Radar object detection method, device, and storage medium

Abstract

The application discloses a radar object detection method and device and a storage medium, and belongs to the technical field of radars. The method comprises the following steps: acquiring a depth feature map sent by at least one radar node; mapping the depth feature map to a world coordinate system according to preset spatial alignment parameters, to generate a perspective feature map; weighting and fusing the perspective feature map through an attention mechanism according to spatial position information of the radar node and the perspective feature, to generate a fused feature map; and performing target detection based on the fused feature map, to generate a three-dimensional air situation. Through radar networking and power adjustment, the detection effect of the radar is improved through the cooperative detection capability of low-cost devices.

Description

Technical Field

[0001] This application relates to the field of radar technology, and in particular to a radar object detection method, device and storage medium. Background Technology

[0002] Radar detection technology is widely used in the field of low-altitude target surveillance. In related technologies, in order to effectively detect low-altitude, slow-moving, and small targets such as drones, high-precision, high-power radar equipment is usually used. By increasing the transmission power and receiving sensitivity, the ability to detect weak echoes is enhanced, thereby obtaining information such as the target's position and speed.

[0003] However, in order to achieve sufficient target detection capabilities, high-performance radars need to be equipped with high-power transmitters, high-sensitivity receivers, and high-precision servo mechanisms, resulting in excessively high costs per unit. This makes it difficult to deploy on a large scale in scenarios requiring large-area coverage, such as urban areas and areas around key facilities, and fails to meet the demand of low-altitude economic development for low-cost, grid-based surveillance.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide a radar object detection method, device, and storage medium, aiming to solve the technical problem of excessively high deployment costs for high-performance radar.

[0006] To achieve the above objectives, this application provides a radar object detection method, which includes the following steps: Acquire view feature maps collected by at least two radar nodes, wherein the scanning angles of the first radar node and the second radar node are not parallel. Based on the spatial location information and viewpoint features of the radar node, the viewpoint feature map is weighted and fused using an attention mechanism to generate a fused feature map. Target detection is performed based on the fused feature map to generate a three-dimensional air situation.

[0007] In one embodiment, the object detection method of the radar further includes: Transmit radar signals and collect the reflected echo data of the radar signals; The reflected echo data is processed by a complex neural network to extract depth features and generate a depth feature map, wherein the depth feature map includes the target's distance features, velocity features, angle features and / or micro-Doppler features. The depth feature map is sent to the central node in the radar node.

[0008] In one embodiment, the step of transmitting radar signals and acquiring reflected echo data of the radar signals includes: Based on the radar network information, determine the radar's scanning output power and scanning angular velocity; The radar antenna rotates based on the scanning angular velocity, and the radar signal is transmitted based on the scanning output power during the rotation. The reflected echo of the radar signal is received, and the reflected echo data of the reflected echo is determined.

[0009] In one embodiment, the step of acquiring the view feature maps collected by at least two radar nodes includes: After receiving the depth feature map sent by the target radar node, the spatial location information of the target radar node is obtained; The transformation matrix between the radar coordinate system and the world coordinate system of the target radar node is determined based on the spatial location information. Based on the transformation matrix, the feature points in the depth feature map are mapped from the radar coordinate system to the world coordinate system to generate the view feature map.

[0010] In one embodiment, before the step of acquiring the view feature maps collected by at least two radar nodes, the method further includes: Acquire topographic elevation data and feature distribution data of the area to be detected; Using a line-of-sight analysis algorithm, the field-of-sight coverage of candidate points in the area to be detected is calculated based on the terrain elevation data and the ground feature distribution data. Based on the field of view coverage, the radar node is selected from the candidate locations as the deployment location.

[0011] In one embodiment, the step of performing target detection based on the fused feature map to generate a three-dimensional air situation further includes: Target detection processing is performed on the fused feature map to extract the target's three-dimensional spatial coordinates and motion state information in the world coordinate system; Based on the three-dimensional spatial coordinates and the motion state information, the radar cross-sectional area characteristics and motion trajectory characteristics of the target are obtained; Based on the radar cross-sectional area characteristics and the motion trajectory characteristics, the target is classified and identified in combination with the preset target classification rules to determine the target type information.

[0012] In one embodiment, after the step of performing target detection based on the fused feature map to generate a three-dimensional air situation, the method further includes: The three-dimensional air situation is overlaid onto the digital twin electronic map to generate a visualized air situation interface; The visual air situation interface is displayed on the target terminal; In response to the target terminal's selection of a target icon in the visualized air situation interface, detailed information about the target corresponding to the target icon is displayed.

[0013] In one embodiment, the step of generating a fused feature map by weighting and fusing the view feature map using an attention mechanism based on the spatial location information and view features of the radar node includes: The view feature map corresponding to the radar node is obtained, and the view feature map is input into the attention fusion module; The attention fusion module calculates the correlation between feature vectors in the view feature map at spatial locations and determines the fusion weight of the view feature map at the spatial location based on the correlation. The fused feature map is generated by weighting and summing the feature vectors of the viewpoint feature map at the spatial location according to the fusion weights.

[0014] In addition, to achieve the above objectives, this application also provides a radar object detection device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the radar object detection method as described above.

[0015] In addition, to achieve the above objectives, this application also provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the radar object detection method as described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: This application acquires depth feature maps transmitted by at least one radar node, maps these depth feature maps to the world coordinate system according to preset spatial alignment parameters to generate view feature maps, and then uses an attention mechanism to weightedly fuse the view feature maps based on the spatial position information and view features of the radar nodes to generate a fused feature map. Based on the fused feature map, target detection is performed to generate a three-dimensional air situation. Thus, through the collaborative perception of multiple low-cost radar nodes, original information from multiple perspectives is fused at the feature level. The attention mechanism adaptively enhances effective features and suppresses noise interference, making the collaborative detection capability of multiple low-cost devices equivalent to or even exceeding the detection effect of a single high-performance radar. This fundamentally solves the problem of high cost and difficulty in large-scale deployment of high-performance radar, providing a feasible technical path for gridded and low-cost surveillance of low-altitude airspace. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the radar object detection method of this application; Figure 2 This is a schematic diagram of the radar angles in the radar network of the object detection method of this application. Figure 3 This is a flowchart illustrating the second embodiment of the radar object detection method of this application; Figure 4 This is a schematic diagram of the structure of a radar object detection device in the hardware operating environment involved in the embodiments of this application.

[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0022] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application embodiment is: to acquire a depth feature map sent by at least one radar node, to map the depth feature map to the world coordinate system according to a preset spatial alignment parameter, to generate a view feature map, to perform weighted fusion of the view feature map through an attention mechanism based on the spatial position information and view features of the radar node, to generate a fused feature map, and to perform target detection based on the fused feature map to generate a three-dimensional air situation.

[0024] In existing technologies, to effectively detect low-altitude, slow-moving, and small targets such as drones, high-precision, high-power radar equipment is typically used. This is achieved by increasing transmission power and receiving sensitivity to enhance the detection capability of weak echoes and obtain information such as the target's position and velocity. However, to achieve sufficient target detection capabilities, high-performance radar requires high-power transmitters, high-sensitivity receivers, and high-precision servo mechanisms, resulting in excessively high costs per unit. This makes large-scale deployment difficult in scenarios requiring extensive coverage, such as urban areas and areas surrounding key facilities, and fails to meet the demands of low-altitude economic development for low-cost, grid-based surveillance.

[0025] This application acquires depth feature maps transmitted by at least one radar node, maps these depth feature maps to the world coordinate system according to preset spatial alignment parameters to generate view feature maps, and then uses an attention mechanism to weightedly fuse the view feature maps based on the spatial position information and view features of the radar nodes to generate a fused feature map. Based on the fused feature map, target detection is performed to generate a three-dimensional air situation. Thus, through the collaborative perception of multiple low-cost radar nodes, original information from multiple perspectives is fused at the feature level. The attention mechanism adaptively enhances effective features and suppresses noise interference, making the collaborative detection capability of multiple low-cost devices equivalent to or even exceeding the detection effect of a single high-performance radar. This fundamentally solves the problem of high cost and difficulty in large-scale deployment of high-performance radar, providing a feasible technical path for gridded and low-cost surveillance of low-altitude airspace.

[0026] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0027] It should be noted that the executing entity in this embodiment can be a radar control system, or a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or radar object detection device capable of the above functions. This embodiment does not specifically limit the specific implementation. The following uses a radar control system as an example to describe this embodiment and the following embodiments.

[0028] Based on this, embodiments of this application provide a radar object detection method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the radar object detection method of this application.

[0029] In this embodiment, the object detection method of the radar includes steps S10~S30: Step S10: Obtain the view feature maps collected by at least two radar nodes, wherein the scanning angles of the first radar node and the second radar node are not parallel. In this embodiment, multiple radar devices form a radar network through distributed connections. The radar control system deployed in the radar devices or other radar control equipment can acquire data transmitted by radar nodes deployed at key points in the area to be detected via a wireless communication network. This data can be a view feature map of the corresponding radar node, or a depth feature map used to extract the view feature map. The first radar node and the second radar node are radar nodes with different scanning angles. The scanning angles between the first radar nodes are parallel or non-parallel, and the scanning angles between the second radar nodes are also parallel or non-parallel; that is, at least two radar nodes have non-parallel scanning angles.

[0030] For example, such as Figure 2 As shown, Figure 2 This is a schematic diagram of the radar angles in the radar network of this application. In the diagram, the X-axis is parallel to the ground, and the Z-axis is perpendicular to the ground. a, b, and c each represent a radar node, and their extensions represent the corresponding scanning angles. It can be seen that radar a scans at an angle parallel to the ground, while radars b and c have angle differences from a. Based on the different scanning angles of multiple radars, the radar nodes can achieve complementarity of three-dimensional spatial information, enabling the radar network to detect objects in three-dimensional space through feature-level fusion and spatial geometric calculation using two-dimensional radar.

[0031] Specifically, each radar node is deployed at high points, boundaries, or key passages within the area to be detected, with their detection areas overlapping to achieve comprehensive surveillance of key airspace. After extracting depth features locally, each radar node transmits depth feature map data packets to the radar control system via a wireless link. Each data packet contains a radar node identifier, a timestamp, the feature map data volume, and the node's own spatial location information. Upon receiving the data packets, the radar control system synchronizes the data from each node based on the timestamp, aligning feature maps collected by different nodes at the same time to the same time reference plane. Simultaneously, it establishes a mapping relationship between the radar node identifier and each node to ensure accurate differentiation of feature data from different sources during subsequent processing. The radar control system also performs integrity checks on the data packets, initiating retransmission requests to the corresponding nodes for packets that fail the check, thus ensuring the integrity and reliability of the received data.

[0032] Optionally, a central node can be selected through an election mechanism among the radar nodes, with other nodes designated as edge nodes. Alternatively, the user can designate a specific radar node or other control device connected to the radar network, such as a computer, as the central node, and the other radar nodes as edge nodes. The central node receives data transmitted from the edge nodes and performs object detection and recognition based on this data.

[0033] In one example, each radar node employs a lightweight complex neural network to extract features from the raw echo. This complex neural network directly processes the radar's I / Q dual-channel data as complex input. The generated depth feature map has a size of 64×64×32, where 64×64 represents the spatial dimension of the feature map corresponding to the grid division of the detection area in the horizontal and vertical directions, and 32 represents the number of feature channels, each corresponding to a radar feature. The range channel records the radial distance between the target and the radar, calculated by measuring the time delay between the transmitted signal and the echo signal. The velocity channel records the radial velocity of the target relative to the radar, calculated by measuring the Doppler frequency shift of the echo signal. The angle channel records the azimuth and elevation angles of the target, calculated using phase-to-amplitude or amplitude-to-amplitude angle measurement methods. The micro-Doppler channel records the micro-motion characteristics generated by rotor modulation, extracted through time-frequency analysis of the echo signal. During reception, the radar control system verifies the feature map data and initiates retransmission requests for data packets that fail verification, ensuring the integrity of the received data. Meanwhile, the radar control system identifies the feature map of the overlapping coverage area based on the positional relationship of each radar node.

[0034] In another example, the radar control system receives depth feature maps from each radar node via a mesh ad hoc network. Each radar node is deployed on a vehicle-mounted or drone-based platform and is in motion. Each radar node obtains its real-time latitude and longitude coordinates and altitude via a Global Positioning System (GPS) receiver, and its roll, pitch, and yaw attitude data via an Inertial Measurement Unit (IMU). The radar node then combines its position and attitude information with the depth feature maps. Figure 1The data is packaged and sent to the radar control system. The radar control system dynamically adjusts the data reception priority based on the real-time location of each node. For radar nodes located in critical areas, such as the edge of key airspace or near important facilities, higher data reception bandwidth and lower transmission latency are assigned to ensure priority transmission and processing of detection data in key areas. For radar nodes that are moving and whose location changes frequently, the radar control system uses an incremental transmission method, transmitting only the changes in location and feature maps, reducing data transmission volume and improving communication efficiency. The radar control system also performs spatiotemporal consistency checks on the data from each node based on the timestamps and location information reported by each node, eliminating data that is out of sync due to node movement.

[0035] It should be noted that among the relevant radars, air defense early warning radars include high-power transmit / receive (T / R) components, large antenna arrays, cooling systems, and high-precision servo mechanisms. Their cost is too high, making large-scale deployment in urban areas and around key facilities impossible. Furthermore, maintenance costs are high, requiring specialized teams. Kilometer-level low-altitude radars include solid-state transmitters, mechanical scanning mechanisms, and industrial control computer platforms. Compared to the required coverage area (square kilometers), the coverage radius of a single unit is limited (3-5 km). Achieving seamless area coverage would require a large number of units, resulting in excessively high overall costs. Vehicle-mounted radars, on the other hand, include consumer-grade monolithic microwave integrated circuits (MMICs) and low-cost antennas. However, due to performance limitations such as low power and narrow elevation angles, they are unsuitable for low-altitude surveillance and cannot be directly applied to air defense early warning scenarios.

[0036] In this embodiment, the radar control system transforms the received depth feature map from the radar coordinate system of each radar node to a unified world coordinate system, generating a view feature map. This depth feature map is a multi-dimensional feature representation generated by the radar nodes after extracting depth features from the original echo data. It is organized in tensor form, with each feature point corresponding to a specific location in the detection space and containing feature information in multiple dimensions such as range, velocity, angle, and micro-Doppler. During the reception process, the radar control system performs time alignment and integrity verification on the feature map data of each node. The preset spatial alignment parameters are a set of data describing the transformation relationship between the radar coordinate system and the world coordinate system, including rotation matrices and translation vectors. Through the spatial representation of the depth feature map in the world coordinate system, the radar control system enables features from different radar nodes to be compared and fused under the same spatial reference frame.

[0037] As an optional implementation, the radar control system acquires the spatial location information of a target radar node, which may be an edge radar node. Based on the spatial location information, a transformation matrix between the radar coordinate system and the world coordinate system of the target radar node is determined. Using this transformation matrix, feature points in the depth feature map are mapped from the radar coordinate system to the world coordinate system, generating a view feature map.

[0038] Specifically, the radar control system acquires the spatial location information of the radar node transmitting the depth feature map. This spatial location information includes the radar node's geographic coordinates and attitude angles; for example, geographic coordinates are obtained via GPS, and attitude angles are obtained via IMU. Based on the radar node's spatial location information, the radar control system determines the transformation matrix between the radar node's radar coordinate system and the world coordinate system. This transformation matrix includes rotation and translation components. The rotation component transforms the direction vector in the radar coordinate system to the direction vector in the world coordinate system. It is constructed by determining the rotation around the Z-axis based on the radar node's heading angle, the rotation around the Y-axis based on the pitch angle, and the rotation around the X-axis based on the roll angle. These three rotation matrices are then multiplied to obtain the final rotation matrix. The translation component transforms the position coordinates in the radar coordinate system to the position coordinates in the world coordinate system; its value represents the radar node's three-dimensional coordinates in the world coordinate system. The radar control system traverses every feature point in the depth feature map. For each feature point, it calculates the spatial direction of the feature point in the radar coordinate system according to the beam scanning law of the radar node. It then transforms the direction to the spatial direction in the world coordinate system using a rotation matrix. Based on the position coordinates of the radar node and the spatial direction, it calculates the three-dimensional coordinates of the feature point in the world coordinate system and associates and stores the feature vector of the feature point with the calculated spatial coordinates to form a view feature map.

[0039] Optionally, the radar control system uses pre-calibrated spatial alignment parameters for mapping. When each radar node is deployed in a fixed position, its position coordinates and antenna pointing angle are accurately measured using measuring equipment such as a total station or laser rangefinder, and the measurement results are stored in the radar control system as preset spatial alignment parameters. After receiving the depth feature map, the radar control system performs coordinate mapping according to the pre-stored calibration parameters. For radar nodes deployed on mobile platforms, real-time acquired position information is used for dynamic mapping, and during the mapping process, a motion compensation algorithm is used to calculate the displacement and rotation of the node within the time interval between radar signal transmission and reception based on the radar node's movement speed and attitude change rate, compensating and correcting the spatial coordinates of the feature points to eliminate the influence of platform motion on the calculation of feature point coordinates. For radar nodes deployed on UAV platforms, the motion compensation algorithm also considers the vibration and sway of the UAV itself, and removes high-frequency noise through a filtering algorithm to improve the accuracy of coordinate mapping.

[0040] Step S20: Based on the spatial location information and viewpoint features of the radar nodes, the viewpoint feature maps are weighted and fused using an attention mechanism to generate a fused feature map; In this embodiment, the attention mechanism is a deep learning module capable of adaptively calculating the importance of features from different sources, determining the fusion weights by calculating the correlation between features. The fused feature map is a comprehensive feature representation generated by integrating feature maps from multiple perspectives through a weighted summation method. The feature vectors at each spatial location in this fused feature map incorporate observation information from multiple radar nodes at that location, enabling multi-view information to complement and corroborate each other. The radar control system uses the attention mechanism to adaptively weight and fuse feature maps from multiple perspectives to generate the fused feature map.

[0041] Specifically, the radar control system acquires multi-view feature maps corresponding to multiple radar nodes. These feature maps are all located in the world coordinate system and have different perspectives. The radar control system inputs the multi-view feature maps into the attention fusion module. The attention fusion module includes a multi-head attention layer and a normalization layer. It first calculates the correlation between feature vectors at various spatial locations in the feature maps from different perspectives. A correlation matrix is ​​obtained by calculating the dot product or cosine similarity between the feature vectors to reflect the consistency and complementarity of information from different perspectives at the same spatial location. When the feature vectors of two perspectives are similar in direction, it indicates that the two nodes have observed consistent information at that location. When the feature vectors of two perspectives are orthogonal or opposite in direction, it indicates possible interference or occlusion. The attention fusion module determines the fusion weight of each feature map at each spatial location based on the correlation matrix, assigning higher fusion weights to features with high correlation and lower fusion weights to features with low correlation, in order to suppress noise and interference. Based on the determined fusion weights, the attention fusion module performs a weighted summation of the feature vectors of each feature map at each spatial location to generate a fused feature map.

[0042] For example, the radar control system acquires three view feature maps generated by three radar nodes, each with a spatial size of 64×64 and 32 feature channels. The radar control system concatenates the three view feature maps in the channel dimension into an input tensor of dimension 3×64×64×32, which is then input to the multi-head attention fusion module. The multi-head attention fusion module uses four attention heads to compute in parallel, with each attention head independently calculating the correlation between features from different viewpoints. During the computation, each attention head first transforms the input tensor through a linear transformation to generate a query matrix, a key matrix, and a value matrix. The query matrix represents the feature to be queried, the key matrix represents the queried feature, and the value matrix represents the feature to be weighted and fused. The attention head calculates the dot product of the query matrix and the key matrix and divides it by a scaling factor to obtain an attention score matrix. Each element in the attention score matrix represents the correlation between two feature vectors. The attention head normalizes the attention score matrix using the Softmax function to obtain an attention weight matrix, and then multiplies the attention weight matrix by the value matrix to obtain the weighted features. The outputs of the four attention heads are concatenated along the channel dimension and linearly transformed through a fully connected layer to generate a fused feature map. The feature vector at each spatial location in this fused feature map is a weighted combination of the observation features of the three radar nodes at that location, with the weights reflecting the reliability of the information from each node at that location.

[0043] Optionally, the radar control system acquires the view feature map corresponding to the radar node and inputs it into the attention fusion module. The attention fusion module calculates the correlation between feature vectors at various spatial locations in the view feature map and determines the fusion weight of the view feature map at the corresponding spatial location based on the correlation. The radar control system then uses this fusion weight to perform a weighted summation of the feature vectors in the view feature map at that spatial location to generate a fused feature map.

[0044] Optionally, the radar control system enhances the attention mechanism based on the spatial location information of the radar nodes. The radar control system first calculates the spatial distribution of each radar node relative to the detection area. For each spatial location within the detection area, the radar control system calculates the distance between each radar node and that location, as well as the observation angle. Nodes closer to the target location have a higher signal-to-noise ratio, and nodes with better observation angles have a lower probability of obstruction. The radar control system calculates the geometric prior weights for each node based on the distance and angle. These geometric prior weights are calculated using a Gaussian or exponential function; the smaller the distance and the better the angle, the larger the weight. The radar control system introduces these geometric prior weights as initial weights into the attention fusion module. When calculating the attention score, the attention fusion module multiplies the initial weights by the correlation score to obtain the fusion weights. When feature correlation is unclear, the geometric prior weights can provide an effective basis for fusion. When feature correlation is clear, feature correlation plays a dominant role, and the geometric prior weights serve only as an auxiliary reference. The radar control system also dynamically updates the geometric prior weights. When a node is assigned a low weight for multiple consecutive frames, the geometric prior weight of that node is reduced to avoid the continuous low-quality data of a single node affecting the fusion result.

[0045] In one implementation, since two-dimensional radar cannot measure altitude, multi-radar cross-positioning 3D perception technology utilizes multiple two-dimensional radar nodes to observe the same target from different angles. Through feature-level fusion and spatial geometric calculation, it achieves low-cost two-dimensional radar network altitude measurement. Due to the high requirements for spatiotemporal synchronization among multiple radars, dynamic spatial alignment and coordinate system mapping technology employs a two-layer alignment mechanism combining preset calibration with real-time GPS / INS correction to accurately map the feature maps of different nodes to a unified world coordinate system. Furthermore, feature-level fusion replaces track-level fusion technology, overcoming the limitations of shallow traditional track-level fusion. It fuses multi-radar data at the feature level, preserving fine-grained information such as phase and amplitude in the original echoes, thus improving information utilization. The distributed and scalable networking architecture technology adopts an edge-center distributed architecture, with edge nodes responsible for feature extraction and central nodes responsible for fusion detection. It supports smooth expansion from a few to hundreds of nodes, overcoming the scale bottleneck of centralized processing.

[0046] Step S30: Target detection is performed based on the fused feature map to generate a three-dimensional air situation.

[0047] In this embodiment, the radar control system performs target detection processing on the fused feature map, extracting the target's position, motion state, and type information in three-dimensional space to generate a three-dimensional air situation. The three-dimensional air situation is a digital description of the target distribution and motion state within the detection area, including the target's three-dimensional spatial coordinates, motion state, type information, and spatiotemporal relationships, which can be used for subsequent airspace management, threat assessment, and countermeasure decision-making.

[0048] Furthermore, the radar control system can generate continuous temporal information about the target based on its position and motion state in three-dimensional space over a period of time, in order to obtain information such as the target's flight trajectory and behavior.

[0049] Furthermore, the system achieves target detection and classification through signal processing and AI. Specifically, the end-to-end processing technology of complex neural networks directly processes radar in-phase / quadrature (I / Q) dual-channel data as complex input, avoiding the phase information loss caused by splitting real numbers in traditional methods due to the small radar cross section (RCS) and low echo signal-to-noise ratio of the target. This improves target detection sensitivity under low SNR conditions. In addition, due to the abundance of ground clutter, the system constructs a multi-head attention mechanism through cross-modal attention feature fusion technology. This adaptively weights and fuses features from different radar nodes, assigning higher weights to high-confidence areas and effectively suppressing ground clutter using multi-view information.

[0050] Furthermore, the system integrates target trajectory features (speed, altitude, route deviation) with airspace management rules (electronic fences, no-fly zones, time windows), uses graph neural networks to detect trajectory anomalies, outputs compliance confidence and reasons for violations, and distinguishes between legal and illegal flights. Through feature + rule fusion intelligent discrimination technology, the accuracy of "black flight" detection is improved.

[0051] In one embodiment, the radar control system performs target detection processing on the fused feature map, which includes target classification and target localization. The radar control system inputs the fused feature map into a detection head network, which is constructed based on a convolutional neural network and contains multiple convolutional and fully connected layers. The detection head network performs layer-by-layer convolution processing on the fused feature map, extracting high-level semantic features. Through a classification branch, it outputs the probability of a target's presence at each spatial location and the probability distribution of the target's type. Through a regression branch, it outputs the target's 3D bounding box parameters, including center point coordinates, length, width, height, and orientation angle. On the target confidence map output by the classification branch, the radar control system uses a non-maximum suppression algorithm to filter out target candidate points with confidence scores higher than a preset threshold. For each selected candidate point, it extracts the corresponding bounding box parameters from the bounding box parameter map output by the regression branch and calculates the target's 3D spatial coordinates and size information in the world coordinate system. Based on the detection results of multiple consecutive frames of fused feature maps, the radar control system tracks the target using a Kalman filter algorithm to obtain the target's motion state information, including velocity and acceleration vectors.

[0052] Optionally, the radar control system performs target detection processing on the fused feature map, extracts the target's three-dimensional spatial coordinates and motion state information in the world coordinate system, obtains the target's radar cross-sectional area features and motion trajectory features based on the three-dimensional spatial coordinates and motion state information, and classifies and identifies the target based on the radar cross-sectional area features and motion trajectory features, combined with preset target classification rules, to determine the target's type information.

[0053] For example, in the process of feature-based classification and recognition, the radar control system can combine flight trajectory features for identification. For instance, drones have relatively stable flight attitudes, while flying objects such as birds exhibit wave-like flight trajectories due to wing flapping and other behaviors. The system can combine relevant feature information such as cross-sectional area and flight trajectory to identify the object type. Through micro-Doppler feature enhancement and classification technology, a micro-Doppler enhancement module is designed in a complex neural network to amplify the micro-motion modulation features of rotary-wing drones, extract the target's "micro-Doppler fingerprint," and improve the classification accuracy of drones / birds.

[0054] Optionally, the radar control system may also include a digital twin-based visualization interface, which generates a visualization interface by overlaying a three-dimensional air situation onto a digital twin electronic map, and displays the visualization interface on the target terminal. Simultaneously, in response to the user's selection of a target icon in the visualization interface on the target terminal, detailed information about the target corresponding to the icon is displayed.

[0055] In one example application scenario, when disasters such as earthquakes or fires occur and communication infrastructure is severely damaged, rescue teams carry multiple portable radar nodes into the disaster area, quickly setting up radar nodes at high points and in open areas. After powering on, each radar node automatically completes network formation and elects a central node via a wireless ad hoc network protocol. The central node receives depth feature maps reported by each edge node and fuses them to generate a three-dimensional air situation covering the disaster area. When rescue drones enter the disaster area to perform material delivery missions, the radar network system uses multi-view collaborative detection. Even if a single node cannot continuously track a drone due to obstruction by mountains, other nodes can still relay the observation, ensuring the integrity of the drone's flight path. Command personnel can intuitively view the drone's position, altitude, and speed through a digital twin electronic map, achieving full-process monitoring of the rescue drone.

[0056] In another application scenario, multiple patrol vehicles or speedboats carrying radar nodes move along border lines or waterways. Each node acquires its real-time position and attitude via GPS and IMU, automatically forming a mobile radar network system through a wireless ad hoc network. When an illegal target suddenly appears from a valley or water surface, multiple mobile nodes simultaneously detect the target from different angles. Each edge node extracts depth feature maps locally and sends them to the central node. The central node fuses multi-view features through an attention mechanism, accurately identifies the target type, calculates the target's 3D coordinates, and pushes the target's position, speed, and heading to each mobile terminal in real time. When a node leaves the network, the network system automatically triggers a re-election, with the node with the best network connectivity among the remaining nodes taking over the responsibilities of the central node. This ensures the continuity and reliability of the network system during movement, enabling all-weather tracking and detection of low-altitude or surface targets.

[0057] It should be noted that individual radars typically have detection blind spots due to limitations imposed by their installation location, beam shape, elevation angle coverage, and environmental obstructions such as buildings, mountains, and trees. This results in low-altitude targets not being continuously and stably detected in specific airspaces. Therefore, by networking multiple radar nodes collaboratively, the differences in spatial location, scanning angle, and detection perspective among different nodes can be utilized to compensate for the blind spots of a single radar. Specifically, different radar nodes are deployed in different geographical locations, such as high points, boundaries, and key passages, and their respective detection blind spots do not overlap spatially. Through system-level field-of-view coverage planning, the effective detection areas of multiple nodes overlap and interlock, eliminating blind spots that a single node cannot cover. When a radar node cannot detect a specific airspace due to obstruction by terrain features or buildings, radar nodes located at other angles may be completely unaffected by the obstruction due to different line-of-sight paths. Through multi-view joint observation, effective penetration or bypassing of obstructed areas can be achieved, significantly reducing detection losses caused by environmental obstruction. In addition, the central node assesses the detection coverage status of each radar node in real time, identifies the location and range of blind spots, and dynamically schedules adjacent nodes to adjust scanning parameters or increase detection resources to actively compensate for blind spots and ensure the detection continuity of the overall network system.

[0058] This application embodiment acquires depth feature maps sent by at least one radar node, maps the depth feature maps to the world coordinate system according to preset spatial alignment parameters to generate view feature maps, and then uses an attention mechanism to weightedly fuse the view feature maps based on the spatial position information and view features of the radar nodes to generate fused feature maps. Target detection is then performed based on the fused feature maps to generate a three-dimensional air situation awareness. Thus, through the collaborative perception of multiple low-cost radar nodes, original information from multiple perspectives is fused at the feature level. The attention mechanism adaptively enhances effective features and suppresses noise interference, making the collaborative detection capability of multiple low-cost devices equivalent to or even exceeding the detection effect of a single high-performance radar. This fundamentally solves the problem of high cost and difficulty in large-scale deployment of high-performance radars, providing a feasible technical path for gridded, low-cost surveillance of low-altitude airspace.

[0059] Based on the same inventive concept, this application also provides a second embodiment, referring to... Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the radar object detection method of this application.

[0060] In this embodiment, the object detection method of the radar further includes steps S11 to S13: Step S11: Send radar signals and collect the reflected echo data of radar signals; Step S12: Extract depth features from the reflected echo data using a complex neural network to generate a depth feature map, wherein the depth feature map includes the target's distance features, velocity features, angle features and / or micro-Doppler features. Step S13: Send the depth feature map to the central node in the radar node.

[0061] In this embodiment, when a radar node is elected as an edge node, its deployed radar control system can perform radar signal transmission and echo acquisition at the radar edge node. It then uses a complex neural network to extract depth features from the reflected echo data, generating a depth feature map which is sent to the central node. The complex neural network is a deep learning network that directly processes complex domain data. Its input is the radar's I / Q dual-channel data, treating the real and imaginary parts as the real and imaginary parts of a complex number for joint processing, thus fully preserving the phase information of the echo signal. The depth feature map is a multi-dimensional feature representation generated by this node after extracting depth features from the original echo data. It is organized in tensor form, with each feature point corresponding to a specific location in the detection space and containing feature information in multiple dimensions such as distance, velocity, angle, and micro-Doppler.

[0062] It should be noted that during the networking process, beam pointing jitter occurs due to platform movement. Jitter cancellation and motion compensation techniques are employed, along with real-time acquisition of platform attitude changes using IMU / INS data. A jitter propagation model is constructed to predict beam pointing deviations. Motion compensation is performed on the echo at the signal layer, and geometric correction is applied to the feature map at the feature layer to reduce dynamic platform angle measurement errors. Furthermore, since the relative positions of nodes change dynamically, the system also employs dynamic self-organizing networking and real-time spatial registration techniques. A lightweight self-organizing networking protocol is designed to support automatic node discovery, authentication, and time synchronization, with nodes broadcasting real-time position / attitude information. The central node dynamically updates the spatial transformation matrix, supporting rapid node joining / leaving. When target handover is prone to interruption during movement, mobile relay tracking technology is used. Based on target motion prediction and node position changes, smooth target handover between mobile nodes is achieved. A collaborative filtering algorithm is employed to maintain track continuity during the handover process, improving the handover success rate.

[0063] Specifically, after determining its own scanning output power and scanning angular velocity based on radar network information, the radar antenna is driven to transmit radar signals in continuous scanning mode. Simultaneously with signal transmission, the receiving antenna continuously collects reflected echo data after the radar signal is reflected by the target. The collected I / Q dual-channel data is used as input to a complex neural network. The convolutional layers of the complex neural network use complex convolution kernels to perform convolution operations on the input complex data, extracting features in the complex domain. Through the stacking of multiple complex convolutional layers and pooling layers, the complex neural network progressively extracts the target's range, velocity, angle, and micro-Doppler features. The range feature is calculated by measuring the time delay between the transmitted and echo signals, reflecting the radial distance between the target and the local node. The velocity feature is calculated by measuring the Doppler frequency shift of the echo signal, reflecting the target's radial velocity relative to the local node. The angle feature is calculated using phase-to-amplitude or amplitude-to-amplitude angle measurement methods, reflecting the target's azimuth and elevation angles. The micro-Doppler feature is extracted through time-frequency analysis of the echo signal, reflecting the micro-motion modulation characteristics generated by the target's rotating components. The complex neural network organizes the extracted multidimensional features into a deep feature map, with each feature point carrying the four types of feature information mentioned above. The generated deep feature map is then transmitted to the central node via a wireless ad hoc network link.

[0064] Optionally, the radar node can determine the radar's scanning output power and scanning angular velocity based on the radar network information, rotate the radar antenna based on the scanning angular velocity, and transmit radar signals based on the scanning output power during the rotation, receive the reflected echoes of the radar signals, and determine the reflected echo data.

[0065] It should be noted that radar scanning waves are typically pulse scanning waves. After the radar antenna transmits a pulse signal, the radar receives the reflected echo of the pulse signal and then rotates the radar antenna accordingly. In this situation, fast-flying objects such as drones can easily pass through the radar's blind spots, leading to inaccurate detection results.

[0066] Furthermore, radar nodes determine radar scanning accuracy through continuous radar scanning. Based on the mapping relationship between electromagnetic power and actual radar performance, the radar power supply output power is designed to minimize scanning blind spots in a networked configuration. Optionally, the radar's scanning output power can be dynamically calculated based on the actual radar network configuration. Simultaneously, to avoid interference from radar wavelengths specified in relevant protocols on aviation and other applications, adjustments need to be made to the actual application wavelength of the radar.

[0067] Furthermore, to avoid interference from radar wavelengths in relevant protocols on aviation and other situations, it is necessary to adjust the actual application wavelength of the radar. The system acquires the current scanning mode and transmit power information of each radar edge node. Based on the target distribution in the three-dimensional air situation, it determines whether there are blind spots in the scanning coverage area of ​​the current radar network. When a blind spot is determined, a scanning mode adjustment command or a transmit power adjustment command is generated and sent to the corresponding radar edge node to control the corresponding radar edge node to adjust its scanning mode or transmit power.

[0068] Specifically, by acquiring the spatial location information and detection area coverage information of adjacent nodes within the network, and calculating the distance and relative azimuth between the radar node and its adjacent nodes, it is determined whether the radar's detection area overlaps with that of the adjacent nodes. If the detection area overlaps with the adjacent nodes, the scanning output power is set to be lower than the rated power when a single node operates independently, so that the edge of the detection area is connected to or partially overlaps with the edge of the detection area of ​​the adjacent node. This ensures coverage continuity while reducing energy consumption and electromagnetic interference.

[0069] Optionally, based on the scanning cycle coordination command issued by the central node, the scanning angular velocity is set to a value matching that of other nodes in the network, ensuring consistent scanning cycles across all nodes and facilitating time alignment and fusion processing of multi-node data. The radar antenna is controlled to rotate continuously and uniformly at a determined scanning angular velocity, causing the beam pointing angle to change continuously over time without beam turning pauses, thus eliminating scanning blind spots caused by beam turning gaps. During the radar antenna rotation, the radar control system controls the radar transmitter to continuously transmit radar signals at a determined scanning output power. The transmitted signal is a continuous wave signal or a high-repetition-frequency pulse train, ensuring the beam always illuminates the detection space during the scanning process. The receiving antenna synchronously acquires the reflected echoes of the radar signal after reflection from the target, amplifies, filters, mixes, and performs analog-to-digital conversion on the echo signals to determine the reflected echo data.

[0070] Optionally, the system can acquire terrain elevation data and ground feature distribution data of the area to be detected, and calculate the field of view coverage of the candidate points in the area to be detected based on the terrain elevation data and ground feature distribution data using a line-of-sight analysis algorithm. Based on the field of view coverage, the system selects the radar node deployment points from the candidate points.

[0071] Specifically, the system acquires the boundary of the area to be detected, discretizes the area into multiple spatial sampling points, each with three-dimensional coordinates, and the distribution density is determined according to the detection accuracy requirements. Coordinates of multiple candidate points are acquired; these candidate points can be high points, ridgelines, building rooftops, or road intersections within the area to be detected, used for deploying radar edge nodes. For each candidate point, all spatial sampling points are traversed, and the line connecting the candidate point and other spatial sampling points is calculated. Based on terrain elevation data and ground feature distribution data, it is determined whether the line intersects with the terrain surface or ground features. Interpolation points are taken along the line at a certain step size, and the elevation value of each interpolation point is obtained. If the elevation value of any interpolation point is higher than the height of the line at that point, the spatial sampling point is determined to be occluded; otherwise, it is determined to be visible. The number of visible spatial sampling points for each candidate point is counted, which is used as the field of view coverage area corresponding to that candidate point. Based on the field of view coverage of each candidate point, a greedy algorithm or integer programming method is used to select combinations of candidate points. The combination ensures that the union of the visible spatial sampling points of each candidate point covers all spatial sampling points, and the distance between any two candidate points in the combination is less than twice the maximum detection range of the radar edge nodes, thus ensuring that the detection areas of adjacent candidate points overlap. The selected combinations of candidate points are then designated as deployment points, and deployment instructions for each deployment point are generated for deployment personnel.

[0072] Since the system described in Embodiment 2 of this application is a system used to implement the method of Embodiment 1 of this application, those skilled in the art can understand the specific structure and variations of the system based on the method described in Embodiment 1 of this application, and therefore will not be described again here. All systems used in the method of Embodiment 1 of this application fall within the scope of protection of this application.

[0073] This application provides a radar object detection device, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the radar object detection method of the above embodiment 1.

[0074] The following is for reference. Figure 4 The diagram illustrates a structural schematic of an object detection device suitable for implementing the radar embodiments of this application. The radar object detection device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The radar object detection device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this application.

[0075] like Figure 4As shown, the radar object detection device may include a processing unit 1001 (e.g., a core processor, graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The random access memory 1004 also stores various programs and data required for the operation of the radar object detection device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the radar's object detection equipment to communicate wirelessly or wiredly with other devices to exchange data. Although radar object detection equipment with various systems is shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0076] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0077] The radar object detection device provided in this application, employing the radar object detection method described in the above embodiments, can solve the technical problem of excessively high deployment costs for high-performance radars. Compared with the prior art, the beneficial effects of the radar object detection device provided in this application are the same as those of the radar object detection method provided in the above embodiments, and other technical features of this radar object detection device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0078] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0079] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0080] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the radar object detection method in the above embodiments.

[0081] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0082] The aforementioned computer-readable storage medium may be included in the radar's object detection device; or it may exist independently and not be assembled into the radar's object detection device.

[0083] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the radar's object detection device, the radar's object detection device: acquires a depth feature map sent by at least one radar node; maps the depth feature map to the world coordinate system according to preset spatial alignment parameters to generate a view feature map; performs weighted fusion of the view feature map through an attention mechanism based on the spatial position information and view features of the radar node to generate a fused feature map; and performs target detection based on the fused feature map to generate a three-dimensional air situation.

[0084] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0085] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation that may be implemented in systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0086] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0087] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the object detection method of the radar described above, which can solve the technical problem of excessively high deployment costs for high-performance radar. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the object detection method of the radar provided in the above embodiments, and will not be repeated here.

[0088] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A radar object detection method, characterized in that, The method includes the following steps: Acquire view feature maps collected by at least two radar nodes, wherein the radar nodes are two-dimensional radar nodes, and the scanning angles of the first radar node and the second radar node are not parallel, so that the radar nodes can complement each other in three-dimensional spatial information. Based on the spatial location information and viewpoint features of the radar node, the viewpoint feature map is weighted and fused using an attention mechanism to generate a fused feature map. Target detection is performed based on the fused feature map to generate a three-dimensional air situation. The three-dimensional air situation includes the type information of the target. Specifically, target detection processing is performed on the fused feature map to extract the three-dimensional spatial coordinates and motion state information of the target in the world coordinate system. Based on the three-dimensional spatial coordinates and the motion state information, the radar cross-sectional area features and motion trajectory features of the target are obtained. Based on the radar cross-sectional area features and the motion trajectory features, the target is classified and identified in combination with a preset target classification rule to determine the type information.

2. The radar object detection method as described in claim 1, characterized in that, The object detection method of the radar also includes: Transmit radar signals and collect the reflected echo data of the radar signals; The reflected echo data is processed by a complex neural network to extract depth features and generate a depth feature map, wherein the depth feature map includes the target's distance features, velocity features, angle features and / or micro-Doppler features. The depth feature map is sent to the central node in the radar node.

3. The radar object detection method as described in claim 2, characterized in that, The steps of transmitting radar signals and collecting the reflected echo data of the radar signals include: Based on the radar network information, determine the radar's scanning output power and scanning angular velocity; The radar antenna rotates based on the scanning angular velocity, and the radar signal is transmitted based on the scanning output power during the rotation. The reflected echo of the radar signal is received, and the reflected echo data of the reflected echo is determined.

4. The radar object detection method as described in claim 2, characterized in that, The step of acquiring the view feature maps obtained from at least two radar nodes includes: After receiving the depth feature map sent by the target radar node, the spatial location information of the target radar node is obtained; The transformation matrix between the radar coordinate system and the world coordinate system of the target radar node is determined based on the spatial location information. Based on the transformation matrix, the feature points in the depth feature map are mapped from the radar coordinate system to the world coordinate system to generate the view feature map.

5. The radar object detection method as described in claim 1, characterized in that, Before the step of acquiring the view feature maps collected by at least two radar nodes, the method further includes: Acquire topographic elevation data and feature distribution data of the area to be detected; Using a line-of-sight analysis algorithm, the field-of-sight coverage of candidate points in the area to be detected is calculated based on the terrain elevation data and the ground feature distribution data. Based on the field of view coverage, the radar node is selected from the candidate locations as the deployment location.

6. The radar object detection method as described in claim 1, characterized in that, After the step of performing target detection based on the fused feature map and generating a three-dimensional air situation, the method further includes: The three-dimensional air situation is overlaid onto the digital twin electronic map to generate a visualized air situation interface; The visual air situation interface is displayed on the target terminal; In response to the target terminal's selection of a target icon in the visualized air situation interface, detailed information about the target corresponding to the target icon is displayed.

7. The radar object detection method as described in claim 1, characterized in that, The step of generating a fused feature map by weighting and fusing the view feature map using an attention mechanism based on the spatial location information and view features of the radar node includes: The view feature map corresponding to the radar node is obtained, and the view feature map is input into the attention fusion module; The attention fusion module calculates the correlation between feature vectors in the view feature map at spatial locations and determines the fusion weight of the view feature map at the spatial location based on the correlation. The fused feature map is generated by weighting and summing the feature vectors of the viewpoint feature map at the spatial location according to the fusion weights.

8. A radar object detection device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the object detection method of the radar as described in any one of claims 1 to 7.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the radar object detection method as described in any one of claims 1 to 7.

Images