Unmanned aerial vehicle data fusion processing system based on distributed stream computing framework

Abstract

The application provides a UAV data fusion processing system based on a distributed streaming computing framework, comprising: a data access unit configured to receive multi-source UAV data streams; a data analysis and standardization unit configured to analyze the UAV data streams and convert the UAV data streams into uniform standard data structures and then input the uniform standard data structures into a distributed streaming processing engine; the distributed streaming processing engine is configured to distribute the UAV data streams to distributed nodes, each of the distributed nodes comprises a feature extraction unit configured to extract feature vectors from the UAV data within a sliding time window; a target association determination unit configured to cluster the feature vectors to form target association clusters; a data fusion unit configured to perform data fusion on the target association clusters; and a trajectory smoothing and fusion result output unit configured to perform smoothing estimation and position correction on fusion trajectories in the fusion results by using robust Kalman filtering, and distribute the fusion results containing the filtered fusion trajectories. The application significantly improves monitoring efficiency and data fusion accuracy.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) data processing technology, and in particular to a UAV data fusion processing system based on a distributed streaming computing framework. Background Technology

[0002] With the rapid development of drone technology, drones are increasingly widely used in both military and civilian fields. During flight, drones generate a large amount of multi-source heterogeneous data, including location, speed, altitude, and equipment status information. How to effectively integrate this multi-source data to provide accurate and real-time information support for drone monitoring and management has become a significant challenge in current technological development.

[0003] In the prior art, Chinese Patent No. CN112964246B discloses a method, apparatus, storage medium, and device for multi-sensor data fusion of unmanned aerial vehicles (UAVs). The technical solution involves acquiring data from each type of sensor in each sensor combination, and extracting features from the continuously collected data using signal processing. If a significant change occurs, the sensor is considered faulty, and the sensor combination is deemed ineffective. Otherwise, based on the mathematical relationships existing in the data collected by different sensors, the data collected by each type of sensor in the sensor combination is processed to obtain a unified data type, thereby determining the validity of the data collected by each type of sensor in the sensor combination. The main purpose of this invention is to ensure the reliability of the sensor data. Therefore, fault detection of the sensors is necessary before using the collected data to eliminate data from faulty sensors. This invention does not involve real-time processing of large-scale concurrent data streams. Chinese patent application CN116399327A discloses a UAV positioning system based on multi-source data fusion, which mainly uses a federated filtering algorithm for data fusion. However, the multi-source data must be homogeneous, and there are performance bottlenecks when processing high-concurrency data streams. In the field of big data processing, distributed stream processing frameworks such as Flink have been widely used. For example, Chinese patent application CN119311729A discloses a stream data processing method, apparatus, device, and storage medium based on Flink windows. However, this method is mainly aimed at general stream data processing and lacks specific optimization for the characteristics of UAV data.

[0004] Existing technologies lack efficient processing systems for multi-source data fusion from unmanned aerial vehicles (UAVs), particularly in terms of real-time performance, accuracy, and scalability, which fail to meet the requirements of modern UAV monitoring systems. Therefore, a system and method are needed that can efficiently process large-scale UAV data streams and effectively fuse multi-source data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a UAV data fusion processing system based on a distributed streaming computing framework, which can achieve efficient fusion processing of multi-source heterogeneous UAV data in large-scale concurrent UAV data stream processing scenarios.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A UAV data fusion processing system based on a distributed streaming computing framework includes: The data access unit is used to receive drone data streams from different data sources; The data parsing and standardization unit is used to parse the UAV data stream and convert it into a unified standard data structure. The UAV data stream with the standard data structure is then input into the distributed stream processing engine through a message queue. The distributed stream processing engine is used to distribute drone data streams to distributed nodes. Each distributed node contains a feature extraction unit, which is used to traverse the drone data within a sliding time window and extract feature vectors. The target association determination unit is used to cluster each feature vector, and the feature vectors that meet the association conditions form a target association cluster; The data fusion unit is used to perform data fusion on each target associated cluster to generate a unified fusion result; The trajectory smoothing and fusion result output unit uses robust Kalman filtering to smooth and correct the position of the fused trajectory in the fusion result, and distributes the fusion result containing the filtered fused trajectory to the message queue, database storage interface and real-time monitoring interface to realize UAV monitoring.

[0008] To optimize the above technical solution, the specific measures also include: Furthermore, the data sources include radar data sources, radio protocol parsing data sources, TDOA positioning data sources, and 5G sensing data sources. After data parsing and standardization, the drone data streams from different data sources are input into the distributed stream processing engine in parallel through a message queue, providing raw records containing timestamps, locations, and speeds.

[0009] Furthermore, the data parsing and standardization unit includes a data cleaning subunit, a coordinate system transformation subunit, a data type normalization subunit, and a timestamp synchronization subunit; The specific steps of parsing the UAV data stream and converting it into a unified standard data structure are as follows: The data cleaning subunit is used to clean the UAV data stream, including removing invalid data, handling missing values, and detecting outliers. Removing invalid data specifically involves eliminating records with null values ​​or those that are out of range. Handling missing values ​​specifically involves filling missing values ​​with interpolation or default values. Outlier detection is specifically based on statistical thresholds to identify and correct anomalies. The coordinate system transformation subunit is used to standardize position data, including identifying the input coordinate system, converting to a standard coordinate system, and maintaining accuracy. The data type normalization subunit is used to unify the representation of fields with the same name in different data sources. This includes unifying the location, speed and altitude numerical fields into a predetermined numerical type, unifying the equipment identifier, batch number and target identifier fields into a standard string format, and performing unified conversion and formatting on the time field. The timestamp synchronization subunit is used to parse timestamps, convert timestamps to the same time zone, perform clock offset correction based on network latency, and align the accuracy of timestamps. The method for clock offset correction based on network latency is as follows: Corrected timestamp = Original timestamp + Clock offset - Network transmission delay, where the clock offset is obtained through NTP synchronization and the network transmission delay is measured by round-trip time.

[0010] Furthermore, the specific process of the feature extraction unit extracting the feature vector is as follows: The receiver window receives trajectory data grouped by timestamp, including longitude, latitude, speed, and altitude. Sort the trajectory data in ascending order by timestamp; Iterate through the trajectory data within the window, determine whether the current trajectory data is the starting point of the track, if so, initialize the starting point features to set the reference position, and update the feature statistics; otherwise, calculate the features of adjacent points, which include speed and heading, and update the feature statistics, until all trajectory data has been traversed. Calculate the final characteristic values, including longitude range, latitude range, average speed, maximum speed, flight altitude, heading change, and flight duration; Output the extracted feature vector.

[0011] Furthermore, the specific steps of clustering each feature vector and forming a target association cluster from the feature vectors that satisfy the association conditions are as follows: The feature vectors within the same time window are uniformly standardized to obtain standardized feature vectors. The distance value is calculated for any two standardized feature vectors and mapped to a similarity value. When the similarity value is higher than a preset threshold, the two corresponding standardized feature vectors are considered to be from the same UAV. Combined with time and space constraints, it is determined whether they come from the same UAV and the standardized feature vectors from the same UAV are grouped into the same target association cluster.

[0012] Furthermore, the specific steps of performing data fusion on each target associated cluster to generate a unified fusion result are as follows: For each target cluster, information representing the earliest occurrence time and reference position of the cluster is selected as the fusion benchmark. Combined with the previous fusion identifier and previous fusion reference position saved in the historical fusion state, a unified fusion reference time, fusion reference latitude and longitude, and fusion reference altitude are determined. A unique fusion identifier is generated using a hash algorithm based on the fusion reference time, fusion reference latitude and longitude, and fusion reference altitude. According to the preset field merging rules, the device model, pilot position, frequency information, speed information, and target category information are selected and integrated from the target cluster to form a unified fusion result, including a unified fusion identifier, fusion trajectory, target category, and a list of associated signals. The fusion trajectory includes the fusion position and fusion speed. A distributed state memory is used to store the historical fusion information of each UAV, including the previous fusion identifier, the previous fusion start time, and the previous fusion reference position. In subsequent windows, the identifier of the same target is maintained within consecutive windows by comparing the matching relationship between the current cluster and the historical fusion state.

[0013] Furthermore, the distributed stream processing engine is specifically the Flink stream processing engine, and the system also includes a state management unit, an anomaly detection unit, and a load balancing unit; The State Management Unit maintains the Flink state backend and performs state recovery through a checkpoint-based fault tolerance mechanism. The error detection unit is used to detect and handle runtime errors; The load balancing unit is used for task scheduling, resource allocation, and performance monitoring to dynamically adjust node load.

[0014] Furthermore, the allocation of UAV data streams to distributed nodes specifically involves allocating UAV data streams to distributed nodes through KeyBy partitioning, data routing, and load balancing mechanisms.

[0015] The beneficial effects of this invention are: (1) It realizes efficient processing of large-scale UAV data streams and supports real-time data processing needs in high-concurrency scenarios; (2) Through multi-source data fusion processing, the accuracy and completeness of UAV data have been significantly improved; (3) It adopts a distributed architecture design, which has good scalability and fault tolerance; (4) It provides rich flight feature extraction functions, providing comprehensive data support for subsequent analysis and decision-making; (5) Supports dynamic configuration and flexible deployment to adapt to different scales and needs of drone monitoring scenarios. Attached Figure Description

[0016] Figure 1 This is an overall architecture diagram of the UAV data fusion processing system based on a distributed streaming computing framework proposed in this invention.

[0017] Figure 2 This is a flowchart of the data parsing and standardization unit processing.

[0018] Figure 3 This is a diagram of the internal structure of a distributed stream processing engine.

[0019] Figure 4 This is a flowchart of the feature extraction unit.

[0020] Figure 5 This is a schematic diagram of the processing flow for target association determination, data fusion, trajectory smoothing, and fusion result output.

[0021] Figure 6 This is a flowchart of the UAV data fusion processing method based on a distributed streaming computing framework proposed in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0023] Example 1 This invention proposes a UAV data fusion processing system based on a distributed streaming computing framework. The overall architecture of the system is as follows: Figure 1 As shown, it includes: The data access unit is used to receive UAV data streams from different data sources, including radar data sources, radio protocol parsing data sources, TDOA positioning data sources, and 5G sensing data sources. After data parsing and standardization, the UAV data streams from different data sources are input into the distributed stream processing engine in parallel through a message queue, providing raw records containing timestamps, locations, and speeds. The design of this unit ensures the diversity and real-time nature of the data and supports unified processing of heterogeneous signals.

[0024] The data parsing and standardization unit is used to parse the UAV data stream and convert it into a unified standard data structure. The UAV data stream with the standard data structure is input into the distributed stream processing engine through a message queue. The data parsing and standardization unit includes a data cleaning subunit, a coordinate system transformation subunit, a data type normalization subunit, and a timestamp synchronization subunit. Figure 2The processing flow of data parsing and standardization units is shown.

[0025] The specific process of parsing the drone data stream and converting it into a unified standard data structure is as follows: The data cleaning subunit cleans the UAV data stream, including removing invalid data, handling missing values, and detecting outliers. Removing invalid data specifically involves eliminating records with null values ​​or those out of range. Handling missing values ​​involves filling missing values ​​with interpolation or default values. Outlier detection is based on statistical thresholds to identify and correct anomalies. The data cleaning subunit optimizes data integrity and prevents the amplification of errors in downstream algorithms.

[0026] The coordinate system transformation subunit is used to standardize position data, including identifying the input coordinate system, converting to a standard coordinate system (such as UTM), and maintaining accuracy (such as floating-point truncation control). The coordinate system transformation subunit ensures the uniformity of multi-source coordinates, for example, using the PyProj library to convert from WGS84 to EPSG:32651.

[0027] The data type normalization subunit is used to unify the representation of fields with the same name in different data sources. This includes unifying the location, speed and altitude numerical fields into a predetermined numerical type, unifying the equipment identifier, batch number and target identifier fields into a standard string format, and performing unified conversion and formatting on the time field. The timestamp synchronization subunit is used to parse timestamps, convert timestamps to the same time zone (such as UTC), perform clock offset correction based on network latency, and align the timestamp accuracy. The method for clock offset correction based on network latency is as follows: Corrected timestamp = original timestamp + clock offset - network transmission latency, where the clock offset is obtained through NTP (Network Time Protocol) synchronization and the network transmission latency is measured by round-trip time.

[0028] The distributed stream processing engine is used to distribute drone data streams to distributed nodes. Each distributed node contains a feature extraction unit, which is used to traverse the drone data within a sliding time window and extract feature vectors. Figure 3 This diagram illustrates the internal structure of the distributed stream processing engine of this invention. The diagram employs a hierarchical, modular design, explaining the engine's core components and their interactions. The distributed stream processing engine is implemented based on a distributed framework, integrating units such as data distribution, window computation, feature extraction, state management, and anomaly detection, supporting high-concurrency real-time signal fusion. In the diagram, the master node (stream processing engine) connects to each sub-unit via directional arrows, demonstrating the parallel dependencies and data flow logic between units.

[0029] The specific method for distributing drone data streams to distributed nodes is as follows: the drone data streams are distributed to distributed nodes through KeyBy partitioning, data routing, and load balancing mechanisms.

[0030] Figure 4 The processing flow of the feature extraction unit of the present invention is shown. The specific process of the feature extraction unit extracting feature vectors is as follows: The receiver window receives trajectory data grouped by timestamp, including longitude, latitude, speed, and altitude. Sort the trajectory data in ascending order by timestamp; Iterate through the trajectory data within the window, determine whether the current trajectory data is the starting point of the track, if so, initialize the starting point features to set the reference position, and update the feature statistics; otherwise, calculate the features of adjacent points, which include speed and heading, and update the feature statistics, until all trajectory data has been traversed. Calculate the final characteristic values, including longitude range, latitude range, average speed, maximum speed, flight altitude, heading change, and flight duration; Output the extracted feature vector.

[0031] The distributed stream processing engine is specifically the Flink stream processing engine. The system also includes a state management unit, an anomaly detection unit, and a load balancing unit. The State Management Unit maintains the Flink state backend and performs state recovery through a checkpoint-based fault tolerance mechanism. The error detection unit is used to detect and handle runtime errors; The load balancing unit is used for task scheduling, resource allocation, and performance monitoring to dynamically adjust node load.

[0032] The target association determination unit is used to cluster each feature vector, and feature vectors that meet the association conditions form target association clusters. The specific process of clustering each feature vector and forming target association clusters from feature vectors that meet the association conditions is as follows: The feature vectors within the same time window are uniformly standardized to obtain standardized feature vectors. The distance value is calculated for any two standardized feature vectors and mapped to a similarity value. When the similarity value is higher than a preset threshold, the two corresponding standardized feature vectors are considered to be from the same UAV. Combined with time and space constraints, it is determined whether they come from the same UAV and the standardized feature vectors from the same UAV are grouped into the same target association cluster.

[0033] The data fusion unit is used to perform data fusion on each target association cluster to generate a unified fusion result. The specific process of performing data fusion on each target association cluster to generate a unified fusion result is as follows: For each target cluster, information representing the earliest occurrence time and reference position of the cluster is selected as the fusion benchmark. Combined with the previous fusion identifier and previous fusion reference position saved in the historical fusion state, a unified fusion reference time, fusion reference latitude and longitude, and fusion reference altitude are determined. A unique fusion identifier is generated using a hash algorithm based on the fusion reference time, fusion reference latitude and longitude, and fusion reference altitude. According to the preset field merging rules, the device model, pilot position, frequency information, speed information, and target category information are selected and integrated from the target cluster to form a unified fusion result, including a unified fusion identifier, fusion trajectory, target category, and a list of associated signals. The fusion trajectory includes the fusion position and fusion speed. A distributed state memory is used to store the historical fusion information of each UAV, including the previous fusion identifier, the previous fusion start time, and the previous fusion reference position. In subsequent windows, the identifier of the same target is maintained within consecutive windows by comparing the matching relationship between the current cluster and the historical fusion state.

[0034] The trajectory smoothing and fusion result output unit uses robust Kalman filtering to smooth and correct the position of the fused trajectory in the fusion result, and distributes the fusion result containing the filtered fused trajectory to the message queue, database storage interface and real-time monitoring interface to realize UAV monitoring.

[0035] Figure 5 The process flow for target association determination, data fusion, trajectory smoothing, and fusion result output is shown.

[0036] Example 2 This invention proposes a UAV data fusion processing method based on a distributed streaming computing framework, corresponding to the system in Embodiment 1. Figure 6 The processing flow of the UAV data fusion processing method based on a distributed streaming computing framework is shown, including the following stages: Data access phase: This stage, located at the very top of the process, is used to collect and access real-time data, supporting various sensor inputs (such as radar and GPS) to ensure the timeliness and diversity of the data. The design of this stage optimizes access efficiency and is suitable for high-frequency data scenarios.

[0037] Data preprocessing stage: Following the data ingestion, this stage standardizes the input data, including coordinate system transformation, data format standardization, and timestamp synchronization. This stage addresses heterogeneity issues; for example, coordinate transformation is unified to a standard system; format standardization ensures data consistency; and timestamp synchronization aligns event times.

[0038] Distributed stream processing phase: This stage introduces a distributed stream processing engine to achieve data distribution and distributed data processing. This stage embodies the distributed innovation of this invention and ensures parallel processing of large-scale data.

[0039] Feature extraction stage: This stage calculates multidimensional features, including spatial features, velocity features, altitude features, and heading features. Statistical vectors (such as range and average) are generated through window aggregation. The parallel design of this stage optimizes the computational load.

[0040] Data fusion phase: This stage integrates the features extracted in the previous stage, clusters each feature vector, and forms target association clusters from the feature vectors that meet the association conditions; data fusion is then performed on each target association cluster to generate a unified fusion result.

[0041] Result output stage: This stage, located at the very bottom of the process, is used to distribute the fusion results, including monitoring system interfaces, database storage, and analytics application interfaces.

[0042] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0043] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A UAV data fusion processing system based on a distributed streaming computing framework, characterized in that, include: The data access unit is used to receive drone data streams from different data sources; The data parsing and standardization unit is used to parse the UAV data stream and convert it into a unified standard data structure. The UAV data stream with the standard data structure is then input into the distributed stream processing engine through a message queue. The distributed stream processing engine is used to distribute drone data streams to distributed nodes. Each distributed node contains a feature extraction unit, which is used to traverse the drone data within a sliding time window and extract feature vectors. The target association determination unit is used to cluster each feature vector, and the feature vectors that meet the association conditions form a target association cluster; The data fusion unit is used to perform data fusion on each target associated cluster to generate a unified fusion result; The trajectory smoothing and fusion result output unit uses robust Kalman filtering to smooth and correct the position of the fused trajectory in the fusion result, and distributes the fusion result containing the filtered fused trajectory to the message queue, database storage interface and real-time monitoring interface to realize UAV monitoring.

2. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The data sources include radar data sources, radio protocol parsing data sources, TDOA positioning data sources, and 5G sensing data sources. After data parsing and standardization, the drone data streams from different data sources are input into the distributed stream processing engine in parallel through a message queue, providing raw records containing timestamps, locations, and speeds.

3. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The data parsing and standardization unit includes a data cleaning subunit, a coordinate system transformation subunit, a data type normalization subunit, and a timestamp synchronization subunit. The specific steps of parsing the UAV data stream and converting it into a unified standard data structure are as follows: The data cleaning subunit is used to clean the UAV data stream, including removing invalid data, handling missing values, and detecting outliers. Removing invalid data specifically involves eliminating records with null values ​​or those that are out of range. Handling missing values ​​specifically involves filling missing values ​​with interpolation or default values. Outlier detection is specifically based on statistical thresholds to identify and correct anomalies. The coordinate system transformation subunit is used to standardize position data, including identifying the input coordinate system, converting to a standard coordinate system, and maintaining accuracy. The data type normalization subunit is used to unify the representation of fields with the same name in different data sources. This includes unifying the location, speed and altitude numerical fields into a predetermined numerical type, unifying the equipment identifier, batch number and target identifier fields into a standard string format, and performing unified conversion and formatting on the time field. The timestamp synchronization subunit is used to parse timestamps, convert timestamps to the same time zone, perform clock offset correction based on network latency, and align the accuracy of timestamps. The method for clock offset correction based on network latency is as follows: Corrected timestamp = Original timestamp + Clock offset - Network transmission latency, where the clock offset is obtained through NTP synchronization and the network transmission latency is measured by round-trip time.

4. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The specific process by which the feature extraction unit extracts feature vectors is as follows: The receiver window receives trajectory data grouped by timestamp, including longitude, latitude, speed, and altitude. Sort the trajectory data in ascending order by timestamp; Iterate through the trajectory data within the window, determine whether the current trajectory data is the starting point of the track, and if so, initialize the starting point features to set the reference position and update the feature statistics; Otherwise, calculate the features of neighboring points, including speed and heading, and update the feature statistics until all trajectory data has been traversed; Calculate the final characteristic values, including longitude range, latitude range, average speed, maximum speed, flight altitude, heading change, and flight duration; Output the extracted feature vector.

5. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The specific steps of clustering each feature vector and forming a target association cluster from the feature vectors that satisfy the association conditions are as follows: The feature vectors within the same time window are uniformly standardized to obtain standardized feature vectors. The distance value is calculated for any two standardized feature vectors and mapped to a similarity value. When the similarity value is higher than a preset threshold, the two corresponding standardized feature vectors are considered to be from the same UAV. Combined with time and space constraints, it is determined whether they come from the same UAV and the standardized feature vectors from the same UAV are grouped into the same target association cluster.

6. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The specific steps for fusing data for each target cluster to generate a unified fusion result are as follows: For each target cluster, information representing the earliest occurrence time and reference position of the cluster is selected as the fusion benchmark. Combined with the previous fusion identifier and previous fusion reference position saved in the historical fusion state, a unified fusion reference time, fusion reference latitude and longitude, and fusion reference altitude are determined. A unique fusion identifier is generated by hashing algorithm based on the fusion reference time, fusion reference latitude and longitude, and fusion reference altitude. According to the preset field merging rules, the device model, pilot position, frequency information, speed information, and target category information are selected and integrated from the target cluster to form a unified fusion result, including a unified fusion identifier, fusion trajectory, target category, and list of associated signals. The fusion trajectory includes the fusion location and fusion speed; A distributed state memory is used to store the historical fusion information of each UAV, including the previous fusion identifier, the previous fusion start time, and the previous fusion reference position. In subsequent windows, the identifier of the same target is maintained within consecutive windows by comparing the matching relationship between the current cluster and the historical fusion state.

7. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The distributed stream processing engine is specifically the Flink stream processing engine. The system also includes a state management unit, an anomaly detection unit, and a load balancing unit. The State Management Unit maintains the Flink state backend and performs state recovery through a checkpoint-based fault tolerance mechanism. The error detection unit is used to detect and handle runtime errors; The load balancing unit is used for task scheduling, resource allocation, and performance monitoring to dynamically adjust node load.

8. The UAV data fusion processing system based on a distributed streaming computing framework as described in claim 1, characterized in that, The specific method for allocating drone data streams to distributed nodes involves using KeyBy partitioning, data routing, and load balancing mechanisms to distribute drone data streams to distributed nodes.

Images