation; the specific method is as follows:

[0109] Step 2-1, let Θ be the identification frame, containing all mutually exclusive and complete basic propositions (i.e., the set of all possible ship types, such as: Θ = {container ship, cargo ship, tanker, fishing boat}), introduce the interference quantifier do(·), for sensor S i Set ship type and marine environment Calculate the basic probability distribution of counterfactual evidence :

[0110]

[0111] in, This indicates that the maritime vessel type T is forcibly set as the specified vessel type. , This indicates that the marine environment E is the set marine environment. A is a proposition. Θ (e.g., A={container ship} indicates that the identified ship type is a container ship, and A={container ship, cargo ship} indicates that the identified ship type is either a container ship or a cargo ship). To enable sensor S to operate under intervention conditions and set marine environmental conditions. i The observation results support the proposition. The probability of;

[0112] Step 2-2, for each ship type Repeat step 2-1 to obtain counterfactual evidence under all the assumed conditions;

[0113] Step 3: Calculate the confidence level of true evidence and the weight of counterfactual evidence; the specific method is as follows:

[0114] Step 3-1, calculate the degree of mutual support C among the true evidence. real :

[0115]

[0116]

[0117] Where n is the number of sensors. and Sensors and The observation results represent the basic probability allocation for the corresponding propositions;

[0118] Step 3-2, calculate the confidence level α of the true evidence:

[0119]

[0120]

[0121] in, , Under normal historical conditions The mean and standard deviation, For adjustment coefficients, The value ranges from 2 to 3, where λ is the sensitivity coefficient. This applies to scenarios involving civilian vessel traffic management and port monitoring. The value ranges from 1 to 2, for maritime safety monitoring scenarios. The value is between 2 and 3, for security-sensitive scenarios. The value ranges from 3 to 5; the value of α is (0,1), where α=1 indicates complete trust in the evidence and α=0 indicates complete distrust of the evidence.

[0122] Step 3-3, calculate sensor S i causal path strength :

[0123]

[0124] in, For all sensors S i The set of causal paths ending at p, where p represents a causal path consisting of directed edges b. Let b be the weight of edge b. ;

[0125] Steps 3-4: Calculate sensor S i Counterfactual evidence weight :

[0126]

[0127] in, For sensor S j The causal path strength, where n is the number of sensors;

[0128] Step 4: Calculate the causal spatial conflict degree and the real evidence spatial conflict degree; the specific method is as follows:

[0129] Step 4-1: Based on the causal graph model from Step 1 and the counterfactual evidence from Step 2, calculate the sensor S. i In a given marine environment Expected observation allocation :

[0130]

[0131] in, The ship types are derived from historical statistics. The prior probability, Assuming the ship type is The basic probability distribution of counterfactual evidence obtained at that time;

[0132] Step 4-2, calculate sensor S i True observational basic probability allocation Basic probability allocation of counterfactual evidence The difference between them, i.e. the degree of causal spatial conflict :

[0133]

[0134] Step 4-3: Calculate the spatial conflict of true evidence among all sensors. The average value is then used as the global real space conflict index. :

[0135]

[0136]

[0137] Where A and B are propositions, and n is the number of sensors;

[0138] Step 5: Construct a conflict type identification table and perform adaptive collaborative fusion based on the conflict type; the specific method is as follows:

[0139] Step 5-1: Calculate the causal space conflict threshold and the evidence space conflict threshold, and construct an identification table for information conflict types. The specific method is as follows:

[0140] Step 5-1-1: Calculate the causal spatial conflict threshold based on historical normal data. :

[0141]

[0142] in, and All sensors under normal historical conditions The population mean and population standard deviation, This refers to the corresponding adjustment coefficient;

[0143] Step 5-1-2: Calculate the spatial conflict threshold of true evidence based on historical normal data. :

[0144]

[0145] in, and These represent conflicts in the global real evidence space under normal historical conditions. The mean and standard deviation, This refers to the corresponding adjustment coefficient;

[0146] Step 5-1-3: Construct an identification table for information conflict types between counterfactual evidence and true evidence, and provide the characteristics, diagnostic conclusions and handling strategies for each conflict type;

[0147] Among them, the characteristic of type I is: the presence of sensor S i , making ,and The diagnosis is: environmental mutation; the treatment strategy is: update the cause-effect graph structure and adjust the state of environmental nodes.

[0148] Type II is characterized by the presence of sensor S. i , making ,and The diagnosis was: sensor failure or local interference. The treatment strategy was: use a discount factor to reduce the weight of evidence from the suspected sensor.

[0149] Type III is characterized by the presence of sensor S. i , making ,and The diagnosis was: the actual state of the ship has changed, and the handling strategy is: retain the conflict information and trigger a reassessment of the ship's state.

[0150] Type IV is characterized by the following: for all sensors, it satisfies ,and The diagnosis conclusion is: normal situation. The treatment strategy is: no additional treatment is required, and routine collaborative fusion is performed directly.

[0151] Step 5-2, define Dempster combination rules and discount operations:

[0152] For the two basic probability assignments and Dempster combination rules Defined as:

[0153]

[0154]

[0155] The discount operation γ·m is defined as:

[0156]

[0157] Step 5-3: Perform adaptive collaborative fusion based on conflict type:

[0158] Step 5-3-1, for type I:

[0159] Identify all sensor nodes affected by the environment;

[0160] Using historical window data, the weights of the "Environment → Sensor" edge are updated via linear regression. ;

[0161] Update the causal graph model and recalculate the expected observation assignments using step 4-1. ;

[0162] Computational fusion results :

[0163]

[0164]

[0165]

[0166] Where α is the degree of trust in the true evidence, and its value ranges from (0,1). α=1 indicates that the evidence is completely trusted, and α=0 indicates that the evidence is completely distrusted. For sensor S n The basic probability distribution of true observations; For sensor S n The basic probability assignment weights of counterfactual evidence;

[0167] Step 5-3-2, for Type II:

[0168] The sensor with the highest average conflict rate with other sensors is located;

[0169] Calculate the discount factor ;

[0170] Correcting the basic probability assignment of the sensor: ;

[0171] Computational fusion results :

[0172]

[0173] Step 5-3-3, for Type III:

[0174] Mark the state change points and reinitialize the tracking filter;

[0175] Based on the re-evaluated tracking filter, the results of true evidence fusion and discounted counterfactual evidence fusion are recalculated. :

[0176]

[0177]

[0178] Step 5-3-4, for type IV:

[0179] Without re-estimating, we directly perform collaborative fusion to obtain the fusion result. :

[0180]

[0181] .

[0182] The following is an example of a multi-sensor ship identification scenario in a maritime traffic management system. The data below is only an example to describe the implementation process of this method and is not the actual value during the collection and processing.

[0183] 1. System Settings:

[0184] Sensor configuration: radar, AIS, visible light, infrared;

[0185] Deployment location: Coastal monitoring stations, covering the sea area near ports;

[0186] The maritime vessel type identification framework Θ = {container ships, cargo ships, oil tankers, fishing vessels}.

[0187] 2. Implementation process:

[0188] 1) Constructing an identification framework and a causal graph model of ship type, sensor, and marine environment.

[0189] (1) Constructing a recognition framework

[0190] Define sensor type ( For radar, For AIS, For visible light, (For infrared), the identification frame Θ = {container ship, cargo ship, oil tanker, fishing boat} is constructed using historical data to construct sensor S. iBasic Probability Assignment (BPA) function The following conditions must be met:

[0191] Empty set does not allocate trust

[0192] The sum of the trust levels of all subsets is 1.

[0193] in, This indicates the degree of confidence that the observations from sensor i support proposition A (i.e., "the true type of the target belongs to subset A"). For example, Indicates sensor S i The observations support the level of trust in the target, container ships. Indicates sensor S i The observations support the level of trust in the target as a container ship or cargo ship.

[0194] Assuming that based on historical data, the basic probability allocation (BPA) of the actual observation results of the four types of sensors is obtained, an example is shown in Table 1:

[0195] Table 1. Examples of Basic Probability Allocation (BPA) for Real Observations from Four Types of Sensors

[0196]

[0197] (2) Constructing a cause-effect diagram of marine vessel type-sensor-marine environment

[0198] a) Constructing causal graph nodes

[0199] Construct a node T for different types of ships at sea: {container ship, cargo ship, oil tanker, fishing vessel}; construct a node for sensor observation. (radar), (AIS) (Visible light) (Infrared); Construct marine environment node E1 (sea state: calm, light waves, moderate waves, large waves), marine environment node E2 (weather: sunny, fog, rain, snow), marine environment node E3 (light: daytime, dusk, night);

[0200] b) Set the edge and causal path strength between nodes based on historical data.

[0201] Assuming that based on historical data, the influence strength of the edges in the causal graph and the corresponding causal paths is statistically obtained, an example is shown in Table 2:

[0202] Table 2 Examples of the influence strength of causal graph edges and corresponding causal paths

[0203]

[0204] 2) Generating counterfactual evidence based on interference calculations

[0205] (1) Setting up a marine environment

[0206] Current marine environment: E1 = light waves, E2 = fog, E3 = daytime. Visible light is affected by sea fog, resulting in blurred images; infrared is less affected by fog; radar is affected by sea clutter but is acceptable; AIS signal is normal.

[0207] (2) For all ship types Θ, generate counterfactual evidence for that marine environment.

[0208] Data was selected from historical data under conditions of light waves, fog, and daytime at sea. The average probability of visible light identifying vessel types was calculated and used as the basic probability allocation for counterfactual evidence. ,

[0209]

[0210] The above method was used to generate counterfactual data for four types of sensors in this marine environment. Examples are shown in Table 3:

[0211] Table 3 Counterfacts about visible light in the current marine environment Example

[0212]

[0213] (3) For all ship types Θ={container ship, cargo ship, tanker, fishing vessel}, generate counterfactual scenarios for all marine environments.

[0214] Repeat (1) and (2) to generate counterfactual evidence data for all sensors in all marine environments.

[0215] 3) Calculate the confidence level of true evidence and the weight of counterfactual evidence.

[0216] (1) Calculate the confidence level α of true evidence.

[0217] Calculate the pairwise collision coefficients for the four types of sensors: radar, AIS, visible light, and infrared.

[0218]

[0219] Table 4 shows an example of calculating the collision coefficient between radar and visible light:

[0220] Table 4. Examples of Calculation Results for Radar and Visible Light Conflict Coefficients

[0221]

[0222] Calculate the average collision degree among all sensors The degree of mutual support between sensor evidence The confidence level α of the evidence, assuming it can be obtained from historical data:

[0223] = (0.577 + 0.695 + 0.643 + 0.692 + 0.565 + 0.726) / 6 = 0.650

[0224] =0.350

[0225] Calculate the confidence level of true evidence. :

[0226]

[0227] in, , Under normal historical conditions The mean and standard deviation, For adjustment coefficients, in this example Assuming based on historical data, , ,but =0.75, let the sensitivity coefficient be... =2, ≈ 0.310.

[0228] (2) Calculate the counterfactual weights of the sensors

[0229] Since all causal paths in this embodiment are one-sided, the summation is performed directly, and the counterfactual weights of each sensor are calculated. Calculation examples are shown in Table 5:

[0230] Table 5 Counterfactual Weights of Each Sensor Calculation Example Explanation

[0231]

[0232] 4) Calculate the information conflict between the causal space and the evidence space.

[0233] Based on the causal graph model in step 1 and the counterfactual evidence in step 2, the sensor S is calculated. i In a given marine environment Expected observation allocation :

[0234]

[0235] in, Let be the prior probability of the ship type (assuming a uniform distribution in this example, each with a probability of 0.25). Assuming the ship type is The basic probability distribution of counterfactual evidence obtained at that time.

[0236] According to Table 3:

[0237] θ = container ship: (Container ship, Container ship) = 0.60

[0238] θ = cargo ship: (Container ship, cargo ship) = 0.15

[0239] θ = oil tanker: (Container ships, oil tankers) = 0.15

[0240] θ = fishing boat: (Container ships, fishing boats) = 0.05

[0241] =0.25×0.6+0.25×0.15+0.25×0.15+0.25×0.05=0.2375

[0242] Similarly, the basic probability distribution of counterfactual evidence for other propositions can be calculated.

[0243] Calculate the four types of sensors Assuming the conclusion is based on historical data, =0.10, =0.50, =0.42, =0.08.

[0244] 5) Adaptive processing driven by conflict type

[0245] Assuming the conclusion is based on historical normal data:

[0246] =0.12, =0.04, =2, therefore =0.20

[0247] =0.35, =0.10, =2, therefore =0.55.

[0248] current, ,and =0.55, therefore it is determined to be conflict type III.

[0249] Mark the current moment as the state change point, reinitialize the tracking filter, use the current observation as the initial value for observation, and re-fuse based on the re-estimated state.

[0250] 6) Output the results of marine vessel type identification.

[0251] Assuming that after a period of re-tracking, the final identification result is obtained: =0.62, =01.4, =0.09, =0.05

[0252] Based on the maximum confidence principle of DS evidence theory, the ship type with the highest basic probability allocation is selected as the final identification result, i.e.:

[0253]

[0254] Calculate the overall trust level of the recognition results If all focal elements are single-element propositions, If a non-single-element focal element exists, then , To support the type of maritime vessel The minimum level of trust;

[0255] Calculate the realism of the recognition results. , , All and Incompatible propositions To support the type of maritime vessel The highest level of trust;

[0256] Use recognition results and credibility range [ , ], supporting subsequent decision-making (such as maritime vessel traffic management, maritime safety monitoring, or autonomous navigation), within the [section] , The width of the symbol reflects the degree of uncertainty in the identification result; the wider the width, the more likely the type of vessel is to be identified. The more ambiguous the certainty;

[0257] like ( For set values, such as The system can directly accept the recognition results;

[0258] like ( For set values, such as ),or If the result is uncertain, it is marked as an uncertain identification result.

[0259] In this example, all focal elements are single-element propositions, therefore =0.62, An interval width of 0 indicates that the recognition result is certain.

[0260] Due to 0.4 This yields a fusion identification conclusion of "ship type = container ship, confidence level = medium".

[0261] This invention fully integrates the causal model into the DS evidence theory framework, enabling counterfactual generation, constructing a measure of information conflict between the causal and evidence spaces, and achieving causal-evidence synergistic fusion. This helps solve long-standing problems such as sea state interference and difficulty in distinguishing the root causes of information conflicts. The method is applicable to the fusion identification of ship types using multi-source heterogeneous sensors in scenarios such as maritime traffic management, port monitoring, and maritime safety.

Claims

1. A multi-sensor fusion identification method for marine vessels, characterized in that, Includes the following steps: Step 1: Construct a causal graph model of ship type, sensor, and marine environment; Step 2: Generate counterfactual evidence based on the interference calculation. Step 3: Calculate the confidence level of true evidence and the weight of counterfactual evidence; Step 4: Calculate the causal spatial conflict degree and the real evidence spatial conflict degree; Step 5: Construct a conflict type identification table and perform adaptive collaborative fusion based on the conflict type.

2. The multi-sensor fusion identification method for marine vessels according to claim 1, characterized in that, The specific method for step 1 is as follows: Step 1-1: Construct the node set V of the cause-effect graph, including the ship type variable T, sensor variable S, and marine environmental variable E; Steps 1-2: Construct the directed edge set L of the causal graph, where the directed edges include: edges from ship type variables to sensor observation variables, and edges from marine environmental variables to sensor observation variables. Steps 1-3: Construct the edge weight set W of the directed edges, where W contains elements... This represents the causal influence strength of directed edge b. The value range is [0,1]; Steps 1-4: Construct a causal graph model G=(V,L,W) of the marine vessel-sensor-marine environment.

3. The multi-sensor fusion identification method for marine vessels according to claim 2, characterized in that, The specific method for step 2 is as follows: Step 2-1, let Θ be the identification frame, containing all mutually exclusive and complete basic propositions, i.e., all possible ship types. Introduce the interference quantifier do(·) for sensor S. i Set ship type and marine environment Calculate the basic probability distribution of counterfactual evidence. : in, This indicates that the maritime vessel type T is forcibly set as the specified vessel type. , This indicates that the marine environment E is the set marine environment. A is a proposition. Θ, To enable sensor S to operate under intervention conditions and set marine environmental conditions. i The observation results support the proposition. The probability of; Step 2-2, for each ship type Repeat step 2-1 to obtain counterfactual evidence under all the assumed conditions.

4. The multi-sensor fusion identification method for marine vessels according to claim 3, characterized in that, The specific method for step 3 is as follows: Step 3-1, calculate the degree of mutual support C among the true evidence. real : Where n is the number of sensors. and Sensors and The observation results represent the basic probability allocation for the corresponding propositions; Step 3-2, calculate the confidence level α of the true evidence: in, , Under normal historical conditions The mean and standard deviation, For adjustment coefficients, The value ranges from 2 to 3, where λ is the sensitivity coefficient. This applies to scenarios involving civilian vessel traffic management and port monitoring. The value ranges from 1 to 2, for maritime safety monitoring scenarios. The value is between 2 and 3, for security-sensitive scenarios. The value ranges from 3 to 5; the value of α is (0,1), where α=1 indicates complete trust in the evidence and α=0 indicates complete distrust of the evidence. Step 3-3, calculate sensor S i causal path strength : in, For all sensors S i The set of causal paths ending at p, where p represents a causal path consisting of directed edges b. Let b be the weight of edge b. ; Steps 3-4: Calculate sensor S i Counterfactual evidence weight : in, For sensor S j The causal path strength, where n is the number of sensors.

5. The multi-sensor fusion identification method for marine vessels according to claim 4, characterized in that, The specific method for step 4 is as follows: Step 4-1: Based on the causal graph model from Step 1 and the counterfactual evidence from Step 2, calculate the sensor S. i In a given marine environment Expected observation allocation : in, The ship types are derived from historical statistics. The prior probability, Assuming the ship type is The basic probability distribution of counterfactual evidence obtained at that time; Step 4-2, calculate sensor S i True observational basic probability allocation Basic probability allocation of counterfactual evidence The difference between them, i.e. the degree of causal spatial conflict : Step 4-3: Calculate the spatial conflict of true evidence among all sensors. The average value is then used as the global real space conflict index. : Where A and B are propositions, and n is the number of sensors.

6. The multi-sensor fusion identification method for marine vessels according to claim 1, characterized in that, The specific method for step 5 is as follows: Step 5-1: Calculate the causal space conflict threshold and the evidence space conflict threshold, and construct an identification table for information conflict types. The specific method is as follows: Step 5-1-1: Calculate the causal spatial conflict threshold based on historical normal data. : in, and All sensors under normal historical conditions The population mean and population standard deviation, This refers to the corresponding adjustment coefficient; Step 5-1-2: Calculate the spatial conflict threshold of true evidence based on historical normal data. : in, and These represent conflicts in the global real evidence space under normal historical conditions. The mean and standard deviation, This refers to the corresponding adjustment coefficient; Step 5-1-3: Construct an identification table for information conflict types between counterfactual evidence and true evidence, and provide the characteristics, diagnostic conclusions and handling strategies for each conflict type; Among them, the characteristic of type I is: the presence of sensor S i , making ,and The diagnosis is: environmental mutation; the treatment strategy is: update the cause-effect graph structure and adjust the state of environmental nodes. Type II is characterized by the presence of sensor S. i , making ,and The diagnosis was: sensor failure or local interference. The treatment strategy was: use a discount factor to reduce the weight of evidence from the suspected sensor. Type III is characterized by the presence of sensor S. i , making ,and The diagnosis was: the actual state of the ship has changed, and the handling strategy is: retain the conflict information and trigger a reassessment of the ship's state. Type IV is characterized by the following: for all sensors, it satisfies ,and The diagnosis conclusion is: normal situation. The treatment strategy is: no additional treatment is required, and routine collaborative fusion is performed directly. Step 5-2, define Dempster combination rules and discount operations: For the two basic probability assignments and Dempster combination rules Defined as: The discount operation γ·m is defined as: Step 5-3: Perform adaptive collaborative fusion based on conflict type: Step 5-3-1, for type I: Identify all sensor nodes affected by the environment; Using historical window data, the weights of the "Environment → Sensor" edge are updated via linear regression. ; Update the causal graph model and recalculate the expected observation assignments using step 4-1. ; Computational fusion results : Where α is the degree of trust in the true evidence, and its value ranges from (0,1). α=1 indicates that the evidence is completely trusted, and α=0 indicates that the evidence is completely distrusted. For sensor S n The basic probability distribution of true observations; For sensor S n The basic probability assignment weights of counterfactual evidence; Step 5-3-2, for Type II: The sensor with the highest average conflict rate with other sensors is located; Calculate the discount factor ; Correcting the basic probability assignment of the sensor: ; Computational fusion results : Step 5-3-3, for Type III: Mark the state change points and reinitialize the tracking filter; Based on the re-evaluated tracking filter, the results of true evidence fusion and discounted counterfactual evidence fusion are recalculated. : Step 5-3-4, for type IV: Without re-estimating, we directly perform collaborative fusion to obtain the fusion result. : 。

Images