Space-air-ground cooperative intelligent low-altitude target perception method based on channel state information
By collaborating with a three-layer perception network and neural network in a combined air-space-ground approach, the problems of perception blind spots and limited coverage in low-altitude scenarios have been solved, achieving high-precision low-altitude target positioning and wide-area coverage while reducing transmission overhead.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2025-12-11
- Publication Date
- 2026-06-19
AI Technical Summary
In low-altitude scenarios, existing technologies struggle to achieve seamless perception over a wide area, resulting in numerous blind spots and limited coverage. Furthermore, existing node deployment methods are susceptible to interference, leading to insufficient perception reliability and stability.
A three-layer sensing network integrating air, space, and ground is constructed. Channel state information and neural networks are used to achieve two-stage collaborative sensing. Multi-level communication facilities and multiple sensing nodes are integrated. Through the collaborative work of remote sensing satellites, UAV swarms, ground base stations, and user equipment, independent detection and fusion positioning of channel state information are carried out.
It significantly expands the sensing range, eliminates sensing blind spots, and effectively reduces the transmission overhead of sensing data while ensuring high positioning accuracy.
Smart Images
Figure CN121619652B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, and in particular to a space-air-ground collaborative intelligent low-altitude target sensing method based on channel state information. Background Technology
[0002] The low-altitude economy refers to various economic activities conducted in airspace below 3000 meters above sea level. Its development has led to the rapid application of unmanned aerial vehicles (UAVs) in commerce, inspection, and logistics. However, unauthorized UAV flights pose a serious threat to public safety and privacy, necessitating reliable detection and location methods. As a core scenario in the 6G vision, integrated communication and sensing technology provides a feasible path for integrating sensing and communication functions on the same infrastructure. Wireless signal-based sensing schemes can reuse existing equipment, offering advantages such as low cost, wide coverage, and non-intrusiveness. In the device-free sensing paradigm, Channel State Information (CSI) can encode finer-grained propagation characteristics, making it more suitable for environmental reasoning. On the other hand, Artificial Intelligence (AI) methods demonstrate strong potential in fully exploiting the rich features of CSI, exhibiting stronger robustness and higher sensing accuracy compared to traditional model-based methods.
[0003] Low-altitude scenarios require vast surveillance areas, demanding sufficient coverage of sensing signals in both horizontal and vertical directions. However, limitations in antenna directivity, array elevation, and line-of-sight path obstruction caused by tall buildings create significant blind spots, reducing sensing reliability. Furthermore, the coverage of a single sensing node pair is limited and susceptible to interference in multipath-rich or heavily obstructed channels. To overcome these limitations, a multi-node collaborative paradigm is urgently needed to improve sensing reliability. Regarding node selection, existing technologies mostly rely on multi-base station (BS) collaboration to sense low-altitude targets. While BSs possess high-precision sensing capabilities, their sparse deployment limits spatial coverage, and antenna tilt towards the ground makes high-altitude coverage difficult to guarantee. Introducing user equipment can expand coverage, but its mobility can compromise sensing stability. Therefore, achieving seamless sensing over a wide low-altitude area, effectively integrating heterogeneous nodes, and reducing transmission overhead are the core challenges for reliable low-altitude monitoring. Summary of the Invention
[0004] This invention provides a space-air-ground collaborative intelligent low-altitude target perception method based on channel state information. It integrates multi-level communication facilities and multiple sensing nodes into a unified sensing framework, and uses the channel state information generated during the communication process to achieve low-altitude target positioning, which can significantly expand the sensing range and eliminate sensing blind spots. In addition, a neural network is used to realize a two-stage collaborative sensing process, which can effectively reduce the transmission overhead of sensing data while ensuring high positioning accuracy.
[0005] This invention provides a space-air-ground coordinated intelligent low-altitude target sensing method based on channel state information, comprising the following steps:
[0006] Step 1: Construct a three-layer collaborative sensing network consisting of a space-air-ground layer, an air-ground layer, and a ground-ground layer. The space-ground layer includes remote sensing satellites, the air-ground layer communicates with the ground base stations of the ground layer through a swarm of drones or high-altitude airships, and the ground layer deploys a fixed wireless access network, including multiple base stations and multiple user-side devices.
[0007] Step 2: After detecting a low-altitude target at the ground level, the data is reported to the central processing unit (CPU). The CPU then transmits the data to the ground level. M Each base station sends a command, and the base station obtains the channel status information between the base station and all air-level nodes and ground-level user-side equipment respectively;
[0008] Step 3: Base stations and airborne ground-level nodes, and base stations and ground-level user-side equipment are used as sensing pairs for monitoring low-altitude targets. Each base station independently detects channel state information and filters out... L The system identifies several sensing pairs affected by low-altitude targets and their node indices, obtains corresponding channel state information, independently locates each base station for each sensing pair, estimates the low-altitude target location, and extracts lightweight channel features, resulting in a total of [number missing] sensing pairs. L The location estimation results of an independent low-altitude target and the corresponding channel characteristics and node index;
[0009] Step 4, each base station will L The position estimation results of each independent low-altitude target, along with the corresponding channel features and node indexes, are uploaded to the central processing unit. The data is then fused using a neural network to obtain high-precision collaborative positioning results for the low-altitude targets.
[0010] Optionally, in one embodiment of the present invention, in step 1, the empty base station node and the user-side equipment perform uplink transmission with the ground base station in a multiple-input multiple-output orthogonal frequency division multiplexing system.
[0011] Optionally, in one embodiment of the present invention, in step 1, the empty base station node communicates with the ground base station through a reconfigurable smart surface.
[0012] Optionally, in one embodiment of the present invention, in step 2, channel state information is obtained at the base station through pilot transmission and channel estimation, and each node operates at a different frequency.
[0013] Optionally, in one embodiment of the present invention, in step 3, each base station checks the channel state information. Conduct independent testing. For channel state information between the base station and airborne base station nodes or ground-based user-side equipment, a preprocessing function is first used. Channel state information Preprocessing is performed to obtain the preprocessed channel. :
[0014] ;
[0015] The independent detection process uses a neural network. To achieve this, the preprocessed channel Input Neural Network The output is the independent detection result. Neural Networks It consists of convolutional layers, batch normalization layers, and fully connected layers, for the () m , n The independent detection process for each of the 10 perception pairs is described as follows:
[0016] ;
[0017] in, This represents the neural network parameters. All perception pairs are independently detected using the same neural network architecture, but with different neural network parameters, based on the independent detection results. Filter out L A sensing pair affected by low-altitude targets and its corresponding channel state information. , M For the number of base stations, N It is the sum of the number of empty base station nodes and the number of user-side devices at the ground level.
[0018] Optionally, in one embodiment of the present invention, in step 3, the independently detected neural network is trained using gradient descent, and the cost function is minimized by iteratively optimizing the neural network parameters. The cost function is the binary classification cross-entropy, as described below:
[0019] ;
[0020] in,( k ) is the sample index. K For the sample size, For the Euclidean norm, for The probability of.
[0021] Optionally, in one embodiment of the present invention, in step 3, a preprocessing function is used. Perception l Corresponding channel state information Preprocessing is performed to obtain the preprocessed channel. :
[0022] ;
[0023] Each base station uses a neural network for independent positioning. To achieve this, the preprocessed channel Input Neural Network The output is a low-altitude target position estimate. and channel characteristics These two figures represent the final output and intermediate layer output of the neural network, respectively; neural network It mainly consists of convolutional layers, batch normalization layers, and fully connected layers, for the first... l The independent localization process for each sensing pair is described as follows:
[0024] ;
[0025] in, This represents the neural network parameters. Each sensor uses the same neural network architecture to independently locate objects, but employs different neural network parameters.
[0026] Optionally, in one embodiment of the present invention, in step 3, L Each independent localization neural network is trained on its own dataset using gradient descent. The cost function, which is the mean squared error, is minimized by iteratively optimizing the neural network parameters. The cost function is described as follows:
[0027] ;
[0028] in, K For the sample size, ( k ) is the sample index. The true coordinates of the low-altitude target. It is the Euclidean norm.
[0029] Optionally, in one embodiment of the present invention, in step 4, the central processing unit employs a neural network when fusing multi-base station sensing data. The implementation takes as input group sensing data obtained and uploaded from each base station, including... , and ,in, For the sake of perception lThe obtained low-altitude target position estimation results, For the sake of perception l The obtained channel features This indicates a perception pair index. N The sum of the number of empty base layer nodes and the number of user-side devices at the ground base layer is given for the first... l The sensing pairs affected by low-altitude targets, the base station and the air-to-ground layer, and the base station and the ground-to-ground layer user-side equipment are numbered as follows: and The output is a high-precision collaborative positioning result for low-altitude targets. Neural Network Using the Transformer encoder as the basic framework, the collaborative localization process is described as follows:
[0030] ;
[0031] in, This represents the parameters of the neural network.
[0032] Optionally, in one embodiment of the present invention, in step 4, the cooperative localization neural network is trained using gradient descent, and the cost function is the mean squared error, as described below:
[0033] ;
[0034] in, K For the sample size, The true coordinates of the low-altitude target. It is the Euclidean norm.
[0035] The air-space-ground collaborative intelligent low-altitude target perception method based on channel state information in this invention integrates multi-level communication facilities and multiple sensing nodes into a unified perception framework. It utilizes the channel state information generated during the communication process to achieve low-altitude target positioning, which can significantly expand the perception range and eliminate perception blind spots. In addition, it adopts a neural network to realize a two-stage collaborative perception process, which can effectively reduce the transmission overhead of perception data while ensuring high positioning accuracy.
[0036] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0037] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0038] Figure 1 A flowchart of an air-space-ground coordinated intelligent low-altitude target sensing method based on channel state information provided in an embodiment of the present invention;
[0039] Figure 2 This is a schematic diagram of a space-air-ground collaborative low-altitude target sensing architecture according to an embodiment of the present invention;
[0040] Figure 3 This is a schematic diagram of a two-stage collaborative low-altitude target localization based on AI, according to an embodiment of the present invention. Detailed Implementation
[0041] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.
[0042] In low-altitude scenarios, the required surveillance area is vast, demanding sufficient coverage of sensing signals in both horizontal and vertical directions. However, limitations in antenna directivity, array elevation, and line-of-sight path obstruction caused by tall buildings create significant blind spots, reducing sensing reliability. Furthermore, the coverage of a single sensing node pair is limited and susceptible to interference in multipath-rich or heavily obstructed channels. To overcome this limitation, a multi-node collaborative paradigm is urgently needed to improve sensing reliability. Regarding node selection, existing technologies mostly rely on multi-BS (Base Station) collaboration for low-altitude target sensing. While BS possesses high-precision sensing capabilities, its sparse deployment limits spatial coverage, and antenna tilt towards the ground makes high-altitude coverage difficult to guarantee. Therefore, this invention designs a space-air-ground collaborative low-altitude target sensing framework. It utilizes Communication Instability (CSI) between communication facilities at different altitudes for low-altitude sensing (including high-altitude platforms such as airships, UAVs, and ground-based BS and CPEs), employing AI methods and neural networks to achieve two-stage collaborative target localization, thus improving the framework's performance and robustness.
[0043] Figure 1 This is a flowchart of a space-air-ground coordinated intelligent low-altitude target perception method based on channel state information, according to an embodiment of the present invention.
[0044] like Figure 1 As shown, the air-space-ground coordinated intelligent low-altitude target perception method based on channel state information includes the following steps:
[0045] Step 1: Construct a three-layer collaborative sensing network consisting of air, space, and ground. The air layer includes remote sensing satellites, enabling coarse-grained wide-area monitoring. The ground layer communicates with ground base stations via UAV swarms or High-Altitude Platform Stations (HAPS) airships, comprising... N spEach node; a fixed wireless access (FWA) network is deployed at the ground level. M One base station and N gd Each customer-side device (CPE) can communicate. N = N sp + N gd Communication signals between the air-base layer and the ground-base layer can assist in the detection of low-altitude targets. The coordinates of low-altitude targets are represented as follows: .
[0046] The uplink transmission between airborne base station nodes and user-side equipment and the ground base station is considered and performed within a Multiple-Input Multiple-Output (MIMO) Orthogonal Frequency Division Multiplexing (OFDM) system. Uplink channel state information is used to detect low-altitude targets, which can affect the wireless channel. Airborne base station nodes communicate with the ground base station via a Reconfigurable Intelligent Surface (RIS) to avoid the problem of poor signal coverage at high altitudes caused by base station antenna downtilt.
[0047] Step 2: After the airborne satellite detects a low-altitude target, it reports to the Central Processing Unit (CPU). The CPU then sends instructions to each base station, which in turn obtains the channel state information between itself and all airborne base station nodes and ground-based user-side equipment. .
[0048] Channel state information is obtained at the base station through pilot transmission and channel estimation, with each node operating at a different frequency.
[0049] Step 3: Using base stations and airborne ground-level nodes, and base stations and ground-level user-side equipment as sensing pairs for monitoring low-altitude targets, each base station performs independent detection and filters out... L A perception pair affected by low-altitude targets, with node index as follows: and obtain the corresponding channel state information. For each perception pair Each base station performs independent positioning to obtain a location estimate. Simultaneous extraction of lightweight features This step yielded a total of L Each independent location result and its corresponding channel features and index.
[0050] In one embodiment of the present invention, in step 3, each base station checks the channel state information. Conduct independent testing. For channel state information between the base station and airborne base station nodes or ground-based user-side equipment, a preprocessing function is first used. Channel state information Preprocessing is performed to obtain the preprocessed channel. :
[0051] ;
[0052] The independent detection process uses a neural network. To achieve this, the preprocessed channel Input Neural Network The output is the independent detection result. Neural Networks It consists of convolutional layers, batch normalization layers, and fully connected layers, for the () m , n The independent detection process for each of the 10 perception pairs is described as follows:
[0053] ;
[0054] in, This represents the neural network parameters. All perception pairs are independently detected using the same neural network architecture, but with different neural network parameters, based on the independent detection results. Filter out L A sensing pair affected by low-altitude targets and its corresponding channel state information. , M For the number of base stations, N It is the sum of the number of empty base station nodes and the number of user-side devices at the ground level.
[0055] The independent detection neural network is trained using gradient descent. The cost function, which is the binary cross-entropy, is minimized by iteratively optimizing the neural network parameters. The cost function is described as follows:
[0056] ;
[0057] in,( k ) is the sample index. K For the sample size, For the Euclidean norm, for The probability of.
[0058] In one embodiment of the present invention, in step 3, a preprocessing function is used. Perception l Corresponding channel state information Preprocessing is performed to obtain the preprocessed channel. :
[0059] ;
[0060] Each base station uses a neural network for independent positioning. To achieve this, the preprocessed channel Input Neural Network The output is a low-altitude target position estimate. and channel characteristics These two figures represent the final output and intermediate layer output of the neural network, respectively; neural network It mainly consists of convolutional layers, batch normalization layers, and fully connected layers, for the first... l The independent localization process for each sensing pair is described as follows:
[0061] ;
[0062] in, This represents the neural network parameters. Each sensor uses the same neural network architecture to independently locate objects, but employs different neural network parameters.
[0063] L Each independent localization neural network is trained on its own dataset using gradient descent. The cost function, which is the mean squared error, is minimized by iteratively optimizing the neural network parameters. The cost function is described as follows:
[0064] ;
[0065] in, K For the sample size, ( k ) is the sample index. The true coordinates of the low-altitude target. It is the Euclidean norm.
[0066] Step 4: Each base station uploads its positioning results, channel characteristics, and the index of the sensing nodes to the central processing unit (CPU). The CPU then fuses all the sensing data and utilizes a neural network. Obtain high-precision cooperative positioning results .
[0067] In one embodiment of the present invention, in step 4, the central processing unit employs a neural network when fusing multi-base station sensing data. The implementation takes as input group sensing data obtained and uploaded from each base station, including... , and ,in, For the sake of perception lThe obtained low-altitude target position estimation results, For the sake of perception l The obtained channel features This indicates a perception pair index. N The sum of the number of empty base layer nodes and the number of user-side devices at the ground base layer is given for the first... l The sensing pairs affected by low-altitude targets, the base station and the air-to-ground layer, and the base station and the ground-to-ground layer user-side equipment are numbered as follows: and The output is a high-precision collaborative positioning result for low-altitude targets. Neural Network Using the Transformer encoder as the basic framework, the collaborative localization process is described as follows:
[0068] ;
[0069] in, This represents the parameters of the neural network.
[0070] The cooperative localization neural network is trained using gradient descent, with the cost function being the mean squared error, as described below:
[0071] ;
[0072] in, K For the sample size, The true coordinates of the low-altitude target. It is the Euclidean norm.
[0073] The following is a detailed description of the air-space-ground coordinated intelligent low-altitude target perception method based on channel state information according to a specific embodiment of the present invention. Specifically, it includes the following steps:
[0074] Step 1: In an uplink MIMO-OFDM system, M =2 BS configurations N r =64 antennas N =16 ground-level CPEs, air-level UAVs, and airship configurations N t =64 receiving antennas, carrier frequency of 2.8GHz, bandwidth of 20MHz, subcarrier spacing of 30kHz, and a total of 512 subcarriers. The low-altitude scene size is set at 200m×200m, and the above facilities are distributed within the area for communication transmission. The low-altitude target flies within a low-altitude range of 150m×150m. It is assumed that the ground-level satellite can accurately detect the presence of the low-altitude target through remote sensing technology and send a positioning request to the base station via the CPU.
[0075] Step 2, Figure 2This demonstrates a space-air-ground collaborative low-altitude target perception architecture. Specifically, the low-altitude target perception process unfolds from top to bottom, gradually transitioning from coarse-grained detection to fine-grained localization. The space-air layer includes remote sensing satellites, which initially achieve wide-area monitoring through coarse-grained remote sensing. When a satellite detects a low-altitude target, it reports it to the CPU, which then... M Each local base station (BS) sends a low-altitude target location request.
[0076] Step 3, the empty base layer includes N sp A collaborative UAV or high-altitude platform airship, with a ground-level component N gd One CPE, N = N sp + N gd All can communicate uplink with ground-based BSs; furthermore, air-based BSs can communicate with ground-based BSs via RIS to avoid the problem of poor signal coverage at high altitudes caused by BS antenna downtilt. Low-altitude targets can affect the radio channel; therefore, each BS obtains its CSI with all air-based and ground-based BS nodes separately. This is used for subsequent sensing. CSI is obtained at the BS through pilot transmission and channel estimation, with each node operating at a different frequency.
[0077] Step 4: Each BS performs independent testing, and a neural network is used to filter out [the problematic BS]. L A perception pair affected by the target, with node index as and obtain the corresponding CSI .
[0078] Step 5, Figure 3 The diagram illustrates a two-stage collaborative low-altitude target localization process based on AI. In the first stage of collaborative localization, for each sensing pair... Each BS performs independent localization and uses a neural network to obtain a location estimate. Simultaneous extraction of lightweight features This step yields a total of 10 independent positioning results and their corresponding channel features and indices.
[0079] Step 6: In the second stage of cooperative localization, each BS uploads its localization results, channel features, and the index of the sensing node to the CPU. The CPU fuses all the sensing data and uses a neural network to obtain high-precision cooperative localization results. When fusing multi-BS sensing data, a Transformer is used, with the input being the group sensing data obtained and uploaded by each BS. This method can effectively estimate more accurate low-altitude target locations.
[0080] The air-space-ground collaborative intelligent low-altitude target sensing method based on channel state information proposed in this invention integrates multi-level communication facilities and multiple sensing nodes into a unified sensing framework, utilizing channel state information generated during the communication process to achieve low-altitude target positioning. This invention significantly expands the sensing range and eliminates sensing blind spots while reducing the overhead of additional facility deployment, thereby improving monitoring reliability. Furthermore, the use of neural networks to implement a two-stage collaborative sensing process effectively reduces the transmission overhead of sensing data while ensuring high positioning accuracy.
[0081] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0082] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0083] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.
Claims
1. A space-air-ground collaborative intelligent low-altitude target sensing method based on channel state information, characterized in that, Includes the following steps: Step 1: Construct a three-layer collaborative sensing network consisting of a space-air-ground layer, an air-ground layer, and a ground-ground layer. The space-ground layer includes remote sensing satellites, the air-ground layer communicates with the ground base stations of the ground layer through a swarm of drones or high-altitude airships, and the ground layer deploys a fixed wireless access network, including multiple base stations and multiple user-side devices. Step 2: After a low-altitude target is detected at the air-to-ground level, it is reported to the central processing unit. The central processing unit sends instructions to the M base stations at the ground level. The base stations obtain the channel status information between the base station and all air-to-ground level nodes and ground-to-ground user-side equipment, respectively. Step 3: The base station and the air-to-ground node, and the base station and the ground-to-ground user-side equipment are considered as sensing pairs for monitoring low-altitude targets. Each base station independently detects channel state information, selecting L sensing pairs affected by low-altitude targets and their node indices, and obtaining the corresponding channel state information. For each sensing pair, each base station independently locates the target, obtaining a low-altitude target location estimate while extracting lightweight channel features. A total of L independent low-altitude target location estimates and corresponding channel features and node indices are obtained. In step 3, each base station detects channel state information... Conduct independent testing. For channel state information between the base station and airborne base station nodes or ground-based user-side equipment, a preprocessing function is first used. Channel state information Preprocessing is performed to obtain the preprocessed channel. : ； The independent detection process uses a neural network. To achieve this, the preprocessed channel Input Neural Network The output is the independent detection result. Neural Networks It consists of convolutional layers, batch normalization layers, and fully connected layers, for the first... The independent detection process for each sensing pair is described as follows: in, This represents the neural network parameters. All perception pairs are independently detected using the same neural network architecture, but with different neural network parameters, based on the independent detection results. L sensing pairs affected by low-altitude targets and their corresponding channel state information were selected. M represents the number of base stations, and N represents the sum of the number of airborne base station nodes and the number of ground-based user-side devices. Step 4: Each base station uploads the location estimation results of L independent low-altitude targets, along with the corresponding channel features and node indexes, to the central processing unit. The data is then fused using a neural network to obtain high-precision collaborative positioning results for the low-altitude targets.
2. The method according to claim 1, characterized in that, In step 1, the empty base station node and user-side equipment perform uplink transmission with the ground base station in the multiple-input multiple-output orthogonal frequency division multiplexing system.
3. The method according to claim 1, characterized in that, In step 1, the empty base station node communicates with the ground base station through the reconfigurable smart surface.
4. The method according to claim 1, characterized in that, In step 2, channel state information is obtained at the base station through pilot transmission and channel estimation, with each node operating at a different frequency.
5. The method according to claim 1, characterized in that, In step 3, the independent detection neural network is trained using gradient descent. The cost function, which is the binary cross-entropy, is minimized by iteratively optimizing the neural network parameters. The cost function is described as follows: Where (k) is the sample index and K is the number of samples. For the Euclidean norm, for The probability of.
6. The method according to claim 1, characterized in that, In step 3, a preprocessing function is used. Channel state information corresponding to sensing pair l Preprocessing is performed to obtain the preprocessed channel. : Each base station uses a neural network for independent positioning. To achieve this, the preprocessed channel Input Neural Network The output is a low-altitude target position estimate. and channel characteristics These two values represent the final output and intermediate layer output of the neural network, respectively. The neural network consists of convolutional layers, batch normalization layers, and fully connected layers. For the l-th perceptual pair, the independent localization process is described as follows: in, Represents the parameters of the neural network. Each sensor uses the same neural network architecture for independent localization, but employs different neural network parameters.
7. The method according to claim 6, characterized in that, In step 3, the L independent localization neural networks are trained on their respective datasets using gradient descent. The cost function, which is the mean squared error, is minimized by iteratively optimizing the neural network parameters. The cost function is described as follows: Where K is the number of samples, and (k) is the sample index. The true coordinates of the low-altitude target. It is the Euclidean norm.
8. The method according to claim 1, characterized in that, In step 4, the central processing unit uses a neural network when fusing sensing data from multiple base stations. To achieve this, the input consists of data obtained and uploaded from each base station. Group-aware data, including , and ,in, This is the result of low-altitude target position estimation obtained from sensing pair l. The channel characteristics obtained from the sensing pair l, This represents the sensing pair index, where N is the sum of the number of air base station nodes and the number of ground base station user-side devices. For the l-th sensing pair affected by low-altitude targets, the base station and air base station, and the base station and ground base station user-side devices are numbered respectively. and The output is a high-precision collaborative positioning result for low-altitude targets. The neural network uses a Transformer encoder as its basic framework, and the cooperative localization process is described as follows: in, This represents the parameters of the neural network.
9. The method according to claim 8, characterized in that, In step 4, the cooperative localization neural network is trained using gradient descent, with the cost function being the mean squared error, as described below: Where K is the sample size. The true coordinates of the low-altitude target. It is the Euclidean norm.