Method and system for enhancing robustness of multi-source data fusion algorithm

Abstract

The invention discloses a method and a system for enhancing the robustness of a multi-source data fusion algorithm, and aims to improve the adaptability to noise and outliers in a multi-sensor data fusion process. According to the method, an adaptive noise mechanism and an outlier detection mechanism are introduced, so that the influence of abnormal observation data is effectively suppressed, and the fusion precision and the system stability are improved. The method comprises the following steps: firstly, a system carries out preliminary state prediction by initializing a target state, a covariance matrix and observation noise; then, the mahalanobis distance of each observation data is calculated based on chi-square test through an outlier detection mechanism, if the observation data deviates greatly, the observation data is judged as an outlier, and the influence of the observation noise matrix on a fusion result is reduced by increasing the observation noise matrix of the observation data; then, an adaptive noise adjustment method is adopted, adaptive factors are calculated in real time according to observation data of each sensor, and observation noise is dynamically adjusted to adapt to changes of the noise level; and finally, state estimation and covariance updating are carried out in combination with fusion algorithms such as Kalman filtering, and data fusion and system feedback are completed. According to the scheme, noise interference and abnormal values in multi-source data can be effectively dealt with, the robustness and the real-time response capability of the system are improved, and the method has wide application prospects.

Description

Technical Field

[0001] This invention belongs to the field of data fusion technology, specifically relating to a method and system for enhancing the robustness of multi-source data fusion algorithms. Background Technology

[0002] In multi-source data fusion processes, such as collaborative perception of UAV targets, it is typically necessary to fuse data from multiple sensors, including radar, 5G-A, RID, and spectrum sensing. However, real-world application environments are highly complex. Significant differences in the accuracy of different sensors, external environmental interference (such as electromagnetic interference and airflow) can lead to outliers in the observation data, and target maneuvering or changes in the sensor's own state can cause model mismatch and other problems, affecting the performance of traditional fusion algorithms (such as Kalman filtering). This contaminated data can significantly reduce fusion accuracy and even cause filter divergence, resulting in completely erroneous outputs. Therefore, enhancing the robustness of data fusion algorithms—that is, their ability to operate stably under conditions of inaccurate models or abnormal data—is a key technical challenge in this field. Summary of the Invention

[0003] To address the problems of unstable fusion accuracy, high computational complexity, insufficient response to sudden outliers and dynamic noise changes, and lack of specificity in existing data fusion schemes, this invention discloses a method and system for enhancing the robustness of multi-source data fusion algorithms. It provides a computationally lightweight technical solution that can simultaneously address two typical problems: dynamic changes in sensor noise and interference from outlier data. Without altering the existing fusion framework, it significantly improves fusion stability, anomaly resistance, and scene adaptability.

[0004] To solve the technical problem, the technical solution of the present invention is as follows:

[0005] A method for enhancing the robustness of multi-source data fusion algorithms, the method comprising:

[0006] S1: Initialize the state vector, state covariance matrix and observation noise covariance matrix, and establish the system parameters for multi-source data fusion;

[0007] S2: Based on the system parameters initialized in step S1, the target state is recursively calculated over time through the state transition process to obtain the prior state estimate and prior covariance at the current moment.

[0008] S3: Configure robust mechanisms to enhance noise immunity, including: outlier detection switch, adaptive noise adjustment switch, chi-square threshold parameter, and fitness factor range;

[0009] S4: Based on the prior state obtained in step S2, calculate the innovation of each sensor observation and its statistical distance, and use the decision threshold configured in step S3 to identify outliers. For observations identified as outliers, amplify and adjust the corresponding observation noise covariance matrix.

[0010] S5: Based on the observation noise covariance matrix and the corresponding innovation vector output in step S4, calculate the fitness factor for each observation and dynamically adjust the observation noise covariance matrix within a preset range.

[0011] S6: Input the observation noise covariance matrix processed in step S5 into the fusion framework to complete the state estimation and covariance update at the current time. The update result is fed back to step S2 to realize the closed-loop continuous multi-source data fusion.

[0012] Furthermore, step S1 includes:

[0013] Based on prior knowledge of the target's motion, an initial state vector is set, which includes the target's position and velocity information;

[0014] Based on the sensor's own measurement accuracy and characteristics, the state covariance matrix, which reflects the uncertainty of state estimation, and the observation noise covariance matrix, which characterizes the sensor's measurement noise level, are initialized respectively.

[0015] To simplify the initial model, the two covariance matrices mentioned above can be set as diagonal matrices. By completing the initialization of the state and noise statistical characteristics, a complete system parameter foundation and iterative starting point are established for the entire multi-source data fusion process.

[0016] Furthermore, step S2 includes:

[0017] After obtaining the state estimate of the target at the previous moment, time recursion is performed based on the preset target motion model. This process updates the state vector of the previous moment through the state transition matrix and expands the uncertainty of the prediction by using the process noise covariance matrix to calculate the prior state estimate of the current moment and its corresponding prior error covariance matrix.

[0018] The specific forms of the state transition matrix and the process noise covariance matrix are determined by the selected motion model; this prediction step provides the necessary prior estimation basis for subsequent outlier detection and noise adaptive adjustment.

[0019] Furthermore, step S4 includes:

[0020] Online anomaly identification and suppression are performed on the observation data of each sensor;

[0021] First, based on the prior state estimate obtained in step S2, the innovation and covariance matrix of each observation data are calculated, and then the Mahalanobis distance characterizing the degree of deviation is obtained.

[0022] Subsequently, the Mahalanobis distance is compared with a preset decision threshold based on the chi-square test principle; when the Mahalanobis distance exceeds the threshold, the corresponding observation is determined to be an outlier.

[0023] For the identified outliers, their observation noise covariance matrix is ​​amplified according to the relationship between their Mahalanobis distance and the square of the threshold ratio to suppress their weight in subsequent fusion.

[0024] For normal observations, the noise covariance remains unchanged;

[0025] The final output is the observation noise covariance matrix after outlier suppression, which provides input for subsequent adaptive adjustment and data fusion.

[0026] Furthermore, step S5 includes:

[0027] Based on the current deviation between observation and prediction, the observation noise covariance matrix processed in step S4 is dynamically adjusted.

[0028] First, the corresponding adaptation factor is calculated based on the innovation amplitude of the observation data of each sensor. This factor is limited to a preset upper and lower limit range, and its magnitude directly reflects the degree of mismatch between prediction and observation.

[0029] Subsequently, the observation noise covariance matrix is ​​proportionally adjusted using the aforementioned adaptation factor: when the prediction bias is large, the noise level is increased by a larger adaptation factor to reduce the weight of the observation in the fusion.

[0030] When the prediction bias is small, a low noise level is maintained to preserve filter stability;

[0031] Through the above mechanism, real-time adaptive correction of observation noise covariance is achieved, providing noise parameters that match the environment for data fusion.

[0032] Furthermore, step S6 includes:

[0033] The observation information and noise parameters processed by steps S4 and S5, combined with the prior state and prior covariance generated in step S2, are input into the fusion algorithm framework to solve for the optimal state estimate and the updated error covariance matrix at the current time.

[0034] The fusion process calculates the fusion weights of each sensor's information based on the observation noise level and prediction uncertainty, and supports sequential or parallel multi-source update strategies.

[0035] After the update is completed, the output posterior state estimate and posterior covariance matrix are used as inputs for the state prediction at the next time step, thus forming a continuously iterative closed-loop fusion process to achieve robust state estimation under multi-source observation and dynamic noise environment.

[0036] A system for enhancing the robustness of multi-source data fusion algorithms, the system being used to perform any of the methods described above, the system comprising:

[0037] System initialization module: Initializes the state vector, state covariance matrix and observation noise covariance matrix, and establishes the system parameters for multi-source data fusion;

[0038] State prediction module: Based on the system parameters initialized by the system initialization module, the target state is recursively extrapolated through the state transition process to obtain the prior state estimate and prior covariance at the current moment;

[0039] Robust Mechanism Configuration Module: Configures robust mechanisms to enhance noise immunity, including: outlier detection switch, adaptive noise adjustment switch, chi-square threshold parameter, and fitness factor range;

[0040] Outlier detection and suppression module: Based on the prior state obtained by the state prediction module, calculate the innovation of each sensor observation and its statistical distance, and use the decision threshold configured by the robust mechanism configuration module to identify outliers. For observations identified as outliers, amplify and adjust the corresponding observation noise covariance matrix.

[0041] Adaptive noise adjustment module: Based on the observation noise covariance matrix and corresponding innovation vector output by the outlier detection and suppression module, calculate the adaptation factor for each observation, and dynamically adjust the observation noise covariance matrix within a preset range;

[0042] Data fusion and state update module: Input the observation noise covariance matrix processed by the adaptive noise adjustment module into the fusion framework to complete the state estimation and covariance update at the current time. The update result is fed back to the state prediction module to realize closed-loop continuous multi-source data fusion.

[0043] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method described above for enhancing the robustness of a multi-source data fusion algorithm.

[0044] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above for enhancing the robustness of a multi-source data fusion algorithm.

[0045] Compared with the prior art, the advantages of the present invention are as follows:

[0046] This invention combines two lightweight yet efficient strategies—an adaptive noise mechanism and an outlier detection mechanism—to achieve the following significant benefits:

[0047] 1. Significantly improves the robustness of the fusion algorithm

[0048] This invention employs a dual-mechanism approach: an outlier detection mechanism handles "point-level elimination" of instantaneous strong interference, while an adaptive noise mechanism adapts to continuous changes in the noise environment. Compared to existing technologies with single, macro-level indicator-based control mechanisms, this invention's response strategy is more targeted and robust.

[0049] 2. Improve the accuracy and reliability of data fusion.

[0050] By dynamically adjusting the observation noise covariance matrix, this invention enables the fusion model to better match the actual noise statistics, thereby improving the accuracy of state estimation. Simultaneously, effective handling of outliers prevents erroneous measurement information from contaminating the fusion results, ensuring the reliability of the final output.

[0051] 3. Enhance the system's adaptability and response speed.

[0052] Both mechanisms employed in this invention perform real-time judgment and adjustment based on current information, eliminating the need for complex historical data windows or optimization solutions. This enables the algorithm to respond extremely quickly to sudden changes in noise levels or the abrupt appearance of outliers, exhibiting strong adaptability and excellent real-time performance.

[0053] 4. Simple calculation and easy to implement in engineering.

[0054] Compared with existing technologies that rely on complex calculations such as information entropy evaluation and optimization, the additional calculations introduced in this invention only involve basic operations such as Mahalanobis distance and vector norm. The increase in computational load is minimal and does not constitute a computational burden. It is particularly suitable for deployment on low-altitude platforms (such as drones and edge computing nodes) where computing resources are limited. Attached Figure Description

[0055] Figure 1 The main flowchart of a method for enhancing the robustness of multi-source data fusion algorithms proposed in this embodiment;

[0056] Figure 2 A comparison diagram of different fusion methods proposed in this embodiment;

[0057] Figure 3 A comparison chart of different fusion errors presented in this embodiment. Detailed Implementation

[0058] The specific implementation of the present invention is described below with reference to embodiments:

[0059] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0060] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0061] Example 1:

[0062] like Figure 1 As shown in an exemplary embodiment, this invention provides a method for enhancing the robustness of multi-source data fusion algorithms. Its core lies in introducing an adaptive noise mechanism and an outlier detection mechanism to improve traditional fusion algorithms. The complete process of this technical solution is as follows: It can be applied to the process of multi-source data fusion, and the specific steps are as follows: Figure 1 As shown.

[0063] (1) System Initialization

[0064] Set the initial state of the system, including the target's initial position, velocity, and other state vectors:

[0065]

[0066] The state covariance matrix P and the observation noise covariance matrix R are initialized based on the sensor's own parameters or characteristics. They are usually set as diagonal matrices (the initial values ​​of each state variable are independent and uncorrelated).

[0067] Initial state covariance matrix for:

[0068]

[0069] Observation noise covariance matrix for:

[0070]

[0071] in:

[0072] The initial values ​​for position and velocity are uncertain;

[0073] To observe the noise variance;

[0074] for identity matrix for Zero matrix.

[0075] (2) State prediction

[0076] It receives raw target measurement data uploaded from external multi-source sensors and interacts with the internal data management system to obtain the target's previous state. Based on the preset target motion model, the target's previous state is recursively calculated to generate the prior state estimate and the covariance of the prior state at the current moment, i.e., the predicted state vector and the predicted state error covariance matrix:

[0077]

[0078]

[0079] in:

[0080] Here is the state transition matrix. The process noise covariance matrix has a form that depends on the motion model chosen by the fusion system.

[0081] For example, in a uniform linear motion model, the conventional state transition matrix for:

[0082]

[0083] Process noise covariance matrix for:

[0084]

[0085] in, The variance representing process noise, This represents the time interval from time k-1 to time k.

[0086] (3) Configure robust mechanism

[0087] Configure independent switches for outlier detection and adaptive noise mechanisms, and configure adjustable parameters, including chi-square confidence, degrees of freedom, sliding window size, and fitness factor range.

[0088] (4) Outlier detection and suppression

[0089] This step identifies and suppresses outlier observations (i.e., outliers) online. Its working mechanism is as follows:

[0090] (a) Calculate innovation: Based on the prior state estimate of the input, for each sensor track Observational data Calculate the difference (i.e., the innovation) between the observed value and the projection of the predicted state onto the observation space:

[0091]

[0092] in:

[0093] The observation matrix has dimensions that depend on the dimensions of the state vector and the observation vector.

[0094] For example, for sensor data with a six-dimensional state vector and observation vector, its observation matrix... for:

[0095]

[0096] (b) Based on the new information Prediction state error covariance Observation matrix and the current observation noise covariance Calculate the covariance matrix of the new information. Then calculate the Mahalanobis distance of the current observation. This distance characterizes the degree to which the current observation deviates from the prediction: Calculate the Mahalanobis distance:

[0097]

[0098]

[0099] (c) Threshold decision: The calculated Mahalanobis distance is compared with a decision threshold set based on the Chi-Squared test theory. For comparison, the calculation formula is:

[0100]

[0101] It is a critical value of the chi-square distribution, determined by the following two parameters:

[0102] df (Degrees of Freedom): This value is determined by the dimension of the observation vector and represents the number of independent variables being compared. For example, if the sensor only measures three-dimensional position, then df = 3; if the sensor can simultaneously measure three-dimensional position and three-dimensional velocity, then df = 6.

[0103] Confidence Level (p): This value is a pre-defined probability (e.g., 0.95, 0.99) used to define a valid region of association. A confidence level of p means that a valid measurement has a probability of falling within this region of p, and the probability of that measurement being incorrectly rejected (i.e., a Type I error) is 1-p. The higher the confidence level, the looser the threshold, and the greater the tolerance for outliers.

[0104] The following table shows the chi-square critical values ​​and corresponding decision thresholds for different combinations of degrees of freedom and confidence levels:

[0105] Table 1: Chi-square critical values ​​and decision thresholds under commonly used degrees of freedom and confidence levels

[0106]

[0107] (d) Observation noise suppression: if and only if the Mahalanobis distance is greater than the decision threshold, i.e. When this happens, the unit determines the current observation data to be an outlier. Besides directly removing outlier data, another method for handling outlier data is to reduce its weight in subsequent fusion. This is achieved by dynamically increasing the corresponding observation noise covariance matrix through a nonlinear gain function. The increase factor is proportional to the square of the Mahalanobis distance, thereby significantly reducing the influence weight of this outlier data in subsequent fusion calculations and suppressing outliers. The specific calculation is as follows:

[0108]

[0109] (5) Adaptive noise adjustment

[0110] This step adjusts the system model parameters in real time to adapt to dynamic changes in sensor noise levels. Its working mechanism is as follows:

[0111] (a) Adaptation factor calculation: For multi-objective or multi-dimensional observation data, the new information is... Grouping data according to different tracks or dimensions, and analyzing data from each sensor. The innovation vector is:

[0112]

[0113] For each New information Calculate its vector norm Based on the magnitude (i.e., norm) of the new information, a normalization factor is calculated to obtain an adaptation factor. And limit the adaptation factor to a reasonable range (which is a configurable parameter) to prevent over-tuning:

[0114]

[0115] in:

[0116] This is the lower limit of the adaptation factor;

[0117] This is the upper limit of the fitness factor;

[0118] This is the normalization coefficient.

[0119] The magnitude of the adaptation factor is proportional to the magnitude of the innovation, reflecting the degree of mismatch between the current prediction model and actual observations.

[0120] (b) Model parameter correction: Adjust the corresponding sensor measurement data using the aforementioned adaptation factor. Observation noise covariance:

[0121]

[0122] The method described above dynamically and smoothly adjusts the observation noise covariance matrix in the fusion algorithm. When the prediction-observation deviation is large, the covariance is increased to improve the response to new observations; when the deviation is small, the filtering remains stable, thereby improving the system's dynamic response and adaptability.

[0123] (6) Data fusion and status update

[0124] This step executes the final data processing flow. It receives the observation information (i.e., including potentially suppressed outlier weights) and system model parameters (i.e., the adaptively adjusted covariance matrix) processed by the robust steps above. Based on the mathematical framework of the fusion algorithm, it performs the predicted state update calculation, solves for the optimal state estimate of the target at the current moment and the updated state covariance matrix, and outputs the results. Simultaneously, it uses the updated state as the prior state for the next moment, returns to step (2), and performs the next round of prediction and fusion, achieving continuous and robust data fusion. Taking Kalman filtering as an example:

[0125] The fusion algorithm is represented as:

[0126]

[0127]

[0128] in:

[0129] The state fusion mapping can be represented by Kalman filtering, weighted average, particle filtering, etc.

[0130] This indicates an update based on uncertainty after fusion;

[0131] The final output can be used as the prediction input for the next time step, enabling continuous and robust fusion.

[0132] If multiple sensors exist Sequential fusion can be used:

[0133]

[0134]

[0135] in:

[0136] Indicates sequential fusion of sensor indexes;

[0137] Output , Used for prediction in the next time step.

[0138] In an alternative implementation, the target motion model can be a uniform motion model, a uniformly accelerated motion model, or other complex motion models, and the data fusion method can be Kalman filtering, or a variety of fusion methods such as extended Kalman filtering, particle filtering, Bayesian estimation, and information filtering.

[0139] Example 2:

[0140] like Figure 1 As shown, this embodiment applies to Embodiment 1 above. This embodiment provides a method to enhance the robustness of multi-source data fusion, which can be applied to multi-source sensor data fusion systems in fields such as low-altitude UAV surveillance, autonomous driving environmental perception, and robot navigation. To clearly illustrate the present invention, a low-altitude UAV detection system is used as a preferred application scenario for description below.

[0141] 1. In a preferred embodiment, the specific parameters of the low-altitude multi-target tracking system are as follows:

[0142] (1) Data type: UAV flight track monitoring data

[0143] (2) Data source: two sensors, Remote ID and 5G-A.

[0144] Sensor 1 (Remote ID): Configured as a high-precision sensor with a measurement noise standard deviation. =5 m.

[0145] Sensor 2 (5G-A): Configured as a relatively low-precision sensor with a measurement noise standard deviation. =30 m.

[0146] (3) Fusion information: a six-dimensional state vector containing three-dimensional spatial position and velocity information.

[0147]

[0148] (4) Fusion method: Sequential fusion based on Kalman filtering

[0149] (5) Motion model: A constant velocity (CV) linear motion model is adopted, and the simulation time step is... .

[0150] State transition matrix for:

[0151]

[0152] Observation matrix for:

[0153]

[0154] (6) Robust configuration parameters:

[0155] Chi-square confidence level: 99%

[0156] Chi-square degrees of freedom: 6

[0157] Adaptive scaling factor: =0.8, =1.5

[0158] Normalization parameter: =5

[0159] 2. In the preferred embodiment, the specific implementation steps are as follows:

[0160] (1) Receive multi-source observation data

[0161] For each moment Receives observation data from sensor 1 (RID) and sensor 2 (5G-A):

[0162]

[0163] The data undergoes basic preprocessing and spatiotemporal alignment to construct a unified six-dimensional state observation:

[0164]

[0165] Their measurement noise covariance matrices are as follows:

[0166]

[0167] (2) Initialize the system state and covariance matrix

[0168] At the start of the fusion process, the initial state vector of the target is set. State covariance matrix and process noise covariance matrix .

[0169]

[0170]

[0171] Process noise standard deviation Process noise covariance matrix:

[0172]

[0173] (3) Predict the state and covariance at the next time step

[0174] Based on the system's motion model, the current state is predicted to obtain the predicted state and prediction error covariance for the next moment.

[0175]

[0176] (4) Outlier detection and processing

[0177] This step aims to identify and suppress anomalous observation data.

[0178] For each sensor track Observations Calculate innovation:

[0179]

[0180] Calculate Mahalanobis distance:

[0181]

[0182]

[0183] For a case with 6 degrees of freedom and a confidence level of 95%, the chi-square threshold is approximately =16.812, its square root is:

[0184]

[0185] Compare the Mahalanobis distance with the threshold; if... If the current observation is considered an outlier, this embodiment uses a direct removal method to handle outliers.

[0186] (6) Adaptive noise adjustment

[0187] This step is also based on the new information. Regardless of whether the current observation is identified as an outlier, the system will perform this step to adapt to potential changes in noise levels. For multi-track scenarios, this step can be performed for each sensor track. New information Calculate its fitness factor separately In a preferred embodiment, the calculation method is as follows:

[0188]

[0189] Subsequently, the observation noise covariance of the track is adjusted using this adaptation factor:

[0190]

[0191] (7) Use the adjusted parameters to merge and update

[0192] Using the covariance matrix adjusted (or not adjusted) by steps (5) and (6), perform the standard Kalman filter update steps, calculate the Kalman gain, and obtain the optimal state estimate at the current time.

[0193] Joint observation vector from the two sensors:

[0194]

[0195] Kalman gain:

[0196]

[0197] New information vector:

[0198]

[0199] Status Update:

[0200]

[0201] Covariance update:

[0202]

[0203] Final output:

[0204]

[0205] 3. In the preferred embodiment, the specific implementation results are as follows:

[0206] Figure 2 and Figure 3The method of this invention (labeled RAKF) and the standard Kalman filter (labeled SKF) are respectively shown to compare the trajectory and error in the scenario of fusing Remote ID and 5G-A data.

[0207] (See) Figure 2 ),like Figure 2 The diagram clearly shows the observation data from the two sensors (Remote ID is represented by a high-precision blue circle, and 5G-A by a low-precision gray x) and the tracking trajectories of the two filters. It can be seen that the RAKF (red solid line) trajectory of this invention consistently closely follows the actual trajectory (black solid line), while the SKF (blue dashed line) shows significant deviations during the target maneuver phase (t=40s-50s) and when the Remote ID device exhibits outliers (t=60s).

[0208] (See) Figure 3 ),like Figure 3 The diagram quantitatively compares the tracking errors of the two methods. At t=40s, when the target maneuvers, the RAKF method of this invention (solid red line) has a lower peak error and converges faster, demonstrating superior maneuver response capability. At t=60s, when the Remote ID device generates an outlier, the error of SKF (dashed blue line) spikes to nearly 55 meters, while the RAKF method of this invention, having successfully eliminated the outlier, is almost unaffected by the error.

[0209] Simulation results show that the average position error (RMSE) of the standard KF (SKF) is 7.24 meters, while the average position error (RMSE) of the robust adaptive KF (RAKF) of this invention is 5.75 meters, representing a performance improvement of 20.63%.

[0210] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0211] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0212] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0213] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0214] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0215] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A method for enhancing the robustness of multi-source data fusion algorithms, characterized in that, The method includes: S1: Initialize the state vector, state covariance matrix and observation noise covariance matrix, and establish the system parameters for multi-source data fusion; S2: Based on the system parameters initialized in step S1, the target state is recursively calculated over time through the state transition process to obtain the prior state estimate and prior covariance at the current moment. S3: Configure robust mechanisms to enhance noise immunity, including: outlier detection switch, adaptive noise adjustment switch, chi-square threshold parameter, and fitness factor range; S4: Based on the prior state obtained in step S2, calculate the innovation of each sensor observation and its statistical distance, and use the decision threshold configured in step S3 to identify outliers. For observations identified as outliers, amplify and adjust the corresponding observation noise covariance matrix. S5: Based on the observation noise covariance matrix and the corresponding innovation vector output in step S4, calculate the fitness factor for each observation and dynamically adjust the observation noise covariance matrix within a preset range. S6: Input the observation noise covariance matrix processed in step S5 into the fusion framework to complete the state estimation and covariance update at the current time. The update result is fed back to step S2 to realize the closed-loop continuous multi-source data fusion.

2. The method for enhancing the robustness of multi-source data fusion algorithms according to claim 1, characterized in that, Step S1 includes: Based on prior knowledge of the target's motion, an initial state vector is set, which includes the target's position and velocity information; Based on the sensor's own measurement accuracy and characteristics, the state covariance matrix, which reflects the uncertainty of state estimation, and the observation noise covariance matrix, which characterizes the sensor's measurement noise level, are initialized respectively. To simplify the initial model, the two covariance matrices mentioned above can be set as diagonal matrices. By completing the initialization of the state and noise statistical characteristics, a complete system parameter foundation and iterative starting point are established for the entire multi-source data fusion process.

3. The method for enhancing the robustness of multi-source data fusion algorithms according to claim 2, characterized in that, Step S2 includes: After obtaining the state estimate of the target at the previous moment, time recursion is performed based on the preset target motion model. This process updates the state vector of the previous moment through the state transition matrix and expands the uncertainty of the prediction by using the process noise covariance matrix to calculate the prior state estimate of the current moment and its corresponding prior error covariance matrix. The specific forms of the state transition matrix and the process noise covariance matrix are determined by the selected motion model; this prediction step provides the necessary prior estimation basis for subsequent outlier detection and noise adaptive adjustment.

4. The method for enhancing the robustness of multi-source data fusion algorithms according to claim 3, characterized in that, Step S4 includes: Online anomaly identification and suppression are performed on the observation data of each sensor; First, based on the prior state estimate obtained in step S2, the innovation and covariance matrix of each observation data are calculated, thus obtaining the Mahalanobis distance characterizing the degree of deviation. Subsequently, the Mahalanobis distance is compared with a preset decision threshold based on the chi-square test principle; when the Mahalanobis distance exceeds the threshold, the corresponding observation is determined to be an outlier. For the identified outliers, their observation noise covariance matrix is ​​amplified according to the relationship between their Mahalanobis distance and the square of the threshold ratio to suppress their weight in subsequent fusion. For normal observations, the noise covariance remains unchanged; The final output is the observation noise covariance matrix after outlier suppression, which provides input for subsequent adaptive adjustment and data fusion.

5. A method for enhancing the robustness of multi-source data fusion algorithms according to claim 4, characterized in that, Step S5 includes: Based on the current deviation between observation and prediction, the observation noise covariance matrix processed in step S4 is dynamically adjusted. First, the corresponding adaptation factor is calculated based on the innovation amplitude of the observation data of each sensor. The adaptation factor is limited to a preset upper and lower limit range, and its magnitude directly reflects the degree of mismatch between prediction and observation. Subsequently, the observation noise covariance matrix is ​​proportionally adjusted using the aforementioned adaptation factor: when the prediction bias is large, the noise level is increased by a larger adaptation factor to reduce the weight of the observation in the fusion. When the prediction bias is small, a low noise level is maintained to preserve filter stability; Through the above mechanism, real-time adaptive correction of observation noise covariance is achieved, providing noise parameters that match the environment for data fusion.

6. A method for enhancing the robustness of multi-source data fusion algorithms according to claim 5, characterized in that, Step S5 includes: calculating the corresponding adaptation factor based on the innovation amplitude of each sensor observation data, including: For observational data with multiple targets or multiple dimensions, new information will be generated. Grouping data according to different tracks or dimensions, and analyzing data from each sensor. The innovation vector is: ； The information represents the difference between the projection of the observed value and the predicted state onto the observation space, where k represents the current time and i represents the i-th sensor. This represents the raw data actually measured by the i-th sensor at time k; Represents the observation matrix; This represents the prior state estimate; For each New information Calculate its vector norm Based on the magnitude of the new information, i.e., the norm, a normalization factor is calculated to obtain an adaptation factor. And the adaptation factor is limited to a preset upper and lower limit range to prevent over-adjustment: ； in: This is the lower limit of the adaptation factor; This is the upper limit of the fitness factor; These are the normalization coefficients; The magnitude of the adaptation factor is proportional to the magnitude of the innovation, reflecting the degree of mismatch between the current prediction model and actual observations.

7. A method for enhancing the robustness of multi-source data fusion algorithms according to claim 5, characterized in that, Step S6 includes: The observation information and noise parameters processed by steps S4 and S5, combined with the prior state and prior covariance generated in step S2, are input into the fusion algorithm framework to solve for the optimal state estimate and the updated error covariance matrix at the current time. The fusion process calculates the fusion weights of each sensor's information based on the observation noise level and prediction uncertainty, and supports sequential or parallel multi-source update strategies. After the update is completed, the output posterior state estimate and posterior covariance matrix are used as inputs for the state prediction at the next time step, thus forming a continuously iterative closed-loop fusion process to achieve robust state estimation under multi-source observation and dynamic noise environment.

8. A system for enhancing the robustness of multi-source data fusion algorithms, characterized in that, The system is used to perform the method according to any one of claims 1-7, the system comprising: System initialization module: Initializes the state vector, state covariance matrix and observation noise covariance matrix, and establishes the system parameters for multi-source data fusion; State prediction module: Based on the system parameters initialized by the system initialization module, the target state is recursively extrapolated through the state transition process to obtain the prior state estimate and prior covariance at the current moment; Robust Mechanism Configuration Module: Configures robust mechanisms to enhance noise immunity, including: outlier detection switch, adaptive noise adjustment switch, chi-square threshold parameter, and fitness factor range; Outlier detection and suppression module: Based on the prior state obtained by the state prediction module, calculate the innovation of each sensor observation and its statistical distance, and use the decision threshold configured by the robust mechanism configuration module to identify outliers. For observations identified as outliers, amplify and adjust the corresponding observation noise covariance matrix. Adaptive noise adjustment module: Based on the observation noise covariance matrix and corresponding innovation vector output by the outlier detection and suppression module, calculate the adaptation factor for each observation, and dynamically adjust the observation noise covariance matrix within a preset range; Data fusion and state update module: Input the observation noise covariance matrix processed by the adaptive noise adjustment module into the fusion framework to complete the state estimation and covariance update at the current time. The update result is fed back to the state prediction module to realize closed-loop continuous multi-source data fusion.

9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method for enhancing the robustness of a multi-source data fusion algorithm as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for enhancing the robustness of a multi-source data fusion algorithm as described in any one of claims 1 to 7.

Images