Target fusion positioning method based on multi-source perception information and decision tree model

Abstract

This invention discloses a target fusion localization method based on multi-source sensing information and a decision tree model, belonging to the technical fields of low-altitude target surveillance and high-precision positioning. It acquires target observation data and ground truth data from three sensors, and performs time alignment on the multi-source data to form samples under the same time reference. Based on the aligned samples, longitude and latitude regression decision tree models are trained separately, using observed latitude and longitude as input and actual latitude and longitude as output. The trained models are then used to correct the observed latitude and longitude of each sensor. Fusion weights are determined based on the variance between the observed data and the ground truth data of each sensor, and these weights are used to perform weighted fusion of the corrected sensor positioning data to output the final positioning result. This invention effectively reduces systematic positioning errors by learning nonlinear error mapping through a decision tree model, and achieves adaptive weighted fusion of multi-source data by combining Bayesian theory, significantly improving the positioning accuracy of low-altitude, slow-moving, and small targets.

Description

Technical Field

[0001] This invention relates to the fields of low-altitude target surveillance and high-precision positioning technology, and in particular to a target fusion positioning method based on multi-source sensing information and decision tree model. Background Technology

[0002] The rapid development of the low-altitude economy has placed increasingly higher demands on the precise positioning of low-altitude, slow-moving, and small aerial targets. Low-altitude, slow-moving, and small targets typically refer to aerial targets with low flight altitude, slow speed, and small radar cross-sections, including UAVs and light aircraft. Traditional radar surveillance systems are limited by factors such as weak target scattering characteristics and significant environmental clutter interference, making it difficult to achieve precise positioning and optimized tracking of low-altitude, slow-moving, and small aerial targets.

[0003] Currently, multi-sensor fusion positioning has become the mainstream technology for improving target positioning accuracy. By fusing sensing information from multiple sensors such as radar, photoelectric, and radio sensors, the shortcomings of a single sensor can be compensated for to some extent. However, existing multi-source positioning systems still face several key technical challenges in practical applications: First, the observation data from different sensors contain systematic errors, including measurement deviations of the sensors themselves and installation and calibration errors. These errors are nonlinear and spatially heterogeneous, and are difficult to eliminate through simple linear correction.

[0004] Second, multi-source sensing data suffers from temporal asynchrony. Different sensors have different sampling frequencies and data transmission delays, resulting in inconsistent observation times of the same target across different sensors. Directly fusing these data will introduce additional time registration errors.

[0005] Third, the measurement accuracy of different sensors varies significantly, and the reliability of each sensor changes dynamically in complex environments. How to reasonably allocate fusion weights to each sensor is a key factor affecting the final positioning accuracy.

[0006] To address the aforementioned issues, some studies have attempted to model and compensate for positioning errors using machine learning methods. For example, neural networks or decision trees are used to fuse multiple data sources for navigation decisions; decision tree classifiers are used to identify driving environment states to dynamically adjust fusion strategies. However, these methods are mostly geared towards specific application scenarios such as vehicle navigation and have failed to systematically solve the joint optimization problems of time asynchrony, nonlinear error modeling, and adaptive weighted fusion in multi-source positioning. Summary of the Invention

[0007] The purpose of this invention is to propose a target fusion localization method based on multi-source sensing information and decision tree model. This method addresses the technical problems of systematic errors, time asynchrony, and large differences in the accuracy of various sensors in existing multi-source localization systems, and achieves high-precision and stable localization of low, slow, and small targets.

[0008] To achieve the above objectives, this invention provides a target fusion localization method based on multi-source sensing information and a decision tree model, comprising the following steps: Step S1: Acquire target observation data and target ground truth data from three sensors. The observation data includes the observation longitude, observation latitude and timestamp information of the target at each sampling time point. The ground truth data includes the actual longitude and actual latitude of the target at each sampling time point and its timestamp information. Time-align the observation data and ground truth data from the three sensors to form multi-source data samples under the same time reference. Step S2: Based on the time-aligned multi-source data samples, using the observed longitude and observed latitude of each sensor as input features, and the corresponding true longitude and true latitude as output targets, construct longitude regression decision tree models and latitude regression decision tree models respectively, and train them to obtain a positioning correction model for point-by-point correction of the positioning data observed by each sensor. Step S3: Use the positioning correction model to correct the observed longitude and observed latitude of the three sensors respectively, and output the corrected longitude and latitude data of each sensor; Step S4: Determine the fusion weight of each sensor based on the variance between the observed data and the true data of each sensor. Use the fusion weight to perform weighted fusion of the positioning data of each sensor after the correction in step S3, and output the target fusion positioning result.

[0009] Preferably, in step S1, time alignment specifically includes: S11: Extract the second-level time series T1, T2, T3 corresponding to the observation data from the three sensors, and the second-level time series T4 corresponding to the true value data; S12: Perform a stepwise intersection operation on the four time series. First, perform a time intersection operation on the first time series T1 and the second time series T2 to obtain the first intersection result. Then, find the intersection of the first intersection result with the third time series T3 to obtain the second intersection result. Finally, find the intersection of the second intersection result with the fourth time series T4 to obtain the set of common time points that exist in all four time series. S13: Find the index position corresponding to the common time point in each original time series, complete the time matching, and form multi-source data samples corresponding to the same common time point.

[0010] Preferably, in step S2, the training methods for the longitude regression decision tree model and the latitude regression decision tree model specifically include: S21: Longitude observations from each sensor after time alignment and observation latitude As a two-dimensional input feature vector ,in Indicates a time index; S22: Actual longitude aligned with time As the output target of the longitude regression decision tree model, the true latitude As the output target of the latitudinal regression decision tree model; S23: Construct mapping relationships respectively and ,in This is the longitude error mapping function. This is a latitude error mapping function; S24: During training, the input feature space is recursively partitioned with the goal of minimizing the mean square error of the output within the leaf node region. S25: After training is complete, store the parameters of the longitude regression decision tree model and the latitude regression decision tree model.

[0011] Preferably, during the training process of a regression decision tree model, the objective of each partition is to minimize the mean squared error of the output within the region: ; in, Indicates the first Leaf node regions, Indicates the first The mean of the sample outputs within each leaf node region. Indicates the first The output value of each sample; When the sample falls into the first Leaf node region At that time, the predicted output of the sample is given by the mean of the outputs of the training samples within the leaf node region: ; in, Indicates the first The number of samples within each leaf node region. Indicates falling into the first The first leaf node region 1 input feature vector.

[0012] Preferably, in step S3, the specific method for correcting the observation and positioning data of each sensor is as follows: Extract the observed longitudes from sensor 1, sensor 2, and sensor 3 respectively. , , and observation latitude , , ; For sensor 1, construct the input feature vector. Inputting these values ​​into the trained longitude regression decision tree model and latitude regression decision tree model yields the corrected longitude values. and corrected latitude ; For sensor 2, construct the input feature vector. Input the longitude regression decision tree model and the latitude regression decision tree model respectively to obtain the corrected longitude. and corrected latitude ; For sensor 3, construct the input feature vector. Input the longitude regression decision tree model and the latitude regression decision tree model respectively to obtain the corrected longitude. and corrected latitude .

[0013] Preferably, in step S4, the fusion weights are determined as follows: Calculate the longitude and latitude variances between the observed data and the true data for each sensor: ; ; in, For sensor serial number, These correspond to sensor 1, sensor 2, and sensor 3, respectively. For time indexing, This represents the total number of samples after time alignment. For the first The sensor at the first The observed longitude at each time point For the first The sensor at the first The observation latitude at each time point; For the first The true longitude at each point in time. For the first The true value dimension at each time point; The longitude fusion weights of each sensor are determined based on the longitude variance: ; in, For the first The longitude fusion weights of each sensor satisfy the normalization condition. and ; The latitudinal fusion weights of each sensor are determined based on the latitudinal variance: ; in, For the first The dimensional fusion weights of each sensor satisfy the normalization condition. and .

[0014] Preferably, in step S4, the weighted fusion outputs the target fusion localization result according to the following formula: ; ; in, To integrate and locate the longitude of the target, To integrate positioning latitude and longitude for the target, For the first The longitude values ​​of the sensors after correction in step S3 For the first The latitude value of each sensor after step S3 correction.

[0015] Preferably, the three sensors are radar sensors, photoelectric sensors, or radio sensors.

[0016] Preferably, in step S1, the time information in various types of data is standardized, and the time fields in each data source are uniformly converted into the same time format; a unified time axis is constructed as a common time reference, which adopts the time series of sensor 1, sensor 2, sensor 3 and true value observation data.

[0017] Preferably, step S1 further includes: after completing time matching, counting the number of successfully matched time points to evaluate the effectiveness of multi-source time alignment; performing an integrity check on the index results, and outputting a prompt message when no valid common time point is detected, prompting to check whether the time format or time precision of different data sources are consistent.

[0018] Therefore, the target fusion localization method based on multi-source sensing information and decision tree model described above has the following advantages: (1) This invention learns the nonlinear error mapping relationship between observed positioning data and real location through a regression decision tree model, which can effectively capture the nonlinear system error and spatial heterogeneous error in multi-source positioning system, and has higher error compensation accuracy than traditional linear correction methods. (2) This invention achieves precise time alignment of multi-source sensing data through stepwise time intersection operation, which effectively solves the problem of time asynchrony caused by different sampling frequencies of different sensors and inconsistent data transmission delays, and provides a reliable data foundation for subsequent error modeling and fusion positioning; (3) Based on Bayesian theory, this invention dynamically allocates fusion weights according to the variance between the measured data and the true value of each sensor. The smaller the variance, the higher the accuracy and reliability of the sensor data, and the higher the fusion weight is assigned, thus realizing the adaptive optimal weighted fusion of multi-source positioning results. (4) The present invention splits the two-dimensional positioning output into two independent one-dimensional regression problems of longitude and latitude, and trains independent regression decision tree models respectively, thereby reducing the model complexity and improving the prediction stability and accuracy; (5) This invention is particularly suitable for the precise positioning of low, slow and small targets. It can accurately assess the latitude and longitude position of low, slow and small targets based on multiple types of monitoring sensor information, providing strong technical support for target monitoring and management in my country's low-altitude economic development.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the target fusion localization method based on multi-source sensing information and decision tree model in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0023] Example like Figure 1 As shown in the figure, this embodiment proposes a target fusion localization method based on multi-source sensing information and decision tree model. The specific steps are as follows: In this embodiment, the detection data of low, slow, and small targets (such as drones) using three sensors (sensor 1, sensor 2, and sensor 3) is used as an example. The three sensors can be a radar sensor, an electro-optical sensor, and a radio sensor, respectively providing the target's observation longitude, observation latitude, and timestamp information. Meanwhile, RTK (Real-Time Kinematic) data acquired through a high-precision GNSS receiver is used as the ground truth.

[0024] Step S1: Time alignment based on three sensors and truth data information: Precise target localization utilizes observation data provided by three different sensors, with each of the three data sources providing three columns of time information. The target's latitude and longitude ground truth data is used to verify the fusion information error. In target fusion localization, time alignment is performed first to address the inconsistency in the time dimension of multi-source sensing data from different sensor systems.

[0025] Specifically, the following three types of data sources are acquired: sensor data from Sensor 1, Sensor 2, and Sensor 3, including the observation longitude, observation latitude, and timestamp information of the target at different time points. The time information in the three types of sensor data is standardized by converting the time fields from each data source into the same time format (datatime). A unified timeline is then constructed as the system's common time reference. This common time reference is preferably the time series of observation data from Sensor 1, to ensure that subsequent models can directly infer from actual observation data.

[0026] The system performs second-level time matching on time series from multiple sensing systems using a time function. Specifically, it filters out common time points that exist simultaneously in multiple data sources by finding the intersection of multiple time sets, and then returns the index positions of these common time points in each original time series.

[0027] The time-aligned input includes four time series parameters: T1: the second-level time series corresponding to the data sensed by sensor 1; T2: the second-level time series corresponding to the data sensed by sensor 2; T3: the second-level time series corresponding to the data sensed by sensor 3; and T4: the second-level time series corresponding to the ground truth data. Each time series is represented in the form of a second-level timestamp, with consistent time units and formats.

[0028] First, a time intersection operation is performed on the first time series T1 and the second time series T2 to obtain a set of time points that are completely consistent in the time dimension. After obtaining the intersection of the first two time series, the intersection result is intersected with the third time series T3, and then the result is intersected with the fourth time series T4. Through the stepwise intersection method, a set of common time points that exist in all four time series is finally obtained. This set of common time points represents the time points where there is valid data in all sensing systems.

[0029] After obtaining the set of common time points, the index positions corresponding to these common time points are further searched in each original time series. Specifically, this includes obtaining the indexes of the common time points in the three types of sensor and true value data information sequences.

[0030] After completing time matching, the number of successfully matched time points is counted to evaluate the effectiveness of multi-source time alignment. Simultaneously, an integrity check is performed on the index results. If no valid common time point is detected, a prompt message is output, reminding users to check whether the time formats or time precision of different data sources are consistent to avoid matching failures due to inconsistent time standards.

[0031] After time alignment is completed, multi-source data samples corresponding to the same time point are formed. Each sample includes at least: sensor 1 observation location information, sensor 2 observation location information and sensor 3 observation location information at the same time point.

[0032] In this embodiment, after time alignment processing, a total of 222 precise second-level matching points were obtained. These matching points constitute the basic dataset for subsequent model training and fusion localization.

[0033] Step S2: Training method for multi-source localization error learning model based on regression decision tree: Based on the completion of time alignment of multi-source sensing data, machine learning methods are used to model the mapping relationship between observation and positioning data and reference ground truth.

[0034] After time alignment is completed in step S1, it can be done at the same time point. The following data was obtained: The positioning results output by the observation system (such as sensor 1): ; High-precision reference true value: ; in Indicates longitude components. Indicates the latitude component.

[0035] This model transforms the multi-source localization error modeling problem into a supervised regression problem. To reduce model complexity and improve prediction stability, this model decomposes the two-dimensional output into two one-dimensional regression problems: ; Regression decision trees are a type of nonparametric regression model based on data partitioning. The core idea is to recursively partition the input feature space so that the output variable in each partition region has the smallest possible variance.

[0036] For the input feature vector: ; During training, regression decision trees divide the sample space into several sub-regions by selecting feature dimensions and splitting thresholds. The goal of each partition is to minimize the mean square error of the output within the region. ; in: Indicates the first Leaf node region; This represents the mean of the sample outputs within this region.

[0037] When the sample falls into a certain node region At that time, its predicted output is given by the mean of the training samples in that region: ; For the longitude and latitude components, construct independent regression decision tree models respectively: Longitude regression model: ; Latitude regression model: ; The model training process involves using time-aligned observation and positioning data as input features and the ground truth values ​​at the corresponding time points as the guiding targets. Based on the principle of minimizing regression error, a regression decision tree model is trained to obtain positioning correction models in both longitude and latitude directions. After training, the model parameters are stored.

[0038] In this embodiment, the specific training parameters of the regression decision tree model are set as follows: the maximum depth of the decision tree is set to 10 layers, the minimum number of samples per leaf node is set to 5, and the splitting stopping condition is that the number of samples in a node is less than the minimum number of samples or the mean square error in the node is less than a preset threshold of 1×10. -12 The parameters mentioned above were optimized using grid search and five-fold cross-validation, with the root mean square error (RMSE) on the validation set used as the model performance evaluation metric.

[0039] The training and validation sets were divided as follows: 222 time-aligned matching points were used as both the training and validation sets to train the latitude and longitude regression model. Other data were then used to correct the model and observe its correction capabilities. Autocorrelation training was used during the training process.

[0040] Step S3: Correct the measured latitude and longitude values ​​based on the true data: After completing the training of the positioning error learning model based on regression decision tree in step S2, the measured latitude and longitude values ​​are corrected by loading the model trained with high-precision ground truth data.

[0041] In this embodiment, the correction steps specifically include: For sensor 1, extract its observed longitude. and observation latitude Construct the input feature vector Input it into the longitude regression decision tree model The corrected longitude is obtained. Input it into the latitudinal regression decision tree model The corrected latitude is obtained. .

[0042] Similarly, for sensor 2, its observed longitude is extracted. and observation latitude Construct the input feature vector The corrected longitude is obtained. and corrected latitude .

[0043] For sensor 3, extract its observed longitude. and observation latitude Construct the input feature vector The corrected longitude is obtained. and corrected latitude .

[0044] At this point, the observation and positioning data from the three sensors have all been corrected by the same regression decision tree model to obtain their respective corrected latitude and longitude values, which are then used for weighted fusion in step S4.

[0045] Step S4: Calculate the variance allocation weights based on Bayesian theory: The measurement data from the three types of sensors have different accuracies, and the target precise positioning weight is calculated by the variance between the measured value and the true value.

[0046] First, the sample data is time-aligned (step S1), and the time-aligned sample data is read into column vectors: ; in, , , These are the measured latitude and longitude coordinates of sensor 1, sensor 2, and sensor 3, respectively. These are true latitude and longitude data.

[0047] Calculate the variance between the sample data and the true value respectively: ; ; in, The number of time-aligned samples (in this embodiment) ), For the first The sensor at the first The observed longitude at each time point For the first The true longitude at each point in time. For the first The sensor at the first The latitude of each observation point in time, For the first The true dimension of each time point.

[0048] In this embodiment, the following variance values ​​were calculated: Longitude variance: Sensor 1: ; Sensor 2: ; Sensor 3: ; Latitude variance: Sensor 1: ; Sensor 2: ; Sensor 3: ; Assign weights to the data from sensor 1, sensor 2, and sensor 3, and the weights must meet the normalization condition: ; ; in, Assign weights to longitude. Assign weights to latitudes.

[0049] The specific weight values ​​are determined by the variance: ; ; The calculation yielded the following results: Longitude fusion weight: Sensor 1: ; Sensor 2: ; Sensor 3: ; Dimensional fusion weight: Sensor 1: ; Sensor 2: ; Sensor 3: ; As can be seen from the above weight allocation, sensor 3 has the smallest longitude variance (59.4577×10). -10 Therefore, it was assigned the highest longitude fusion weight (0.7064); sensor 3 also had the smallest latitude variance (5.385×10). -10 The sensor 3 was also assigned the highest latitude fusion weight (0.4949). This indicates that sensor 3 has the highest measurement accuracy in this scenario, and the fusion system adaptively gives it the highest level of trust.

[0050] It should be noted that the measured latitude and longitude of each sensor used to calculate the variance in step S4 are... The original observation data after time alignment in step S1 This is used to evaluate the statistical accuracy of each sensor relative to the true value; while the weighted fusion formula uses... The positioning data after correction in step S3 In other words, the statistical error variance of each sensor relative to the true value determines its weight in the fusion, and the data participating in the weighted fusion are the corrected data. This achieves the dual optimization effect of "giving high weight to high-precision sensors and correcting all data for errors".

[0051] Finally, the sensor positioning data corrected by the regression decision tree model in step S3 are weighted and fused with the above weights to obtain the positioning latitude and longitude, which is then output: ; ; in, To integrate and locate the longitude of the target, To integrate positioning latitude and longitude for the target, For the first The longitude values ​​of the sensors after correction in step S3 For the first The latitude value of each sensor after step S3 correction.

[0052] Through the above four steps, this invention realizes a complete target localization process from time alignment of multi-source sensing data, regression decision tree error modeling and correction, to Bayesian weighted fusion, effectively improving the localization accuracy of low, slow and small targets under multi-source sensing information.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A target fusion localization method based on multi-source sensing information and decision tree model, characterized in that, Includes the following steps: Step S1: Acquire target observation data and target ground truth data from three sensors. The observation data includes the observation longitude, observation latitude and timestamp information of the target at each sampling time point. The ground truth data includes the actual longitude and actual latitude of the target at each sampling time point and its timestamp information. Time-align the observation data and ground truth data from the three sensors to form multi-source data samples under the same time reference. Step S2: Based on the time-aligned multi-source data samples, using the observed longitude and observed latitude of each sensor as input features, and the corresponding true longitude and true latitude as output targets, construct longitude regression decision tree models and latitude regression decision tree models respectively, and train them to obtain a positioning correction model for point-by-point correction of the positioning data observed by each sensor. Step S3: Use the positioning correction model to correct the observed longitude and observed latitude of the three sensors respectively, and output the corrected longitude and latitude data of each sensor; Step S4: Determine the fusion weight of each sensor based on the variance between the observed data and the true data of each sensor. Use the fusion weight to perform weighted fusion of the positioning data of each sensor after the correction in step S3, and output the target fusion positioning result.

2. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: In step S1, time alignment specifically includes: S11: Extract the second-level time series T1, T2, T3 corresponding to the observation data from the three sensors, and the second-level time series T4 corresponding to the true value data; S12: Perform a stepwise intersection operation on the four time series. First, perform a time intersection operation on the first time series T1 and the second time series T2 to obtain the first intersection result. Then, find the intersection of the first intersection result with the third time series T3 to obtain the second intersection result. Finally, find the intersection of the second intersection result with the fourth time series T4 to obtain the set of common time points that exist in all four time series. S13: Find the index position corresponding to the common time point in each original time series, complete the time matching, and form multi-source data samples corresponding to the same common time point.

3. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: In step S2, the training methods for the longitude regression decision tree model and the latitude regression decision tree model specifically include: S21: Longitude observations from each sensor after time alignment and observation latitude As a two-dimensional input feature vector ,in Indicates a time index; S22: Actual longitude aligned with time As the output target of the longitude regression decision tree model, the true latitude As the output target of the latitudinal regression decision tree model; S23: Construct mapping relationships respectively and ,in This is the longitude error mapping function. This is a latitude error mapping function; S24: During training, the input feature space is recursively partitioned with the goal of minimizing the mean square error of the output within the leaf node region. S25: After training is complete, store the parameters of the longitude regression decision tree model and the latitude regression decision tree model.

4. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 3, characterized in that: During the training process of a regression decision tree model, the goal of each partition is to minimize the mean squared error of the output within that region. ； in, Indicates the first Leaf node regions, Indicates the first The mean of the sample outputs within each leaf node region. Indicates the first The output value of each sample; When the sample falls into the first Leaf node region At that time, the predicted output of the sample is given by the mean of the outputs of the training samples within the leaf node region: ； in, Indicates the first The number of samples within each leaf node region. Indicates falling into the first The first leaf node region 1 input feature vector.

5. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: In step S3, the specific method for correcting the positioning data observed by each sensor is as follows: Extract the observed longitudes from sensor 1, sensor 2, and sensor 3 respectively. , , and observation latitude , , ; For sensor 1, construct the input feature vector. Inputting these values ​​into the trained longitude regression decision tree model and latitude regression decision tree model yields the corrected longitude values. and corrected latitude ; For sensor 2, construct the input feature vector. Input the longitude regression decision tree model and the latitude regression decision tree model respectively to obtain the corrected longitude. and corrected latitude ; For sensor 3, construct the input feature vector. Input the longitude regression decision tree model and the latitude regression decision tree model respectively to obtain the corrected longitude. and corrected latitude .

6. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: In step S4, the fusion weights are determined as follows: Calculate the longitude and latitude variances between the observed data and the true data for each sensor: ； ； in, For sensor serial number, These correspond to sensor 1, sensor 2, and sensor 3, respectively. For time indexing, This represents the total number of samples after time alignment. For the first The sensor at the first The observed longitude at each time point For the first The sensor at the first The observation latitude at each time point; For the first The true longitude at each point in time. For the first The true value dimension at each time point; The longitude fusion weights of each sensor are determined based on the longitude variance: ； in, For the first The longitude fusion weights of each sensor satisfy the normalization condition. and ; The latitudinal fusion weights of each sensor are determined based on the latitudinal variance: ； in, For the first The dimensional fusion weights of each sensor satisfy the normalization condition. and .

7. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 6, characterized in that: In step S4, the weighted fusion outputs the target fusion localization result according to the following formula: ； ； in, To integrate and locate the longitude of the target, To integrate positioning latitude and longitude for the target, For the first The longitude values ​​of the sensors after correction in step S3 For the first The latitude value of each sensor after step S3 correction.

8. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: The three types of sensors are radar sensors, photoelectric sensors, or radio sensors.

9. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: In step S1, the time information in various types of data is standardized, and the time fields in each data source are uniformly converted into the same time format; a unified time axis is constructed as a common time reference, which adopts the time series of sensor 1, sensor 2, sensor 3 and true value observation data.

10. The target fusion localization method based on multi-source sensing information and decision tree model according to claim 1, characterized in that: Step S1 also includes: after completing time matching, counting the number of successfully matched time points to evaluate the effectiveness of multi-source time alignment; performing an integrity check on the index results, and outputting a prompt message when no valid common time point is detected, prompting to check whether the time format or time precision of different data sources are consistent.

Images