Non-line-of-sight scene based mobile terminal positioning system and method using external radio frequency load of unmanned aerial vehicle

Abstract

The application discloses a mobile terminal positioning system and method based on an unmanned aerial vehicle external radio frequency load in a non-line-of-sight scene, and belongs to the technical field of mobile terminal radio frequency signal detection, which comprises an unmanned aerial vehicle platform, an aerial sensing load, a ground receiving station and a ground command workstation; the aerial load passively intercepts a mobile phone WiFi sniffing frame, extracts radio frequency features and binds unmanned aerial vehicle space-time pose data, and returns to the ground through a low-frequency long-distance link; the ground workstation adopts a thread decoupling architecture, constructs a space-time physical fingerprint aggregation random MAC sampling point, establishes a weighted residual error model to realize multi-point positioning, determines observation geometry through GDOP and other indexes, and automatically switches a single-point peak degradation model when the geometry is degraded; and the radio frequency thermal field is constrained by relying on an unmanned aerial vehicle flight path segmentation mask. The application does not need active cooperation of a terminal, solves technical problems such as MAC randomization, positioning drift, thermal misjudgment and interface lag, and hierarchically outputs a visual search and rescue situation with confidence, thereby improving search and rescue reliability in a complex shielding environment.

Description

Technical Field

[0001] This invention relates to the field of mobile terminal radio frequency signal detection technology, and in particular to a mobile terminal positioning system and method based on an external radio frequency payload of a UAV in non-line-of-sight scenarios. Background Technology

[0002] In emergency rescue scenarios such as building collapses, landslides, dense forest cover, and thick smoke, trapped personnel are often completely obscured by rock, concrete, or vegetation. Conventional visible light and infrared detection equipment rely on line-of-sight propagation, and their signals become completely ineffective in such obstructed environments, making target searching difficult. To address this, the industry has developed drone-based WiFi radio frequency search and rescue technology, which relies on the penetrating characteristics of mobile terminal wireless signals to achieve non-line-of-sight detection. However, existing mainstream solutions suffer from several unavoidable technical flaws.

[0003] Current mainstream WiFi search and rescue methods employ an active inducement mode, where drones simulate wireless access point broadcast signals to induce mobile phones to initiate scanning and access interactions to collect radio frequency data. This method heavily relies on the active response behavior of the terminal. With the upgrading of mobile phone privacy protection and power-saving sleep mechanisms, terminals will actively limit the frequency of WiFi scanning. In low-power and complex electromagnetic interference environments, it is difficult to output stable response signals, making it impossible to support long-term, large-scale continuous search and rescue operations.

[0004] Meanwhile, modern mobile terminals generally enable MAC address randomization mechanisms, and devices periodically change the MAC address of the sniffing frame source. Existing systems only use MAC as a unique target identifier, which cannot associate different random addresses with the same terminal. This results in a large number of short-term false targets, causing trajectory breaks, an inflated number of targets, severely damaging the ability to continuously track targets, and interfering with rescuers' judgment of the size and activity range of trapped people.

[0005] The positioning calculation suffers from a dual error problem: firstly, multipath and occlusion effects caused by rubble and walls significantly distort the RSSI signal, leading to huge ranging errors when using the free space loss model to calculate distance; secondly, the reciprocating straight flight path of UAVs easily results in concentrated sampling point distribution, narrow observation angles, ill-conditioned observation geometry, and exceeding GDOP limits, making multi-point positioning solutions prone to coordinate drift and divergence. Current technologies lack a geometric degradation judgment mechanism, directly outputting distorted positioning results that can easily mislead search and rescue directions.

[0006] In the radio frequency (RF) thermal visualization stage, existing systems directly interpolate discrete sampling points globally to generate heat maps without binding them to the constraints of actual UAV flight paths. Blank areas far from the flight path and without measured data will generate false signal hotspots, resulting in low reliability of the heat maps and increasing the risk of on-site misjudgments. In addition, existing ground command software performs data parsing, positioning calculation, thermal rendering, and interface refresh simultaneously. The massive amounts of RF data transmitted at high frequencies are prone to causing interface lag and situation update delays, making it difficult to guarantee the real-time decision-making needs of rescue operations.

[0007] Therefore, there is an urgent need for a mobile terminal positioning system and method for UAV external radio frequency payloads that uses passive listening as the front-end acquisition method and randomized address aggregation, geometric degradation adaptive degradation, and physical trajectory constraint visualization as the core processing mechanism, in order to improve the continuity of target recognition, the stability of positioning output, and the reliability of heat map in non-line-of-sight search and rescue environments. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a mobile terminal positioning system and method based on an external radio frequency payload of a UAV in non-line-of-sight scenarios.

[0009] The technical solution adopted by the present invention to solve its technical problem is: a mobile terminal positioning system based on an external radio frequency payload of a UAV in a non-line-of-sight scenario, including a UAV platform, an airborne sensing payload, a ground receiving station and a ground command workstation;

[0010] The aerial sensing payload is mounted on an unmanned aerial vehicle (UAV) platform, and the aerial sensing payload establishes a wireless communication connection with the ground receiving station via a low-frequency long-range wireless link; the ground receiving station is connected to the ground command workstation.

[0011] The airborne sensing payload is used to passively intercept WiFi sniffing management frames sent by mobile terminals, extract radio frequency feature information, synchronously bind the spatial pose information and time information of the UAV platform, encapsulate and generate a return message and send it to the ground receiving station; the ground receiving station is used to receive and demodulate the return message and inject the structured message data into the ground command workstation.

[0012] The ground command workstation is used to perform randomized MAC address entity cluster aggregation, weighted positioning calculation, observation geometry degradation determination, and spatial radio frequency scalar field reconstruction on the injected message data, and output visualized search and rescue results with confidence status.

[0013] Furthermore, the airborne sensing payload includes a wireless sniffing main control module, a positioning and timing module, a low-frequency long-range communication module, an independent power supply module, a storage / cache module, and an antenna unit;

[0014] The independent power supply modules provide independent power to the wireless sniffing main control module, the positioning and timing module, the low-frequency long-range communication module, the storage / caching module, and the antenna unit. The positioning and timing module is connected to the wireless sniffing main control module and is used to transmit spatiotemporal reference data to the wireless sniffing main control module. The wireless sniffing main control module is connected to the low-frequency long-range communication module and is used to transmit the encapsulated return message. The wireless sniffing main control module is electrically connected to the storage / caching module and is used to temporarily store the sampling data that was not successfully returned. The antenna unit is radio frequency connected to the wireless sniffing main control module and the low-frequency long-range communication module respectively to complete WiFi signal reception and low-frequency wireless data transmission.

[0015] Furthermore, the radio frequency feature information extracted by the wireless sniffing master control module includes the source MAC address, received signal strength RSSI, WiFi channel, reception time, and signal-to-noise ratio SNR; the wireless sniffing master control module periodically performs frequency hopping scanning according to a preset channel strategy, only capturing WiFi sniffing management frames spontaneously sent by the mobile terminal, and does not actively transmit wireless signals to induce terminal interaction;

[0016] The spatial pose and time information output by the positioning and timing module includes the longitude, latitude, altitude, and timestamp of the UAV; the positioning and timing module synchronously outputs spatial pose and time data to the wireless sniffing main control module.

[0017] The positioning and timing module adopts any one of GNSS positioning module, RTK high-precision positioning module or inertial navigation fusion positioning module; the low-frequency long-distance communication module adopts any one of LoRa communication, narrowband data transmission or self-organizing network low-speed wireless communication standard.

[0018] Furthermore, the ground command workstation includes a main display and interaction thread and a background asynchronous calculation thread;

[0019] The main display and interaction threads independently execute human-computer interaction, UAV trajectory rendering, entity cluster list display, thermal field visualization, and alarm result output.

[0020] The background asynchronous solution thread independently performs message parsing, randomized MAC address aggregation, weighted location solution, GDOP geometric degradation index calculation, and spatial radio frequency scalar field reconstruction.

[0021] The main display and interaction thread and the background asynchronous calculation thread complete data interaction through a thread-safe queue.

[0022] Furthermore, the specific logic of the background asynchronous calculation thread performing MAC address derandomization aggregation is as follows: randomized MAC addresses are distinguished by the local management bits of the MAC address, a multi-dimensional spatiotemporal physical fingerprint feature vector containing time difference, RSSI deviation, spatial displacement, channel consistency, trajectory continuity, and sampling quality is constructed, the matching score between the sampling point and the active entity cluster is calculated by weighting, and the sampling point is determined based on the dual conditions of the score threshold and the difference between the second-best scores, and the entity cluster is merged or newly created.

[0023] The specific logic of the background asynchronous solution thread for weighted localization and geometric degradation determination is as follows: filter effective independent sampling points within the entity cluster, construct a weighted matrix based on sampling quality and spatial distribution, establish a weighted residual objective function and iteratively solve for candidate target positions; calculate the geometric degradation index, and if the geometric degradation index does not exceed the limit, output a high-confidence multi-point localization result; if the geometric degradation index exceeds the limit, switch to the single-point peak strong feature degradation model to output a fuzzy localization result and mark the low-confidence state.

[0024] The specific logic of the background asynchronous calculation thread for reconstructing the spatial radio frequency scalar field and physical trajectory backoff masking is as follows: a virtual grid of the target region is established based on discrete sampling points of the entity cluster, and an initial radio frequency scalar field is generated through spatial interpolation; the minimum distance from each grid point to the complete set of UAV sampled tracks is calculated, and a piecewise physical trajectory backoff masking function is constructed; the initial radio frequency scalar field is multiplied by the masking function to obtain a constrained reliable thermodynamic field, and the signal extrema are searched only in the effective masking region as candidate target positions.

[0025] This invention also provides a mobile terminal positioning method based on UAV external radio frequency payload in non-line-of-sight scenarios. The mobile terminal positioning method using the aforementioned mobile terminal positioning system based on UAV external radio frequency payload in non-line-of-sight scenarios is divided into: air-to-ground data link and message encapsulation process, MAC address de-randomization and aggregation process, positioning calculation and GDOP adaptive degradation process, and spatial radio frequency scalar field reconstruction and physical trajectory backoff masking process.

[0026] The air-to-ground data link and message encapsulation process is used to complete WiFi radio frequency signal acquisition, spatiotemporal tag binding, standardized message encapsulation, and low-frequency air-to-ground wireless backhaul in the air, and output structured radio frequency sampling data to the ground end.

[0027] The MAC address derandomization aggregation process is used to construct a multidimensional spatiotemporal physical fingerprint for the discrete sampling points corresponding to randomized MAC addresses, and to merge multiple different random MAC sampling points of the same physical mobile terminal into a single device entity cluster, so as to eliminate false targets caused by MAC randomization.

[0028] The positioning solution and GDOP adaptive degradation process are used to perform weighted multi-point positioning solution on the sampling points of the entity cluster, determine the observation quality of the sampling track through multiple geometric degradation indices, and automatically switch to the single-point peak degradation positioning model when geometric degradation occurs.

[0029] The spatial radio frequency scalar field reconstruction and physical trajectory backoff masking process is used to generate a continuous radio frequency thermal field based on discrete sampling points, construct a masking function based on the actual flight path of the UAV to suppress false hotspots in unsampled areas, and output a reliable thermal extreme value as a reference for target auxiliary positioning.

[0030] Furthermore, the air-to-ground data link and message encapsulation process includes the following steps:

[0031] S101, Intercepting WiFi sniffing management frames: The wireless sniffing master control module of the airborne sensing payload passively listens to and intercepts WiFi sniffing management frames spontaneously sent by surrounding mobile terminals.

[0032] S102. Extract target feature information: The wireless sniffing main control module extracts radio frequency feature information from the WiFi sniffing management frame. The radio frequency feature information includes the source MAC address, received signal strength RSSI, WiFi channel and reception time.

[0033] S103. Read UAV pose information: The wireless sniffing main control module reads the UAV pose information output in real time by the positioning and timing module. The UAV pose information includes the UAV's longitude, latitude, altitude, and time information.

[0034] S104. Perform spatiotemporal tag binding: The wireless sniffing main control module performs spatiotemporal tag binding between the radio frequency feature information obtained in step S102 and the UAV pose information obtained in step S103 to generate sampling data;

[0035] S105. Forming a wireless return message: The wireless sniffing main control module encapsulates the bound sampling data into an extended return message;

[0036] S106. Transmission via low-frequency long-distance wireless link: The low-frequency long-distance communication module transmits the extended return message to the ground receiving station over a long distance via the low-frequency long-distance wireless link;

[0037] S107. Ground receiving station receives and demodulates: The ground receiving station receives the radio frequency signals transmitted back from the air and demodulates them into a structured data stream;

[0038] S108, Injecting into Ground Command Workstation 301: The ground receiving station injects the structured data stream into the ground command workstation, and the background asynchronous processing thread reads the structured data stream to carry out subsequent processes.

[0039] Furthermore, the MAC address derandomization and aggregation process includes the following steps:

[0040] S201. Receive and standardize air-to-ground backhaul data packets: The background asynchronous processing thread receives the structured data stream injected by the ground receiving station, performs integrity verification and field parsing, and obtains structured sampling points that include at least the source MAC address, received signal strength RSSI, WiFi channel, reception time, UAV longitude, latitude and altitude.

[0041] S202. Determine if the source MAC address is a randomized MAC: Determine if the source MAC address of the current sampling point is a randomized MAC address based on the local management bits of the MAC address. If the local management bits of the MAC address are not set, then treat it as a relatively stable device identifier; if the local management bits are set, then continue with the following steps.

[0042] S203. Establishing the spatiotemporal physical description of the sampling points: For each sampling point, the system establishes a sampling description item: ;

[0043] in, Indicates the source MAC address. Indicates the receiving time. This indicates that the UAV sampled three-dimensional coordinates. Indicates the received signal strength. Indicates the WiFi channel. Indicates the sampling quality index;

[0044] S204. Maintain the real device entity cluster; the system maintains an entity cluster C for each presumed real mobile terminal. j Single entity cluster C j The stored content includes historical sampling queues, the most recent update time, the RSSI statistical mean, representative spatial coordinates, channel distribution, trajectory change trends, entity confidence, and lifecycle status;

[0045] S205. Generate a candidate entity cluster set: If the current sampling point corresponds to a randomized MAC address, traverse the entity clusters within the preset active time window and add only the entity clusters whose spatial distance, most recent update time, and sampling quality meet the preset coarse screening conditions to the candidate entity cluster set.

[0046] S206. Construct a multi-dimensional spatiotemporal physical fingerprint feature vector: For the current sampling point R k With candidate entity cluster C j Construct a multidimensional spatiotemporal physical fingerprint feature vector:

[0047] ;

[0048] in, Indicates the sampling time difference; Indicates RSSI intensity deviation; Indicates the difference in spatial coordinates; Indicates the channel consistency difference; Indicates deviation in trajectory continuity; Indicates the difference in sampling quality;

[0049] S207. Normalize and penalize the features: Normalize and segmented penalize each component of the multidimensional spatiotemporal physical fingerprint feature vector, and map the original difference to the same direction matching score component.

[0050] S208. Calculate the candidate matching score: Calculate the matching score between the current sampling point and a single candidate entity cluster according to the preset weights. The calculation formula is as follows:

[0051] ;

[0052] In the formula, , , , , , These are the preset weighting coefficients corresponding to the dimensions of time difference, RSSI deviation, spatial distance, channel consistency, trajectory continuity, and sampling quality, respectively.

[0053] The lower the score, the higher the probability that the current sampling point and the candidate entity cluster belong to the same terminal;

[0054] S209. Dual-condition judgment based on execution threshold and suboptimal gap: The optimal matching entity cluster is selected using a dual-matching condition judgment rule. The judgment condition is: the optimal matching score satisfies... The difference between the best and second-best matching scores satisfies ;

[0055] If both of the above matching conditions are met, the current sampling point will be merged into the optimal candidate entity cluster.

[0056] If only the optimal matching score condition is met, but the difference between the optimal and second-best matching scores is not met, then the sampling point is marked as a sample to be confirmed. The sample to be confirmed is written into the confirmation cache, and the matching score is recalculated in combination with the subsequently added sampling points within the preset confirmation time window. When the double matching condition is met again, it is merged into the optimal candidate entity cluster. If the double matching condition is not met when the confirmation time window expires, a new device entity cluster is created with the sample to be confirmed.

[0057] If neither of the two matching conditions is met, a new device entity cluster is created with the current sampling point;

[0058] S210. Update entity cluster status and confidence: If the double matching condition judgment in step S209 passes, merge the current sampling point into the corresponding real device entity cluster, and synchronously update the entity cluster's historical sampling queue, RSSI statistics, representative coordinates, channel distribution, trajectory change trend, latest update time, and entity confidence.

[0059] S211. Create, freeze, or eliminate entity clusters: If step S209 determines that there are no candidate entity clusters that meet the double matching conditions, the current sampling point is used as the initial sampling data to create a new device entity cluster; for entity clusters that have not been updated for more than a preset time, their status is switched from active to frozen; for entity clusters that have been frozen for more than a preset time and whose confidence level is lower than a preset threshold, they are eliminated from the active candidate set, and only historical records are retained for task replay.

[0060] S212: Output Entity Cluster Cache Queue: Pushes the updated entity cluster cache queue of all devices to the background asynchronous calculation thread for subsequent location calculation, GDOP judgment and spatial field reconstruction.

[0061] Furthermore, the localization solution and GDOP adaptive degradation process includes the following steps:

[0062] S301. Read sampling points from the entity cluster cache queue: The background asynchronous calculation thread periodically reads sampling points from the entity cluster cache queue; each sampling point includes at least the UAV coordinates, received signal strength RSSI, timestamp, sampling quality index and corresponding entity cluster identifier;

[0063] S302. Determine whether the number of valid independent sampling points has reached the set threshold. If the number is not met, do not execute the formal localization solution for the time being, and wait for the addition of new sampling points until the number of valid independent sampling points reaches the set threshold.

[0064] S303. Constructing the weighting matrix: For each valid independent sampling point, calculate the single-point weight based on the received signal strength RSSI, time freshness, spatial distribution, and sampling quality. The weighted matrix W is composed of all the individual point weights and forms a diagonal weighted matrix.

[0065] The weight expression for a single valid independent sampling point is:

[0066] ;

[0067] In the formula, , , , These are preset fixed weighting coefficients corresponding to four dimensions: RSSI intensity, sampling freshness, sampling quality, and spatial distribution. For the received signal strength mapping function, This is a sampling time freshness mapping function. The sampling quality mapping function, A mapping function for the contribution of the spatial distribution of sampling points;

[0068] The weighting matrix W is a diagonal matrix, and the elements on the diagonal are the weights corresponding to each valid independent sampling point. All values ​​in the other off-diagonal positions of the matrix are set to zero;

[0069] S304. Constructing the weighted residual objective function: Establish the residual objective function based on the relationship between the coordinates of the effective independent sampling points and the coordinates of the candidate target; the residual for the i-th effective independent sampling point can be expressed as:

[0070] ;

[0071] Where p represents the coordinates of the target object to be determined. Indicates the sampling coordinates of the UAV. This represents the equivalent constraint formed by RSSI or other radio frequency characteristics;

[0072] Integrating the residuals of all valid independent sampling points, the weighted objective function is constructed as follows: ;

[0073] S305. Perform multi-point localization solution: The background asynchronous solution thread solves the weighted objective function using a numerical optimization algorithm to obtain candidate location solutions p. hat ;

[0074] S306. Construct the observation geometry matrix: Solve for candidate positions p hat Construct the observation Jacobian matrix H, and then construct the information matrix based on the Jacobian matrix H and the weight matrix W. ;

[0075] S307. Calculate geometric degradation index: Based on the information matrix G, calculate at least one of the following geometric degradation indices: GDOP geometric precision factor, information matrix condition number κ(G), information matrix minimum eigenvalue λmin(G), and sampling point track coverage angle.

[0076] S308. Determine if the observed geometry has degraded: If any of the geometry degradation indicators calculated in step S307 exceeds the preset threshold, then geometry degradation is determined.

[0077] S309. Output high-confidence multi-point solution results: When step S308 determines that the observed geometry has not degraded and the solved residuals meet the requirements, output the multi-point positioning solution results and mark the result status as high confidence or normal positioning.

[0078] S310. Switch to single-point peak strong feature degradation model: When step S308 determines that the observation geometry has degraded, switch to single-point peak strong feature degradation model. The single-point peak degradation model selects the effective independent sampling point with the highest RSSI intensity from the historical sampling queue of the entity cluster as the peak sampling point.

[0079] S311. Output the downgraded positioning result and the status to be reviewed: The fuzzy positioning area is output using the UAV spatial coordinates and neighborhood range corresponding to the peak sampling point; the reason for the downgrade is simultaneously marked and marked as a low confidence status to be reviewed.

[0080] S312, Spatial field reconstruction result fusion display: The high-confidence multi-point solution results and the downgraded positioning results are pushed to the main display and interaction thread in a unified manner; the main display and interaction thread will display the positioning coordinates, fuzzy areas and radio frequency thermal fields after masking constraints in a fusion and synchronous manner.

[0081] Furthermore, the spatial radio frequency scalar field reconstruction and physical trajectory backoff masking process includes the following steps:

[0082] S401. Obtain historical sampling points of entity clusters: Filter discrete sampling points participating in spatial field reconstruction from the target entity cluster cache queue;

[0083] S402. Establish a virtual grid for the target area: Generate a regular virtual grid within the search and rescue target area. The resolution of the virtual grid is adaptively configured according to the UAV flight altitude, sampling point distribution density, map scale, and the positioning accuracy requirements of the rescue mission.

[0084] S403. Perform spatial interpolation to obtain the initial radio frequency scalar field: Perform spatial interpolation operation on all grid points of the virtual grid based on discrete sampling points to generate the initial radio frequency scalar field. For any grid point x, it can be represented as: ;

[0085] S404. Construct a set of real UAV sampling tracks: Arrange the real coordinates of the UAVs participating in the sampling into a track set in chronological order.

[0086] ;

[0087] And a set of track segments is generated by connecting adjacent sampling points pairwise:

[0088] ;

[0089] S405. Calculate the minimum distance from a grid point to the actual flight path: For any grid point x within the virtual grid, calculate the minimum Euclidean distance from the grid point to the set of actual UAV sampled flight paths. ;

[0090] S406. Constructing a physical trajectory backoff mask according to the minimum Euclidean distance: Based on the minimum Euclidean distance constructing a segmented physical trajectory backoff mask function , the expression of the mask function is:

[0091] ;

[0092] wherein D1 is the distance threshold of the high-confidence region, D2 is the distance threshold of the shielding region, is an attenuation function that decreases monotonically with increasing distance;

[0093] ;

[0094] S407. Obtaining the radio frequency scalar field after track constraint: multiplying the initial radio frequency scalar field and the mask function grid by grid to generate a track-constrained trusted radio frequency scalar field , the formula is:

[0095] ;

[0096] S408. Dividing a trusted region, a backoff region and a shielding region: dividing the heat map into three types of differentiated display regions according to the output value of the mask function Mask(x):

[0097] Mask(x)=1 corresponds to the trusted region, 0<Mask(x)<1 corresponds to the transition attenuation backoff region, and Mask(x)=0 corresponds to the unsampled shielding region;

[0098] The trusted region renders the thermal gradient completely, the transition attenuation backoff region is displayed with reduced transparency, and the unsampled shielding region does not participate in thermal extreme value retrieval;

[0099] S409: Outputting the extreme value of the thermal field and the possible position of the target: retrieving the signal strength extreme point only in the masked radio frequency scalar field composed of the trusted region and the transition attenuation backoff region, outputting the extreme point and its surrounding neighborhood as the candidate position of the mobile terminal, and generating comprehensive search and rescue situation visualization data in combination with the multi-point positioning solution result, GDOP geometric state, entity cluster confidence and sampling quality, and pushing the data to the main display and interaction thread.

[0100] Beneficial effects of the present invention:

[0101] 1. This invention passively intercepts mobile terminal WiFi sniffing management frames in the air using an external radio frequency sensing payload mounted on an unmanned aerial vehicle (UAV). It requires no active cooperation from the target terminal, can penetrate obstacles to acquire signals, and improves the signal capture success rate in non-line-of-sight scenarios. This provides a reliable auxiliary search and rescue method in complex rescue environments, in addition to visible light and infrared detection. Compared with existing technologies that actively simulate WiFi networks to induce terminal connections, this invention eliminates the need for mobile terminals to complete connection or authentication interactions. By employing a passive sensing method, it avoids interference with the target and is more suitable for complex rescue scenarios with low intervention and low interaction, ensuring the safety and concealment of rescue operations.

[0102] 2. This invention abandons the traditional approach of relying solely on MAC addresses as unique target identifiers and constructs a multi-dimensional spatiotemporal physical fingerprint that includes time difference, RSSI deviation, spatial displacement, channel consistency, trajectory continuity, and sampling quality to improve the identity confusion problem caused by MAC address randomization. Through an entity cluster aggregation algorithm that combines threshold judgment and suboptimal gap as dual conditions, multiple randomized MAC addresses of the same physical terminal can be aggregated, reducing the number of false targets and short-lifetime targets, improving the continuity and accuracy of target tracking, and solving the technical problem of target loss in MAC randomization environments in existing technologies.

[0103] 3. After solving the multi-point weighted positioning problem, this invention introduces a comprehensive judgment mechanism for geometric degradation indices. By calculating multi-dimensional indices such as GDOP, information matrix condition number, minimum eigenvalue, and coverage angle, the quality of the observed geometry is evaluated, and positioning scenarios with poor observed geometry are identified. When the observed geometry does not meet the positioning accuracy requirements, the system automatically switches to a single-point peak strong feature degradation model and outputs the results in a degraded state or a state pending verification. This avoids the misuse of ill-conditioned solution results, improves the reliability and stability of positioning results in non-line-of-sight environments, and reduces the risk of rescue decision-making.

[0104] 4. In the process of reconstructing discrete radio frequency sampling points into a spatial continuous scalar field, this invention introduces a physical trajectory backoff masking technology based on the actual sampling trajectory of the UAV, which performs signal attenuation or shielding processing on areas far from the actual trajectory and lacking sampling support. This technology can effectively constrain the interpolation range of the heat map, making the extreme values ​​of the heat map closer to the actual measurement range, reducing false hot spots and misjudged areas caused by the outward expansion of traditional interpolation, improving the spatial accuracy and reliability of the radio frequency thermal field, and providing rescue personnel with more accurate target area guidance.

[0105] 5. This invention uses a low-frequency long-distance wireless link to transmit radio frequency characteristic data, which is suitable for long-distance low-speed transmission scenarios. Compared with traditional high-frequency communication links, it has stronger penetration and anti-interference capabilities in complex terrain or obstructed environments. The system supports flexible selection of LoRa, narrowband data transmission, self-organizing network data transmission or other communication methods according to the site conditions, improving the environmental adaptability and availability of the air-to-ground data link, and ensuring the continuity and stability of data transmission during rescue operations.

[0106] 6. This invention adopts a thread-decoupled architecture, separating the interface display thread from the background positioning calculation thread. This allows time-consuming calculation tasks such as high-frequency radio frequency data processing, entity cluster aggregation, geometric degradation determination, and thermal field reconstruction to be executed asynchronously in the background without directly blocking the main display thread. This architecture design can improve the real-time response speed of ground command software, ensuring that rescue personnel can obtain the latest positioning results and situation information in a timely manner, thereby improving the efficiency of rescue command and the timeliness of decision-making.

[0107] 7. This invention integrates multiple layers of data, including location points, downgraded ambiguity areas, entity cluster confidence states, and masked radio frequency thermal fields, to provide rescue personnel with comprehensive and multi-layered situational information. This enables them to intuitively distinguish between high-confidence targets, targets awaiting verification, and areas lacking sampling support. This visualization solution reduces the difficulty of information interpretation for rescue personnel and improves the efficiency and success rate of rescue operations. Attached Figure Description

[0108] Figure 1 This is a schematic diagram of the overall system structure in this invention.

[0109] Figure 2 This is a schematic diagram of the air-sensing payload structure in this invention.

[0110] Figure 3 This is a flowchart of the air-to-ground data link and message encapsulation process in this invention.

[0111] Figure 4 This is a flowchart of the MAC address derandomization and aggregation process in this invention.

[0112] Figure 5 This is a flowchart of the localization solution and GDOP adaptive degradation process in this invention.

[0113] Figure 6 This is a schematic diagram of spatial radio frequency scalar field reconstruction and physical trajectory backoff masking in this invention.

[0114] Figure 7 This is a schematic diagram of the functional modules of the ground command workstation in this invention.

[0115] In the diagram: 101. UAV platform; 102. Aerial sensing payload; 103. Wireless sniffing main control module; 104. Positioning and timing module; 105. Low-frequency long-range communication module; 106. Independent power supply module; 107. Storage / caching module; 108. Antenna unit; 201. Ground receiving station; 301. Ground command workstation; 302. Main display and interaction thread; 401. Equipment status area; 402. Data receiving area; 403. Target entity cluster list area; 404. Radar / heat map display area; 405. Alarm and result output area; 406. Parameter setting area; 407. Historical record area; 303. Background asynchronous calculation thread. Detailed Implementation

[0116] The present invention will be further described in detail below with reference to the accompanying drawings.

[0117] This invention discloses a mobile terminal positioning system based on an external radio frequency payload of a UAV in non-line-of-sight scenarios.

[0118] Reference Figure 1 This mobile terminal positioning system based on an external radio frequency payload of a UAV in non-line-of-sight scenarios mainly consists of a UAV platform 101, an aerial sensing payload 102, a ground receiving station 201, and a ground command workstation 301. The entire system relies on passive radio frequency sniffing and acquisition, air-to-ground low-frequency long-distance backhaul, multi-dimensional spatiotemporal fingerprint target aggregation, geometric degradation adaptive positioning, trajectory constraint thermal field reconstruction, and front-end and back-end thread decoupling visualization to work together. It forms a complete closed-loop search and rescue positioning link from the hardware acquisition end to the ground solution visualization end, solving the technical defects of existing active-guided WiFi search and rescue solutions, such as unstable terminal response, inability to continuously track random MACs, non-line-of-sight positioning drift, false hotspots in heat maps, and lag in the host computer interface. It is effectively adapted to non-line-of-sight emergency rescue scenarios where visible light and infrared detection fail, such as earthquake rubble burial, landslide and debris flow coverage, dense smoke obstruction, and dense vegetation obstruction.

[0119] The UAV platform 101 can be a multi-rotor UAV, a fixed-wing UAV, a compound-wing UAV, or other flight platforms with standard payload mounting interfaces. In typical low-altitude search and rescue scenarios in ruins, multi-rotor UAVs with stable hovering and precise positioning capabilities are preferred to complete low-altitude traversal searches of the area. The airborne sensing payload 102 can be detached and externally mounted on the fuselage mounting interface of the UAV platform 101 without modifying the UAV's native circuitry. The airborne sensing payload 102 and the ground receiving station 201 achieve long-distance wireless communication via a low-frequency long-range wireless link. The ground receiving station 201 is connected to the ground command workstation 301 via a serial port, USB, network interface, or other communication interface.

[0120] The basic working link of the system is as follows: the airborne sensing payload 102 passively collects the WiFi sniffing signal of the mobile terminal and binds it to the spatiotemporal information of the UAV, encapsulates the message and sends it to the ground receiving station 201; the ground receiving station 201 demodulates and verifies the data and injects it into the ground command workstation 301; the ground command workstation 301 completes target aggregation, positioning calculation, geometric degradation judgment and radio frequency field reconstruction in the background, and outputs the visualized search and rescue results with confidence markers.

[0121] The aerial sensing payload 102 is the core of aerial signal acquisition, integrating a wireless sniffing main control module 103, a positioning and timing module 104, a low-frequency long-range communication module 105, an independent power supply module 106, a storage / cache module 107, and an antenna unit 108. The independent power supply module 106 provides independent power to the wireless sniffing main control module 103, the positioning and timing module 104, the low-frequency long-range communication module 105, the storage / cache module 107, and the antenna unit 108; it is independent of the UAV power supply bus, which can avoid electrical fluctuations of the payload from interfering with the flight control, and also supports individual battery replacement to extend the operating time.

[0122] The positioning and timing module 104 is connected to the wireless sniffing main control module 103 via a UART serial port, continuously outputting spatiotemporal reference data. The output data includes the longitude, latitude, altitude, and unified timestamp of the UAV, providing a time alignment reference for multi-source sampling data. The hardware of this module can be a general GNSS module, an RTK positioning module, or an inertial navigation combined positioning module. It can also directly read the pose information output by the UAV flight control to obtain spatiotemporal data. For scenarios with high coordinate accuracy requirements, such as tunnels and dense buildings, an RTK and visual / inertial combined positioning scheme can be used to improve the accuracy of the UAV sampling position tag.

[0123] The wireless sniffing master control module 103 connects to the low-frequency long-range communication module 105 via the SPI interface to transmit return messages, and is also electrically connected to the storage / caching module 107. The storage / caching module 107 adopts a sliding window caching mechanism to temporarily store sampled data when the air-to-ground link is interrupted, and retransmit it after the link is restored, thus avoiding data loss. The antenna unit 108 is divided into two isolated radio frequency antennas. One is matched with the wireless sniffing master control module 103 to receive WiFi sniffing signals, and the other is matched with the low-frequency long-range communication module 105 to complete air-to-ground low-frequency data transmission and reception. The two antennas are isolated in frequency band to eliminate radio frequency interference.

[0124] The wireless sniffing main control module 103 can be equipped with a microcontroller, embedded Linux device, software radio device, or other wireless RF processing module that supports WiFi management frame monitoring. The actual selection can be flexibly matched with the hardware model based on the overall power consumption, payload weight, number of peripheral interfaces, and real-time processing computing power. During operation, it performs frequency hopping scanning according to a preset channel period, passively capturing only the WiFi sniffing management frames spontaneously generated by the mobile terminal, without actively transmitting signals to induce the terminal to connect to the network, thus reducing dependence on the terminal's working status. After capturing the frame, the module extracts complete RF features: source MAC address, received signal strength RSSI, WiFi channel, reception time, and signal-to-noise ratio SNR, and simultaneously binds them to the UAV pose and timestamp output by the positioning and timing module 104 to generate sampling data.

[0125] The low-frequency long-range communication module 105 encapsulates the sampled data into standardized return messages. Fixed fields in these messages include payload identifier, source MAC address, longitude, latitude, altitude, RSSI, WiFi channel, SNR, timestamp, and CRC checksum. This module can utilize low-rate, long-distance transmission standards such as LoRa, narrowband data transmission, and self-organizing network communication. Low-frequency links offer stronger diffraction and penetration capabilities, adapting to obstructed terrain. If air-to-ground link bandwidth is limited, the system can perform compression, low-value sampling filtering, frequency-limited uploading, and high-priority message scheduling on the return messages, adapting to unstable wireless transmission conditions. The entire process of aerial data acquisition, spatiotemporal binding, and message delivery is described in [reference needed]. Figure 3 During execution, each message carries complete spatial and temporal tags, serving as the basic data source for ground algorithm computation.

[0126] The ground command workstation 301 can be equipped with a laptop, industrial tablet, portable command terminal, or vehicle-mounted command terminal. The supporting software can be implemented based on any development framework that supports graphics display, multi-threaded parallel computing, and serial / network data access, without the limitation of a dedicated development platform. At the software level, a multi-threaded or multi-process architecture can be adopted, internally divided into an independent main display and interaction thread 302 and a background asynchronous calculation thread 303. The two types of threads can complete cross-thread data interaction through thread-safe queues, message buses, or shared cache snapshots, thereby reducing the negative impact of high-frequency radio frequency data batch processing on the interface response speed and solving the problems of interface lag and operation delay in traditional single-threaded software.

[0127] Reference Figure 7 The main display and interaction thread 302 is specifically responsible for real-time refresh of the user interface, human-computer interaction operation, and visualization rendering of the overall search and rescue situation. The interface has seven built-in functional areas, namely the device status area 401, data receiving area 402, target entity cluster list area 403, radar / heat map display area 404, alarm and result output area 405, parameter setting area 406, and history record area 407.

[0128] The equipment status area 401 displays the real-time operating status of the UAV platform 101, the airborne sensing payload 102, the ground receiving station 201, and the air-to-ground wireless link; the data receiving area 402 visualizes the total amount of data packets received in real time, the message parsing success rate, the number of valid sampling points, and abnormal message alarm prompts; the target entity cluster list area 403 displays the information of each mobile terminal entity cluster generated by the system, including the entity cluster number, the last update time, the total number of valid sampling points, the RSSI peak value, and the entity's overall confidence status; the radar / heatmap display area 404 synchronously overlays and renders the UAV flight path and discrete... The system includes: radio frequency sampling points, positioning output coordinates, constrained radio frequency thermal field, and thermal extreme points; alarm and result output area 405 displays high-confidence accurate positioning targets, geometrically degraded positioning targets, blurred targets to be verified, and various equipment / data anomaly alarms; parameter setting area 406 allows rescuers to configure sampling filter thresholds, MAC matching weights, GDOP degradation judgment thresholds, track masking distance thresholds, and related parameters on the interface online; historical record area 407 persistently stores all positioning results and entity cluster lifecycle change records, supporting track data playback, rescue mission report export, and full-process log retrieval.

[0129] The background asynchronous calculation thread 303 independently occupies the device's computing resources and serially completes the entire process of numerical calculation, including message integrity parsing, randomized MAC address entity cluster aggregation, weighted multi-point positioning calculation, GDOP geometric degradation index calculation, and spatial radio frequency scalar field reconstruction. All calculation results are uniformly pushed to the thread interaction channel, where the main display and interaction thread 302 reads and completes the interface rendering. The background calculation will not block the front-end interface refresh.

[0130] The background asynchronous computation thread 303 is equipped with three core processing flows, and the execution logic of each flow is as follows:

[0131] The specific logic of the background asynchronous calculation thread 303 in performing MAC address derandomization aggregation is as follows: First, randomized MAC addresses are distinguished by bitwise AND operation on the first byte of the MAC address; a six-dimensional spatiotemporal physical fingerprint vector containing time difference, RSSI deviation, spatial displacement, channel consistency, trajectory continuity, and sampling quality is constructed for random MAC sampling points; a weighted matching score is calculated by traversing active entity clusters, and sampling points are merged into matching entity clusters or new independent entity clusters are created using a dual judgment condition of scoring threshold and the difference between the best and second-best scores; the system synchronously manages the life cycle of entity clusters, and entity clusters that have not been updated for a long time are switched to a frozen state, and frozen entity clusters with low confidence are removed from the active queue, leaving only historical records, thereby eliminating the target breakage and false target problems caused by MAC randomization.

[0132] The specific logic of the background asynchronous solution thread 303 for weighted positioning and geometric degradation determination is as follows: 1. Filter valid independent sampling points that meet spatiotemporal and quality constraints; 2. Pause the solution when the sampling point distribution does not meet the standards; 3. Construct a weight matrix based on the sampling signal quality and spatial distribution; 4. Establish a weighted residual objective function and solve for the target candidate coordinates through weighted least squares iteration; 5. Calculate multiple geometric degradation indices such as GDOP, information matrix condition number, minimum eigenvalue, and sampling coverage angle to comprehensively judge the observation quality; 6. Output high-confidence multi-point positioning results when the geometric conditions are good; 7. When the indices exceed the limits and geometric degradation is determined, switch to the single-point peak strong feature degradation model, select the UAV neighborhood corresponding to the valid sampling point with the highest RSSI intensity as the fuzzy positioning area, mark the low-confidence state to be reviewed, and avoid positioning result drift distortion.

[0133] The specific logic of the background asynchronous calculation thread 303 for reconstructing the spatial radio frequency scalar field and the physical trajectory backoff masking is as follows: High-confidence valid sampling points are selected, and an adaptive resolution virtual grid is generated in the search and rescue area; an initial radio frequency scalar field is generated using algorithms such as inverse distance weighting, kriging, and radial basis interpolation; the complete set of UAV trajectory segments is extracted, the minimum Euclidean distance from each grid point to the trajectory is calculated, and a piecewise physical trajectory backoff masking function is constructed; the initial radio frequency scalar field is multiplied point by point with the masking function to generate a trajectory-constrained reliable thermal field, where the signal strength in the far-distance unsupported sampling area is attenuated and shielded; during visualization, the reliable area, backoff area, and shielded area are divided according to the masking value, and only the signal extreme values ​​in the valid area are retrieved as target candidate points to eliminate false hotspots caused by interpolation.

[0134] In this embodiment, the replacement schemes for each hardware module are all equivalent extended implementations. Those skilled in the art can adapt the hardware scheme by combining rescue scenarios, equipment mounting restrictions, and radio management regulations. All adjustments do not deviate from the core technical concepts of passive radio frequency sniffing, multi-dimensional fingerprint target aggregation, geometric degradation adaptive positioning, trajectory constraint thermal reconstruction, and front-end and back-end thread decoupling of this invention. The entire system forms a complete closed loop of airborne data acquisition, air-to-ground transmission, ground calculation, and situation visualization, which is suitable for various non-line-of-sight disaster search and rescue scenarios and has good engineering implementation value.

[0135] The present invention also discloses a mobile terminal positioning method based on an external radio frequency payload of a UAV in non-line-of-sight scenarios.

[0136] The mobile terminal positioning method based on UAV external radio frequency payload in non-line-of-sight scenarios is applied to the above-mentioned mobile terminal positioning system. The mobile terminal positioning method is divided into: air-to-ground data link and message encapsulation process, MAC address de-randomization and aggregation process, positioning calculation and GDOP adaptive degradation process, and spatial radio frequency scalar field reconstruction and physical trajectory backoff masking process.

[0137] The air-to-ground data link and message encapsulation process is used to complete WiFi radio frequency signal acquisition, spatiotemporal tag binding, standardized message encapsulation, and low-frequency air-to-ground wireless backhaul in the air, outputting structured radio frequency sampling data to the ground end; specifically, it includes the following steps:

[0138] S101, Intercepting WiFi sniffing management frames: The wireless sniffing master control module 103 of the airborne sensing payload 102 passively listens to and intercepts WiFi sniffing management frames spontaneously sent by surrounding mobile terminals.

[0139] S102. Extract target feature information: The wireless sniffing main control module 103 extracts radio frequency feature information such as source MAC address, received signal strength RSSI, WiFi channel, and reception time from the WiFi sniffing management frame.

[0140] S103, Read UAV pose information: The wireless sniffing main control module 103 reads the UAV pose information output in real time by the positioning and timing module 104. The UAV pose information includes the UAV's longitude, latitude, altitude and time information.

[0141] S104. Perform spatiotemporal tag binding: The wireless sniffing main control module 103 performs spatiotemporal tag binding between the radio frequency feature information obtained in step S102 and the UAV pose information obtained in step S103 to generate sampling data.

[0142] S105. Forming a wireless return message: The wireless sniffing master control module 103 encapsulates the bound sampling data into an extended return message.

[0143] In this embodiment, the extended return message includes the following fields:

[0144] $WIFI,ID,MAC,LON,LAT,ALT,RSSI,CH,SNR,TS,CRC

[0145] Wherein, ID represents the payload or data packet identifier, MAC represents the source MAC address, LON represents longitude, LAT represents latitude, ALT represents altitude, RSSI represents received signal strength, CH represents the WiFi channel, SNR represents the signal-to-noise ratio or link quality parameter, TS represents the timestamp, and CRC represents the check field.

[0146] In other embodiments, message fields can be added or removed based on communication link bandwidth and system requirements. For example, when communication link bandwidth is low, duplicate MACs, low-priority samples, or low-confidence samples can be compressed, merged, or frequency-limited before uploading.

[0147] In one implementation corresponding to the ground program, the serial port data stream uses independent WiFi sampling messages and UAV telemetry messages, respectively:

[0148] $WIFI,MAC,LON,LAT,RSSI,GPS_AGE,CH,CAPTURE_MS and $UAV,LON,LAT,ALT,SATS,GPS_AGE,CH.

[0149] The ground parsing thread parses the packets according to their headers. When the location field in the WiFi sampling packet is invalid, the most recent valid UAV telemetry location is used as the spatiotemporal location of the sampling point, provided that both the GPS age and location jump meet the preset thresholds. Packets exceeding the thresholds are not included in the location calculation.

[0150] S106. Transmission via low-frequency long-distance wireless link: The low-frequency long-distance communication module 105 transmits the extended return message to the ground receiving station 201 over a long distance via the low-frequency long-distance wireless link.

[0151] S107. Ground receiving station 201 receives and demodulates: Ground receiving station 201 receives the radio frequency signal transmitted back from the air and demodulates it into a structured data stream.

[0152] S108, Injecting into ground command workstation 301: Ground receiving station 201 injects structured data stream into ground command workstation 301 through serial port, USB or network interface, and the background asynchronous processing thread 303 reads the structured data stream to carry out subsequent processes.

[0153] The MAC address derandomization aggregation process is used to construct a multi-dimensional spatiotemporal physical fingerprint from discrete sampling points corresponding to randomized MAC addresses. This process merges multiple sets of different randomized MAC sampling points from the same physical mobile terminal into a single device entity cluster, thereby eliminating false targets introduced by MAC randomization. Specifically, it includes the following steps:

[0154] S201. Receive and standardize air-to-ground backhaul data packets: The background asynchronous processing thread 303 receives the structured data stream injected by the ground receiving station 201, performs integrity verification and field parsing on each message within the data stream, and obtains structured sampling points that include at least the source MAC address, received signal strength RSSI, WiFi channel, reception time, UAV longitude, latitude, and altitude; messages that fail verification, have abnormal timestamps, or lack pose information can be marked as low-quality data and discarded.

[0155] S202. Determine if the source MAC address is a randomized MAC address: Extract the source MAC address field carried by the current structured sampling point, and determine whether the source MAC address is a randomized MAC address based on the local management bits of the first byte of the MAC address; in this embodiment, the system performs a bitwise AND operation on the first byte of the MAC address to complete the determination, and the determination formula is: ;

[0156] If the calculation result is false, it means that the local management bit of the MAC address is not set, and the source MAC address is a fixed and stable device identifier. The update operation is directly performed on the corresponding real device entity cluster. If the calculation result is true, it means that the local management bit of the MAC address is set, and the source MAC address is determined to be a privacy randomized MAC address. The following steps are performed to enter the randomized address aggregation process based on multidimensional spatiotemporal physical fingerprint matching.

[0157] S203. Establish the spatiotemporal physical description of the sampling points: For each sampling point, establish a sampling description item:

[0158] ;

[0159] Among them, Mac k Indicates the source MAC address, t k p represents the receiving time. k This indicates the three-dimensional coordinates sampled by the UAV, rssi k Indicates the received signal strength, ch k Indicates the WiFi channel, q k This indicates the sampling quality index; the sampling quality index is jointly determined by message verification status, signal-to-noise ratio, RSSI fluctuation, link quality, or time synchronization status.

[0160] S204. Maintaining Real Device Entity Clusters: The system maintains an entity cluster C for each presumed real mobile terminal. j Single entity cluster C j The stored content includes historical sampling queues, the most recent update time, the RSSI statistical mean, representative spatial coordinates, channel distribution, trajectory change trends, entity confidence, and lifecycle status. Entity clusters can use fixed-length queues or sliding time windows to store recent sampling points, in order to control the computational load in real-time search and rescue missions.

[0161] S205. Generate a candidate entity cluster set: If the current sampling point corresponds to a randomized MAC address, traverse the entity clusters within the preset active time window Tactive, and filter the entity clusters that meet the set spatial distance, most recent update time and sampling quality to add them to the candidate entity cluster set, so as to avoid using entity clusters that are too far apart, have not been updated for a long time or have low quality as candidate objects.

[0162] S206. Construct a multi-dimensional spatiotemporal physical fingerprint feature vector: For the current sampling point R k With each entity cluster C in the candidate entity cluster set j Construct a multidimensional spatiotemporal physical fingerprint feature vector:

[0163] ;

[0164] in, Indicates the sampling time difference; Indicates RSSI intensity deviation; Indicates the difference in spatial coordinates; Indicates the channel consistency difference; Indicates deviation in trajectory continuity; This indicates the difference in sampling quality.

[0165] S207. Normalize and penalize the features: normalize and penalize the six-dimensional multidimensional spatiotemporal physical fingerprint feature vector constructed in step S206. The original difference features of each internal dimension are normalized and penalized, and all dimensional features are uniformly mapped to same-direction matching scoring components. The scoring components corresponding to the six-dimensional features are denoted as f. t Δt, f r Δrssi、f d Δd, f c Δch, f v Δv and f q Δq;

[0166] For four types of deviation features—time difference Δt, RSSI intensity deviation Δrssi, spatial coordinate distance difference Δd, and sampling quality difference Δq—the corresponding score component values ​​increase synchronously with the increase of feature difference. For two types of matching features—channel consistency Δch and trajectory continuity Δv—differential penalty transformations are applied based on whether the channel is consistent and whether the trajectory change trend is continuous. The worse the channel mismatch or trajectory continuity, the higher the corresponding score component value. The basis function used to implement the normalization and penalty mapping can be any one of piecewise linear functions, exponential functions, sigmoid functions, or lookup table functions. Each dimension feature can independently select an appropriate basis function to complete the transformation calculation.

[0167] S208. Calculate the candidate matching score: Calculate the matching score between the current sampling point and a single candidate entity cluster based on the preset weights of each dimension. The calculation formula is as follows:

[0168] ;

[0169] In the formula, , , , , , These are preset weighting coefficients for time difference, RSSI deviation, spatial distance, channel consistency, trajectory continuity, and sampling quality. Each weighting coefficient can be adaptively configured according to the type of on-site rescue scenario, drone sampling density, drone flight speed, and RF signal stability.

[0170] The lower the matching score, the higher the probability that the current sampling point and the candidate entity cluster belong to the same mobile terminal.

[0171] S209. Dual-condition judgment based on threshold and suboptimal gap: Extract all comprehensive matching scores within the candidate entity cluster set, and filter to obtain the optimal matching score. Second-best match score Set entity merging threshold suboptimal difference threshold ;

[0172] A dual-condition decision rule is used to select the optimal matching entity cluster. The decision condition is: the optimal matching score satisfies... The difference between the best and second-best matching scores satisfies ;

[0173] If both of the above conditions are met, the current sampling point will be merged into the optimal candidate entity cluster.

[0174] If only the optimal matching score condition is met, but the difference between the optimal and second-best matching scores is not met, it indicates that the matching distinction between the optimal and second-best entity clusters is insufficient. In this case, the current sampling point is marked as a sample to be confirmed to avoid erroneous merging when multiple targets are close. This sample to be confirmed is written to the confirmation cache and the matching score is recalculated in conjunction with subsequent newly added sampling points within the preset confirmation time window. When the double matching conditions are met again, it is merged into the optimal candidate entity cluster. If the double conditions are not met by the end of the confirmation time window, a new device entity cluster is created with the sample to be confirmed.

[0175] If neither of the two conditions is met, a new device entity cluster is created with the current sampling point.

[0176] S210. Update the status and confidence of the real device entity cluster: If the dual condition judgment in step S209 passes, merge the current sampling point into the corresponding real device entity cluster, and synchronously update the historical sampling queue, RSSI statistics, representative coordinates, WiFi channel distribution, trajectory change trend, latest update time and entity confidence of the entity cluster; the entity confidence increases synchronously with the positive change of matching features, and the improvement judgment dimensions are increased number of consecutive matches, increased number of cumulative entity sampling points, small fluctuation of previous matching scores and good sampling time sequence continuity; if the entity cluster does not receive new structured sampling point matching for a long time, or if the matching scores of multiple consecutive matches show large abnormal fluctuations, the confidence value of the entity cluster is lowered.

[0177] S211. Create, freeze, or eliminate entity clusters: If step S209 determines that there are no candidate entity clusters that meet the double matching conditions, then create a new independent real device entity cluster using the current sampling point as the initial sampling data; for entity clusters that have not been updated for more than a preset time, switch their status from active to frozen; for entity clusters that have been frozen for more than a preset time and whose entity confidence is lower than a preset threshold, eliminate them from the active candidate set and retain only historical records for task replay.

[0178] S212, Output Entity Cluster Cache Queue: Push the updated entity cluster cache queue of all devices to the background asynchronous calculation thread 303 for subsequent positioning calculation, GDOP judgment and spatial field reconstruction.

[0179] The aggregation logic of this process no longer relies solely on the source MAC string to classify device affiliation. Instead, it comprehensively considers multiple physical characteristics, including time dimension, spatial dimension, RSSI signal strength, WiFi channel, and trajectory continuity, to constrain the physical terminal affiliation corresponding to the structured sampling points. Even if the mobile terminal enables the MAC address randomization privacy mechanism, sampling points corresponding to different random MAC addresses can be merged into the same entity cluster, effectively reducing short-term false targets and improving the continuous target tracking capability in search and rescue operations.

[0180] The localization solution and GDOP adaptive degradation process is used to perform weighted multi-point localization solutions on entity cluster sampling points. It determines the observation quality of the sampling track through multiple geometric degradation indices, and automatically switches to a single-point peak degradation localization model when geometric degradation occurs. Specifically, it includes the following steps:

[0181] S301. Read sampling points from the entity cluster cache queue: The background asynchronous calculation thread 303 periodically reads all sampling points from the entity cluster cache queue; each sampling point includes at least the UAV coordinates, received signal strength RSSI, timestamp, sampling quality index and corresponding entity cluster identifier; the system prioritizes sampling points that are within the sliding time window and whose quality index meets the preset requirements.

[0182] S302. Determine if the number of valid independent sampling points has reached the set threshold: Valid independent sampling points must simultaneously meet four constraints: the time interval between sampling points is not less than the preset minimum, the spatial interval between sampling points is not less than the preset minimum, the comprehensive sampling quality index is higher than the preset minimum signal quality threshold, and the difference between the sampling time and the current solution time does not exceed the preset maximum time staleness. If any of the following situations occur: the total number of sampling points does not reach the solution lower limit, the spatial distribution of sampling points is too concentrated, or the positions of multiple sets of sampling points corresponding to the UAV track are approximately overlapping, then the observation geometry is deemed insufficient, and the subsequent weighted positioning solution process will not be started for the time being. The judgment will be re-verified after the addition of new structured sampling data.

[0183] S303. Constructing the weighting matrix: For each valid independent sampling point, calculate the single-point weight based on the received signal strength RSSI, time freshness, spatial distribution, and sampling quality. The weighted matrix W is composed of all the individual point weights and forms a diagonal weighted matrix.

[0184] The weight expression for a single valid independent sampling point is: ;

[0185] In the formula, , , , These are preset fixed weighting coefficients corresponding to four dimensions: RSSI intensity, sampling freshness, sampling quality, and spatial distribution. For the received signal strength mapping function, This is a sampling time freshness mapping function. The sampling quality mapping function, A mapping function for the contribution of the spatial distribution of sampling points;

[0186] The weighting matrix W is a diagonal matrix, and the elements on the diagonal are the weights corresponding to each valid independent sampling point. All values ​​in the other off-diagonal positions of the matrix are set to zero.

[0187] S304. Constructing the weighted residual objective function: Establish the residual objective function based on the relationship between the coordinates of the effective independent sampling points and the coordinates of the candidate target; the residual for the i-th effective independent sampling point can be expressed as:

[0188] ;

[0189] Where p represents the coordinates of the target object to be determined. Indicates the sampling coordinates of the UAV. This represents the equivalent constraint formed by RSSI or other RF characteristics; in one implementation, a logarithmic distance path loss model is used to map RSSI to an equivalent distance.

[0190] ;

[0191] Among them, RSSI ref Let n be the received signal strength at a reference distance of 1m, n be the path loss exponent, and d be the received signal strength at a reference distance of 1m. i Limited to a preset distance range. When corresponding to program parameters, RSSI can be used. ref =−40dBm, n=2.7, and d i Limited to the range of 1 to 500 meters.

[0192] Integrating the residuals of all valid independent sampling points, the weighted objective function is constructed as follows: ;

[0193] S305: Perform multi-point localization solution: Background asynchronous solution thread 303 solves the weighted objective function using weighted least squares, nonlinear least squares, simplex search, or other numerical optimization methods to obtain candidate location solutions p. hat It retains quality indicators such as residuals, iteration count, convergence state, weight distribution, and sampling point distribution. Weighted least squares itself can be implemented using existing numerical algorithms; the focus of this invention is on the subsequent geometric degradation determination and result downgrading processing.

[0194] S306. Construct the observation geometry matrix: Solve for candidate positions p hat Construct an observation Jacobian matrix H, where each row of H corresponds to a local observation constraint of a valid independent sampling point on the target location; construct an information matrix based on the Jacobian matrix H and the weighting matrix W. ;

[0195] If any one of the following conditions is met: G is approximately singular, the condition number of the matrix exceeds a preset threshold, or the minimum eigenvalue of the information matrix is ​​lower than a preset lower limit, then it is determined that the spatial constraint strength of the current UAV sampling observation geometry on the target coordinates is insufficient, and there is a risk of positioning divergence and drift.

[0196] S307. Calculate geometric degradation indices: Based on the information matrix G, calculate at least one of the following geometric degradation indices: GDOP geometric precision factor, information matrix condition number κ(G), information matrix minimum eigenvalue λmin(G), and sampling point track coverage angle. These indices quantify the strength of the current observation's geometric constraints. GDOP can be obtained by combining the trace or diagonal elements of the inverse of the information matrix. The information matrix condition number κ(G) can be used to determine whether the information matrix is ​​ill-conditioned. The sampling point track coverage angle can be used to determine whether the sampling points are concentrated on one side of the target.

[0197] S308. Determine whether the observed geometry has degraded: If any geometric degradation index calculated in step S307 exceeds a preset threshold, then geometric degradation is determined. In some embodiments, degradation determination can also be triggered by conditions such as residual abnormality, iteration non-convergence, or candidate position jump distance exceeding a preset value.

[0198] S309. Output high-confidence multi-point solution results: When step S308 determines that the observed geometry is not degraded and the solution residuals meet the requirements, output the multi-point positioning solution results and mark the result status as high confidence or normal positioning; the output content includes target location, entity cluster number, entity confidence, number of sampling points participating in the solution, GDOP index and residual index.

[0199] S310. Switch to Single-Point Peak Strong Feature Degradation Model: When step S308 determines that the observation geometry has degraded, switch to the single-point peak strong feature degradation model. The single-point peak degradation model traverses the historical sampling queue of the current device entity cluster, filters all valid independent sampling points, and selects the valid independent sampling point with the highest RSSI intensity as the peak sampling point. In one implementation, the valid independent sampling points are sorted from strong to weak RSSI, and the strong signal sample subset at the top of the sort is selected. The weights and degradation positions are defined as follows. When there are insufficient samples, the position corresponding to the sampling point with the highest RSSI is used as the degradation position.

[0200] ;

[0201] ;

[0202] In the formula, The weight of the i-th strong signal sample in the degraded mode; RSSI i Let p be the signal strength at the i-th sampling point. peak The output fuzzy positioning coordinates are shown in the degraded mode; S is the selected set of strong signal samples; p i Let be the coordinates of the UAV corresponding to the i-th sampling point.

[0203] S311. Output downgraded positioning results and pending verification status: The UAV spatial coordinates and neighborhood range corresponding to the peak sampling point are used as the output fuzzy positioning area; the reasons for downgrade are marked simultaneously, such as collinear sampling points, GDOP exceeding the limit, ill-conditioned information matrix or residual abnormality; and marked as a low confidence state pending verification. This state result is displayed in the ground command workstation 301 in the form of downgraded state, pending verification state or fuzzy area to avoid rescue personnel mistaking it for a high-precision point.

[0204] S312, Spatial field reconstruction result fusion display: The high-confidence multi-point solution result or the downgraded positioning result is uniformly pushed to the main display and interaction thread 302; the main display and interaction thread 302 overlays and renders the positioning coordinates, downgraded blurred area and masked radio frequency thermal field layer, and simultaneously completes the fusion display on the interface.

[0205] Through the above methods, the present invention can improve positioning accuracy by using multi-point data when the observation geometry is good, and avoid abnormal drift caused by ill-conditioned solutions when the observation geometry degrades, thereby improving the stability of mobile terminal positioning results in non-line-of-sight environments.

[0206] The spatial radio frequency scalar field reconstruction and physical trajectory retreat masking process is used to generate a continuous radio frequency thermal field based on discrete sampling points, construct a masking function based on the actual flight trajectory of the UAV to suppress false hotspots in unsampled areas, and output reliable thermal extrema as a target auxiliary positioning reference; specifically, it includes the following steps:

[0207] S401. Obtain historical sampling points of entity clusters: Filter sampling points participating in spatial field reconstruction from the target entity cluster cache queue, and extract the UAV spatial coordinates p of each sampling point. i RSSI observation value RSSI i Sampling time t i Sampling quality q i Discrete sampling points participating in spatial field reconstruction are selected based on preset time windows, sampling quality, and entity cluster confidence requirements.

[0208] S402. Establish a virtual grid for the target area: Since UAV flight paths are usually not regular grids and sampling points are spatially discrete and irregularly distributed, a regular virtual grid is generated within the overall search and rescue target area. The resolution of the virtual grid is adaptively configured according to the UAV flight altitude, sampling point distribution density, map scale, and the positioning accuracy requirements of the rescue mission.

[0209] S403. Perform spatial interpolation to obtain the initial radio frequency scalar field: Perform spatial interpolation operation on all grid points of the virtual grid based on discrete sampling points to generate the initial radio frequency scalar field. Interpolation methods can include radial basis function interpolation, inverse distance weighted interpolation, cubic spline interpolation, kriging interpolation, or other spatial interpolation methods.

[0210] Let x be any virtual grid point to be solved. The general interpolation expression for the initial radio frequency scalar field is:

[0211] .

[0212] S404. Construct a set of real UAV sampling tracks: Arrange the real coordinates of the UAVs participating in the sampling into a track set in chronological order.

[0213] ;

[0214] And a set of track segments is generated by connecting adjacent sampling points pairwise:

[0215] ;

[0216] Unlike using only sampling points, a set of track segments can represent the actual measured path of a UAV during continuous flight.

[0217] S405. Calculate the minimum distance from a grid point to the actual flight path: For any grid point x within the virtual grid, calculate the minimum Euclidean distance from the grid point to the set of actual UAV sampled flight paths. ;

[0218] ;

[0219] In one embodiment, is the minimum Euclidean distance from grid point x to all sampling point coordinates p i ; In another embodiment, is the minimum distance from grid point x to the track line segments formed by all adjacent sampling points.

[0220] S406. Constructing a physical trajectory backoff mask according to the minimum Euclidean distance: Based on the minimum Euclidean distance a piecewise physical trajectory backoff mask function is constructed , and the expression of the mask function is:

[0221] ;

[0222] In the formula, D1 is the distance threshold of the high-confidence region, D2 is the distance threshold of the shielding region, is a decay function that decreases monotonically with increasing distance. In one implementation, linear decay is adopted, and its expression and applicable interval are as follows:

[0223] ;

[0224] When corresponding to the program parameters, D1 can be 5m, and D2 can be 15m.

[0225] S407. Obtaining a radio frequency scalar field constrained by the track: Multiply the initial radio frequency scalar field and the mask function grid by grid to generate a track-constrained trusted radio frequency scalar field , and the formula is:

[0226] ;

[0227] After mask processing, the interpolation regions that are too far from the real track and lack sampling support are attenuated or shielded, which can reduce the influence of regions without data support on the judgment of the extreme value of the thermal field.

[0228] S408. Dividing the trusted region, the backoff region and the shielding region: According to the output value of the mask function Mask(x), the heat map is divided into three types of differentiated display regions: Mask(x)=1 corresponds to the trusted region, 0<Mask(x)<1 corresponds to the transition decay backoff region, and Mask(x)=0 corresponds to the unsampled shielding region;

[0229] The trusted region fully renders the thermal gradient; the transition decay backoff region is displayed with reduced transparency to indicate that although the radio frequency scalar value of this region is obtained by interpolation, the sampling support is insufficient; the unsampled shielding region does not participate in the thermal extremum retrieval or is only displayed as a background region.

[0230] S409, Output thermal field extreme values ​​and possible target locations: Only search for signal strength extreme points in the radio frequency scalar field behind the mask composed of the credible region and the transition attenuation retreat region, output the extreme points and their surrounding neighborhoods as candidate locations of the mobile terminal, and combine the multi-point positioning solution results, GDOP geometric state, entity cluster confidence and sampling quality to generate comprehensive search and rescue situation visualization data and push it to the main display and interaction thread 302.

[0231] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A mobile terminal positioning system based on an external radio frequency payload of a UAV in non-line-of-sight scenarios, characterized in that: It includes an unmanned aerial vehicle platform (101), an aerial sensing payload (102), a ground receiving station (201), and a ground command workstation (301); The airborne sensing payload (102) is mounted on the UAV platform (101), and the airborne sensing payload (102) establishes a wireless communication connection with the ground receiving station (201) through a low-frequency long-range wireless link; the ground receiving station (201) is connected to the ground command workstation (301). The airborne sensing payload (102) is used to passively intercept WiFi sniffing management frames sent by mobile terminals, extract radio frequency feature information, synchronously bind the spatial pose information and time information of the UAV platform (101), encapsulate and generate a return message and send it to the ground receiving station (201); the ground receiving station (201) is used to receive and demodulate the return message and inject the structured message data into the ground command workstation (301); The ground command workstation (301) is used to perform randomized MAC address entity cluster aggregation, weighted positioning calculation, observation geometry degradation determination, and spatial radio frequency scalar field reconstruction on the injected message data, and output visualized search and rescue results with confidence status.

2. The mobile terminal positioning system based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 1, characterized in that: The airborne sensing payload (102) includes a wireless sniffing main control module (103), a positioning and timing module (104), a low-frequency long-range communication module (105), an independent power supply module (106), a storage / cache module (107), and an antenna unit (108); The independent power supply module (106) provides independent power to the wireless sniffing master control module (103), the positioning and timing module (104), the low-frequency long-range communication module (105), the storage / cache module (107), and the antenna unit (108). The positioning and timing module (104) is connected to the wireless sniffing master control module (103) and is used to transmit spatiotemporal reference data to the wireless sniffing master control module (103). The wireless sniffing master control module (103) is connected to the low-frequency long-range communication module (105) and is used to transmit the encapsulated return message. The wireless sniffing master control module (103) is electrically connected to the storage / cache module (107) and is used to temporarily store the sampling data that was not successfully returned. The antenna unit (108) is RF connected to the wireless sniffing master control module (103) and the low-frequency long-range communication module (105) respectively, and performs WiFi signal reception and low-frequency wireless data transmission respectively.

3. A mobile terminal positioning system based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 2, characterized in that: The radio frequency feature information extracted by the wireless sniffing master control module (103) includes the source MAC address, received signal strength RSSI, WiFi channel, reception time, and signal-to-noise ratio SNR; the wireless sniffing master control module (103) periodically performs frequency hopping scanning according to a preset channel strategy, only capturing WiFi sniffing management frames spontaneously sent by the mobile terminal, and does not actively transmit wireless signals to induce terminal interaction. The spatial pose and time information output by the positioning and timing module (104) includes the longitude, latitude, altitude and timestamp of the UAV; the positioning and timing module (104) synchronously outputs spatial pose and time data to the wireless sniffing master control module (103); The positioning and timing module (104) adopts any one of GNSS positioning module, RTK high-precision positioning module or inertial navigation fusion positioning module; the low-frequency long-distance communication module (105) adopts any one of LoRa communication, narrowband data transmission or self-organizing network low-speed wireless communication standard.

4. A mobile terminal positioning system based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 3, characterized in that: The ground command workstation (301) includes a main display and interaction thread (302) and a background asynchronous calculation thread (303); The main display and interaction thread (302) independently performs human-computer interaction, UAV trajectory rendering, entity cluster list display, thermal field visualization and alarm result output; The background asynchronous solution thread (303) independently performs message parsing, randomized MAC address aggregation, weighted location solution, GDOP geometric degradation index calculation, and spatial radio frequency scalar field reconstruction. The main display and interaction thread (302) and the background asynchronous calculation thread (303) complete data interaction through a thread-safe queue.

5. A mobile terminal positioning system based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 4, characterized in that: The specific logic of the background asynchronous calculation thread (303) for performing MAC address derandomization aggregation is as follows: randomized MAC addresses are distinguished by the local management bits of the MAC address, a multi-dimensional spatiotemporal physical fingerprint feature vector containing time difference, RSSI deviation, spatial displacement, channel consistency, trajectory continuity and sampling quality is constructed, the matching score between the sampling point and the active entity cluster is calculated by weighting, and the sampling point is determined based on the dual conditions of the score threshold and the difference between the second-best scores, and the entity cluster is merged or newly created. The specific logic of the background asynchronous solution thread (303) for performing weighted localization solution and geometric degradation determination is as follows: filter effective independent sampling points within the entity cluster, construct a weighted matrix based on sampling quality and spatial distribution, establish a weighted residual objective function and iteratively solve the candidate target position; calculate the geometric degradation index, if the geometric degradation index does not exceed the limit, output a high-confidence multi-point localization result, if the geometric degradation index exceeds the limit, switch to the single-point peak strong feature degradation model to output a fuzzy localization result and mark the low-confidence state; The specific logic of the background asynchronous solution thread (303) for reconstructing the spatial radio frequency scalar field and the physical trajectory backoff masking is as follows: a virtual grid of the target area is established based on the discrete sampling points of the entity cluster, and an initial radio frequency scalar field is generated by spatial interpolation; the minimum distance from each grid point to the complete set of UAV sampled tracks is calculated, and a segmented physical trajectory backoff masking function is constructed; the initial radio frequency scalar field is multiplied by the masking function to obtain a constrained reliable thermodynamic field, and the signal extrema are searched only in the effective masking area as candidate positions for the target.

6. A mobile terminal positioning method based on an external radio frequency payload of a UAV in non-line-of-sight scenarios, characterized in that: The mobile terminal positioning system based on UAV external radio frequency payload in non-line-of-sight scenarios as described in claim 5 is divided into the following steps: air-to-ground data link and message encapsulation process, MAC address de-randomization and aggregation process, positioning calculation and GDOP adaptive degradation process, and spatial radio frequency scalar field reconstruction and physical trajectory backoff masking process. The air-to-ground data link and message encapsulation process is used to complete WiFi radio frequency signal acquisition, spatiotemporal tag binding, standardized message encapsulation, and low-frequency air-to-ground wireless backhaul in the air, and output structured radio frequency sampling data to the ground end. The MAC address derandomization aggregation process is used to construct a multidimensional spatiotemporal physical fingerprint for the discrete sampling points corresponding to randomized MAC addresses, and to merge multiple different random MAC sampling points of the same physical mobile terminal into a single device entity cluster, so as to eliminate false targets caused by MAC randomization. The positioning solution and GDOP adaptive degradation process are used to perform weighted multi-point positioning solution on the sampling points of the entity cluster, determine the observation quality of the sampling track through multiple geometric degradation indices, and automatically switch to the single-point peak degradation positioning model when geometric degradation occurs. The spatial radio frequency scalar field reconstruction and physical trajectory backoff masking process is used to generate a continuous radio frequency thermal field based on discrete sampling points, construct a masking function based on the actual flight path of the UAV to suppress false hotspots in unsampled areas, and output a reliable thermal extreme value as a reference for target auxiliary positioning.

7. A mobile terminal positioning method based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 6, characterized in that: The air-to-ground data link and message encapsulation process includes the following steps: S101, Intercepting WiFi sniffing management frames: The wireless sniffing master control module (103) of the airborne sensing payload (102) passively listens to and intercepts WiFi sniffing management frames spontaneously sent by surrounding mobile terminals. S102, Extract target feature information: The wireless sniffing master control module (103) extracts radio frequency feature information from the WiFi sniffing management frame. The radio frequency feature information includes the source MAC address, received signal strength RSSI, WiFi channel and reception time. S103, Read UAV pose information: The wireless sniffing main control module (103) reads the UAV pose information output in real time by the positioning and timing module (104). The UAV pose information includes the UAV's longitude, latitude, altitude and time information. S104, Perform spatiotemporal tag binding: The wireless sniffing main control module (103) performs spatiotemporal tag binding between the radio frequency feature information obtained in step S102 and the UAV pose information obtained in step S103, and generates sampling data; S105, Forming a wireless return message: The wireless sniffing master control module (103) encapsulates the bound sampling data into an extended return message; S106, Transmit via low-frequency long-distance wireless link: The low-frequency long-distance communication module (105) transmits the extended return message to the ground receiving station (201) over a long distance via the low-frequency long-distance wireless link; S107. Ground receiving station (201) receives and demodulates: Ground receiving station (201) receives the radio frequency signal transmitted back from the air and demodulates it into a structured data stream; S108, Injecting into ground command workstation 301: The ground receiving station (201) injects the structured data stream into the ground command workstation (301), and the background asynchronous calculation thread (303) reads the structured data stream to carry out subsequent processes.

8. A mobile terminal positioning method based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 6, characterized in that: The MAC address derandomization and aggregation process includes the following steps: S201. Receive and standardize air-to-ground backhaul data packets: The background asynchronous processing thread (303) receives the structured data stream injected by the ground receiving station (201), performs integrity verification and field parsing, and obtains structured sampling points that include at least the source MAC address, received signal strength RSSI, WiFi channel, receiving time, UAV longitude, latitude and altitude. S202. Determine if the source MAC address is a randomized MAC: Determine if the source MAC address of the current sampling point is a randomized MAC address based on the local management bits of the MAC address. If the local management bits of the MAC address are not set, then treat it as a relatively stable device identifier; if the local management bits are set, then continue with the following steps. S203. Establishing the spatiotemporal physical description of the sampling points: For each sampling point, the system establishes a sampling description item: ; in, Indicates the source MAC address. Indicates the receiving time. This indicates that the UAV sampled three-dimensional coordinates. Indicates the received signal strength. Indicates the WiFi channel. Indicates the sampling quality index; S204. Maintain the real device entity cluster; the system maintains an entity cluster C for each presumed real mobile terminal. j Single entity cluster C j The stored content includes historical sampling queues, the most recent update time, the RSSI statistical mean, representative spatial coordinates, channel distribution, trajectory change trends, entity confidence, and lifecycle status; S205. Generate a candidate entity cluster set: If the current sampling point corresponds to a randomized MAC address, traverse the entity clusters within the preset active time window and add only the entity clusters whose spatial distance, most recent update time, and sampling quality meet the preset coarse screening conditions to the candidate entity cluster set. S206. Construct a multi-dimensional spatiotemporal physical fingerprint feature vector: For the current sampling point R k With candidate entity cluster C j Construct a multidimensional spatiotemporal physical fingerprint feature vector: ； in, Indicates the sampling time difference; Indicates RSSI intensity deviation; Indicates the difference in spatial coordinates; Indicates the channel consistency difference; Indicates deviation in trajectory continuity; Indicates the difference in sampling quality; S207. Normalize and penalize the features: Normalize and segmented penalize each component of the multidimensional spatiotemporal physical fingerprint feature vector, and map the original difference to the same direction matching score component. S208. Calculate the candidate matching score: Calculate the matching score between the current sampling point and a single candidate entity cluster according to the preset weights. The calculation formula is as follows: ； In the formula, , , , , , These are the preset weighting coefficients corresponding to the dimensions of time difference, RSSI deviation, spatial distance, channel consistency, trajectory continuity, and sampling quality, respectively. The lower the score, the higher the probability that the current sampling point and the candidate entity cluster belong to the same terminal; S209. Dual-condition judgment based on execution threshold and suboptimal gap: The optimal matching entity cluster is selected using a dual-matching condition judgment rule. The judgment condition is: the optimal matching score satisfies... The difference between the best and second-best matching scores satisfies ; If both of the above matching conditions are met, the current sampling point will be merged into the optimal candidate entity cluster. If only the optimal matching score condition is met, but the difference between the optimal and second-best matching scores is not met, then the sampling point is marked as a sample to be confirmed. The sample to be confirmed is written into the confirmation cache, and the matching score is recalculated in combination with the subsequently added sampling points within the preset confirmation time window. When the double matching condition is met again, it is merged into the optimal candidate entity cluster. If the double matching condition is not met when the confirmation time window expires, a new device entity cluster is created with the sample to be confirmed. If neither of the two matching conditions is met, a new device entity cluster is created with the current sampling point; S210. Update entity cluster status and confidence: If the double matching condition judgment in step S209 passes, merge the current sampling point into the corresponding real device entity cluster, and synchronously update the entity cluster's historical sampling queue, RSSI statistics, representative coordinates, channel distribution, trajectory change trend, latest update time, and entity confidence. S211. Create, freeze, or eliminate entity clusters: If step S209 determines that there are no candidate entity clusters that meet the double matching conditions, the current sampling point is used as the initial sampling data to create a new device entity cluster; for entity clusters that have not been updated for more than a preset time, their status is switched from active to frozen; for entity clusters that have been frozen for more than a preset time and whose confidence level is lower than a preset threshold, they are eliminated from the active candidate set, and only historical records are retained for task replay. S212: Output Entity Cluster Cache Queue: Push all updated device entity cluster cache queues to the background asynchronous calculation thread (303) for subsequent positioning calculation, GDOP judgment and spatial field reconstruction.

9. A mobile terminal positioning method based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 6, characterized in that: The localization calculation and GDOP adaptive degradation process includes the following steps: S301, Read sampling points from entity cluster cache queue: The background asynchronous calculation thread (303) periodically reads sampling points from the entity cluster cache queue; each sampling point includes at least the UAV coordinates, received signal strength RSSI, timestamp, sampling quality index and corresponding entity cluster identifier; S302. Determine whether the number of valid independent sampling points reaches a set threshold. If the number does not meet the requirement, the formal positioning solution is not executed temporarily, and new sampling points are waited until the number of valid independent sampling points reaches the set threshold; S303. Constructing the weighting matrix: For each valid independent sampling point, calculate the single-point weight based on the received signal strength RSSI, time freshness, spatial distribution, and sampling quality. The weighted matrix W is composed of all the individual point weights and forms a diagonal weighted matrix. The weight expression for a single valid independent sampling point is: ； In the formula, , , , These are preset fixed weighting coefficients corresponding to four dimensions: RSSI intensity, sampling freshness, sampling quality, and spatial distribution. For the received signal strength mapping function, This is a sampling time freshness mapping function. The sampling quality mapping function, A mapping function for the contribution of the spatial distribution of sampling points; The weighting matrix W is a diagonal matrix, and the elements on the diagonal are the weights corresponding to each valid independent sampling point. All values ​​in the other off-diagonal positions of the matrix are set to zero; S304. Construct a weighted residual objective function: establish a residual objective function according to the relationship between the coordinates of the valid independent sampling points and the coordinates of the candidate target; the residual of the i-th valid independent sampling point can be expressed as: ； Where p represents the coordinates of the target object to be determined. Indicates the sampling coordinates of the UAV. This represents the equivalent constraint formed by RSSI or other radio frequency characteristics; Integrating the residuals of all valid independent sampling points, the weighted objective function is constructed as follows: ; S305. Perform multi-point localization solution: The background asynchronous solution thread (303) solves the weighted objective function through a numerical optimization algorithm to obtain the candidate location solution p. hat ; S306. Construct the observation geometry matrix: Solve for candidate positions p hat Construct the observation Jacobian matrix H, and then construct the information matrix based on the Jacobian matrix H and the weight matrix W. ; S307. Calculate geometric degradation indicators: calculate at least one geometric degradation indicator from the GDOP (Geometric Dilution of Precision), the information matrix condition number κ(G), the minimum eigenvalue λmin(G) of the information matrix, and the sampling point track coverage angle based on the information matrix G; S308. Determine whether the observation geometry is degraded: if any geometric degradation indicator calculated in step S307 exceeds a preset threshold, it is determined that the geometry is degraded; S309. Output a high-confidence multi-point solution result: when it is determined in step S308 that the observation geometry is not degraded and the solution residual meets the requirements, output the multi-point positioning solution result, and mark the result status as high-confidence or normal positioning; S310. Switch to the single-point peak strong feature degradation model: when it is determined in step S308 that the observation geometry is degraded, switch to the single-point peak strong feature degradation model, and the single-point peak degradation model selects the valid independent sampling point with the highest RSSI intensity from the historical sampling queue of the entity cluster as the peak sampling point; S311. Output the degraded positioning result and pending review status: output the UAV spatial coordinates and the neighborhood range corresponding to the peak sampling point as a fuzzy positioning area; simultaneously mark the degradation reason, and mark it as a low-confidence status pending review; S312. Fusion display of spatial field reconstruction results: push the high-confidence multi-point solution result and the degraded positioning result to the main display and interaction thread (302) in a unified manner; the main display and interaction thread (302) fuses and synchronously displays the positioning coordinates, the fuzzy area and the radio frequency thermal field after mask constraint.

10. A mobile terminal positioning method based on an external radio frequency payload of a UAV in a non-line-of-sight scenario according to claim 6, characterized in that: The spatial radio frequency scalar field reconstruction and physical track backoff masking process includes the following steps: S401. Obtain historical sampling points of the entity cluster: screen discrete sampling points participating in spatial field reconstruction from the target entity cluster cache queue; S402. Establish a virtual grid of the target area: generate a regular virtual grid in the search and rescue target area, and the resolution of the virtual grid is adaptively configured according to the UAV flight altitude, sampling point distribution density, map scale, and positioning accuracy requirements of the rescue mission; S403. Perform spatial interpolation to obtain the initial radio frequency scalar field: Perform spatial interpolation operation on all grid points of the virtual grid based on discrete sampling points to generate the initial radio frequency scalar field. For any grid point x, it can be represented as: ; S404. Construct a real sampling track set of UAVs: compose a track set by arranging the real coordinate points of the participating UAVs in chronological order: ； and generate a track line segment set by connecting adjacent sampling points in pairs: ； S405. Calculate the minimum distance from a grid point to the actual flight path: For any grid point x within the virtual grid, calculate the minimum Euclidean distance from the grid point to the set of actual UAV sampled flight paths. ; S406. Construct a physical trajectory backoff mask based on the minimum Euclidean distance: Based on the minimum Euclidean distance Constructing a segmented physical trajectory backoff masking function The masking function expression is: ； In the formula, D1 is the distance threshold for the high-confidence area, and D2 is the distance threshold for the shielded area. The decay function is a monotonically decreasing function that decreases with increasing distance, and its formula is: ； S407. Obtaining the radio frequency scalar field after trajectory constraints: Multiply the initial radio frequency scalar field by the masking function grid by grid to generate a reliable radio frequency scalar field with trajectory constraints. The formula is: ； S408. Divide the trusted area, backoff area and shielding area: divide the thermal map into three types of differentiated display areas according to the output value of the mask function Mask(x): Mask(x)=1 corresponds to the trusted area, 0<Mask(x)<1 corresponds to the transition attenuation backoff area, and Mask(x)=0 corresponds to the no-sampling shielding area; The thermal gradient is fully rendered in the trusted region, the transparency of the transitional decay and retreat region is reduced, and the unsampled and shielded region does not participate in the thermal extreme value retrieval. S409: Output thermal field extreme values ​​and possible target locations: Only search for signal strength extreme points in the radio frequency scalar field behind the mask composed of the credible region and the transition attenuation retreat region, output the extreme points and their surrounding neighborhoods as candidate locations of the mobile terminal, and combine the multi-point positioning solution results, GDOP geometric state, entity cluster confidence and sampling quality to generate comprehensive search and rescue situation visualization data and push it to the main display and interaction thread (302).

Images