Cross-domain robot cooperative remote sensing ad hoc network system and method

Abstract

This invention relates to the field of robot ad hoc networking technology, and discloses a cross-domain robot collaborative remote sensing ad hoc networking system and method. The system includes a ground control center, an aerial UAV cluster, a surface unmanned vessel cluster, and an underwater vehicle cluster. During operation, the underwater vehicles encapsulate and upload a spatial positioning covariance scalar, which the surface unmanned vessels use to allocate data forwarding priorities. The ground control center calculates a channel attenuation penalty factor based on the optical data acquired by the UAVs, and feeds it forward to the surface and underwater nodes to adjust spatial maneuvering distances or communication transmission power. The ground control center fuses and reconstructs the converged data, generates a spatiotemporal void feature vector based on error calculation, and issues reverse trajectory control commands to drive the nodes to perform secondary data acquisition and supplementation. This invention maintains the stability of the cross-domain network communication link and ensures the integrity and consistency of the overall water area data fusion parameter field through sensing parameter feedforward compensation and fusion quality reverse feedback mechanisms.

Description

Technical Field

[0001] This invention relates to the field of robot self-organizing network technology, specifically to a cross-domain robot collaborative remote sensing self-organizing network system and method. Background Technology

[0002] With the development of marine exploration technology, cross-domain robotic self-organizing network systems composed of aerial drones, surface unmanned vessels, and underwater vehicles are widely used in aquatic remote sensing and environmental monitoring tasks. However, in complex natural aquatic environments, existing cross-domain collaborative networks still have technical limitations in terms of communication stability, data scheduling mechanisms, and observation integrity.

[0003] In cross-media communication, the quality of underwater acoustic communication links is easily affected by changes in the hydrological environment, especially the concentration of suspended solids in the water, and may be degraded. Existing cross-domain networks usually adopt passive error retransmission mechanisms or maintain a fixed communication transmission power. Due to the lack of prior perception and proactive adaptation to changes in the underwater environment, when local water quality deteriorates, the system cannot adjust the transmission energy of nodes or reduce the physical distance between communication nodes in a timely manner, resulting in the interruption or instability of the communication link between underwater nodes and surface relay nodes.

[0004] In the data aggregation and scheduling stage, most existing systems adopt conventional first-in-first-out or static priority strategies to process the data streams generated by each node. When underwater vehicles are performing exploration missions, the positioning error of their inertial navigation system will accumulate over time. When network bandwidth is limited or communication congestion occurs, there is a lack of a mechanism for dynamic scheduling based on positioning reliability. This causes high-value initial observation data with high positioning accuracy to be delayed or discarded, resulting in some data that is finally transmitted back to the control center losing its practical reference value due to severe spatial positioning drift.

[0005] Current cross-domain observation systems generally adopt a one-way data acquisition and hierarchical reporting mode, lacking a global assessment and low-level control feedback mechanism for the fusion quality of multi-source heterogeneous data. In actual detection processes, track deviations caused by navigation interference or temporary loss of node connectivity will create spatial voids with missing data or large errors in the parameter model of the target water area. The existing fixed route operation mode cannot automatically identify these error-exceeding areas, nor can it autonomously guide the front-end UAVs or underwater vehicles to turn back and conduct secondary supplementary measurements. As a result, the overall water area data fusion parameter field output by the system is incomplete, making it difficult to guarantee the integrity and spatiotemporal consistency of the data. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a cross-domain robot collaborative remote sensing self-organizing network system and method, which solves the problems of unstable underwater acoustic communication links due to changes in the hydrological environment, loss of effective detection data with high positioning reliability due to network congestion, and incomplete global data fusion parameter field due to the lack of dynamic supplementary acquisition of observation blind areas.

[0007] To achieve the above objectives, the first aspect of the present invention provides a cross-domain robot collaborative remote sensing ad hoc network system, comprising:

[0008] The ground control center is responsible for decomposing global tasks, receiving multi-source remote sensing data, and performing high-level data fusion processing. It also serves as the physical origin of the system's spatiotemporal reference, providing the master clock reference and absolute geographic coordinate reference.

[0009] The aerial drone swarm establishes a communication connection with the ground control center through an air-to-ground radio broadband link. Each individual drone in the aerial drone swarm is equipped with a hyperspectral imager and a multispectral camera to acquire optical and spectral data of the water surface.

[0010] A swarm of unmanned surface vessels is deployed on the surface of a water body. Each unmanned surface vessel in the swarm is equipped with an edge computing node, a radio frequency communication gateway, and an underwater acoustic communication gateway. It connects to the air-to-ground broadband radio link through the radio frequency communication gateway and establishes a cross-medium narrowband link through the underwater acoustic communication gateway. The edge computing node performs local data caching, protocol parsing, and queue scheduling operations.

[0011] An underwater vehicle swarm is deployed underwater to perform exploration missions. Each individual vehicle in the swarm is equipped with a multibeam sonar, an underwater acoustic communication unit, and an inertial navigation unit. It accesses the cross-medium narrowband link through the underwater acoustic communication unit to collect water profile physicochemical parameters and seabed mass cloud data.

[0012] Preferably, the ground control center compares the measured values ​​of the verification set with the gridded estimated values ​​of the model under the same spatial coordinates to calculate the root mean square error. When the root mean square error of the target area exceeds the upper limit of the system tolerance error, the three-dimensional spatial coordinate components of the geometric centroid of the target area and the radius of the influence boundary are extracted and combined to generate a spatiotemporal cavity feature vector.

[0013] The ground control center generates reverse trajectory control commands based on the spatiotemporal cavity feature vector, and routes the reverse trajectory control commands through the communication network to the aerial UAV cluster in the corresponding airspace or the underwater vehicle cluster in the corresponding water area for secondary data acquisition based on the elevation reference attribute in the three-dimensional spatial coordinate components of the geometric centroid.

[0014] This configuration transforms the quality assessment results after multi-source data fusion into feature vectors containing specific spatiotemporal attributes, driving sensor nodes in the network to perform adaptive resampling. The system ensures the integrity of the global environmental parameter field by identifying and filling data blind spots caused by node navigation deviations or channel interruptions.

[0015] Preferably, the ground control center receives the optical and spectral data transmitted back by the aerial UAV cluster to invert the three-dimensional suspended matter concentration distribution, converts the three-dimensional suspended matter concentration distribution into the signal attenuation coefficient value of the corresponding frequency band according to the acoustic absorption attenuation model, and calculates the network channel attenuation penalty factor in combination with the three-dimensional spatial topology of each node in the system.

[0016] The ground control center sends the network channel attenuation penalty factor as a feedforward input to the topology reconstruction command to the surface unmanned vessel cluster and the underwater vehicle cluster.

[0017] The system utilizes surface optical features of water bodies acquired by an aerial unmanned aerial vehicle (UAV) platform to infer the physical parameters of the underwater environment, and then extracts the specific obstructive effects of the water environment on acoustic signal transmission. Environmental parameters at the perception layer are mapped to control parameters at the communication layer, enabling prior feedforward of cross-domain information and providing a basis for subsequent network topology and power adjustments.

[0018] Preferably, the edge computing node sends a synchronization request message with a local transmission timestamp through the underwater acoustic communication gateway, and receives a response message containing the receiving timestamp and transmission timestamp of the single-unit aircraft.

[0019] The edge computing node combines the arrival timestamp of the response message it records with the difference between the round-trip time of the message and the internal processing time of the single vehicle to calculate the one-way physical propagation delay time of the underwater acoustic channel. The one-way physical propagation delay time is injected into the local system clock register as a time deviation compensation item to obtain the absolute time. The time message carrying the absolute time is sent to the single vehicle through the cross-medium narrowband link to reset the airborne underlying clock.

[0020] This configuration eliminates the severe propagation delay error caused by the extremely low sound speed in the underwater acoustic channel. By using a two-way message exchange and internal processing time difference subtraction algorithm, it establishes a consistent absolute time reference among air, sea, and underwater nodes in a complex acoustic environment, providing a basic guarantee for the spatiotemporal registration of heterogeneous data.

[0021] Preferably, the single-unit vehicle includes a low-level computing module. The low-level computing module receives measurement data from the inertial navigation unit, extracts the trace of the spatial state prediction covariance matrix to generate a spatial positioning covariance feature scalar during the time update stage of spatial state filtering update, and writes the spatial positioning covariance feature scalar into the frame header of the underwater acoustic communication frame for transmission.

[0022] The edge computing node receives the underwater acoustic communication frame and extracts the spatial positioning covariance feature scalar. Based on the ratio of the spatial positioning covariance feature scalar to the system's preset covariance tolerance threshold, and combined with the upper limit of the highest priority level supported by the radio frequency communication gateway, it calculates and allocates the service quality priority of the data in the transmission buffer queue of the radio frequency communication gateway, and sends the data to the ground control center via the radio frequency communication gateway with priority according to the service quality priority.

[0023] The system directly maps the confidence parameters from the underwater node inertial navigation calculation process to the transmission priority criteria of the communication network layer. When the spatial positioning drift is large, the corresponding data stream automatically receives a higher forwarding weight, enabling the ground control center to quickly acquire high-confidence observation data before positioning degradation and reducing the effective information loss rate caused by network congestion.

[0024] Preferably, when calculating the air-sea cross-layer routing metric, the edge computing node incorporates the network channel attenuation penalty factor into the calculation framework of the comprehensive routing metric;

[0025] When the comprehensive routing metric is lower than the preset link hold threshold, the edge computing node calculates the target two-dimensional plane position vector. The calculation of the target two-dimensional plane position vector is based on the two-dimensional plane position vector of the current unmanned vessel, the two-dimensional plane position vector of the next hop node, and a preset movement step constant. The node then drives the motion controller inside the unmanned vessel to maneuver toward the target two-dimensional plane position vector.

[0026] The above control logic causes the surface node to no longer remain in a static relay state, but to dynamically move closer to the next hop node based on the current link communication quality. Thus, when the hydrological environment changes drastically and the channel deteriorates, the signal attenuation at the communication layer is compensated by reducing the spatial distance at the physical layer.

[0027] Preferably, after receiving the network channel attenuation penalty factor, the underwater vehicle cluster dynamically adjusts the transmission power of the next cycle using the power control model of the underwater acoustic communication unit;

[0028] The calculation of the transmission power includes the following related terms: the minimum operating sensitivity threshold of the underwater acoustic communication gateway, the physical slant distance between the single vehicle and the single unmanned vessel, the frequency-dependent seawater absorption attenuation coefficient calculated based on real-time water temperature and salinity, the penalty factor compensation gain coefficient, and the network channel attenuation penalty factor.

[0029] The underwater vehicle adjusts its energy output in real time by combining environmental feedforward parameters, avoiding bit error retransmissions caused by fixed power transmission in areas with high concentrations of suspended matter, while reducing energy consumption in areas with good water quality, thus optimizing the lifespan of the entire underwater vehicle cluster.

[0030] Preferably, the single vehicle is also equipped with a Doppler log or a forward-looking altimeter. After the underwater vehicle cluster receives the reverse trajectory control command, the controller in the single vehicle analyzes and extracts the three-dimensional coordinates of the geometric centroid and the radius of the influence boundary in the spatiotemporal cavity feature vector, interrupts the current mapping route, and generates a local parallel scanning path.

[0031] In the local parallel scanning path, the calculation correlation items for the horizontal and vertical spacing between adjacent parallel scanning lines include: the actual relative flight altitude measured by the Doppler log or the forward altimeter, the effective beam opening angle of the multibeam sonar, and the overlap rate threshold of adjacent scanning strips.

[0032] Flight path reconstruction technology can dynamically generate secondary mapping paths based on target influence boundaries issued by the ground control center. It calculates spacing using the actual sonar viewpoint and relative altitude relationship to ensure the uniformity of point cloud density in the supplementary data acquisition area.

[0033] Preferably, the individual UAV in the aerial UAV cluster includes a flight control computer. After receiving the reverse trajectory control command, the flight control computer extracts the three-dimensional coordinates of the geometric centroid in the spatiotemporal cavity feature vector and maps them to the latitude and longitude parameters of the target point, and extracts the parameters of the influencing boundary radius.

[0034] The flight control computer, in conjunction with the effective field-of-view model of the hyperspectral imager or multispectral camera, calculates the flight altitude for the reshoot based on the principle of spatial geometric cone projection, and generates a smooth transition route from the current spatial position to the target position.

[0035] The aerial node autonomously adjusts its flight altitude based on the projected geometry of the underlying void region. By adjusting the physical imaging altitude to fully cover the radius of the target's influence boundary, accurate supplementation of cross-media remote sensing data is achieved.

[0036] A second aspect of this invention provides a method for cross-domain robot collaborative remote sensing ad hoc networks, comprising the following steps:

[0037] The ground control center assigns detection mission instructions, and each network node uses a two-way ranging protocol to complete system time synchronization. When the underwater vehicle cluster acquires the water body profile physicochemical parameters and seabed mass cloud data, it performs spatial state filtering update and extracts spatial positioning covariance feature scalar.

[0038] The underwater vehicle cluster uploads a data frame containing the water body profile physicochemical parameters, the seabed mass cloud data, and the spatial positioning covariance feature scalar. The surface unmanned vessel cluster parses the feature scalar and allocates the data forwarding priority of the radio frequency communication gateway in the local queue according to the value of the feature scalar, and then aggregates the data to the ground control center.

[0039] The ground control center processes the optical and spectral data acquired by the aerial drone cluster to perform parameter inversion and calculate and distribute the network channel attenuation penalty factor. The surface unmanned vessel cluster and underwater vehicle cluster incorporate the network channel attenuation penalty factor into the calculation of the comprehensive routing metric or transmission power to adjust the spatial distribution distance or transmission energy compensation.

[0040] The ground control center receives the aggregated water body profile physicochemical parameters and the seabed bottom mass cloud data, performs a unified transformation of the discrete data points in the water body profile physicochemical parameters and the seabed bottom mass cloud data in the global coordinate system, and outputs the data fusion parameter field of the entire water area using interpolation and classification operations.

[0041] The ground control center calculates the root mean square error based on the estimated values ​​of the independent observation set and the data fusion parameter field, extracts the three-dimensional coordinates and influence radius of the region exceeding the error upper limit to generate a spatiotemporal cavity feature vector, and issues reverse trajectory control commands to drive the nodes to perform secondary data acquisition and supplementation.

[0042] The aforementioned method coordinates the sensing, communication, computing, and control processes across a three-layer structure encompassing the airspace, surface, and underwater domains. It utilizes upper-layer sensing parameters to compensate the lower-layer communication network and leverages lower-layer state parameters to adjust the data aggregation strategy. Combined with global quality assessment, this forms a reverse closed loop, maintaining the connectivity stability of the ad hoc network and the consistency of acquired parameters.

[0043] This invention provides a cross-domain robot collaborative remote sensing ad hoc network system and method. It has the following beneficial effects:

[0044] 1. This invention inverts the optical and spectral data acquired by an aerial UAV into a three-dimensional suspended matter concentration, and converts it into an acoustic signal attenuation penalty factor, which is then fed forward to surface and underwater nodes. This enables underwater vehicles to dynamically adjust their communication transmission power based on this factor, while simultaneously driving surface UAVs to physically maneuver towards the next hop node based on channel attenuation. By combining upper-layer optical sensing parameters with lower-layer acoustic communication control, and through reducing the spatial distance between nodes and compensating for transmission energy, the adverse effects of rapid changes in the hydrological environment on the underwater acoustic channel are offset, thus maintaining the stability of the cross-domain ad hoc network communication link.

[0045] 2. This invention utilizes the spatial positioning covariance feature scalar extracted by the underwater vehicle during the inertial navigation state filtering stage as the basis for determining data transmission priority. After the edge computing node on the surface unmanned vessel parses this scalar, it dynamically allocates the forwarding priority of data in the transmission buffer queue of the radio frequency communication gateway, establishing a direct mapping between the spatial positioning confidence of the underlying node and the queue scheduling of the network layer. This ensures that high-value observation data with high positioning accuracy is transmitted back to the ground control center before the positioning error of the underwater node accumulates and increases, thereby reducing the loss rate of effective detection data due to network congestion.

[0046] 3. After completing the global unified conversion and fusion reconstruction of multi-source data at the ground control center, this invention compares the root mean square error of the measured values ​​and the gridded estimated values ​​of the model with the verification set. It then extracts the three-dimensional coordinates and influence radius of the error-exceeding area to generate a spatiotemporal cavity feature vector. Finally, it issues reverse trajectory control commands to drive the UAV or underwater vehicle to generate a local scanning path and perform secondary data acquisition. This enables automatic identification and targeted supplementary detection of blind spots or abnormal areas, ensuring the integrity and consistency of the overall water area data fusion parameter field output by the system. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall architecture of a cross-domain robot collaborative remote sensing self-organizing network system according to the present invention;

[0048] Figure 2 This is a schematic diagram of the overall process of a cross-domain robot collaborative remote sensing self-organizing network method according to the present invention;

[0049] Figure 3 This is a line graph showing the evolution of the three-dimensional spatial data fusion error over time according to the present invention.

[0050] Figure 4 This is a comparison chart showing the relationship between network data delivery rate and changes in channel attenuation penalty factor according to the present invention. Detailed Implementation

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] See attached document Figure 1 The present invention provides a cross-domain robot collaborative remote sensing self-organizing network system, comprising: a ground control center, an aerial drone swarm, a surface unmanned vessel swarm, and an underwater vehicle swarm.

[0053] The ground control center is equipped with a central computing server and a wide-area communication antenna. The ground control center is used to decompose global tasks, receive multi-source remote sensing data and perform high-level data fusion processing, and is responsible for sending scheduling instructions to each execution domain. As the physical origin of the system's spatiotemporal reference, the ground control center provides the master clock reference and absolute geographic coordinate reference.

[0054] The aerial drone swarm establishes a communication connection with the ground control center through an air-to-ground radio broadband link. Each drone in the aerial drone swarm is equipped with a hyperspectral imager and a multispectral camera to acquire optical and spectral data of the water surface. The aerial drone swarm is also equipped with a radio frequency communication unit to support data transmission under line-of-sight or relay network conditions.

[0055] The surface unmanned surface vessel (USV) swarm is deployed on the water surface as a physical relay node for cross-media communication between air and sea. Each USV in the swarm is equipped with an edge computing node, a radio frequency (RF) communication gateway, and an underwater acoustic communication gateway. The edge computing node is used to perform local data caching, protocol parsing, and queue scheduling operations on the water surface. The USV swarm connects to the air-to-ground broadband radio link via the RF communication gateway, thereby connecting to the ground control center and the aerial USV swarm. It also establishes a cross-media narrowband link with underwater nodes via the underwater acoustic communication gateway.

[0056] The underwater vehicle swarm is deployed underwater to perform exploration missions. Each individual vehicle in the swarm is equipped with a multibeam sonar, an underwater acoustic communication unit, and an inertial navigation unit. The multibeam sonar is used to collect water profile physicochemical parameters and seabed mass cloud data. The inertial navigation unit is used to output basic underwater attitude estimation data. The underwater vehicle swarm is connected to a cross-medium narrowband link through the underwater acoustic communication unit.

[0057] At the physical level, the system constitutes a heterogeneous network topology environment that includes air-to-ground broadband radio links and cross-medium narrowband links. The air-to-ground broadband radio links rely on electromagnetic waves for signal transmission and have high availability bandwidth and low latency. The cross-medium narrowband links rely on sound waves for transmission in water and have low availability bandwidth and high transmission latency. The unmanned surface vessel cluster, combined with edge computing nodes and dual-mode communication gateways, realizes data aggregation and protocol conversion of asymmetric physical links. This physical topology architecture provides the hardware support for the system to implement an environment-feedforward network topology reconstruction mechanism and a confidence-driven data compression mechanism.

[0058] See attached document Figure 2 This invention provides a cross-domain robot collaborative remote sensing self-organizing network method, comprising the following steps:

[0059] The ground control center distributes detection mission instructions to the aerial drone cluster, surface unmanned vessel cluster, and underwater vehicle cluster, and performs joint deployment and unified spatiotemporal reference operation across the entire domain. Each network node uses a two-way ranging protocol and hardware clock to synchronize the system time. The underwater vehicle cluster combines externally issued underwater acoustic observations with its own inertial calculation data to perform spatial state filtering and update. During this operation, the underwater vehicle cluster calculates the local error covariance matrix and extracts the trace of the matrix as the spatial positioning covariance feature scalar at the current moment.

[0060] The underwater vehicle cluster performs source-end data compression and cross-domain scheduling operations. It compares the extracted spatial positioning covariance feature scalar with a preset tolerance threshold. When the positioning error causes the scalar to exceed the threshold, the cluster invokes a computing module to downsample the detected point cloud data. The data and corresponding feature scalar are then encapsulated into a data frame of a specified format and transmitted via an underwater acoustic link. The surface unmanned vessel cluster receives the cross-medium data frame and parses the feature scalar. Based on the feature scalar value, the cluster allocates data forwarding priority for the radio frequency communication gateway in its local queue.

[0061] The topology reconstruction operation based on remote sensing features is performed. The ground control center processes the hyperspectral images transmitted by the aerial UAV swarm, inverts the surface flow field and suspended matter parameters of the water area, and calculates the network channel attenuation penalty factor for specific water areas. When calculating the comprehensive routing metric of the next-hop forwarding node, the channel attenuation penalty factor is introduced as a calculation item for the surface UAV swarm and underwater vehicle swarm. When the factor value changes in a specific area, causing the routing metric value to decrease, the corresponding nodes adjust their spatial distribution distance to cope with the channel quality degradation.

[0062] Heterogeneous data hierarchical registration and fusion reconstruction operations are performed. The detection data acquired by each execution domain are aggregated to the ground control center via the communication network. The ground control center uses rigid body transformation matrix and inverse distance weighting algorithm to perform unified transformation of discrete data points in the global coordinate system. After spatial registration is completed, the ground control center performs interpolation and classification operations on the physical properties of water surface image, water body profile and bottom sediment, and outputs the data fusion parameter field of the whole water area.

[0063] The ground control center performs a reverse resampling operation based on fusion quality evaluation. It calculates the root mean square error using the independent observation set and the output parameter field estimate. When the calculation error in a specific area exceeds the system tolerance limit, the ground control center extracts the three-dimensional coordinates and influence radius of the area to generate a spatiotemporal cavity feature vector. Based on the feature vector, the ground control center sends a reverse trajectory control command to the front-end hardware. After receiving the command, the aerial UAV cluster or underwater vehicle cluster changes its current cruise route and moves to the coordinate area specified by the command to perform secondary data acquisition and supplementation.

[0064] This invention provides a specific implementation method for a ground control center to execute a cross-layer collaborative mechanism. The ground control center, as the computing node and spatiotemporal reference origin of the entire system, undertakes data registration, multi-source fusion and inversion, and inverse physical control functions. The specific implementation process of this function includes the following sub-steps.

[0065] The system performs global spatiotemporal reference alignment and spatial discrete data registration. The ground control center generates the system master clock signal via a deployed hardware network time protocol server and broadcasts this signal to each physical node via the communication link. Spatially, because sensors on different platforms have their own independent perspectives and local coordinates when acquiring data, the ground control center collects the local discrete data from these heterogeneous sensors and combines it with the pose data provided by the corresponding node's inertial navigation unit. A rigid body coordinate transformation matrix is ​​then used to uniformly transform the data points in the local sensor coordinate system to the global absolute geographic coordinate system. The coordinate transformation calculation formula is as follows:

[0066] ;

[0067] In the formula, Represents a three-dimensional spatial coordinate vector in the global coordinate system; Represents the three-dimensional direction cosine rotation matrix from the local sensor coordinate system to the global coordinate system; This represents the translation vector between the origins of two coordinate systems. This represents the original three-dimensional coordinate vector in the local sensor coordinate system.

[0068] After completing the rigid body transformation, the ground control center uses an inverse distance weighted interpolation algorithm to map discrete spatial points onto uniformly divided 3D grid nodes. The physical principle of this algorithm is based on spatial similarity; that is, the closer the measured data is to the target grid, the greater its influence weight on the target grid. The formula for calculating the grid node interpolation estimate is as follows:

[0069] ;

[0070] In the formula, Representing the Interpolation estimates for each grid node; Representing the One measured sample data value; Representing the The measured sample up to the first Euclidean space distance between grid nodes; Represents the power exponent of distance decay; This represents the total number of nearby valid samples involved in the interpolation calculation; Indicates the target grid point to be estimated Within the neighborhood, all The sum of the products of the actual measured values ​​of each known observation data point and its corresponding distance weight value; Indicates all The sum of the distance weights corresponding to the known observation data points is used to normalize the weighted sum of the numerator, ensuring that the sum of the weights of all observation points is 1. This represents the summation symbol. Normally, The value range is from 1 to 3, but in practical applications it is often set to 2 to ensure a smooth transition; The value is set by the system based on the limitations of computing resources, which determines the number of samples within a fixed search radius.

[0071] The process involves hyperspectral image inversion and channel attenuation feedforward parameter extraction. The ground control center receives surface hyperspectral image data from airborne nodes and uses an empirical ratio algorithm to invert the concentration distribution of suspended matter on the water surface. Simultaneously, this data is combined with discrete suspended matter data from the water profile collected by an underwater robot swarm. Layered and cross-layer multi-source data fusion is performed using three-dimensional spatial interpolation (such as three-dimensional kriging interpolation) to ultimately construct a three-dimensional suspended matter concentration distribution. Based on this, the ground control center establishes a mapping relationship between the physical environment and the communication channel. Since suspended particles in the medium increase sound wave scattering and absorption losses during underwater acoustic communication signal propagation, the ground control center, based on an acoustic absorption attenuation model, forward converts this three-dimensional suspended matter concentration distribution into signal attenuation coefficient values ​​for the corresponding frequency band, thereby generating a spatially continuous channel fading prediction field.

[0072] The ground control center maps and overlays the predicted field with the current three-dimensional spatial topology of each network node, pre-calculating the network channel attenuation penalty factor for each potential communication node in the network. Specifically, this penalty factor is the value obtained by normalizing the line integral of the signal attenuation coefficient at each point on the inter-node connection. This attenuation penalty factor characterizes the degree of physical channel quality degradation caused by suspended matter and waves in the water under specific three-dimensional coordinates. This parameter is then sent as a feedforward input to the topology reconstruction command to the surface and underwater network nodes via the downlink.

[0073] A hierarchical fusion and reconstruction operation of multi-scale heterogeneous data was performed. For different physical media levels, the ground control center used different data fusion models in parallel. At the above-water level, the ground control center employed principal component analysis (PCA) to fuse multispectral and high spatial resolution images. This operation aimed to inject the spatial geometric features of the high spatial resolution images into the multispectral images. The process included extracting the first principal component of the multispectral images, replacing it with histogram-matched high-resolution image data, and performing inverse principal component transformation to obtain a water surface remote sensing image with high spatial and hyperspectral resolution. At the water body profile level, to address the issue of uneven parameter distribution, the ground control center established a spatial semi-variogram using discretely extracted physicochemical parameters and reconstructed a continuous three-dimensional physicochemical parameter field using a three-dimensional ordinary kriging algorithm. The three-dimensional interpolation equation calculation formula is as follows:

[0074] ;

[0075] In the formula, Represents the optimal interpolation weight coefficients for ordinary Kriging; It represents the spatial semivariance function fitted based on measured data, and usually uses a spherical model or an exponential model to fit the variance relationship of measured point pairs; This represents the Lagrange multiplier used to minimize the variance of the estimation error; and A three-dimensional spatial coordinate vector representing different observation samples; The three-dimensional spatial coordinate vector representing the target mesh to be estimated; The summation symbol is used.

[0076] At the seabed sediment level, the ground control center extracts the backscattering intensity of multibeam sonar echoes and the water depth topographic elevation features. These feature sets are then input into a pre-trained support vector machine (SVM) classifier, which outputs a category label matrix for the seabed sediment. For the hyperplane partitioning and kernel function mapping of the SVM, those skilled in the art can employ conventional open-source machine learning frameworks. The process of optimizing the classification boundary is well-known in the field and will not be elaborated upon here.

[0077] Perform fusion field accuracy assessment and inverse trajectory control operations. After completing data fusion, the ground control center extracts an independent measured dataset from the total sample that did not participate in interpolation and training as a validation set. The ground control center compares the measured values ​​of the validation set with the gridded estimates of the model under the same spatial coordinates and calculates the root mean square error (RMSE), which reflects the overall model accuracy. The formula for calculating the RMSE is as follows:

[0078] ;

[0079] In the formula, This represents the root mean square error value within the calculation region; This represents the total number of independent observations contained in the validation set; Representing the The estimated values ​​of the fusion model corresponding to the coordinates of each verification point; Representing the The actual measured values ​​corresponding to the coordinates of each verification point; This indicates the residual calculated at a single point. Indicates the summation symbol; The square root operator is represented by the symbol. This indicates that the mean is calculated.

[0080] The ground control center compares the calculated root mean square error (RMSE) value with the system's preset tolerance error upper limit, which is determined by the specific requirements of the detection mission. When the calculated error in a particular sub-region exceeds the tolerance error upper limit, the ground control center determines that there is a lack of valid data or insufficient sensor sampling density in that region. The ground control center uses a density-based spatial clustering algorithm to extract the geometric centroid coordinates of the high-error connected region and the radius of the circumscribed sphere covering the connected region, combining them to generate a spatiotemporal cavity feature vector. The spatiotemporal cavity feature vector is defined as follows:

[0081] ;

[0082] In the formula, A column vector representing the spatiotemporal void feature vector; , and These represent the three-dimensional spatial coordinate components of the geometric centroid of the connected domain with the maximum error, respectively. This represents the radius of the influence boundary radiating outward from the center of mass; Delimiters for vectors or matrices; The transpose operator for a matrix.

[0083] The ground control center generates inverse trajectory control commands based on the extracted spatiotemporal cavity feature vectors. To quickly fill data gaps, these control commands bypass the regular periodic task scheduling queue and are assigned the highest system priority. The ground control center then uses the feature vectors... The target domain is determined by the positive and negative attributes of the coordinate components. When the elevation is greater than or equal to the zero water surface datum, the command will be routed to the aerial drone swarm via a radio broadband link; when When the elevation is below the zero water surface datum, the command is routed to the underwater vehicle cluster via the underwater acoustic communication network. After the command is issued to the execution node in the corresponding airspace or water area, it drives the hardware to move to the coordinate area specified by the feature vector to perform supplementary data acquisition tasks.

[0084] This invention provides a specific implementation method for aerial unmanned aerial vehicle (UAV) swarms to perform global remote sensing and cross-layer response. The aerial UAV swarm, acting as a data acquisition node and network topology extension node in a wide-area environment, undertakes the functions of surface optical data acquisition, feedforward route avoidance, and reverse trajectory reconstruction. The specific implementation process of the above functions includes the following sub-steps.

[0085] The system performs wide-area optical feature acquisition and broadband data transmission. The aerial UAV swarm conducts wide-area patrols according to a pre-set global mission route. When the onboard positioning module determines that the current spatial coordinates have entered the mission trigger area—that is, when the three-dimensional coordinates output by the onboard GPS intersect the pre-stored mission polygon boundary—the onboard flight control computer activates the mounted hyperspectral imager and multispectral camera. The onboard sensors acquire multi-band optical and spectral feature data of the water surface.

[0086] Due to the large volume of high-resolution optical data, airborne computing nodes perform frame segmentation, lossless compression, and protocol encapsulation on the original image sequence to generate continuous data stream messages. The airborne UAV swarm transmits these data stream messages to the ground control center via an airborne radio frequency communication unit and an air-to-ground broadband radio link. For frame compression encoding and underlying network transmission control of the image data, those skilled in the art can employ standard video stream coding standards; the encoding and serialization processes are well-known technologies in the field and will not be elaborated upon here.

[0087] The system performs feedforward parametric coupled network routing calculations and topology reconstruction. The airborne communication unit continuously listens for and receives feedforward control messages containing network channel attenuation penalty factors from the ground control center. When searching for the next-hop relay node in a multi-hop network environment, the edge computing nodes within the airborne UAV swarm introduce a perceptually coupled multi-factor network routing decision algorithm. The edge computing nodes change the conventional network layer's mechanism of relying solely on physical layer packet detection, directly incorporating the channel attenuation penalty factor obtained from environment inversion into the calculation framework of the comprehensive routing metric. For potential next-hop relay nodes, the comprehensive routing metric calculation formula is as follows:

[0088] ;

[0089] In the formula, Represents candidate next-hop nodes The comprehensive routing metric; Represents the current physical link quality calculated based on the received signal-to-noise ratio; Representative node Current normalized residual energy level; The data packet passes through the node The rate of progress in distance toward the target convergence node; Representative node Current network queue cache load ratio; Represents the node-specific feedforward injection from the ground control center. Network channel attenuation penalty factor at the spatial coordinates; to These represent the fixed weighting coefficients corresponding to each parameter. Before performing linear weighting, the edge computing nodes pre-calculate using a minimax normalization algorithm. , and Variables with different magnitudes and physical units are uniformly mapped to a dimensionless interval of 0 to 1. For to The values ​​of are all in the range of 0 to 1, and the sum of all weight coefficients equals 1.

[0090] The specific values ​​of the weighting coefficients are pre-initialized based on the emphasis placed on communication energy consumption, transmission delay, and link reliability in the actual detection mission. When a channel attenuation penalty factor is detected due to environmental degradation in a specific water area... When the value increases dramatically, the comprehensive routing metric of the affected nodes before actual packet loss occurs at the physical layer. This will be actively reduced. The underlying routing protocol recalculates the shortest path based on the degraded metric, controlling the flight nodes within the fleet to actively shorten the physical communication hop distance or detour through high-fading airspace, thereby achieving active obstacle avoidance and topology contraction of the communication link.

[0091] The system performs a closed-loop operation of reverse trajectory analysis and secondary supplementary sampling. When quality holes appear in the global data fusion, the airborne UAV swarm receives the reverse trajectory control command, which is given the highest system priority, via radio link. The onboard flight control computer analyzes the control command and extracts the spatiotemporal hole feature vector contained within the command. The onboard flight control computer maps the geometric centroid three-dimensional spatial coordinate components in the feature vector to the latitude and longitude parameters of the target point.

[0092] Simultaneously, the airborne flight control computer extracts the influence boundary radius parameter from the feature vector, and, combined with the field-of-view model of the airborne optical sensor, uses trigonometric functions to inversely calculate the reshoot flight altitude and hovering radius sufficient to cover the influence range of the hole. The calculation principle is based on spatial geometric frustum projection, and the specific calculation formula is as follows:

[0093] ;

[0094] in, To reshoot the flight altitude; The influence boundary radius parameter is extracted from the spatiotemporal cavity feature vector; The effective field of view for mounting optical sensors; Indicates half of the effective field of view; This represents the tangent function operator.

[0095] The airborne flight control computer interrupts the original routine cruise mission scheduling sequence at the hardware level, setting the parsed target location as the sole high-priority waypoint for the current control cycle. Using a Dubins curve or B-spline curve path planning algorithm based on airframe kinematic constraints, the flight control computer generates a smooth transition path from the current spatial location to the target location. After the flight platform reaches the designated area along the generated path, it reactivates the hyperspectral imager to perform secondary supplementary data acquisition, and prioritizes transmitting the newly acquired high-resolution remote sensing data back to the ground control center to complete the closed-loop repair of data gaps across the entire domain.

[0096] This invention provides a specific implementation method for surface unmanned surface vessel (USV) swarms to perform cross-media bridging and edge computing scheduling. The USV swarm is deployed at the water-air interface, serving as a relay node at the physical layer of the air-sea heterogeneous network, undertaking data aggregation, protocol parsing, and topology adaptive adjustment functions for asymmetric physical links. The specific implementation process of this function includes the following sub-steps.

[0097] The system performs physical bridging of heterogeneous gateways and cross-medium time delay compensation. Individual nodes within the surface unmanned vessel cluster hardware-connected the RF communication gateway and the underwater acoustic communication gateway via an internal high-speed data bus, with edge computing nodes acting as master control units mounted on this bus. Because the speed of sound in water is approximately 1500 m / s, far lower than the speed of electromagnetic waves in air, the underwater acoustic communication link exhibits high latency. Directly applying traditional one-way time synchronization protocols would result in a non-negligible time base offset.

[0098] Therefore, edge computing nodes employ a cross-medium bidirectional ranging protocol for time synchronization. The edge computing node sends a synchronization request message with a local timestamp to the underwater node via the underwater acoustic communication gateway, and records the message transmission time. The underwater node receives the message and, after a fixed processing delay, returns a response message. The edge computing node parses the response message, extracts the receive and transmit timestamps recorded by the underwater node, and, combined with its own recorded response message arrival timestamp, calculates the one-way propagation delay time of the underwater acoustic channel. The formula for calculating the propagation delay time is as follows:

[0099] ;

[0100] In the formula, Represents the one-way physical propagation delay time of the underwater acoustic link; The local timestamp representing the synchronization request message sent by the surface node; The local timestamp representing the underwater node receiving the request message; The local timestamp representing the response message sent by the underwater node; The local timestamp representing the response message received by the surface node; This represents the total round-trip time minus the internal processing time, which is used to calculate the total two-way flight time of a message in the underwater acoustic medium through pure physical transmission.

[0101] The calculation formula is based on the physical assumption that sound waves propagate at approximately equal speeds in both the downlink and uplink directions. The pure physical link delay is extracted by subtracting the node's internal processing time from the round-trip time. The edge computing node injects the calculated propagation delay as a time offset compensation term into its local system clock register to eliminate system-level time reference errors introduced by the slow propagation of sound waves.

[0102] Furthermore, the edge computing node sends a time message carrying the compensated absolute time to the target underwater node via an underwater acoustic link. The underwater node uses this message to reset its onboard underlying clock, thereby ensuring that the underlying sensor data it collects has an absolute timestamp consistent with that of the ground control center. Regarding the underlying hardware interrupt response mechanism for the underwater node to receive the message and extract the timestamp, those skilled in the art can implement it using a conventional digital signal processor capture module. Its signal interrupt handling is a well-known technology in the field and will not be elaborated upon here.

[0103] Cross-domain protocol parsing and covariance-driven data scheduling operations are performed. Edge computing nodes continuously monitor the receive buffer of the underwater acoustic communication gateway. When a cross-media data frame is received from an underwater node, the edge computing node performs protocol conversion. This process includes stripping the physical layer and data link layer frame headers of the underwater acoustic communication protocol and extracting the feature scalar field encapsulated in the front end of the data payload. This feature scalar is the trace of the spatial positioning covariance matrix extracted by the underwater node during spatial state filtering, characterizing the spatial uncertainty of the current underwater detection data.

[0104] Edge computing nodes map the extracted feature scalars to the Quality of Service (QoS) priority of that batch of data in the RF communication gateway's transmission queue. The physical principle behind this mapping mechanism is that a larger spatial positioning covariance indicates a greater spatial error in the underlying data; therefore, it needs to be assigned a higher transmission priority to ensure that this data is transmitted back to the ground control center for global calibration as early as possible. The mapping function formula is as follows:

[0105] ;

[0106] In the formula, This represents the calculated priority value of the radio frequency transmission queue. This represents the highest priority level supported by the RF communication gateway hardware. The specific value is determined by the differential service model of the RF communication protocol, and is usually 7 (representing the highest priority at the network control level). This is a preset priority scaling factor used to adjust the priority distribution gradient; its value is typically equal to... ; This represents a scalar of spatial localization covariance features extracted from the data frame. This represents the system's preset covariance tolerance threshold, which is determined by inverse calculation of the spatial positioning accuracy requirements of the specific detection task. This represents the floor operation; This represents the ratio of the actual error parameter to the tolerance threshold. This represents the minimum value operator.

[0107] After the calculation is completed, the edge computing nodes are based on The data payload of the core point cloud data will be stripped and repackaged into standard radio broadband network protocol messages, and inserted into the radio frequency transmission buffer queue of the corresponding priority. Data messages with higher covariance scalars will be given priority in channel time slots for transmission to the ground control center, thereby ensuring that data from high-error areas can be quickly transmitted back for global calibration.

[0108] The system performs channel attenuation feedforward response and two-dimensional topology contraction operations. The surface unmanned surface vessel (USV) swarm receives the channel attenuation penalty factor from the ground control center via an RF communication gateway. This factor is incorporated into the routing cost equation when calculating the air-sea cross-layer routing metric. When the surface environment parameters in a specific area deteriorate, causing the channel attenuation penalty factor to increase, the link maintenance metric of the corresponding surface nodes in that area decreases.

[0109] When this metric falls below a preset link-preservation threshold, the edge computing node triggers topology self-healing logic. The link-preservation threshold is determined based on the minimum physical layer receiver sensitivity required for the RF communication gateway receiver to correctly demodulate the signal. The edge computing node calculates the target spatial coordinate vector for shortening the physical communication distance based on the neighbor node location table. The formula for calculating the target coordinate vector is as follows:

[0110] ;

[0111] In the formula, This represents the target two-dimensional planar position vector set by the water surface node to shorten the communication jump distance; A two-dimensional planar position vector representing the current water surface node itself; A two-dimensional planar position vector representing a next-hop network node whose communication quality has deteriorated due to environmental degradation; The preset movement step constant represents a single topology contraction. The value of this constant is determined based on the product of the maximum speed of the ship's power system and the control cycle, and is usually taken in the range of 2 meters to 10 meters. The Euclidean norm of a vector; This represents the spatial direction vector pointing to the target node; This represents the actual straight-line distance between two nodes.

[0112] The edge computing nodes send the calculated target two-dimensional planar position vector to the motion controller at the bottom of the hull. The motion controller drives the catamaran thrusters to move the unmanned surface vessel towards the target coordinates, overcoming water flow interference. The physical principle behind this operation is to utilize the reduced free-space path loss caused by shortening the physical communication hop distance to offset the channel fading caused by suspended objects or waves, thus maintaining the connectivity of the cross-medium communication link.

[0113] This invention provides a specific implementation method for underwater vehicle swarms to perform deep-water three-dimensional exploration and reverse response. The underwater vehicle swarm is deployed along the water profile and bottom sediment layer, serving as the bottom-level exploration node of a global system, undertaking functions such as physicochemical parameter acquisition, topographic mapping, and underwater acoustic network topology reconstruction. The specific implementation process of this function includes the following sub-steps.

[0114] The underwater vehicle performs deep-water 3D environment detection and spatial state filtering. Equipped with a temperature, salinity, and depth (TDT) sensor and multibeam sonar, it sets its course based on elevation along the water profile. In deep-water environments where GPS signals are blocked, the underwater vehicle utilizes a combined airborne inertial navigation system and a Doppler log for navigation. To suppress the accumulated position drift error in dead reckoning over time, the underlying computation module employs an extended Kalman filter algorithm for spatial state estimation. During the time update phase, the prediction formula for the state covariance matrix is ​​as follows:

[0115] ;

[0116] In the formula, Representing the The spatial state prediction covariance matrix at time t; The Jacobian matrix representing the state transition matrix; Representing the The optimal estimated covariance matrix at time t; Represents the system process noise covariance matrix; The transpose operator for a matrix; The transpose of the state transition matrix is ​​used to satisfy the linear algebra dimension matching and symmetry requirements during covariance matrix propagation.

[0117] in, The nonlinear kinematic state equations of the underwater vehicle are obtained by partial linearization of the underlying computing module. The value matrix is ​​initialized based on the white noise variance parameters calibrated by the airborne inertial navigation system and Doppler log hardware. The physical principle of this formula is to transfer the error uncertainty of the previous moment to the current moment through the system dynamics model, and to add process noise caused by external environmental disturbances. The underwater vehicle extracts the trace of this predicted covariance matrix in real time as a characteristic scalar representing the uncertainty of the current position, providing a quantitative basis for the cross-media scheduling of subsequent data.

[0118] The underwater vehicle performs underwater acoustic data encapsulation and protocol upload operations. It extracts local features from the collected point cloud data and physicochemical parameters, and uses a voxel grid filtering algorithm to downsample and compress redundant point clouds to adapt to the extremely low bandwidth characteristics of the underwater acoustic channel. The underlying software combines the compressed data payload with the covariance feature scalar extracted by spatial state filtering. The underwater vehicle then encapsulates the combined data into underwater acoustic communication frames via an acoustic modem.

[0119] Specifically, the calculation module writes the covariance feature scalar into a custom extended field in the header of the underwater acoustic physical layer frame, and places the probed data in the data payload area. The underwater vehicle transmits this communication frame to the surface unmanned vessel swarm via the uplink through the underwater acoustic medium. For the orthogonal frequency division multiplexing modulation and channel coding of the underwater acoustic channel, those skilled in the art can use standard underwater acoustic communication protocol stacks for implementation. The underlying signal processing is well-known in the field and will not be described in detail here.

[0120] The underwater vehicle performs channel attenuation feedforward response and transmit power compensation operations. It receives a pre-calculated network channel attenuation penalty factor from the ground control center via the downlink acoustic link. This factor reflects the additional acoustic loss caused by underwater suspended matter or flow field changes obtained from environmental inversion. The underwater vehicle incorporates this penalty factor into the power control model of the airborne acoustic transmitter to dynamically adjust the transmit power for the next cycle. The formula for calculating the target transmit power is as follows:

[0121] ;

[0122] In the formula, This represents the calculated target acoustic emission power level; The minimum operating sensitivity threshold for the surface acoustic receiver; Represents the physical slant distance between the underwater vehicle and the surface node; The frequency-dependent seawater absorption attenuation coefficient; The penalty factor compensation gain coefficient set by the system; This represents the received network channel attenuation penalty factor; This represents the logarithmic term, which indicates the geometric spread and diffusion loss of sound waves propagating in underwater space. According to the physical model of spherical wave propagation, sound energy decreases logarithmically with increasing distance.

[0123] Of the parameters mentioned above, The arrival time of the timed detection frames sent between the underwater vehicle and the surface node is calculated through interactive measurement. The water temperature, salinity, and hydrostatic pressure parameters collected in real time by the airborne temperature, salinity, and depth gauge are extracted by the underlying calculation module and then substituted into Thorp's acoustic empirical formula for calculation. The value of is determined based on the power amplification linearity of the airborne acoustic transducer, typically ranging from 1.5 to 3.0. The physical principle of this calculation logic is to add a feedforward attenuation margin obtained from the inversion of macroscopic environmental parameters to the spread loss and absorption loss included in the standard sonar equations. Through this mechanism, the system increases the acoustic emission energy in advance before the actual physical link is lost, ensuring the connectivity reliability of the underwater acoustic network.

[0124] The underwater vehicle performs reverse trajectory analysis and blind spot coverage control. When the underwater vehicle receives a reverse trajectory control command from the ground control center, the onboard controller analyzes the command and extracts the spatiotemporal cavity feature vector. This feature vector contains the three-dimensional coordinates of the geometric centroid of the data-missing region and the radius of the influence boundary. The underwater vehicle interrupts the currently preset mapping route and calls the kinematic controller to generate a local parallel scanning path covering the cavity region. To ensure seamless coverage of the underwater topographic mapping, the physical spacing between adjacent scanning routes is calculated inversely based on the opening angle of the onboard multibeam sonar and the flight altitude. The formula for calculating the route spacing is as follows:

[0125] ;

[0126] In the formula, This represents the horizontal and vertical spacing between two adjacent parallel scanning lines. This represents the actual relative altitude of an underwater vehicle above the seabed. This represents the effective beam opening angle of a multibeam sonar transducer. The threshold representing the overlap rate of adjacent scan strips; This means dividing the effective beam opening angle into two, constructing a right-angled triangle geometric model based on the optical axis / acoustic axis perpendicularly downwards, and calculating the vertex angle of this right-angled triangle; This represents the tangent function operator; This means that by subtracting the required lateral overlap rate from the number 1, the effective lateral displacement ratio that the next route can actually advance outward is calculated, after deducting the overlapping redundancy.

[0127] in, Provided by real-time measurements from an airborne Doppler log in bottom-tracking mode or an airborne forward-looking altimeter. To ensure edge data feature matching and stitching quality, The value range is set to 0.3 to 0.5. The underwater vehicle generates a reciprocating parallel course based on the calculated spacing, driving the tail thruster and control surfaces to maneuver to the cavity area to perform secondary data acquisition. The newly acquired high-precision terrain and physicochemical parameter data are assigned the highest transmission priority and enter the underwater acoustic transmission queue, completing the physical layer closed-loop repair of the underwater data blind zone.

[0128] Specific application examples:

[0129] The cross-domain robot collaborative remote sensing self-organizing network system provided by this invention will be described below in conjunction with a specific application scenario. This embodiment selects a near-shore sewage outlet sudden pollution source tracing and three-dimensional monitoring project as the analysis object. This area includes a sea area with a length and width of 2000 meters and a maximum water depth of 100 meters. The operating environment is characterized by strong ocean current disturbances, sudden changes in underwater suspended particulate matter concentration, and fluctuating communication channel fading. In this embodiment, a maximum priority level is set for surface radio frequency communication. Covariance tolerance threshold Priority scaling factor Set the minimum operating sensitivity threshold for the underwater acoustic receiver. Penalty factor compensation gain coefficient Setting the effective beam opening angle of the multi-beam sonar The threshold for the overlap rate of the scanned strips .

[0130] Experimental preparation and procedure:

[0131] The system establishes the initial spatial topology of the cross-domain heterogeneous network nodes. The system loads the kinematic constraint parameters and sensor performance files of each physical node into the storage component, defining parameterized calculation matrices for communication hop distance, transmit power level, and reverse mapping route spacing. The global task allocation module issues the cruise path and records the baseline communication delay, baseline data delivery rate, and ideal mapping coverage under interference-free conditions.

[0132] The system initiates global time synchronization and data preprocessing. The surface unmanned surface vessel and underwater vehicle execute a cross-media two-way ranging protocol to acquire timestamp data. Within a specific time step synchronization cycle, the surface node sends a request message with the local timestamp. underwater node received message timestamp underwater response message timestamp The water surface received a response timestamp The system's edge computing nodes perform propagation delay time calculations:

[0133] ;

[0134] The system will Injected into the clock register as a deviation compensation term.

[0135] Edge computing nodes perform covariance-driven scheduling operations on underwater data transmission. When an underwater vehicle passes through a highly turbidity polluted area, the spatial positioning covariance feature scalar extracted by the onboard filtering model... After the unmanned surface vessel receives and analyzes the data, it substitutes it into the priority mapping model to calculate the priority of the radio frequency transmission queue.

[0136] ;

[0137] The system will prioritize point cloud data containing pollution mutation characteristics. Insert messages into the sending queue at a higher level, and prioritize their transmission over regular background messages.

[0138] The system performs channel attenuation feedforward response and topology shrinkage operations based on remote sensing characteristics. The ground control center inverts the suspended matter plume and calculates the channel attenuation penalty factor for the distribution network. The underwater vehicle invokes a power control model, combined with the currently calculated physical slant range. With absorption attenuation coefficient Calculate the target's transmit power:

[0139] ;

[0140] ;

[0141] The system synchronously triggers a two-dimensional topology contraction of the surface nodes, and the current coordinates of the surface vessel are shown. Next hop node coordinates Set the movement step size constant. Edge computing nodes calculate the target plane vector:

[0142] ;

[0143] The controller drives the thruster to move the surface vessel to... Coordinates are used to offset the loss of physical communication links.

[0144] Ground-based computing nodes perform inverse supplementary measurement calculations based on fusion quality assessment. For the void feature vectors generated by the interpolation connected components, the system issues control commands to the backup underwater vehicle. The actual relative navigation altitude is calculated from the bottom layer of the target water area. The airborne computing unit performs reverse mapping spacing calculations:

[0145] ;

[0146] ;

[0147] The underwater vehicle generates a local parallel resample path based on this spacing to perform closed-loop resampling.

[0148] Experimental verification data and comparison results:

[0149] By performing the above steps, the system obtained performance comparison data of cross-domain heterogeneous networks under different cooperative mechanisms. The experimental statistical results are shown in the table below:

[0150] Table 1. Performance of cross-domain collaborative observation networks under different mechanisms

[0151]

[0152] Note: In the table, a negative sign indicates a reduction in prediction error or a decrease in response delay, while a positive sign indicates an improvement in the overall delivery rate evaluation index.

[0153] Experimental results show that at long-term three-dimensional monitoring nodes for nearshore sewage discharge, the cross-medium data delivery rate improved by 39.5% under high fading conditions after considering feedforward environmental parameter coupling and adaptive topology reconstruction. In the specific detection process, the response delay for spatial data blind zone repair achieved a dramatic order-of-magnitude reduction (from 1440 minutes relying on the next day's periodic patrol mission to 11.5 minutes for immediate response), effectively capturing the nonlinear evolution characteristics of underwater blind zone physicochemical parameters in the spatial dimension under sudden pollution diffusion scenarios. Traditional static mechanisms, by forcibly applying a purely physical layer link breakage passive response and fixed mapping routes, mask the impact of rapid deterioration of the aquatic medium on communication attenuation, leading to significant data loss and model reconstruction distortion in high-turbidity areas.

[0154] Experimental conclusion:

[0155] See attached document Figure 3 With appendix Figure 4 Based on the data in Table 1, the experimental conclusions of the cross-domain robot collaborative remote sensing self-organizing network system provided by this invention are as follows:

[0156] The System Evaluation Center performed statistical analysis on the test set containing the pollution diffusion cycle, as shown in the attached figure. Figure 3 As shown, under the baseline condition without considering inverse resampling and adaptive path planning, the root mean square error curve of the global 3D data fusion of the entire system tends to flatten and stagnate after encountering underwater occlusion over time, eventually maintaining at 14.65%. After the system using the present invention incorporates spatiotemporal cavity feature extraction and inverse trajectory closed-loop update, the root mean square error curve exhibits a step decrease and stabilizes at 2.41%. Experimental data demonstrate that the present invention identifies and corrects 12.24% of the global reconstruction error, eliminating the low fidelity problem of the digital base caused by the traditional calculation model neglecting the underlying terrain occlusion and insufficient sensor density.

[0157] As attached Figure 4 As shown, the network management module extracted the connectivity maintenance characteristics of the system when dealing with severe channels by comparing the delivery rate line data of different protocols under high suspended matter concentration. In the harsh test area where the penalty factor increases in a stepwise manner, the data delivery rate curve calculated by this invention is consistently much higher than the baseline static routing curve. In this area, due to the introduction of the channel attenuation feedforward factor and topology contraction calculation, the advance compensation of transmit power and the physical reduction of node distance form a link-break protection mechanism in the spatial dimension, and the cross-medium data delivery rate is maintained at an average of over 90%.

[0158] In the later stages of monitoring, where the contamination flow field was severely diffused, the system's sensitivity to extracting and transmitting high-error anomaly data was dramatically enhanced due to the introduction of priority scheduling logic linked to spatial positioning covariance. The communication gateway detected a significant reduction in the loss rate of core features in this area. (Appendix) Figure 4 The distribution curves show that the routing contraction nodes, after considering environmental parameters and feedforward, exhibit more stable channel locking characteristics at the critical attenuation boundary, reflecting the actual computational behavior of the algorithm in actively self-healing feedforward when crossing the deterioration threshold of multi-medium heterogeneous channels.

[0159] In summary, the experimental results demonstrate the effectiveness of the technical solution of this invention. This invention achieves high-precision digital reconstruction of highly dynamic marine environmental processes by dynamically and physically correcting the underlying links of heterogeneous networks through feedforward parameter routing, combined with accurate remeasurement calculations for spatial data blind spots using an inverse trajectory control model. Experimental conclusions show that this invention can correct information fragmentation in complex channel fading and data acquisition blind spot environments, providing a highly reliable data foundation that conforms to the evolutionary characteristics of the physical environment for the construction of a global digital twin system.

Claims

1. A cross-domain robot collaborative remote sensing self-organizing network system, characterized in that, include: The ground control center provides the system's spatiotemporal reference and performs multi-source data fusion. It generates reverse trajectory control commands for areas where the error exceeds the limit after fusion calculation, drives nodes to perform local supplementary detection, receives surface optical data to invert the suspended matter concentration, and converts the suspended matter concentration into a channel attenuation penalty factor. An aerial drone swarm connects to the ground control center via an air-to-ground broadband link, collects surface optical data, and transmits it back to the ground control center. The unmanned surface vessel swarm bridges the air-to-ground broadband link and the cross-medium narrowband link through the radio frequency communication gateway and the underwater acoustic communication gateway. It is equipped with edge computing nodes. The edge computing nodes establish a mapping between the spatial positioning confidence of the underlying nodes and the network queue, prioritize forwarding high-value data, and perform maneuvering approximation to the next hop communication node based on the feedforward channel attenuation penalty factor. An underwater vehicle cluster accesses the cross-medium narrowband link to perform underwater detection, extracts the spatial positioning covariance feature scalar of the inertial navigation unit as the positioning confidence, and dynamically adjusts the transmission power of the underwater acoustic communication unit in combination with the feedforward channel attenuation penalty factor.

2. The cross-domain robot collaborative remote sensing self-organizing network system according to claim 1, characterized in that, The ground control center will extract independent measured datasets that did not participate in interpolation from the total sample as the measured values ​​of the validation set, and compare the measured values ​​of the validation set with the gridded estimated values ​​of the model under the same spatial coordinates to calculate the root mean square error. When the root mean square error of the target area exceeds the upper limit of the tolerance error, the three-dimensional spatial coordinate components of the geometric centroid of the target area and the radius of the influence boundary are extracted and combined to generate a spatiotemporal cavity feature vector. The ground control center generates reverse trajectory control commands based on the spatiotemporal cavity feature vector, and routes the reverse trajectory control commands via the air-to-ground broadband link or the cross-medium narrowband link to the airborne UAV cluster in the corresponding airspace or the underwater vehicle cluster in the corresponding water area for secondary data acquisition based on the elevation reference attribute in the three-dimensional spatial coordinate components of the geometric centroid.

3. The cross-domain robot collaborative remote sensing self-organizing network system according to claim 1, characterized in that, The ground control center converts the suspended matter concentration into the signal attenuation coefficient value of the corresponding frequency band and generates a spatially continuous channel fading prediction field. It then maps and overlaps the channel fading prediction field with the three-dimensional spatial topology of each node in the system, and normalizes the line integral of the signal attenuation coefficient of each point on the connection between nodes to calculate the network channel attenuation penalty factor. The ground control center sends the network channel attenuation penalty factor as a feedforward input to the topology reconstruction command to the surface unmanned vessel cluster and the underwater vehicle cluster.

4. The cross-domain robot collaborative remote sensing self-organizing network system according to claim 1, characterized in that, The edge computing node sends a synchronization request message with a local transmission timestamp through the underwater acoustic communication gateway, and receives a response message containing the reception timestamp and transmission timestamp of the individual underwater vehicles in the underwater vehicle cluster. The edge computing node combines the arrival timestamp of the response message it records with the difference between the round-trip time of the message and the internal processing time of the single vehicle to calculate the one-way physical propagation delay time of the underwater acoustic channel. The one-way physical propagation delay time is injected into the local system clock register as a time deviation compensation item to obtain the absolute time. The time message carrying the absolute time is sent to the single vehicle through the cross-medium narrowband link to reset the airborne underlying clock.

5. A cross-domain robot collaborative remote sensing self-organizing network system according to claim 1, characterized in that, The individual underwater vehicle in the underwater vehicle cluster contains a low-level computing module. The low-level computing module receives the measurement data of the inertial navigation unit. During the time update stage of the spatial state filtering update, it extracts the trace of the spatial state prediction covariance matrix to generate a spatial positioning covariance feature scalar and writes the spatial positioning covariance feature scalar into the frame header of the underwater acoustic communication frame for transmission. The edge computing node receives the underwater acoustic communication frame and extracts the spatial positioning covariance feature scalar. Based on the ratio of the spatial positioning covariance feature scalar to the system's preset covariance tolerance threshold, and combined with the upper limit of the highest priority level supported by the radio frequency communication gateway, it calculates and allocates the service quality priority of the data in the transmission buffer queue of the radio frequency communication gateway, and sends the data to the ground control center via the radio frequency communication gateway with priority according to the service quality priority.

6. A cross-domain robot collaborative remote sensing self-organizing network system according to claim 3, characterized in that, When calculating the air-sea cross-layer routing metric, the edge computing node incorporates the network channel attenuation penalty factor into the calculation framework of the comprehensive routing metric. When the comprehensive routing metric is lower than the preset link hold threshold, the edge computing node calculates the target two-dimensional plane position vector. The calculation of the target two-dimensional plane position vector is based on the two-dimensional plane position vector of the individual unmanned vessel in the surface unmanned vessel cluster, the two-dimensional plane position vector of the next-hop communication node, and a preset movement step constant. The node then drives the motion controller in the individual unmanned vessel to maneuver toward the target two-dimensional plane position vector.

7. A cross-domain robot collaborative remote sensing ad hoc network system according to claim 3, characterized in that, After receiving the network channel attenuation penalty factor, the underwater vehicle cluster dynamically adjusts the transmission power of the next cycle using the power control model of the underwater acoustic communication unit. The calculation of the transmission power includes: the minimum operating sensitivity threshold of the underwater acoustic communication gateway, the physical slant distance between a single underwater vehicle in the underwater vehicle cluster and a single unmanned surface vessel in the surface vessel cluster, the frequency-dependent seawater absorption attenuation coefficient calculated based on real-time water temperature and salinity, the penalty factor compensation gain coefficient, and the network channel attenuation penalty factor.

8. A cross-domain robot collaborative remote sensing ad hoc network system according to claim 2, characterized in that, The individual underwater vehicles in the cluster are also equipped with ranging sensors and multibeam sonar. After receiving the reverse trajectory control command, the controller in the individual vehicle analyzes and extracts the three-dimensional coordinates of the geometric centroid and the radius of the influence boundary in the spatiotemporal cavity feature vector, interrupts the current mapping route, and generates a local parallel scanning path. In the local parallel scanning path, the calculation correlation items for the horizontal and vertical distance between adjacent parallel scanning lines include: the actual relative flight altitude measured by the ranging sensor, the effective beam opening angle of the multibeam sonar, and the overlap rate threshold of adjacent scanning strips.

9. A cross-domain robot collaborative remote sensing self-organizing network system according to claim 2, characterized in that, The individual UAVs in the aerial UAV cluster are equipped with hyperspectral imagers or multispectral cameras and include flight control computers. After receiving the reverse trajectory control command, the flight control computer extracts the three-dimensional coordinates of the geometric centroid in the spatiotemporal cavity feature vector and maps them to the latitude and longitude parameters of the target point, and extracts the parameters of the influence boundary radius. The flight control computer, in conjunction with the effective field-of-view model of the hyperspectral imager or multispectral camera, calculates the flight altitude for the reshoot based on the principle of spatial geometric cone projection, and generates a smooth transition route from the current spatial position to the target position.

10. A method for cross-domain robot cooperative remote sensing ad hoc networking, implemented based on the cross-domain robot cooperative remote sensing ad hoc networking system described in claims 1-9, characterized in that, Includes the following steps: The ground control center assigns detection mission instructions, and each network node uses a two-way ranging protocol to complete system time synchronization. When the underwater vehicle cluster acquires water body profile physicochemical parameters and seabed mass point cloud data, it performs spatial state filtering update and extracts spatial positioning covariance feature scalar. The underwater vehicle cluster uploads a data frame containing the water body profile physicochemical parameters, the seabed mass cloud data, and the spatial positioning covariance feature scalar. The surface unmanned vessel cluster parses the feature scalar and allocates the data forwarding priority of the radio frequency communication gateway in the local queue according to the value of the feature scalar, and then aggregates the data to the ground control center. The ground control center processes the surface optical data acquired by the aerial drone cluster to perform parameter inversion and calculate and distribute the network channel attenuation penalty factor. The surface unmanned vessel cluster and underwater vehicle cluster incorporate the network channel attenuation penalty factor into the calculation of the comprehensive routing metric or transmission power to adjust the spatial distribution distance or transmission energy compensation. The ground control center receives the aggregated water body profile physicochemical parameters and the seabed bottom mass cloud data, performs a unified transformation of the discrete data points in the water body profile physicochemical parameters and the seabed bottom mass cloud data in the global coordinate system, and outputs the data fusion parameter field of the entire water area using interpolation and classification operations. The ground control center calculates the root mean square error based on the measured values ​​of the validation set and the model gridded estimate of the data fusion parameter field. It extracts the three-dimensional coordinates and influence radius of the region exceeding the tolerance error upper limit to generate a spatiotemporal cavity feature vector. Based on the spatiotemporal cavity feature vector, it issues reverse trajectory control commands to drive the nodes to perform secondary data acquisition and supplementation.

Images