Position precision verification method based on deep learning feature extraction and fusion

Abstract

The invention discloses a position precision verification method based on deep learning feature extraction and fusion, and relates to the technical field of precise position positioning. Firstly, multi-source position and environment data are collected by using satellite positioning, an IMU, a wireless communication base station and the like; deep learning models such as RNN and CNN are constructed for different data to extract features, and preprocessing such as normalization is carried out; and then fusing the features by adopting a weighting or attention mechanism, finally comparing the fused features with a high-precision reference, calculating an error to evaluate the precision, analyzing and correcting if the error does not reach the standard, and feeding back and optimizing each step to improve the position precision verification effect. According to the method, multi-source data and deep learning are fused, features are accurately extracted, errors are effectively reduced, the position precision is greatly improved, the method can adapt to different environments, and prediction and optimization are carried out with the help of historical data; reliable position information is provided for multiple industries, intelligent decision support is enhanced, and practicability and flexibility are high.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of precise positioning technology, and in particular to a position accuracy verification method based on deep learning feature extraction and fusion. BACKGROUND

[0002] In today's digital and intelligent era, location information plays a crucial role in many fields. Industries such as autonomous driving, logistics transportation, and intelligent security have high requirements for target position accuracy. However, existing location positioning technologies have many limitations. Satellite positioning systems can provide relatively accurate location information in open areas, but in urban environments with high-rise buildings, signals can be easily blocked, causing multipath effects and increasing positioning errors. In indoor environments, satellite signals are difficult to effectively cover. Inertial measurement units (IMU) can provide real-time motion state information of the target, but their cumulative error increases over time, making them unable to provide high-precision position information alone for a long time. Wireless communication base station positioning is affected by factors such as signal strength and base station distribution density, resulting in varying positioning accuracy.

[0003] To improve position accuracy, researchers have tried to use various technology fusion methods. Early fusion methods were mostly simple data stacking or rule-based fusion, which had poor adaptability to complex and variable environments and could not fully utilize the advantages of each data source. Moreover, these methods lacked in-depth mining of data features and could not accurately capture the internal relationships between different data sources, resulting in low-quality fused data and affecting the accuracy of position accuracy verification.

[0004] With the rapid development of deep learning technology, it has shown great potential in feature extraction and pattern recognition. However, applying deep learning to the field of position accuracy verification still faces many challenges. Different types of location data have different characteristics and dimensions, and how to build appropriate deep learning models to effectively extract features from these data is a difficult problem. In addition, in the process of multi-source data fusion, how to use deep learning to automatically learn the weights and fusion strategies of different features to adapt to different environments and application scenarios is also a problem that needs to be solved urgently. SUMMARY

[0005] The position accuracy verification method based on deep learning feature extraction and fusion proposed by the present application aims to solve the problems mentioned in the above prior art.

[0006] To achieve the above purpose, the present application adopts the following technical solution: a position accuracy verification method based on deep learning feature extraction and fusion, comprising:

[0007] Multi-source position data collection: Obtain the latitude, longitude and altitude data of the target using satellite positioning system, collect the acceleration and angular velocity information of the target through inertial measurement unit (IMU), and use the signal strength and timestamp information of wireless communication base station to assist positioning; record the environmental information at the time of data collection;

[0008] Deep learning feature extraction: For satellite positioning data, construct a recurrent neural network (RNN) model to learn the time series features and capture the change rule of position over time; for IMU data, use convolutional neural network (CNN) to extract its spatial features and analyze the motion state of the target; for wireless communication base station data, use a special feature extraction network to mine the correlation features between signal and position;

[0009] Feature fusion processing: fuse the features extracted from different data sources; use weighted fusion method to assign appropriate weights to each feature according to the reliability of each data source in different scenarios; use attention mechanism-based fusion method to automatically learn the importance of different features;

[0010] Precision verification and analysis: input the fused features into the precision verification model and compare them with the known position reference data; calculate the error between the predicted position and the reference position, and judge whether the current position meets the requirements according to the set precision standard. If not, analyze the error causes.

[0011] Further, it further comprises:

[0012] Environment adaptive adjustment step: real-time perception of the environmental information of the target, increase the weight of wireless communication base station data in feature fusion when in indoor environment, optimize the processing and analysis of IMU data under poor terrain or weather conditions.

[0013] History data learning and prediction step: collect historical position data and corresponding precision verification results, construct historical data set; use machine learning algorithm to learn historical data, mine the rules and trends of position change.

[0014] Further, in the multi-source position data collection step, to improve the reliability of data, constellation joint positioning technology is used for satellite positioning system, including fusion of GPS, Beidou and GLONASS constellation data; at the same time, the IMU is calibrated regularly, and the output parameters are adjusted according to the calibration results.

[0015] Further, in the environment adaptive adjustment step, reinforcement learning mechanism is introduced; the environmental information, current feature fusion weight and verification precision are input as state, the weight and processing strategy are adjusted as action, and the improvement of verification precision is taken as reward.

[0016] Further, in the historical data learning and prediction step, the long short-term memory network (LSTM) in deep learning and the traditional gray prediction model are combined; the time sequence characteristics in the historical position data are processed by using the LSTM, and the trend of the data is quickly analyzed by using the gray prediction model.

[0017] Further, in the deep learning feature extraction step, the transfer learning technology is adopted for different types of environment data; the model is pre-trained on general position data, and then the model is fine-tuned by using specific data according to the application scene and environment characteristics.

[0018] Further, in the feature fusion processing step, the graph neural network (GNN) is used to mine the potential relationship between different features; the features of each data source are regarded as nodes in the graph, and the correlation between the features is regarded as an edge, and the node features are updated by using the graph convolution operation.

[0019] Further, in the precision checking and analysis step, different requirements of different application scenes for position precision are considered; for the scene required by automatic driving, the precision standard and error tolerance range are set; for the general navigation application, the relative conventional standard is adopted.

[0020] Further, in the whole position precision checking process, a data feedback mechanism is established; the results of the precision checking and the conclusions of the error analysis are fed back to the data acquisition, feature extraction and fusion steps; according to the feedback information, the frequency and mode of data acquisition, the parameters of the feature extraction model or the strategy of the feature fusion are adjusted, so that a closed-loop optimization system is formed to test the performance of the position precision checking method.

[0021] Compared with the existing technology, the beneficial effects of the present application are:

[0022] In terms of precision improvement, by fusing satellite positioning, IMU, wireless communication base station and other multi-source data, and by using the powerful feature extraction capability of deep learning, the potential information of each data source can be fully mined. According to different data characteristics, corresponding deep learning models are constructed to extract more representative features, and then through scientific fusion strategies, the fused features can more comprehensively and accurately describe the target position, effectively reduce the error influence of a single data source, and greatly improve the position precision.

[0023] In terms of environmental adaptability, the environmental self-adaptive adjustment and reinforcement learning mechanism are introduced. The environmental information is perceived in real time, and the feature fusion strategy and data processing mode are automatically adjusted according to different environments. The role of the indoor enhanced wireless communication base station data is highlighted, and the processing of the IMU data in complex terrain is highlighted, so that the checking method can maintain high precision in various environments, and is free from the dependence of the traditional method on specific environments.

[0024] In intelligent decision-making support, leveraging historical data learning and prediction, along with data feedback mechanisms, not only can we uncover patterns in location changes and predict future accuracy, but we can also optimize the entire process based on verification results. This allows us to proactively identify accuracy issues and proactively adjust them, providing reliable location information to support intelligent decision-making in related industries, such as path planning in autonomous driving and vehicle scheduling in logistics and transportation, thereby improving operational efficiency and safety across various sectors.

[0025] In addition, this method also uses technologies such as transfer learning and graph neural networks to reduce model training costs and better explore feature relationships. At the same time, it dynamically adjusts the verification standards considering the needs of different scenarios, making position accuracy verification more practical and flexible. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Fig. 1 This is a schematic block diagram of the position accuracy verification method based on deep learning feature extraction and fusion proposed by the present invention;

[0027] Fig. 2 This is a schematic diagram showing the comparison of the position accuracy of the position accuracy verification method based on deep learning feature extraction and fusion proposed in the present invention under different environments. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0029] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0030] Moreover, the terms "first", "second", etc. are used herein only to describe different instances, and do not imply or suggest relative importance or a number of the technical features indicated. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connecting", "connection" should be broadly understood, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through an intermediate medium, or internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances, and the present application will be further described in detail below with reference to the drawings.

[0031] Reference Figs. 1-2 A position accuracy verification method based on deep learning feature extraction and fusion, comprising the following steps:

[0032] Multi-source position data acquisition: The longitude and latitude and altitude data of the target are obtained by using satellite positioning system (such as GPS, Beidou), the acceleration and angular velocity information of the target are collected by using inertial measurement unit (IMU), and the signal strength and time stamp information of wireless communication base station are used to assist positioning. At the same time, the environmental information such as weather and terrain during data collection is recorded. These multi-source data provide rich basic information for subsequent accuracy verification, and can reflect the target position from different angles.

[0033] Deep learning feature extraction: For satellite positioning data, a recurrent neural network (RNN) model is constructed to learn its time series features to capture the change rule of position over time. For IMU data, convolutional neural network (CNN) is used to extract its spatial features to analyze the motion state of the target. For wireless communication base station data, a special feature extraction network is used to mine the correlation features between signal and position. A large number of labeled position data are used to train these models to make them have the ability to accurately extract features.

[0034] Feature fusion processing: The features extracted from different data sources are fused. Weighted fusion is used, and each feature is given a corresponding weight according to the reliability of each data source in different scenarios. The attention mechanism based fusion method can also be used to automatically learn the importance of different features. The fused features integrate multi-source information and can more comprehensively describe the position state of the target, providing better data for accuracy verification.

[0035] Precision verification and analysis: the fused features are input into a precision verification model and compared with known high-precision position reference data. The error between the predicted position and the reference position is calculated, such as the Euclidean distance, angular deviation, etc. According to the set precision standard, it is judged whether the current position precision meets the requirements. If not, analyze the possible reasons for the error, such as data source error, feature extraction deviation or fusion strategy problem, etc.

[0036] In the present application, it also includes:

[0037] Environment adaptive adjustment step: real-time perception of the environment information of the target, such as indoor, outdoor, city or countryside, etc. When in indoor environment, increase the weight of wireless communication base station data in feature fusion, because the satellite positioning signal may be weak at this time. In complex terrain or bad weather conditions, strengthen the processing and analysis of IMU data, and use its relatively stable characteristics to improve the position precision. Through this adaptive adjustment, the verification method can maintain good precision in different environments.

[0038] History data learning and prediction step: collect historical position data and corresponding precision verification results, and construct a historical data set. Use machine learning algorithms such as time series analysis, decision tree, etc. to learn historical data, and mine the rules and trends of position changes. Based on these rules, predict future position precision, and find possible precision decline problems in advance, and adjust the verification strategy in time to improve the forward-looking and reliability of position precision verification.

[0039] In the present application, in the multi-source position data acquisition step, in order to improve the accuracy and reliability of the data, the multi-constellation joint positioning technology is used for satellite positioning system, and the data of multiple constellations such as GPS, Beidou, GLONASS, etc. is fused. At the same time, the IMU is calibrated regularly, and the output parameters are adjusted according to the calibration results. For wireless communication base station data, filter the base station data with stable signal strength and small positioning error, and filter the interference in the signal transmission process, so as to ensure that high-quality multi-source position data is collected.

[0040] In the present application, in the environment adaptive adjustment step, the reinforcement learning mechanism is introduced. The environment information, the current feature fusion weight and the verification precision are input as the state, the weight adjustment and the processing strategy are taken as the action, and the improvement of the verification precision is taken as the reward. Through continuous interaction with the environment, the feature fusion and processing method in different environments are automatically optimized, so that the position precision verification method can adapt to environmental changes faster and more accurately, and further improve the verification precision.

[0041] In the history data learning and prediction step of the present application, the long short-term memory network (LSTM) in deep learning and the traditional gray prediction model are combined. The complex time sequence characteristics in the historical position data are processed by LSTM, and the gray prediction model quickly analyzes the trend of the data. Through model fusion, the advantages of the two are integrated, and more accurate prediction of future positions is realized, providing a more reliable basis for taking measures to optimize position accuracy in advance.

[0042] In the deep learning feature extraction step of the present application, transfer learning technology is used for different types of environmental data. The model is pre-trained on large-scale general position data, and then fine-tuned using a small amount of specific data according to the specific application scenario and environmental characteristics. This can reduce the amount of data and time required for model training, while improving the accuracy and generalization ability of the model in extracting features in a specific environment.

[0043] In the feature fusion processing step of the present application, graph neural network (GNN) is used to mine the potential relationship between different features. The features of each data source are regarded as nodes in the graph, and the association between features is regarded as an edge. The node features are updated through graph convolution operation. This method can better capture the complex dependency relationship between features, making the fused features more expressive, thereby improving the effect of position accuracy verification.

[0044] In the accuracy verification and analysis step of the present application, different application scenarios have different requirements for position accuracy. For high-precision scenarios such as autonomous driving, strict accuracy standards and error tolerance ranges are set; for ordinary navigation applications, relatively loose standards are used. According to the specific application scenario, the parameters and methods of accuracy verification are dynamically adjusted to make the verification results more in line with actual needs and improve the practicality of position accuracy verification.

[0045] In the present application, a data feedback mechanism is established throughout the entire position accuracy verification process. The results of accuracy verification and the conclusions of error analysis are fed back to each step of data collection, feature extraction and fusion. According to the feedback information, the frequency and method of data collection, the parameters of the feature extraction model or the strategy of feature fusion are adjusted to form a closed-loop optimization system, continuously improving the performance and accuracy of the position accuracy verification method.

[0046] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical solution and inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A position accuracy verification method based on deep learning feature extraction and fusion, characterized in that: include: Multi-source location data acquisition: Utilizes satellite positioning systems to obtain the target's latitude, longitude, and altitude data. The inertial measurement unit (IMU) collects the target's acceleration and angular velocity information. Signal strength and timestamp information from wireless communication base stations are used to assist in positioning. Environmental information during data collection is also recorded. Deep Learning Feature Extraction: For satellite positioning data, a recurrent neural network (RNN) model is constructed to learn its time series features to capture the temporal changes in position. For IMU data, a convolutional neural network (CNN) is used to extract spatial features and analyze the target's motion state. For wireless communication base station data, a specialized feature extraction network is used to mine the correlation between signal and position. Feature fusion processing: Features extracted from different data sources are fused. A weighted fusion approach is used to assign corresponding weights to each feature based on the reliability of each data source in different scenarios. At the same time, an attention-based fusion method is used to automatically learn the importance of different features. Accuracy verification and analysis: The fused features are input into the accuracy verification model and compared with the known position reference data; the error between the predicted position and the reference position is calculated, and based on the set accuracy standard, it is determined whether the accuracy of the current position meets the requirements. If not, the cause of the error is analyzed.

2. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1 is characterized in that: Also includes: Environmental adaptive adjustment steps: Real-time perception of the target's environmental information. When in an indoor environment, increase the weight of wireless communication base station data in feature fusion. In cases of poor terrain or weather conditions, optimize the processing and analysis of IMU data.

3. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1 is characterized in that: Also includes: Historical data learning and prediction steps: Collect historical location data and corresponding accuracy verification results to build a historical data set; Use machine learning algorithms to study historical data and explore the patterns and trends of location changes.

4. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1 is characterized in that: In the multi-source position data collection step, to improve data reliability, the satellite positioning system adopts constellation joint positioning technology, including the integration of GPS, Beidou, and GLONASS constellation data; at the same time, the IMU is regularly calibrated and its output parameters are adjusted according to the calibration results.

5. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 2 is characterized in that: In the environmental adaptive adjustment step, a reinforcement learning mechanism is introduced; environmental information, current feature fusion weights and verification accuracy are used as state inputs, weight adjustment and processing strategies are used as actions, and the improvement of verification accuracy is used as a reward.

6. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 3 is characterized in that: In the historical data learning and prediction steps, the long short-term memory network (LSTM) in deep learning and the traditional grey prediction model are combined; LSTM is used to process the time series features in the historical location data, and the grey prediction model analyzes the trend of the data.

7. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1 is characterized in that: In the deep learning feature extraction step, transfer learning technology is used for different types of environmental data; the model is first pre-trained on general location data, and then the data is used to fine-tune the model according to the application scenario and environmental characteristics.

8. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1 is characterized in that: In the feature fusion processing step, graph neural network (GNN) is used to explore the relationship between different features; the features of each data source are regarded as nodes in the graph, and the associations between features are regarded as edges, and the node features are updated through graph convolution operations.

9. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1 is characterized in that: In the accuracy verification and analysis steps, the different requirements for position accuracy in different application scenarios are taken into consideration; for scenarios requiring autonomous driving, accuracy standards and error tolerance ranges are set; for ordinary navigation applications, conventional standards are adopted.

10. The position accuracy verification method based on deep learning feature extraction and fusion according to claim 1, characterized in that: During the entire position accuracy verification process, a data feedback mechanism is established; the results of the accuracy verification and the conclusions of the error analysis are fed back to the various steps of data acquisition, feature extraction and fusion; based on the feedback information, the frequency and method of data acquisition are adjusted, the parameters of the feature extraction model are optimized, or the feature fusion strategy is improved, forming a closed-loop optimization system to test the performance of the position accuracy verification method.

Images