Low-altitude unmanned aerial vehicle track tracking and monitoring method based on 5g-a integrated base station

Abstract

The application discloses a low-altitude unmanned aerial vehicle track tracking and monitoring method based on a 5G-A integrated base station, relates to the technical field of low-altitude traffic management and communication and perception fusion, and first collects multi-source data such as integrated perception and fusion signals, environmental interference and unmanned aerial vehicle attributes, and encapsulates the data into synchronous data frames through unified timestamp and coordinate system alignment. Then, the data frames are subjected to deep fusion and anti-interference processing, noise is filtered out, and pure fusion data is generated. Further, the pure data is subjected to real-time track solving and motion trend prediction by using multi-base-station cooperative solving and prediction. Based on this, independent individual tracks are accurately stripped from complex mixed data streams by using a multi-target feature recognition and clustering separation mechanism, and the tracks are subjected to compliance verification and abnormality determination in combination with a space domain rule library. In this way, the problems of signal interference and multi-target aliasing in a complex environment can be effectively overcome, so that high-precision global tracking of low-altitude unmanned aerial vehicles and real-time monitoring of abnormal behaviors are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of low-altitude traffic management and communication and perception fusion, and more particularly, to a low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on a 5G-A communication and perception integrated base station. BACKGROUND

[0002] With the vigorous development of low-altitude economy, unmanned aerial vehicles are increasingly widely used in scenarios such as urban logistics, environmental monitoring, and emergency rescue, and low-altitude airspace has become an important space for economic activities. Unmanned aerial vehicle flight trajectory data, which covers key information such as position coordinates and flight speed, is a core element of airspace management, and has irreplaceable value in ensuring low-altitude safety, preventing illegal intrusion, and optimizing airspace resource allocation. At present, unmanned aerial vehicle monitoring technology based on cellular networks has become a research hotspot, and existing technical solutions mostly use a multi-base station cooperation approach, mainly determining the unmanned aerial vehicle coordinates by measuring the 4G base station signal strength and combining the multi-base station cooperation data.

[0003] Specifically, the existing technology mainly relies on the signal characteristics of the communication link for calculation, although it utilizes existing base station resources, but there are significant technical defects when facing complex urban low-altitude environments. First, such a solution is essentially still a communication positioning method, lacking active perception capabilities, and when the unmanned aerial vehicle is in a silent state or the communication signal is shielded, the system will not be able to effectively locate it through signal strength, resulting in monitoring failure. Secondly, in terms of anti-interference and data fusion, the communication function and the perception function are often physically and logically separated in the existing technology, with the base station only responsible for data transmission, and unable to accurately align and deeply fuse radar perception data and communication data on a microsecond timestamp, forming a data island of "communication can be connected but does not know the position, perception can be measured but difficult to connect the identity". In addition, the existing target identification mechanism usually uses a feature weight similarity algorithm, ignoring the dynamic influence of complex environments (such as rain or strong electromagnetic interference) on feature stability, resulting in insufficient identification accuracy and robustness, making it difficult to meet the requirements of low-altitude supervision. Finally, in a multi-target scenario, the existing technology lacks a fine clustering and separation mechanism for individual signal characteristics, and when multiple unmanned aerial vehicles are flying at the same time, signal aliasing and trajectory misjudgment are likely to occur, making it difficult to meet the monitoring needs of millimeter-level high-precision tracking and real-time identification of abnormal behavior.

[0004] Therefore, there is an urgent need for an optimized low-altitude unmanned aerial vehicle trajectory tracking and monitoring method. SUMMARY

[0005] To solve the above technical problems, the present application is proposed.

[0006] According to an aspect of the present application, a low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on a 5G-A communication and perception integrated base station is provided, which includes:

[0007] The data alignment, coordinate unification and packaging are performed on the common sense fusion signal, environmental interference data set and unmanned aerial vehicle basic attribute data to obtain a synchronized original data frame;

[0008] The common sense data fusion and anti-interference processing are performed on the synchronized original data frame to obtain pure fusion data;

[0009] The trajectory calculation and prediction are performed on the pure fusion data provided by multiple base stations to obtain real-time trajectories and predicted trajectories;

[0010] Based on the pure fusion data stream, multi-target identification and trajectory separation are performed on the real-time trajectories to obtain independent individual trajectories;

[0011] Based on the predicted trajectories and the airspace rule library, abnormal trajectory identification is performed on the independent individual trajectories to obtain abnormal events.

[0012] Compared with the prior art, the low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on the 5G-A common sense integrated base station provided by the application first collects multi-source data such as common sense fusion signals, environmental interference and unmanned aerial vehicle attributes, and encapsulates them into synchronized data frames through unified timestamp and coordinate system alignment. Subsequently, the data frames are subjected to deep fusion and anti-interference processing to filter out noise and generate pure fusion data. Further, the pure data is subjected to real-time trajectory calculation and motion trend prediction by using multi-base station collaborative calculation and prediction. Based on this, through multi-target feature identification and clustering separation mechanism, independent individual trajectories are accurately stripped from complex mixed data stream, and the trajectories are checked for compliance and determined for abnormality in combination with the airspace rule library. In this way, the signal interference and multi-target aliasing problems in complex environments can be effectively overcome, so that high-precision global tracking and real-time monitoring of abnormal behavior of low-altitude unmanned aerial vehicles are realized. BRIEF DESCRIPTION OF DRAWINGS

[0013] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of embodiments of the present application and constitute a part of the specification, together with the description, to explain the present application and, together with the description, to explain the present application. The drawings do not constitute a limitation on the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 The flowchart of the low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on the 5G-A common sense integrated base station according to the embodiments of the present application.

[0015] Figure 2 The data flowchart of the low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on the 5G-A common sense integrated base station according to the embodiments of the present application.

[0016] Figure 3Flow chart of sub-step S2 of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station according to the embodiment of the present application.

[0017] Figure 4 Flow chart of sub-step S3 of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station according to the embodiment of the present application.

[0018] Figure 5 Flow chart of sub-step S4 of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station according to the embodiment of the present application.

[0019] Figure 6 Flow chart of sub-step S42 of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station according to the embodiment of the present application. DETAILED DESCRIPTION

[0020] Embodiments of the present disclosure will be described in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be interpreted as being limited to the embodiments set forth herein, but rather these embodiments are provided so as to more thoroughly and completely understand the present disclosure. It is understood that the drawings and embodiments of the present disclosure are for exemplary purposes only and are not intended to limit the scope of protection of the present disclosure.

[0021] In view of the problems in the above background art, the present application proposes a low-altitude unmanned aerial vehicle track tracking and monitoring method based on a 5G-A sensing integrated base station. Figure 1 Flow chart of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station according to the embodiment of the present application. Figure 2 Data flow chart of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station according to the embodiment of the present application. As shown in Figure 1 and Figure 2 The low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station includes the following steps: S1, data alignment, coordinate unification and packaging are performed on the sensing fusion signal, the environmental interference data set and the unmanned aerial vehicle basic attribute data to obtain a synchronized original data frame; S2, sensing data fusion and anti-interference processing are performed on the synchronized original data frame to obtain pure fusion data; S3, track calculation and prediction are performed on the pure fusion data provided by multiple base stations to obtain real-time track and predicted track; S4, based on the pure fusion data flow, multi-target recognition and track separation are performed on the real-time track to obtain independent individual track; S5, based on the predicted track and the airspace rule library, abnormal track recognition is performed on the independent individual track to obtain abnormal events.

[0022] In the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A integrated base station of the above-mentioned integrated sensing, the step S1 is to align, unify the coordinates and package the integrated sensing signal, the environmental interference data set and the unmanned aerial vehicle basic attribute data to obtain the synchronized original data frame. It should be understood that, since the integrated sensing signal, the environmental interference data and the unmanned aerial vehicle basic attribute data come from different sources, there are problems of inconsistent time stamps and different spatial coordinate systems, which lead to ineffective association in subsequent data processing. Therefore, the application aligns the time dimension, unifies the spatial coordinates and structures the package of the integrated sensing signal, the environmental interference data set and the unmanned aerial vehicle basic attribute data, so as to build a standardized original data basis. In this way, the heterogeneity of multi-source data can be eliminated, and subsequent fusion processing, track calculation and other links can be carried out based on consistent data benchmarks, providing a prerequisite guarantee for the accuracy of the entire monitoring process.

[0023] In particular, in one possible embodiment, the implementation process of step S1 is as follows: first, multi-source data acquisition is carried out, and the 5G-A integrated base station deployed at the key nodes of the city starts the function modules: the millimeter wave radar module transmits a detection signal and receives the echo signal reflected by the unmanned aerial vehicle, capturing core perception information such as distance, angle and speed; the Massive MIMO communication module captures the 5G-A communication signal of the unmanned aerial vehicle uplink in real time, demodulates and extracts data such as device identification and flight state; the environmental perception sensor synchronously collects environmental interference data such as electromagnetic interference intensity, temperature and humidity, visibility and terrain shielding parameters. At the same time, through a standardized interface, the registered information associated with the device identification, the model parameters, the maximum flight speed, the endurance time and other basic attribute data are retrieved from the database of the unmanned aerial vehicle management system. Then, the high-precision GPS synchronous clock source built-in the base station edge computing node adds a unified format time stamp to the collected integrated sensing signal and environmental interference data, respectively, to ensure that the time benchmarks of all data streams are consistent, with an accuracy of within 1 ms. Subsequently, coordinate unification is carried out, and all spatial position related data such as the position of the unmanned aerial vehicle and the position of the base station involved in the integrated sensing signal are converted to the WGS-84 geodetic coordinate system, completely eliminating the coordinate system differences of different acquisition modules. Finally, data packaging is completed, and the integrated sensing signal, the environmental interference data and the associated unmanned aerial vehicle basic attribute data after time alignment and coordinate unification are integrated and packaged according to the preset structured data specification, forming an original data frame containing data source identification, time stamp, spatial coordinates and various attribute information, ensuring the integrity, consistency and traceability of the data, and providing standardized and high-quality input data for subsequent integrated sensing data fusion, anti-interference processing and other links.

[0024] In the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A integrated sensing base station, the step S2 is to perform sensing data fusion and anti-interference processing on the synchronized original data frame to obtain pure fusion data. It should be understood that, since the synchronized original data frame contains redundant information such as environmental noise and electromagnetic interference, and the sensing signal and the communication signal are not effectively associated, the data quality is difficult to support high-precision track monitoring. Therefore, the application further carries out deep correlation of sensing data and targeted filtering of environmental interference on the original data frame, so as to obtain fusion data with integrity and reliability. In this way, the fragmented state of sensing and communication data can be broken, the influence of complex environment on data can be weakened, and high-quality data support can be provided for subsequent accurate calculation and prediction of track.

[0025] In particular, in one specific embodiment, Figure 3 The flowchart of the sub-step S2 of the low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A integrated sensing base station according to the embodiment of the application is shown in FIG. 2. Figure 3 As shown in FIG. 2, the step S2 includes: S21, cross-domain data correlation matching is performed on the synchronized original data frame to obtain an associated data object; and S22, fusion filtering and state estimation are performed on the associated data object to obtain pure fusion data.

[0026] Specifically, the step S21 is to perform cross-domain data correlation matching on the synchronized original data frame to obtain an associated data object. It should be understood that, since the synchronized original data frame contains core perception information such as distance, angle and speed collected by the millimeter wave radar module, and data such as device identification and flight state captured and demodulated by the Massive MIMO communication module are in a cross-domain separated state, there is a lack of clear correspondence, and multi-dimensional data cannot be accurately attributed to a specific unmanned aerial vehicle, which affects the pertinence of subsequent processing. Therefore, the application further carries out accurate correlation and matching of cross-domain data based on the space-time characteristics in the data frame, so as to establish a unique mapping relationship between signal characteristics and unmanned aerial vehicle individuals. In this way, the attribution subject of each group of data can be determined, the data confusion problem in a multi-target scene can be eliminated, accurate data correlation basis can be provided for subsequent anti-interference processing, state estimation and multi-target separation, and the pertinence and effectiveness of the whole monitoring process can be ensured.

[0027] In particular, in one possible embodiment, the step S21 is implemented as follows: first, the spatial position, velocity information and corresponding time stamp collected and calculated by the millimeter wave radar module, and the device identification, flight state and other key data demodulated by the Massive MIMO communication module are extracted from the original data frame. Then, taking the high-precision time stamp given by the base station edge computing node as the reference, the position-velocity information and the device identification-flight state information within the same time window are screened out to ensure the consistency of the data in the time dimension. Finally, based on the correlation of the spatial position, the position-velocity information and the device identification-flight state information belonging to the same unmanned aerial vehicle are bound to form a structured associated data object, and the data that is not successfully matched is marked as a to-be-identified target and its characteristic information is completely retained to ensure the comprehensiveness and accuracy of the association and matching.

[0028] Specifically, the step S22 is to perform fusion filtering and state estimation on the associated data object to obtain pure fusion data. It is worth mentioning that the pure fusion data includes identity, optimal state estimation and association attribute, the optimal state estimation includes estimated position, estimated velocity and covariance matrix, and the association attribute includes model information, registration information and data source identification. It can be understood that, since the associated data object has established the correspondence between the identity and the data, but still contains abnormal measurement values caused by meteorological interference, electromagnetic noise and the like, and lacks a comprehensive and accurate description of the motion state of the unmanned aerial vehicle, it cannot directly support high-precision trajectory calculation. Therefore, the application further performs fusion filtering on the associated data object to eliminate noise, carries out state estimation to obtain complete motion parameters, and integrates the identity and the association attribute, so as to form high-quality pure fusion data. In this way, data with identity uniqueness, state accuracy and attribute integrity can be obtained, which effectively improves the data reliability, provides a solid data foundation for subsequent multi-base station cooperative positioning, multi-target trajectory separation and abnormal behavior recognition, and guarantees the high-precision operation of the whole monitoring system.

[0029] In particular, in one possible embodiment, the step S22 is implemented as follows: first, for complex environmental interference, a double processing strategy is adopted for data preprocessing: narrow beams are generated by focusing the direction of the UAV through Massive MIMO beamforming technology to suppress sidelobe interference signals. Second, taking the identity in the associated data object as the core index, the corresponding position, speed and other original measurement data are extracted, and a Kalman filter-particle filter fusion algorithm is used to process these data layer by layer to filter abnormal data caused by weather noise and terrain shielding and correct the positioning deviation. Specifically, this algorithm mainly aims at complex environmental interference, extracts the corresponding position, speed and other original measurement data from the identity in the associated data object, and processes these data layer by layer. Its core function is to filter abnormal data caused by weather noise (such as rain and fog attenuation) and terrain shielding (such as high-rise reflection) to correct the positioning deviation. Based on the filtered effective data, the estimated position and velocity of the UAV are calculated using a state estimation model, and the covariance matrix is calculated to quantify the reliability of the state estimation. Finally, the identity, the optimal state estimation result, and the model information, registration information and data source identifier extracted from the associated data object are structured and integrated to form complete and pure fusion data, ensuring that the data dimensions are correlated and accurate and effective.

[0030] In the above low-altitude UAV trajectory tracking and monitoring method based on 5G-A integrated base station, the step S3 is to calculate and predict the trajectory based on the pure fusion data provided by multiple base stations to obtain real-time trajectory and predicted trajectory. It should be understood that due to the coverage limitations of pure fusion data from a single base station, and the fact that it can only reflect the motion state of the UAV from a certain perspective, it cannot meet the needs of global high-precision tracking and risk avoidance in advance. At the same time, the high-speed movement characteristics of the UAV require high real-time performance for trajectory updates. Therefore, the present application further integrates the pure fusion data from multiple base stations to perform real-time calculation and short-term motion trend prediction of the trajectory, thereby constructing complete and forward-looking trajectory information. In this way, the limitations of single base station monitoring can be overcome, and the precise description of the UAV motion trajectory and the early prediction of the future path can be achieved, providing sufficient time window for subsequent anomaly identification and emergency disposal, and ensuring the timeliness and effectiveness of the regulatory response.

[0031] In particular, in one specific embodiment, Figure 4 The flowchart of the sub-step S3 of the low-altitude UAV trajectory tracking and monitoring method based on the 5G-A integrated base station according to the embodiment of the present application. As Figure 4As shown, the step S3 comprises: S31, setting the pure fusion data from the base station as local pure fusion data and setting the pure fusion data from at least two adjacent cooperative base stations as cooperative multi-source positioning data; S32, performing multi-station cooperative positioning calculation on the local pure fusion data and the cooperative multi-source positioning data to obtain a current position point; S33, performing real-time trajectory state updating on the current position point based on a historical trajectory point sequence to obtain a real-time trajectory; and S34, performing time sequence trajectory prediction on the real-time trajectory to obtain a predicted trajectory.

[0032] Specifically, the step S31 sets the pure fusion data from the base station as local pure fusion data and sets the pure fusion data from at least two adjacent cooperative base stations as cooperative multi-source positioning data. It can be understood that, since the pure fusion data provided by multiple base stations is of different sources, if not explicitly distinguished, it will cause confusion in the subsequent cooperative calculation process, cannot fully exert the complementary advantages of data of different base stations, and affects the positioning accuracy. Therefore, the pure fusion data provided by multiple base stations is further classified and identified in the application, the local data and the cooperative data of at least two adjacent base stations are explicitly distinguished, so as to establish a clear data cooperative basis. In this way, the data of different sources can be accurately called in the subsequent positioning calculation process, the real-time of local data and the spatial complementarity of data of adjacent base stations are fully utilized, orderly and controllable data support is provided for multi-station cooperative positioning, and the accuracy and reliability of the positioning result are ensured.

[0033] Particularly, in a possible embodiment, the implementation process of the step S31 is as follows: first, the pure fusion data collected and processed by the base station is marked as local pure fusion data through the base station identification information of the data, so as to ensure the quick calling of the local data. Then, based on a preset base station cooperative networking scheme, at least two adjacent base stations with overlapping coverage areas and stable communication links with the base station are screened out. Subsequently, the pure fusion data sent by the adjacent base stations is received through a standardized data transmission protocol, and is uniformly marked as cooperative multi-source positioning data. Finally, the two types of data are respectively stored in different data cache areas, and are attached with collection time stamps and base station position information, so as to facilitate the quick association and calling in the subsequent cooperative calculation.

[0034] Specifically, the step S32 is to perform multi-station cooperative positioning calculation on the local pure fusion data and the cooperative multi-source positioning data to obtain the current position point. It should be understood that, due to the influence of factors such as measurement angle and environmental interference, the positioning error of the pure fusion data of a single base station is relatively large, and it is difficult to meet the regulatory requirements of the unmanned aerial vehicle meter-level positioning. However, the multi-base station data has spatial distribution complementarity, and can correct the measurement deviation. Therefore, the application further integrates the local and cooperative multi-source positioning data, and performs joint calculation through a multi-station cooperative positioning model, so as to improve the positioning accuracy of the current position of the unmanned aerial vehicle. In this way, the spatial cooperation advantage of the multi-base station can be fully utilized, the measurement error of the single base station can be effectively offset, and the positioning accuracy can be improved to the meter level. The accurate position benchmark is provided for subsequent real-time trajectory generation and abnormality judgment, and the core data quality of the entire monitoring process is ensured.

[0035] In particular, in one possible embodiment, the implementation process of the step S32 is as follows: first, the angle of arrival parameter and the signal strength measured by the local base station are extracted from the local pure fusion data, and the time difference of arrival parameter and the respective signal strength of the adjacent cooperative base station relative to the local base station are extracted from the cooperative multi-source positioning data. Specifically, a local Cartesian coordinate system needs to be constructed, and the coordinates of the local base station are set as the origin coordinates zero meters, zero meters, and thirty meters, and the coordinates of the two cooperative base stations are set as five hundred meters, zero meters, thirty meters, and two hundred fifty meters, four hundred thirty-three meters, thirty meters, respectively. The system reads the unmanned aerial vehicle azimuth angle measured by the local base station as forty-five degrees and the elevation angle as sixty degrees, and reads the time difference of arrival between the first cooperative base station and the local base station as negative zero point five microseconds and the time difference of arrival between the second cooperative base station and the local base station as zero point two microseconds. Then, the above time difference is converted into a distance difference constraint, and it is calculated that the distance difference between the unmanned aerial vehicle and the first cooperative base station and the local base station is negative one hundred fifty meters, and the distance difference between the unmanned aerial vehicle and the second cooperative base station and the local base station is sixty meters. In order to eliminate the error of a single data source and solve the three-dimensional coordinates, the system constructs an over-determined observation equation set containing the angle of arrival ray constraint and the hyperbolic surface constraint of the time difference of arrival, and uses the weighted least squares method for calculation. In this process, the signal strength index is used for weighting, for example, the weight coefficient corresponding to the negative seventy decibel milliwatt signal strength received by the local base station is set to zero point nine, and the weight coefficient corresponding to the negative seventy-five decibel milliwatt signal strength received by the cooperative base station is set to zero point eight, so as to reduce the influence of low-quality signals on the positioning result. Finally, the system uses the Gauss-Newton iterative algorithm to numerically solve the above weighted observation equation. The system takes the rough distance two hundred meters estimated by the signal strength combined with the angle information as the initial iteration value, and stops iteration when the modulus of the position update quantity is less than the set convergence threshold zero point one meter after three iteration calculations, and finally outputs the calculated three-dimensional coordinate value one hundred point two meters, one hundred and ninety-nine point eight meters, one hundred and fifty point one meters as the current position point of the unmanned aerial vehicle, thereby realizing the meter-level high-precision positioning with the error controlled within three meters.

[0036] Specifically, the step S33 updates the real-time trajectory state of the current position point based on the historical trajectory point sequence to obtain a real-time trajectory. It can be understood that the current position point alone can only reflect the spatial position of the UAV at a certain time, and cannot reflect the continuity and change trend of the motion trajectory, while the regulatory work needs to master the complete motion path of the UAV to determine its flight intention. Therefore, the application further integrates and updates the trajectory state of the newly added current position point in combination with the stored historical trajectory point sequence, so as to generate a continuous and complete real-time trajectory. In this way, discrete position points can be connected into trajectory information with time sequence and motion logic, clearly presenting the flight path, speed change and heading adjustment of the UAV, providing complete motion data support for subsequent multi-target separation and abnormal behavior identification, and ensuring the comprehensiveness of regulatory judgment.

[0037] In particular, in one possible embodiment, the implementation process of the step S33 is as follows: first, the stored historical trajectory point sequence is called from the data cache, which contains information such as position, speed, time stamp of each time point. Then, the newly calculated current position point is spliced with the historical trajectory point sequence in time stamp order. Then, based on the spatial distance and time interval between the current position point and the adjacent historical position point, the instantaneous speed and heading angle of the UAV are calculated. Finally, the updated position, speed, heading angle and corresponding time stamp are integrated into a new trajectory state node, replacing the outdated historical data, forming an updated real-time trajectory, ensuring the timeliness and continuity of the trajectory information.

[0038] Specifically, the step S34 performs time sequence trajectory prediction on the real-time trajectory to obtain a predicted trajectory. It can be understood that due to the high speed and strong maneuverability of the UAV, relying only on the real-time trajectory for regulation may cause response lag, and it is difficult to avoid sudden flight risks in time, and the deviation trend needs to be perceived in advance for trajectory anomaly determination. Therefore, the application further analyzes the historical motion data in the real-time trajectory, and mines the motion law through a time sequence prediction model, so as to predict the flight path of the UAV in the future period of time. In this way, the potential flight trajectory of the UAV can be obtained in advance, the trajectory deviation risk can be discovered in time, sufficient emergency disposal time can be reserved for the regulatory department, the response efficiency to sudden violations, trajectory mutations and other situations can be effectively improved, and the safety and controllability of the low-altitude airspace can be ensured.

[0039] In particular, in one possible embodiment, the step S34 is implemented as follows: first, historical trajectory data in the recent period of time, such as historical 10-second trajectory data, is extracted from the real-time trajectory, including characteristic information such as position, speed, heading angle, etc. at each time point, and standardized processing is performed to eliminate the influence of data dimension. Then, the processed historical trajectory data is input into a pre-trained LSTM time series prediction model, which captures the flight mode of the UAV by learning the historical movement law. Subsequently, the model outputs the predicted position in the next 3 seconds (prediction error ≤0.5 meters) based on the input historical features. Finally, these predicted positions are concatenated in time sequence to form a complete predicted trajectory, and the prediction confidence is labeled, so as to perceive the trajectory deviation risk in advance and provide a clear reference benchmark for subsequent abnormality judgment.

[0040] In the low-altitude UAV trajectory tracking and monitoring method based on the 5G-A integrated base station described above, the step S4 is to perform multi-target recognition and trajectory separation on the real-time trajectory based on the pure fusion data stream to obtain independent individual trajectories. It should be understood that, when multiple UAVs fly simultaneously in a low-altitude scene, the real-time trajectory is prone to confusion due to signal aliasing. The existing technology lacks an effective individual distinguishing mechanism, which leads to the inability to accurately track the flight path of each UAV. Therefore, the present application further combines multi-target recognition algorithms and trajectory separation strategies based on the exclusive features in the pure fusion data stream to accurately bind individuals and trajectories. In this way, the trajectory confusion bottleneck in the multi-target scene is broken, an independent trajectory file is established for each UAV (including unregistered models), and the regulatory department can clearly grasp the flight intention of a single target, thereby providing accurate individual data support for subsequent abnormality judgment and emergency disposal.

[0041] In particular, in one specific embodiment, Figure 5 The flowchart of the sub-step S4 of the low-altitude UAV trajectory tracking and monitoring method based on the 5G-A integrated base station according to the embodiment of the present application is shown in FIG. 4. Figure 5 As shown in FIG. 4, the step S4 includes: S41, extracting features from each frame of pure fusion data in the pure fusion data stream to obtain a set of individual signal feature vectors, the individual signal feature vectors including communication features, radar features and kinematic features; S42, performing target recognition on each individual signal feature vector in the set of individual signal feature vectors based on a UAV individual feature library to obtain identified target data and a set of unidentified target feature vectors; and S43, performing trajectory clustering and association assignment on the set of unidentified target feature vectors and the mixed trajectory points of the unidentified targets corresponding to the set of unidentified target feature vectors to obtain independent individual trajectories.

[0042] Specifically, the step S41, feature extraction is performed on each frame of pure fusion data in the pure fusion data stream to obtain a set of individual signal feature vectors, the individual signal feature vectors including communication features, radar features and kinematic features. It should be understood that, since a single dimension of data features is difficult to uniquely identify an individual UAV, and is susceptible to environmental interference to cause feature distortion, which cannot support accurate differentiation in a multi-target scene. Therefore, the application further extracts exclusive features from the communication, radar and kinematics of the pure fusion data, and constructs a set of individual signal feature vectors after standardization processing. In this way, the complementary advantages of different dimension features can be integrated to form an individual feature identification with uniqueness and stability, effectively resisting the influence of environmental interference on feature recognition, providing a high-identification feature basis for subsequent target recognition and trajectory separation, and ensuring the accuracy of multi-target differentiation.

[0043] In particular, in a possible embodiment, the implementation process of the step S41 is as follows: for each frame of pure fusion data, first, the signal modulation mode, channel quality indicator and other communication features are decoded and extracted from the 5G-A communication signal. Then, the mean and variance of the radar reflection cross-sectional area and other radar features are calculated through radar echo signal analysis. Combined with the change of adjacent position points in the real-time trajectory, the flight speed variation law, turning angular velocity and other kinematic features are calculated. Then, the three types of features are normalized to eliminate the dimensional differences, and then integrated in series according to the preset structure to form a standardized individual signal feature vector, and finally a complete feature vector set is constructed.

[0044] Specifically, the step S42, based on the UAV individual feature library, each individual signal feature vector in the set of individual signal feature vectors is subjected to target recognition to obtain identified target data and a set of un-identified target feature vectors. It should be understood that, since the set of individual signal feature vectors contains feature information of registered and unregistered UAVs, if not distinguished, it will lead to the inability to confirm the legal target identity, and at the same time affect the subsequent processing efficiency of unregistered targets. Therefore, the application further matches and identifies the feature vectors by calculating the similarity between the real-time captured signal features and the known target features in the feature library based on the pre-set UAV individual feature library, realizing the classification of known and unknown targets. In this way, the legal identity of the registered UAV can be quickly confirmed, the correspondence between the identity and the trajectory is established, and the feature vectors of the unregistered targets are accurately separated, providing a clear data basis for subsequent clustering and separation, and improving the pertinence and efficiency of multi-target recognition.

[0045] In particular, the aforementioned UAV target recognition mechanisms have a fundamental weakness in calculating the similarity between real-time captured signal features and known target features in the feature library. These mechanisms typically employ standard cosine similarity or Euclidean distance algorithms, which implicitly assume equal weighting of features, treating all dimensions in the feature vector equally. However, this assumption does not hold true in complex and variable low-altitude environments. The discriminative power and stability of UAV signal features are not static but closely related to specific external environments. For example, in rainy weather, the measurement of radar cross-section fluctuates drastically, significantly reducing its reliability as a recognition criterion; similarly, features demodulated from communication signals in areas with complex electromagnetic environments may be distorted. Existing mechanisms lack consideration for this dynamic relationship between features and the environment, failing to adaptively adjust the importance of different feature dimensions according to the real-time scenario. This makes their similarity calculation results highly susceptible to interference from unstable feature dimensions, especially under non-ideal conditions such as severe weather or strong electromagnetic interference, significantly reducing the accuracy and robustness of recognition. Meanwhile, the method of globally comparing real-time feature vectors with the feature database incurs a huge computational burden as the number of targets and the database size increase, making it difficult to meet the high timeliness requirements of low-altitude surveillance. To address the above issues, in a preferred embodiment of this application, a scene-adaptive dynamic feature weighted recognition method is proposed. This method abandons the static model of equal feature weights, dynamically generates weights by evaluating the stability of the environment and the features themselves in real time, and uses these weights to optimize the similarity calculation process. Simultaneously, a hierarchical screening strategy is introduced to improve computational efficiency.

[0046] In particular, in one specific embodiment, Figure 6 This is a flowchart of sub-step S42 of the low-altitude UAV trajectory tracking and monitoring method based on a 5G-A integrated sensing base station according to an embodiment of this application. Figure 6 As shown, step S42 includes: S421, performing real-time environmental perception and feature stability assessment based on environmental interference data and the sequence of individual signal feature vectors of the same target to obtain a dynamic weight vector; S422, performing dynamic weighted similarity calculation and matching on individual signal feature vectors based on the dynamic weight vector and the UAV individual feature library to obtain a weighted similarity score; S423, based on the weighted similarity score, performing candidate set screening and data separation on the individual signal feature vector set based on hierarchical quantization to obtain the identified target data and the unidentified target feature vector set.

[0047] More specifically, the step S421 performs real-time environment perception and feature stability evaluation based on the environmental interference data and the sequence of individual signal feature vectors of the same target to obtain a dynamic weight vector. It should be understood that, to achieve dynamic weighting of the features, it is necessary to obtain the basis for determining the weight size, which comes from the real-time influence of the external environment and the historical performance of the feature data itself. Specifically, first, real-time environmental interference data is collected, and the historical feature flow of the target in the recent time window is retrieved. By analyzing these data, on the one hand, the influence factor of the environmental factor on the stability of each feature dimension is calculated, and on the other hand, the inherent stability of each feature dimension is evaluated by calculating the dispersion degree of each feature dimension in the historical data. Subsequently, the two evaluation results are fused to generate a dynamic weight vector that can reflect the credibility of each feature dimension in the current scene. The calculation formula is as follows:

[0048]

[0049] wherein, represents the final generated dynamic weight vector, the dimension of which is consistent with that of the feature vector; is the environmental influence factor vector calculated according to the real-time environment; is the feature inherent stability vector obtained by calculating the reciprocal of the normalized standard deviation of each dimension in the historical feature flow; is a hyperparameter for balancing the weights of the external environment influence and the feature inherent stability; is the total dimension of the feature vector. That is, the prior knowledge from the external environment (reflected by ) and the statistical characteristics from the data itself (reflected by ) are combined, so that the generation of the weight has both the ability to quickly respond to environmental changes and the robustness based on historical data. That is, a weight vector that is no longer fixed is output, but a dynamic weight that can accurately quantify the credibility of each feature dimension in the current specific scene is output, laying a foundation for subsequent accurate identification.

[0050] More specifically, in one specific example of the present application, a specific numerical calculation example is given to assist in the description: the balance coefficient = 0.6 (emphasis on real-time environmental influence), the environmental influence factor vector = [0.8, 0.9, 0.7], corresponding to the communication feature, the radar feature, and the kinematics feature respectively, the larger the value, the smaller the feature affected by environmental interference, the feature inherent stability vector = [0.75, 0.85, 0.65], which is obtained by statistical analysis of 100 frames of historical data, and the total dimension of the feature = 3. First, the sum of the feature inherent stability vector is calculated: = 0.75 + 0.85 + 0.65 = 2.25, and the normalized feature stability vector is obtained: = [0.75 / 2.25, 0.85 / 2.25, 0.65 / 2.25] ≈ [0.333, 0.378, 0.289]. Then the weighted components are calculated: the environmental impact weighted component = 0.6 x [0.8, 0.9, 0.7] = [0.48, 0.54, 0.42], the feature stability weighted component = 0.4 x [0.333, 0.378, 0.289] ≈ [0.133, 0.151, 0.116]. Finally, the dynamic weight vector is synthesized: = [0.48 + 0.133, 0.54 + 0.151, 0.42 + 0.116] = [0.613, 0.691, 0.536]. As can be seen, the radar feature has the highest weight (0.691) in the current scenario, followed by the communication feature (0.613), and the kinematics feature is greatly affected by the environment (0.536). Subsequent similarity calculation will focus on the radar feature, improving the robustness of target recognition in complex environments.

[0051] More specifically, the step S422, based on the dynamic weight vector and the individual feature library of the unmanned aerial vehicle, performs dynamic weighted similarity calculation and matching on the individual signal feature vector to obtain a weighted similarity score. That is, the dynamic weight generated in the previous step is applied to the similarity calculation, fundamentally changing the feature equal weight mode of the traditional algorithm. Specifically, in the execution process, instead of directly calculating the similarity between the original individual signal feature vector and the features in the library , the dynamic weight vector generated in the previous step is used to weight the two vectors before calculation. The weighting operation is realized by element-by-element multiplication (Hadamard product) of the vectors, and then the cosine similarity is calculated based on the weighted new vectors. The calculation formula is as follows:

[0052]

[0053] wherein, is the weighted similarity score; is the individual signal feature vector captured in real time; is one of the feature vectors in the individual feature library of the unmanned aerial vehicle; denotes the Hadamard product; represents the vector dot product; Euclidean norm of the representative vector. That is, through the weighting operation, the contribution of the feature dimension judged as high credibility in the current scene is amplified in the similarity calculation, while the influence of the low credibility feature dimension is effectively suppressed. In this way, the similarity calculation process can intelligently focus on the most critical and reliable feature information, so as to realize accurate and robust matching of the UAV identity in a complex dynamic scene with noise and interference.

[0054] More specifically, in one specific example of the present application, a specific numerical calculation example is given to assist in the description: the current extracted individual signal feature vector =[0.7,0.8,0.6] (communication feature: modulation mode matching degree; radar feature: RCS mean normalized value; kinematics feature: velocity change law similarity), the library feature vector in the UAV individual feature library =[0.65,0.78,0.55] (pre-stored registered UAV feature), the dynamic weight vector =[0.613,0.691,0.536] (follow the above calculation results). First, calculate the Hadamard product: , =[0.613×0.65,0.691×0.78,0.536×0.55]≈[0.398,0.539,0.295]. Then calculate the vector dot product: (0.429×0.398)+(0.553×0.539)+(0.322×0.295)≈0.171+0.298+0.095=0.564. Then calculate the Euclidean norm: = ≈ ≈0.770, = ≈ ≈0.732. Finally, calculate the weighted similarity score: =0.564 / (0.770×0.732)≈0.564 / 0.564=1.0. As can be seen, the similarity score is close to 1.0, indicating that the currently detected UAV is highly matched with the registered UAV in the feature library, and the system can accurately determine it as a legal target, avoiding false identification caused by environmental interference.

[0055] More specifically, the step S423 filters the individual signal feature vector set based on the weighted similarity score based on the hierarchical quantization candidate set to separate the data to obtain the identified target data and the unidentified target feature vector set to solve the high computational complexity problem caused by global traversal search and improve the recognition efficiency. Specifically, the execution process adopts a strategy of first rough and then fine. First, the huge individual feature library of the unmanned aerial vehicle is preprocessed by offline vector quantization to cluster it into multiple clusters represented by prototype features. When a real-time individual signal feature vector enters the system, it will first be compared with these prototype features with extremely small calculation to quickly lock a few candidate clusters that are most likely to match. Then, the system only performs fine matching inside the greatly reduced candidate set by calling the dynamic weighted similarity algorithm defined in the previous step with relatively intensive calculation. Finally, whether the highest weighted similarity score obtained by fine matching exceeds the preset threshold is determined to determine the target identity and separate the data. This step significantly reduces the number of unnecessary fine comparisons by introducing hierarchical quantization screening, thereby greatly reducing the calculation delay while ensuring recognition accuracy, so that the entire recognition mechanism can meet the stringent requirements of low-altitude traffic management for high timeliness, ensuring the real-time response capability of the system.

[0056] Through the implementation of the above technical means, the improved mechanism realizes significant optimization of the traditional target recognition method, and achieves the comprehensive technical purpose of improving recognition accuracy, robustness and real-time performance. Specifically, the introduced scene-adaptive dynamic feature weighting similarity calculation method enables the system to intelligently evaluate and utilize the most reliable signal features in a specific environment, effectively overcoming the recognition rate decline problem caused by unstable features in traditional equal weight algorithms in complex scenes such as bad weather and strong electromagnetic interference, thereby greatly enhancing the accuracy and environmental adaptability of recognition. At the same time, the candidate set screening strategy based on hierarchical quantization optimizes large-scale global search to an efficient coarse screening + fine screening mode, significantly reducing the computational complexity of the algorithm, ensuring that even in the scene of multiple targets and large capacity feature library, the identity of the unmanned aerial vehicle can be quickly responded to within seconds. Finally, the mechanism builds an accurate and efficient unmanned aerial vehicle identity recognition system that can provide more reliable and timely technical support for low-altitude safety supervision and effectively address the safety challenges posed by black flying and chaotic flying unmanned aerial vehicles.

[0057] Specifically, the step S43, the mixed track points of the unidentified target feature vector set and the unidentified target corresponding to the unidentified target feature vector set are subjected to track clustering and association assignment to obtain independent individual tracks. It should be understood that, since the unidentified target (such as a black flying unmanned aerial vehicle) lacks a registered identity, the mixed track points corresponding to the target cannot be directly attributed to a specific individual, resulting in a blind area in the track tracking of such targets. Therefore, the application further adopts a density-based clustering algorithm to cluster the unidentified target feature vectors, and combines the spatial and temporal information of the mixed track points to perform association assignment. In this way, the feature aggregation rules of different individuals can be mined from the mixed data without an identity, and the mixed track points can be accurately assigned to the corresponding clustering cluster, an independent track is generated for each unidentified target, the effective tracking of the unregistered target such as the black flying unmanned aerial vehicle is realized, and the supervision blind area is filled.

[0058] In particular, in a possible embodiment, the implementation process of the step S43 is as follows: first, taking the set of unidentified target feature vectors as input, adopting a density-based DBSCAN clustering algorithm to perform clustering analysis on the signal features of the multi-unmanned aerial vehicle sensing, automatically dividing clustering clusters according to the feature similarity, and each cluster corresponding to a potential black flying unmanned aerial vehicle target. Then, the mixed track points corresponding to the unidentified target feature vectors are extracted, and the track points are matched based on the spatio-temporal correlation and the clustering cluster features. Subsequently, a unique temporary identity is assigned to each clustering cluster, and the matched track points are bound to the temporary identity. Finally, through cross verification of multi-base station positioning data, the track point attribution error is corrected, and the complete independent individual track of each unidentified target is integrated, the multi-aircraft tracking problem is solved, and the precise tracking of the black flying unmanned aerial vehicle is realized.

[0059] In the above low-altitude unmanned aerial vehicle track tracking and monitoring method based on the 5G-A sensing integrated base station, the step S5, based on the predicted track and the airspace rule library, the independent individual track is subjected to abnormal track identification to obtain an abnormal event. It should be understood that, since the independent individual track only presents the actual flight path of the unmanned aerial vehicle, and is not combined with the airspace use specification and the future flight trend for judgment, the risks such as airspace violation, sudden track mutation or equipment abnormality cannot be discovered in time, and it is difficult to meet the real-time supervision demand. Therefore, the application further performs multi-dimensional abnormality checking on the independent individual track based on the compliance standards of the airspace rule library and the foresight of the predicted track, so as to comprehensively capture various types of violation and abnormal scenarios. In this way, the change from passive recording to active early warning can be realized, and multiple types of abnormalities such as airspace violation, track mutation and equipment disconnection can be accurately identified, complete abnormal information is provided for the supervision department, and the timeliness and pertinence of emergency disposal are ensured.

[0060] In particular, in one possible embodiment, the implementation of the step S5 is as follows: firstly, the compliance parameters such as flight restricted area and flight restricted height in the airspace rule library and the predicted trajectory data of the corresponding unmanned aerial vehicle are called. Then, the real-time position, speed, heading and other information of the independent individual trajectory are respectively compared with the airspace rule library for compliance and analyzed for deviation from the predicted trajectory. Subsequently, it is determined whether there is an abnormality according to the preset standard, for example, when the flight speed suddenly changes (acceleration / deceleration ≥ 5 m / s), the heading angle suddenly changes (≥ 30° / s) or deviates from the predicted trajectory by ≥ 2 meters, it is determined that the trajectory is abnormal, and when the communication signal is interrupted for more than a certain time and there is no radar signal matching, it is determined that the equipment is abnormal. Finally, for the confirmed abnormality, the key information such as the unmanned aerial vehicle identity, abnormal occurrence time, position and abnormal type is integrated to generate a structured abnormal event, ensuring that the abnormal information is complete and can directly support subsequent emergency response work.

[0061] In summary, the low-altitude unmanned aerial vehicle trajectory tracking and monitoring method based on the 5G-A integrated sensing base station according to the embodiments of the present application is illustrated, which first collects multi-source data such as sensing fusion signals, environmental interference and unmanned aerial vehicle attributes, and encapsulates them into synchronous data frames after aligning the unified timestamp and coordinate system. Then, the data frames are subjected to deep fusion and anti-interference processing to filter out noise and generate pure fusion data. Further, the pure data is subjected to real-time trajectory solving and motion trend prediction by multi-base station collaborative solving and prediction. Based on this, the independent individual trajectory is accurately stripped from the complex mixed data stream by multi-target feature recognition and clustering separation mechanism, and the trajectory is subjected to compliance verification and abnormality determination in combination with the airspace rule library. In this way, the problems of signal interference and multi-target aliasing in complex environments can be effectively overcome, thereby realizing high-precision global tracking of low-altitude unmanned aerial vehicles and real-time monitoring of abnormal behaviors.

[0062] As described above, the low-altitude unmanned aerial vehicle trajectory tracking and monitoring system 100 based on the 5G-A integrated sensing base station according to the embodiments of the present application can be implemented in various wireless terminals, such as a server having a low-altitude unmanned aerial vehicle trajectory tracking and monitoring algorithm based on the 5G-A integrated sensing base station. In one possible implementation, the low-altitude unmanned aerial vehicle trajectory tracking and monitoring system 100 based on the 5G-A integrated sensing base station according to the embodiments of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the low-altitude unmanned aerial vehicle trajectory tracking and monitoring system 100 based on the 5G-A integrated sensing base station can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the low-altitude unmanned aerial vehicle trajectory tracking and monitoring system 100 based on the 5G-A integrated sensing base station can also be one of the many hardware modules of the wireless terminal.

[0063] Alternatively, in another example, the low-altitude unmanned aerial vehicle trajectory tracking monitoring system 100 based on the 5G-A integrated base station can also be a separate device from the wireless terminal, and the low-altitude unmanned aerial vehicle trajectory tracking monitoring system 100 based on the 5G-A integrated base station can be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in an agreed data format.

Claims

1. A method for tracking and monitoring the trajectory of low-altitude unmanned aerial vehicles (UAVs) based on a 5G-A integrated sensing base station, characterized in that, include: Data alignment, coordinate unification, and packaging are performed on the synthetic fusion signal, environmental interference dataset, and UAV basic attribute data to obtain synchronized raw data frames; The synchronized raw data frames are subjected to sensory data fusion and anti-interference processing to obtain clean fused data, including: cross-domain data association matching of the synchronized raw data frames to obtain associated data objects; fusion filtering and state estimation of the associated data objects to obtain clean fused data; the clean fused data includes identity identifiers, optimal state estimates and association attributes, the optimal state estimates include estimated position, estimated speed and covariance matrix, and the association attributes include aircraft type information, registration information and data source identifiers; Trajectory calculation and prediction are performed on clean fusion data provided by multiple base stations to obtain real-time and predicted trajectories, including: setting clean fusion data from the local base station as local clean fusion data and setting clean fusion data from at least two neighboring cooperating base stations as cooperative multi-source positioning data; performing multi-station cooperative positioning calculation on the local clean fusion data and cooperative multi-source positioning data to obtain the current location point; updating the real-time trajectory status of the current location point based on historical trajectory point sequences to obtain the real-time trajectory; and performing time-series trajectory prediction on the real-time trajectory to obtain the predicted trajectory. Based on a clean fused data stream, multi-target identification and trajectory separation are performed on real-time trajectories to obtain independent individual trajectories. This includes: extracting features from each frame of clean fused data in the clean fused data stream to obtain a set of individual signal feature vectors, wherein the individual signal feature vectors include communication features, radar features, and kinematic features; based on the UAV individual feature library, target identification is performed on each individual signal feature vector in the set of individual signal feature vectors to obtain identified target data and a set of unidentified target feature vectors; trajectory clustering and association assignment are performed on the set of unidentified target feature vectors and the mixed trajectory points of unidentified targets corresponding to the set of unidentified target feature vectors to obtain independent individual trajectories. Based on predicted trajectories and a spatial rule base, abnormal trajectories of independent individuals are identified to obtain abnormal events.

2. The method for tracking and monitoring the trajectory of low-altitude unmanned aerial vehicles based on a 5G-A integrated sensing base station according to claim 1, characterized in that, Based on the individual feature library of UAVs, target identification is performed on each individual signal feature vector in the individual signal feature vector set to obtain identified target data and an unidentified target feature vector set, including: Real-time environmental perception and feature stability assessment are performed based on environmental interference data and sequences of individual signal feature vectors of the same target to obtain dynamic weight vectors. Based on dynamic weight vectors and UAV individual feature database, dynamic weighted similarity calculation and matching are performed on individual signal feature vectors to obtain weighted similarity scores; Based on weighted similarity scores, candidate set screening and data separation of individual signal feature vector sets using hierarchical quantization are performed to obtain identified target data and unidentified target feature vector sets.

3. The method for tracking and monitoring the trajectory of low-altitude unmanned aerial vehicles based on a 5G-A integrated sensing base station according to claim 2, characterized in that, Based on a dynamic weight vector and an individual UAV feature database, dynamic weighted similarity calculation and matching are performed on individual signal feature vectors to obtain a weighted similarity score. This includes: performing dynamic weighted similarity calculation and matching on individual signal feature vectors using the following formula: in, The weighted similarity score; It is the feature vector of an individual signal; It is a stock feature vector in the individual feature library of drones; It represents the Hadamardi (or Hadama) stack; Represents the dot product of vectors; The Euclidean norm of a vector. This is a dynamic weight vector.

Images