An unmanned aerial vehicle reconnaissance method, system and storage medium

Abstract

The embodiment of the application provides a kind of unmanned vehicle reconnaissance method, system and storage medium, method includes: obtaining reconnaissance range and the initial image corresponding to reconnaissance range;After target authenticity identification processing is carried out to initial image, the target position information corresponding to enemy target is obtained, and initial image is obtained by remote sensing satellite collection;According to target position information, the deployment position of reconnaissance module is determined with the preset reconnaissance radius of center of target position information and associated environmental information;Control reconnaissance module carries out reconnaissance to enemy target in deployment position. Through obtaining the initial image in reconnaissance range, enemy target identification and positioning are carried out, and enemy target in reconnaissance range is accurately identified;At the same time, the deployment position of reconnaissance module is obtained by calculation, to ensure the position accuracy of reconnaissance, and then improve the timeliness and reliability of reconnaissance.

Description

Technical Field

[0001] This invention relates to, but is not limited to, the field of unmanned aerial vehicle (UAV) technology, and particularly to a UAV reconnaissance method, system, and storage medium. Background Technology

[0002] The evolution of modern warfare has led to drones playing an increasingly important role in future wars. Drones, with their unique combat methods and effectiveness, have played a crucial role in warfare.

[0003] Although the deployment of existing unmanned combat equipment can theoretically replace human reconnaissance with unmanned aerial vehicle (UAV) systems, the rapid development of counter-UAV technology has led to challenges such as difficulty in intelligence gathering and reduced survivability in enemy-controlled areas. As a result, UAV systems are unable to obtain enemy information in a timely manner when replacing human reconnaissance, which reduces the reliability of UAV system reconnaissance. Summary of the Invention

[0004] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0005] The main objective of this invention is to provide a UAV reconnaissance method, system, and storage medium that can obtain enemy information in a timely manner and improve the reliability of UAV system reconnaissance.

[0006] In a first aspect, embodiments of the present invention provide a drone reconnaissance method applied to a target control terminal of a drone reconnaissance system, wherein the drone reconnaissance system includes at least one reconnaissance module connected to the target control terminal, and the drone reconnaissance method includes:

[0007] Obtain the reconnaissance range corresponding to the reconnaissance mission and acquire the initial image corresponding to the reconnaissance range through remote sensing satellites;

[0008] After performing target recognition on the initial image, abnormal targets and their corresponding initial location information are obtained;

[0009] Multi-angle imaging information of the anomalous target was detected by synthetic aperture radar remote sensing.

[0010] After verifying the authenticity of the abnormal target based on the multi-angle imaging information, the enemy target and the target location information corresponding to the enemy target are obtained.

[0011] The deployment location of the reconnaissance module is determined based on the target location information, a preset reconnaissance radius centered on the target location information, and associated environmental information. The associated environmental information represents the environmental information corresponding to the reconnaissance target range constructed by the preset reconnaissance radius and the target location information.

[0012] The reconnaissance module is controlled to conduct reconnaissance of the enemy target at the deployment location.

[0013] In some optional embodiments, the step of obtaining abnormal targets and their corresponding initial location information after target recognition of the initial image includes:

[0014] After performing grayscale and filtering processes on the initial image in sequence, a filtered image is obtained. A data template is determined based on the median of the filtered image. The data template is then moved in the filtered image to obtain the template image corresponding to the data template in the filtered image.

[0015] Obtain the minimum and maximum data values ​​corresponding to the template image;

[0016] When the target median is between the minimum and maximum data values, a first threshold corresponding to the template image is obtained based on the mean between the minimum and maximum data values, wherein the target median represents the median of the filtered image;

[0017] If the target value is not located between the minimum and maximum values ​​of the data, the data template is expanded according to a preset rule to obtain an updated data template. The updated data template is then moved in the filtered image to obtain the updated template image corresponding to the updated data template in the filtered image.

[0018] Obtain the minimum and maximum values ​​of the updated data corresponding to the updated template image, so that the target median value is located between the minimum and maximum values ​​of the updated data, and obtain the second threshold corresponding to the updated template image, wherein both the first threshold and the second threshold are filtering thresholds;

[0019] The template image is denoised according to the filtering threshold to obtain a denoised image.

[0020] The denoised image is divided into multiple image regions using initial segmentation lines;

[0021] After adjusting the position of the initial segmentation line based on the similarity between multiple image regions, the target edge information is obtained;

[0022] The abnormal target and its corresponding initial position information are obtained based on the target edge information.

[0023] In some optional embodiments, obtaining target edge information after adjusting the position of the initial segmentation line based on the similarity between the plurality of image regions includes:

[0024] The comparison result is obtained by comparing the first similarity between the first region and the third region and the second similarity between the second region and the third region. The third region is the region covered by the initial segmentation line. The first region is located on the first side of the third region, and the second region is located on the second side of the third region. The first region, the second region, and the third region all represent the image region.

[0025] If the comparison result indicates that the first similarity is greater than or equal to the second similarity, the initial segmentation curve is adjusted to the position of the first region.

[0026] If the comparison result indicates that the first similarity is less than the second similarity, the initial segmentation curve is adjusted to the position where the second region is located;

[0027] When the difference between the two sides of the initial segmentation curve is the largest, the target edge information is obtained.

[0028] In some optional embodiments, the step of obtaining the enemy target and its corresponding target location information after verifying the authenticity of the abnormal target based on the multi-angle imaging information further includes:

[0029] Based on the target edge information and affine transformation, the relevant position information of the abnormal target at different times is obtained;

[0030] The movement information of the abnormal target is determined based on the initial position information and the relevant position information, wherein the movement information includes turning information, speed information and trajectory information;

[0031] Construct a false target feature library and a false target prediction model;

[0032] The false target feature library, the movement information, and the false target prediction model are used to predict the true or false probability of the abnormal target to obtain the true or false probability of the target. The abnormal target with the true or false probability greater than the probability threshold is configured as the enemy target, and the initial position information of the abnormal target is configured as the target position information.

[0033] When the probability of the target being real or fake is less than a probability threshold, the multi-angle imaging information of the abnormal target is detected by synthetic aperture radar remote sensing. After verifying the authenticity of the abnormal target based on the multi-angle imaging information, the enemy target and the target's location information are obtained.

[0034] In some optional embodiments, the probability of a target being true or false is obtained by predicting its probability of being true or false based on the false target feature library, the movement information, and the false target prediction model, including:

[0035] Obtain the digital elevation model and road vector data corresponding to the reconnaissance range;

[0036] After mapping the trajectory information to the digital elevation model, the elevation data associated with the trajectory points is obtained. The elevation data indicates the terrain elevation, slope, and land cover type.

[0037] Obtain real-time map data corresponding to the reconnaissance range;

[0038] Based on the real-time map data, the road vector data, and the trajectory information, terrain travel cost information and road distance data are determined. The road distance data represents the distance between the trajectory point indicated by the trajectory information and the road. The terrain travel cost information represents the travel cost of the trajectory point indicated by the trajectory information.

[0039] After the target trajectory of the abnormal target is split into multiple time-series trajectories according to a preset time period, the target trajectory represents the trajectory indicated by the trajectory information;

[0040] The elevation data, terrain access cost information, and road distance data corresponding to each time segment trajectory are sequentially input into the false target prediction model. The false target prediction model calls the false target feature library and then performs a true / false prediction on the time segment trajectory to obtain the true / false probability of the segment.

[0041] The false target prediction model integrates the true and false probabilities of all time-series segment trajectories to obtain the first true and false probability.

[0042] After performing cluster analysis on the target edge information of all the abnormal targets, different sub-clusters with different cluster ranges are obtained. The cluster range represents the regional range of all the abnormal targets contained in the same sub-cluster.

[0043] The density ratio is determined based on the distribution of all the anomalous targets in the same sub-cluster within the cluster range;

[0044] The second true / false probability of the anomalous target is determined based on the density ratio;

[0045] The true / false probability of the abnormal target is determined based on the first true / false probability and the second true / false probability.

[0046] In some optional embodiments, determining the deployment location of the reconnaissance module based on the target location information, a preset reconnaissance radius centered on the target location information, and associated environmental information includes:

[0047] Using the location indicated by the target location information as the target center, the preset reconnaissance radius is used to construct the reconnaissance target range around the target center;

[0048] The associated environmental information corresponding to the range of the reconnaissance target is obtained, and the associated environmental information includes LiDAR terrain data, electromagnetic spectrum information and visible light image information;

[0049] A three-dimensional threat situation field is generated by fusing the LiDAR terrain data, the electromagnetic spectrum information, and the visible light image information through a convolutional network.

[0050] The associated environmental information is divided into multiple spatial grid cells after being segmented into grids.

[0051] The elevation and slope information corresponding to the spatial grid cell are determined based on the LiDAR terrain data, the signal strength corresponding to the spatial grid cell is determined based on the electromagnetic spectrum information, and the target type corresponding to the spatial grid cell is determined based on the visible light image information.

[0052] Calculate the correlation value between any two spatial grid cells based on the LiDAR terrain data;

[0053] The grid diagram information is obtained by configuring the spatial grid cells as nodes, configuring the associated values ​​as edges of the corresponding nodes, and configuring the elevation information, the slope information, the signal strength, and the target type as node features of the corresponding nodes.

[0054] Multiple node features corresponding to any target grid cell are obtained from the grid graph information, and the multiple node features are fused to obtain the first feature information;

[0055] Based on the first feature information and the association value corresponding to the target grid cell, the target neighborhood of the target grid cell is determined in the grid graph information;

[0056] After assessing the threat of the target grid cell based on the second feature information of other grid cells in the target's neighborhood and the first feature information, a target score is obtained.

[0057] A three-dimensional threat situation field is determined based on the target grid cells and the target score;

[0058] Obtain the temporal location coordinates of the enemy target;

[0059] After inputting the temporal location coordinates and the associated environmental information into the spatiotemporal feature encoder, kinematic features and radiation features are obtained. The abnormal features of the enemy target include the kinematic features and the radiation features.

[0060] Obtain the real-time pose information of the reconnaissance module;

[0061] The real-time pose information, the kinematic features, the radiation features, and the three-dimensional threat situation field are input into the deep reinforcement learning framework;

[0062] The deep reinforcement learning framework is used to dynamically optimize based on the observation clarity target and the exposure risk target, and a Pareto front solution set is generated based on the non-dominated sorting genetic algorithm. The optimal approach distance is obtained by selecting the optimal solution from the Pareto front solution set according to the tactical stage requirements.

[0063] The deployment location is determined based on the optimal proximity distance.

[0064] In some optional embodiments, calculating the correlation value between any two spatial grid cells based on the LiDAR terrain data includes:

[0065] The terrain information corresponding to each spatial grid cell is determined based on the LiDAR terrain data.

[0066] Based on the terrain information, determine the terrain connectivity information corresponding to any two spatial grid cells, where the terrain connectivity information represents the connectivity coefficient between the two spatial grid cells;

[0067] Calculate the Euclidean distance between any two of the spatial grid cells;

[0068] The correlation value between any two spatial grid cells is determined based on the Euclidean distance and the terrain connectivity information.

[0069] In some optional embodiments, determining the deployment location based on the optimal proximity distance includes:

[0070] Obtain the location range corresponding to the optimal approach distance;

[0071] The fastest travel path for the reconnaissance module to reach the location range is obtained by performing path planning based on the nearest approach range using a dynamic programming algorithm.

[0072] The deployment location within the location range is determined based on the fastest travel path.

[0073] In some optional embodiments, the unmanned aerial vehicle (UAV) reconnaissance system further includes a fire support module, which comprises a reconnaissance vehicle and a reconnaissance aircraft, a strike swarm, and a ground terminal deployed on the reconnaissance vehicle. Controlling the reconnaissance module to conduct reconnaissance of the enemy target at the deployment location includes:

[0074] The target control terminal controls the reconnaissance vehicle to reach the deployment location;

[0075] After the reconnaissance aircraft conducts reconnaissance of the enemy target, it obtains first image information and transmits the first image information back to the ground terminal, so that the ground terminal can send the first image information to the target control terminal.

[0076] The target control terminal obtains the first image information and judges the attack on the bee swarm based on the first image information to obtain a first judgment result;

[0077] When the first judgment result indicates that the strike can be effectively carried out, the target control terminal determines the strike location information corresponding to the strike bee swarm based on the first image information, and sends the strike location information to the ground terminal.

[0078] The ground terminal sends the strike location information to the strike swarm so that the strike swarm can reach the strike location indicated by the strike location information to carry out the strike. After the reconnaissance aircraft acquires images of the strike location, it sends the second image information to the target control terminal.

[0079] The target control terminal obtains the second image information and, based on the target tracking algorithm, performs a destruction judgment on the second image information to obtain a second judgment result.

[0080] When the second judgment result indicates that the destruction is complete, the target control terminal sends a return command to the ground terminal.

[0081] If the second judgment result indicates that the destruction is not complete, the target control terminal controls the strike swarm to strike the enemy target again until the enemy target is completely destroyed.

[0082] In some optional embodiments, the reconnaissance vehicle is also equipped with a decoy drone, and after controlling the reconnaissance vehicle to reach the deployment location, the process further includes:

[0083] Acquire historical reconnaissance behavior, which represents the reconnaissance behavior of the reconnaissance aircraft when its environment was most similar to that of the enemy target in history;

[0084] The decoy drone is controlled to approach the enemy target using the historical reconnaissance behavior. The decoy drone navigates according to the target location information and does not communicate with the ground terminal while approaching the enemy target.

[0085] In some alternative embodiments, the process of the decoy drone approaching the enemy target with the aforementioned historical reconnaissance behavior further includes:

[0086] When the decoy drone encounters strong electromagnetic suppression within the data link frequency band, the first strike unit, which is composed of a swarm of decoy drones, locks onto and strikes the suppression antenna. The first strike unit is equipped with an anti-radiation module.

[0087] By attracting and occupying the detection, interference, suppression, and attack resources of the anti-drone system through other drones in the feint swarm, the anti-drone system can be saturated.

[0088] The strikeable duration of the decoy drone and the first time when the second strike unit of the decoy drone begins to strike are determined based on the initial strike resources of the decoy drone, the real-time strike resources corresponding to the attraction and occupation of the detection, interference, suppression and strike resources of the anti-drone system.

[0089] Electromagnetic characteristic data is obtained by collecting electromagnetic characteristic data emitted by the enemy's anti-navigation radar and electromagnetic suppression antenna in real time using the decoy drone.

[0090] The decoy drone generates an anti-electromagnetic map of the corresponding area under the target location information based on the electromagnetic feature data, and autonomously avoids strong detection signals and electromagnetic suppression areas based on the anti-electromagnetic map to determine the target communication mode. Based on the target communication mode, the anti-electromagnetic map, the first time, and the strikeable duration are sent to the ground terminal.

[0091] In some optional embodiments, the UAV reconnaissance system further includes a fire support module; before determining the attack on the strike swarm based on the first image information to obtain a first determination result, the method further includes:

[0092] The ground terminal determines the reconnaissance route of the reconnaissance aircraft based on the anti-electromagnetic map, and calculates the second time corresponding to the reconnaissance aircraft conducting reconnaissance along the reconnaissance route;

[0093] If the time difference between the first time and the second time is less than the strikeable duration, the reconnaissance aircraft is controlled to collect information on the enemy target according to the reconnaissance route to obtain the first image information;

[0094] If the time difference between the first time and the second time is greater than or equal to the strikeable duration, a strike control request command sent by the ground terminal is received.

[0095] The enemy target and the target location information corresponding to the enemy target are obtained from the request for strike control command;

[0096] The fire support module is controlled to strike the target location indicated by the target location information.

[0097] In some optional embodiments, the method further includes:

[0098] If the first judgment result indicates that an effective strike cannot be carried out, the target control terminal controls the fire support module to strike the strike location indicated by the strike location information.

[0099] The ground terminal acquires images of the strike location via the reconnaissance aircraft and then sends the third image information to the target control terminal.

[0100] The target control terminal obtains a third judgment result after determining whether the third image information has been destroyed based on the target tracking algorithm.

[0101] When the third judgment result indicates that the destruction is complete, the target control terminal sends a return command to the ground terminal.

[0102] If the third judgment result indicates that the destruction is not completed, the target control terminal controls the fire support module to strike the enemy target again.

[0103] In a second aspect, embodiments of the present invention provide an unmanned aerial vehicle (UAV) reconnaissance system, including a controller. The controller includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the UAV reconnaissance method described in the first aspect.

[0104] Thirdly, embodiments of the present invention provide a computer storage medium storing computer-executable instructions, which are used to execute the UAV reconnaissance method described in the first aspect.

[0105] The beneficial effects of this invention include: when conducting reconnaissance of enemy targets, this invention acquires the reconnaissance range corresponding to the reconnaissance mission and obtains an initial image corresponding to the reconnaissance range through remote sensing satellites; after target identification of the initial image, it acquires abnormal targets and their corresponding initial position information; it detects multi-angle imaging information of the abnormal targets through synthetic aperture radar remote sensing; after verifying the authenticity of the abnormal targets based on the multi-angle imaging information, it obtains the enemy targets and their corresponding target position information; it determines the deployment position of the reconnaissance module based on the target position information, a preset reconnaissance radius centered on the target position information, and associated environmental information, wherein the associated environmental information represents the environmental information corresponding to the reconnaissance target range constructed through the preset reconnaissance radius and the target position information; and it controls the reconnaissance module to conduct reconnaissance of the enemy targets at the deployment position. By acquiring initial images within the reconnaissance range through remote sensing satellites, enemy targets are identified and located based on the initial images, thereby accurately identifying enemy targets within the reconnaissance range; simultaneously, the deployment position of the reconnaissance module is calculated to ensure accurate reconnaissance location, thereby improving the timeliness and reliability of reconnaissance. Therefore, this application can obtain enemy information in a timely manner, improving the reliability of UAV system reconnaissance.

[0106] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0107] Figure 1 This is a schematic diagram of the unmanned aerial vehicle (UAV) reconnaissance system provided in an embodiment of the present invention;

[0108] Figure 2 This is a flowchart illustrating the steps of an unmanned aerial vehicle (UAV) reconnaissance method provided in an embodiment of the present invention.

[0109] Figure 3 This is a flowchart of another UAV reconnaissance method provided in an embodiment of the present invention;

[0110] Figure 4 This is a schematic diagram of a controller provided in one embodiment of the present invention.

[0111] Reference numerals: Controller 1000, Processor 1100, Memory 1200. Detailed Implementation

[0112] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0113] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0114] Reference Figure 1 The UAV reconnaissance system of this application includes a target control terminal and at least one reconnaissance module. The target control terminal controls one or more reconnaissance modules to conduct reconnaissance of enemy targets and returns the reconnaissance results to the target control terminal via the reconnaissance modules. The reconnaissance modules move, collect data, and engage enemy targets according to the control commands from the target control terminal. Specifically, the required number of reconnaissance modules can be determined based on the reconnaissance requirements of the reconnaissance mission, and then the reconnaissance modules can be positioned around the enemy targets to enable multi-directional reconnaissance and engagement. A fire support module is also positioned near the target control terminal. If the strike swarm on the reconnaissance modules cannot effectively engage the enemy targets, or if the strike swarm's engagement time is less than the available engagement time, the target control terminal controls the fire support module to destroy the enemy targets.

[0115] like Figure 2 As shown, an embodiment of the present invention provides a drone reconnaissance method, including:

[0116] S100. Obtain the reconnaissance range corresponding to the reconnaissance mission and acquire the initial image corresponding to the reconnaissance range through remote sensing satellite.

[0117] For example, the target control terminal of this application is a control terminal for controlling an unmanned aerial vehicle (UAV) reconnaissance system. The target control terminal can generate a corresponding reconnaissance range based on the specific reconnaissance requirements. Based on the required reconnaissance range, the target control terminal acquires initial images collected by remote sensing satellites within that reconnaissance range, and then performs subsequent reconnaissance based on the initial images.

[0118] In some optional embodiments, when the target control terminal receives a reconnaissance mission: the reconnaissance area is determined according to the reconnaissance mission and a preset area division model; the reconnaissance range is obtained after correcting the reconnaissance area according to the reconnaissance rules corresponding to the reconnaissance database.

[0119] For example, after receiving a specific reconnaissance mission, the target control terminal obtains the reconnaissance area from the reconnaissance mission (the specific reconnaissance area can be an area included in the reconnaissance mission or an area generated according to reconnaissance requirements), and determines the reconnaissance range based on the reconnaissance rules in the reconnaissance database and the reconnaissance area. The reconnaissance rules are rules generated based on the reconnaissance experience of reconnaissance experts.

[0120] S200. After performing target recognition on the initial image, obtain the abnormal target and the initial position information corresponding to the abnormal target.

[0121] For example, the target control terminal acquires initial images gathered by remote sensing satellites within the required reconnaissance range. Then, existing deep learning models such as YOLO, Faster R-CNN, and SSD are used to perform target identification on the initial images, thereby obtaining the corresponding anomalous targets in the initial images and their corresponding bounding box location information. The bounding box location information represents the location of the anomalous targets within the initial images.

[0122] For example, the position transformation is performed based on the position information in the block diagram, combined with the camera intrinsic and extrinsic parameters of the image sensor at the time of initial image acquisition, thereby obtaining the initial position information of the abnormal target in the real environment.

[0123] S300 uses synthetic aperture radar to remotely detect multi-angle imaging information of anomalous targets.

[0124] For example, after obtaining the anomalous target and its corresponding initial location information by target identification on the initial image, a multi-angle observation scheme for the SAR radar is planned based on the initial location information of the anomalous target. The observation angle parameters of the radar are determined according to the specific initial location information and detection requirements. The SAR radar transmits microwave signals from different angles to the area where the anomalous target is located according to the planned observation scheme, and obtains multi-angle imaging information by receiving the echo signals reflected by the target.

[0125] For example, ensure the timestamps of multiple cameras / sensors are synchronized to avoid displacement errors of abnormal targets due to time differences. Unify the acquisition of image resolution, frame rate, and coordinate system (such as world coordinate system or camera coordinate system) for each viewpoint of the abnormal target. Then, obtain the intrinsic parameters (focal length, distortion, etc.) and extrinsic parameters (rotation matrix, translation vector) for each viewpoint through multi-camera calibration. Map the corresponding positions of abnormal targets detected from different viewpoints to a unified coordinate system (such as 3D space or a 2D plane of the reference viewpoint) to obtain multi-angle imaging information.

[0126] S400: After verifying the authenticity of abnormal targets based on multi-angle imaging information, the enemy target and its corresponding target location information are obtained.

[0127] For example, based on visual features (such as shape, texture, and color histograms) or motion trajectories in multi-angle imaging information, the same anomalous target in different viewpoints is associated. A first detection result is obtained by verifying the consistency of the anomalous target's projection across different viewpoints using epipolar geometry. Then, the 2D detection box of the anomalous target in the multi-angle imaging information is back-projected into 3D space, and the existence of reasonable 3D intersections (such as triangulation) is checked to obtain a second detection result. A third detection result is obtained by verifying whether the anomalous target's motion trajectory in the multi-viewpoints conforms to physical laws (such as sudden velocity changes or obstacle penetration) based on the multi-angle imaging information. Finally, information fusion is performed based on the first, second, and third detection results to obtain the true / false target type corresponding to the anomalous target. When the true / false target type is false, the anomalous target is not identified as an enemy target; when the true / false target type is true, the anomalous target is identified as an enemy target, and the initial position information of the anomalous target is determined as the target position information corresponding to the enemy target.

[0128] In some optional embodiments, after target recognition of the initial image, abnormal targets and their corresponding initial location information are obtained, including:

[0129] S310. After performing grayscale processing and filtering processing on the initial image in sequence, a filtered image is obtained. The data template is determined based on the median of the filtered image. The data template is moved in the filtered image to obtain the template image corresponding to the data template in the filtered image.

[0130] For example, after the target control terminal acquires the initial image, it performs grayscale processing on the initial image to obtain a grayscale image. Next, it filters the grayscale image according to a filtering algorithm to obtain a filtered image. All pixel values ​​contained in the filtered image are obtained, and then the median information corresponding to the filtered image is determined by analyzing all pixel values ​​and identified as the target median. Subsequently, a data template is determined; this template can be circular or square, and the specific style is not limited. Then, the data template is moved within the filtered image to obtain the template image corresponding to the data template in the filtered image information.

[0131] S320. Obtain the minimum and maximum data values ​​corresponding to the template image.

[0132] For example, a template image is obtained by sliding the template image over an initial image, and then the corresponding pixel information in the template image is obtained from the initial image. The minimum value corresponding to the template image is determined as the minimum data value based on the pixel information, and the maximum value corresponding to the corresponding pixel in the template image is determined as the maximum data value.

[0133] S330. When the target median is between the minimum and maximum data values, the first threshold corresponding to the template image is obtained based on the mean between the minimum and maximum data values. The target median represents the median of the filtered image.

[0134] For example, when the target median is between the minimum and maximum data values, the first threshold corresponding to the template image is obtained based on the mean between the minimum and maximum data values.

[0135] S340. When the target median is not located between the minimum and maximum data values, expand the data template according to the preset rules to obtain an updated data template. Move the updated data template in the filtered image to obtain the updated template image corresponding to the updated data template in the filtered image.

[0136] For example, when the data value is not between the minimum and maximum values, the filter template is expanded (specifically, it is expanded according to a preset expansion rule, which may be to expand outward by a preset range or to expand in a specific direction; the specific expansion method is not specifically limited here), and the corresponding latest template image is obtained again from the filtered image based on the expanded filter template (updated data template).

[0137] S350. Obtain the minimum and maximum values ​​of the updated data corresponding to the updated template image, so that the target median is between the minimum and maximum values ​​of the updated data, and obtain the second threshold corresponding to the updated template image. Both the first and second thresholds are filtering thresholds.

[0138] For example, the minimum and maximum data values ​​corresponding to the latest template image are obtained again (updating the minimum and maximum data values) until the target median is between the minimum and maximum data values, thereby obtaining the second threshold corresponding to the expanded filter template.

[0139] S360. The template image is denoised after being processed according to the filtering threshold.

[0140] For example, noise identification processing is performed on the template image according to the first threshold or on the latest template image according to the second threshold. When the filter value corresponding to the template image is less than the first threshold, the filter value is determined to be a noise point; otherwise, it is a non-noise point. The denoising processing of the latest template image is the same as that of the previous template image, thereby obtaining all the corresponding noise points in the filtered image. Then, noise removal is performed on the filtered image based on all the noise points to obtain the denoised filtered image.

[0141] S370. The denoised image is divided into multiple image regions by the initial segmentation lines.

[0142] For example, edge information is obtained by performing edge processing on the denoised image using algorithms such as edge detection and superpixel segmentation. Then, initial segmentation lines are determined based on this edge information. Finally, image segmentation is performed on the denoised image based on these initial segmentation lines to obtain multiple image regions corresponding to the denoised image.

[0143] S380. After adjusting the position of the initial segmentation line based on the similarity between multiple image regions, the target edge information is obtained.

[0144] For example, the similarity between adjacent image regions is calculated using Euclidean distance. When the similarity is greater than a preset value, the two adjacent image information are merged. When the similarity is less than or equal to the preset value, the two adjacent image information are retained until the similarity between any two processed image regions is less than or equal to the preset value, thereby obtaining the target edge information.

[0145] S390. Obtain the abnormal target and the initial position information corresponding to the abnormal target based on the target edge information.

[0146] For example, a target classification model is used to classify the target edge information to obtain the target type corresponding to each sub-edge information. If the target type is a preset type such as a person, vehicle or other type, the target type corresponding to the sub-edge information is determined as an abnormal target, and the minimum bounding rectangle corresponding to the sub-edge information is determined as the block diagram position information corresponding to the abnormal target. Then, based on the block diagram position information and the camera intrinsic and extrinsic parameters of the image sensor when the initial image is acquired, position transformation is performed to obtain the initial position information of the abnormal target in the real environment.

[0147] In some optional embodiments, after adjusting the position of the initial segmentation line based on the similarity between multiple image regions, target edge information is obtained, including:

[0148] S381. The first similarity between the first region and the third region and the second similarity between the second region and the third region are compared to obtain the comparison result. The third region is the region covered by the initial segmentation line. The first region is located on the first side of the third region and the second region is located on the second side of the third region. The first region, the second region and the third region all represent image regions.

[0149] For example, an initial segmentation curve is determined, and the filtered image information is segmented according to the initial segmentation curve to obtain a first region corresponding to one side of the initial segmentation curve and a second region corresponding to the other side of the initial segmentation curve, as well as a third region where the initial segmentation curve is located, that is, the region covered by the initial segmentation curve is the third region.

[0150] For example, the first similarity between the first region and the third region is calculated using cosine similarity or Euclidean distance. Similarly, the second similarity between the second region and the third region is calculated using cosine similarity or Euclidean distance. Then, the first similarity and the second similarity are compared. If the first similarity is greater than or equal to the second similarity, the comparison result is determined to be that the first similarity is greater; if the first similarity is less than the second similarity, the comparison result is determined to be that the second similarity is greater.

[0151] S382. If the comparison result indicates that the first similarity is greater than or equal to the second similarity, the initial segmentation curve is adjusted to the position of the first region.

[0152] For example, when the comparison result is that the first similarity is greater than the second similarity, it indicates that the correlation between the third region and the first region is greater, and the initial segmentation curve is adjusted to the position of the first region.

[0153] S383. If the comparison result indicates that the first similarity is less than the second similarity, the initial segmentation curve is adjusted to the position of the second region.

[0154] For example, if the comparison result shows that the second similarity is larger, that is, if the comparison result indicates that the second similarity is smaller than the second similarity, it indicates that the correlation between the third region and the second region is greater, and the initial segmentation curve is adjusted to the position of the second region.

[0155] S384. When the difference between the two sides of the initial segmentation curve is the largest, obtain the target edge information.

[0156] For example, the initial segmentation curve is adjusted by the first similarity and the second similarity until the difference between one side and the other side of the initial segmentation curve is the largest, thereby obtaining the target edge information. Then, the minimum bounding rectangle corresponding to the abnormal target is obtained based on the target edge information, and the initial position information corresponding to the abnormal target is obtained based on the minimum bounding rectangle.

[0157] In some optional embodiments, after verifying the authenticity of abnormal targets based on multi-angle imaging information, the enemy target and its corresponding target location information are obtained, including:

[0158] S410. Obtain the relevant position information of the abnormal target at different times based on the target edge information and affine transformation.

[0159] For example, the target edge information corresponding to the abnormal target is combined with affine transformation to obtain the real position of the abnormal target in the real environment, and then the real position of the abnormal target at different times is used to determine the relevant position information of the abnormal target.

[0160] S420. Determine the movement information of the abnormal target based on the initial position information and relevant position information. The movement information includes turning information, speed information and trajectory information.

[0161] For example, the initial position information and related position information of the abnormal target at different times are obtained. The initial position information is the actual position information of the abnormal target at the current time, and the related position information is the actual position information of the abnormal target at other times. Then, based on the initial position information and related position information and the corresponding time information, a corresponding time-position curve is formed. Then, based on the time-position curve of the abnormal target, data analysis is performed to determine the turning information, speed information and trajectory information of the abnormal target.

[0162] S430. Construct a false target feature library and a false target prediction model.

[0163] For example, by collecting historical battlefield data (such as GPS trajectories, speeds, and turning angles of real tanks / vehicles), terrain matching records (such as road / off-road path preferences), and tactical rules (such as marching distances and the accompanying situation of logistics units), and then constructing a false target feature library through simulation exercises and known enemy camouflage cases (such as trailer-moving fake tanks and corner reflector arrays), a false target trajectory (such as irregular speed changes and anti-terrain movement) can be generated using GAN, thereby enhancing the diversity of the behavior library.

[0164] S440. Based on the false target feature library, movement information and false target prediction model, predict the true and false probabilities of abnormal targets to obtain the true and false probability of the targets, configure abnormal targets with a true and false probability greater than the probability threshold as enemy targets, and configure the initial position information of abnormal targets as target position information.

[0165] For example, the movement pattern corresponding to the abnormal target is determined based on the movement information, and then the first probability corresponding to the abnormal target is determined based on the movement pattern and the false target feature library. The second probability is obtained by predicting the true and false probabilities of the abnormal target based on the false target prediction model, and then the first probability and the second probability are fused to determine the true and false probability of the abnormal target object.

[0166] For example, a probability threshold is determined based on expert experience or actual needs, and then abnormal targets with a probability greater than the probability threshold are configured as enemy targets, and the initial position information of abnormal targets is configured as target position information.

[0167] S450. When the probability of a target being real or fake is less than the probability threshold, multi-angle imaging information of an abnormal target is detected by synthetic aperture radar remote sensing. After verifying the authenticity of the abnormal target based on the multi-angle imaging information, the enemy target and target location information are obtained.

[0168] For example, when the probability of a target being real or fake is less than or equal to a probability threshold, the SAR radar's multi-angle observation scheme is planned based on the actual position corresponding to the initial position information of the abnormal target. The radar's observation angle parameters are determined according to the specific actual position and detection requirements. Following the planned observation scheme, the SAR radar transmits microwave signals from different angles towards the area where the abnormal target is located and receives the echo signals reflected by the target to obtain multi-angle imaging information. Therefore, based on the visual features (such as shape, texture, and color histogram) or motion trajectory in the multi-angle imaging information, the same abnormal target from different viewpoints is associated. The first detection result is obtained by verifying the consistency of the projection of the abnormal target in different viewpoints using epipolar geometry. Then, the 2D detection box of the abnormal target in the multi-angle imaging information is back-projected into 3D space to check for reasonable 3D intersections (e.g., triangulation) to obtain the second detection result. Next, the third detection result is obtained by verifying whether the abnormal target's motion trajectory in multiple viewpoints conforms to physical laws (e.g., sudden velocity changes, obstacle penetration) based on the multi-angle imaging information. Finally, the first, second, and third detection results are fused to obtain the true / false target type corresponding to the abnormal target. When the true / false target type is false, the abnormal target will not be identified as an enemy target. When the true / false target type is true, the abnormal target is identified as an enemy target, and the initial position information of the abnormal target is determined as the target position information corresponding to the enemy target.

[0169] Specifically, by constructing a false target feature library, the accuracy of identifying false target trajectories is improved, thereby enhancing the accuracy of enemy target identification. The probability of anomaly targets being true or false is predicted based on the false target feature library, movement information, and a false target prediction model, thus improving the accuracy of predicting the true or false nature of anomaly targets.

[0170] In some optional embodiments, the probability of an abnormal target being true or false is predicted based on a false target feature library, movement information, and a false target prediction model, including:

[0171] S4401. Obtain the digital elevation model and road vector data corresponding to the reconnaissance range;

[0172] S4402. After mapping the trajectory information to the digital elevation model, the elevation data associated with the trajectory points is obtained. The elevation data indicates the terrain elevation, slope and land cover type.

[0173] S4403. Obtain real-time map data corresponding to the reconnaissance range;

[0174] S4404. Determine terrain travel cost information and road distance data based on real-time map data, road vector data and trajectory information. Road distance data represents the distance between the trajectory point indicated by the trajectory information and the road. Terrain travel cost information represents the travel cost of the trajectory point indicated by the trajectory information.

[0175] S4405. After performing time-series segmentation on the target trajectory of the abnormal target according to the preset time period, multiple time-series segment trajectories are obtained. The target trajectory represents the trajectory indicated by the trajectory information.

[0176] S4406. Input the elevation data, terrain access cost information and road distance data corresponding to each time segment trajectory into the false target prediction model in sequence. After calling the false target feature library, the false target prediction model performs true and false prediction on the time segment trajectory to obtain the true and false probability of the segment.

[0177] S4407. The first true / false probability is obtained by integrating the true / false probabilities of all time-series segment trajectories corresponding to the false target prediction model.

[0178] S4408. After performing cluster analysis on the target edge information of all abnormal targets, different sub-clusters with different cluster ranges are obtained. The cluster range represents the regional range of all abnormal targets contained in the same sub-cluster.

[0179] S4409. Determine the density ratio based on the distribution of all abnormal targets in the same sub-cluster within the cluster range;

[0180] S4410. Determine the second true / false probability of the anomalous target based on the density ratio;

[0181] S4411. Determine the true / false probability of the abnormal target based on the first true / false probability and the second true / false probability.

[0182] For example, using a high-precision digital elevation model (DEM) and road vector data, trajectory information (latitude and longitude) is mapped to terrain elevation, slope, and land cover type (such as vegetation, water, and city). Combined with real-time satellite / UAV-updated map data, temporary obstacles (such as minefields or destroyed bridges) are marked. Based on the map data, the terrain travel cost of the trajectory information (such as the speed attenuation coefficient of a tank on a slope >30°) is calculated, and the distance between the trajectory point and the nearest road is determined. True targets typically move along roads or concealed paths, while false targets may ignore roads. The trajectory of each target object is represented as a timestamp sequence, with each 5-minute segment serving as an inference unit. Windows overlap by 50% to ensure continuity. This yields the associated terrain elevation, slope, road distance, and other information for each trajectory point within each segment, as well as the terrain travel cost and the distance between the trajectory point and the nearest road, forming a multi-dimensional input vector. This multi-dimensional input vector is then fed into the false target prediction model. The false target prediction model includes a bidirectional LSTM (Long Short-Term Memory) network. The forward LSTM in the bidirectional LSTM captures historical motion patterns, while the backward LSTM predicts future plausibility (e.g., whether the target decelerates after a sharp turn; if it doesn't decelerate, it's not plausible). This is combined with the attention layer in the false target prediction model to automatically focus on abnormal segments (e.g., sudden speed changes, floating trajectory points), and output attention weights (e.g., a weight of 0.9 for sudden acceleration on a steep slope). The true / false probability of the segment is then output based on the sigmoid function of the classification head. The true / false probabilities of all time-series segment trajectories are integrated to obtain the first true / false probability.

[0183] For example, after performing cluster analysis on the target edge information of all abnormal targets, different sub-clusters with different cluster ranges are obtained. The similarity of abnormal targets in the same sub-cluster reaches a preset similarity. The cluster range represents the regional range of all abnormal targets contained in the same sub-cluster. The density ratio is determined according to the distribution of all abnormal targets in the same sub-cluster within the cluster range, and the second true / false probability of whether the abnormal target is a true target or a false target object is determined according to the density ratio.

[0184] For example, by using evidence theory to fuse the first and second true / false probabilities under the two methods, the true / false probability of the abnormal target can be obtained.

[0185] Specifically, each trajectory point in each segment is associated with terrain elevation, slope, road distance, and other factors, along with terrain travel costs and the distance between the trajectory point and the nearest road, forming a multi-dimensional input vector. This multi-dimensional input vector is then fed into the false target prediction model to obtain the segment's true / false probability. The first true / false probability is obtained by combining the true / false probabilities of each segment, thus improving the prediction accuracy of the first true / false probability. A second true / false probability is determined based on the density ratio to determine whether an abnormal target is a real or false target. The first and second true / false probabilities from both methods are then fused to obtain the target's true / false probability. This comprehensive judgment of the target's true / false probability across different dimensions improves the accuracy of predicting the true / false nature of abnormal targets.

[0186] S500 determines the deployment location of the reconnaissance module based on the target location information, a preset reconnaissance radius centered on the target location information, and associated environmental information.

[0187] For example, the target control terminal uses the location indicated by the target location information as the target center and a preset length (i.e., a preset reconnaissance radius) as the radius to determine the reconnaissance target range (i.e., to reconnoiter the enemy target within the reconnaissance target range), and obtains the associated environmental information corresponding to the reconnaissance target range. Then, it analyzes the associated environmental information to obtain the deployment position of the reconnaissance module. The reconnaissance module includes a reconnaissance aircraft, a feint swarm, a strike swarm, a ground terminal, and a reconnaissance vehicle. The reconnaissance aircraft, feint swarm, strike swarm, and ground terminal are deployed on the reconnaissance vehicle, and the reconnaissance vehicle deploys the corresponding reconnaissance aircraft, feint swarm, strike swarm, and ground terminal to the corresponding deployment positions.

[0188] In some optional embodiments, the deployment location of the reconnaissance module is determined based on target location information, a preset reconnaissance radius centered on the target location information, and associated environmental information, including:

[0189] S510. Using the location indicated by the target location information as the center, construct a reconnaissance target range around the center with a preset reconnaissance radius;

[0190] S520: Acquire the associated environmental information corresponding to the reconnaissance target range, including LiDAR terrain data, electromagnetic spectrum information and visible light image information;

[0191] S530 generates a three-dimensional threat situation field by fusing LiDAR terrain data, electromagnetic spectrum information and visible light image information through a convolutional network.

[0192] S540: By inputting the three-dimensional threat situation field, tactical phase requirements, and abnormal features of enemy targets into the deep reinforcement learning framework, the optimal approach distance of the reconnaissance module is obtained.

[0193] S550: Determine deployment location based on optimal proximity distance.

[0194] For example, after constructing the reconnaissance target range, the target control terminal acquires the associated environmental information of the reconnaissance target range. The associated environmental information includes LiDAR (Light Detection and Ranging) terrain data, electromagnetic spectrum information, and visible light image information. Then, the target control terminal fuses the LiDAR terrain data, electromagnetic spectrum information, and visible light image information according to a graph convolutional network to generate a three-dimensional threat situation field (intuitively presenting the distribution, intensity, evolution trend, and interrelationship of various threats within the reconnaissance target range, thereby providing visualized and quantitative threat assessment data). Then, the target's abnormal features (including position offset patterns and energy radiation anomalies) are extracted through a spatiotemporal attention mechanism and input into a deep reinforcement learning framework to train a decision model, balancing the dual objectives of "observation clarity" and "exposure risk" (i.e., ensuring sufficient clarity while minimizing exposure). The risk prediction module uses an adversarial generative network to simulate potential threat responses and finally outputs a flexible safety boundary that dynamically adjusts with the environment, thereby obtaining the optimal approach distance for the reconnaissance vehicle. Candidate positions that meet the optimal approach distance are marked within the reconnaissance target range, and specific deployment positions are determined from the candidate positions based on the number of reconnaissance modules and reconnaissance requirements.

[0195] Specifically, a three-dimensional threat situation field is generated using LiDAR terrain data, electromagnetic spectrum information, and visible light imagery, thus visually presenting the distribution, intensity, evolution trends, and interrelationships of various threats within the reconnaissance target area, providing visualized and quantitative threat assessment data. A deep reinforcement learning framework is used for optimized training based on observation clarity and exposure risk, thereby obtaining the optimal close-range distance that ensures sufficient clarity while minimizing exposure, thus determining the deployment location of the reconnaissance module and improving reconnaissance effectiveness.

[0196] In some optional embodiments, the specific method for generating the three-dimensional threat situation field includes: dividing the associated environmental information into grids to obtain multiple spatial grid cells; then obtaining the elevation and slope information corresponding to the spatial grid cells from LiDAR terrain data, obtaining the signal strength corresponding to the spatial grid cells from electromagnetic spectrum information, and obtaining the target type corresponding to the spatial grid cells from visible light imagery; obtaining the terrain information corresponding to each spatial grid cell from the associated environmental information, and determining the terrain connectivity information between any two spatial grid cells based on the terrain information, i.e., whether the spatial grid cells are passable; and calculating the Euclidean distance between any two spatial grid cells, thereby determining the correlation value between any two spatial grid cells based on the terrain connectivity information and the Euclidean distance.

[0197] For example, spatial grid cells are defined as nodes, associated values ​​are defined as edges of the corresponding nodes, and the elevation, slope, signal strength, and target type of the corresponding nodes are defined as node features, thereby constructing a grid map information corresponding to the associated environmental information; the node features corresponding to any target grid cell are obtained from the grid map information and feature fusion is performed to obtain the first feature information corresponding to the target grid cell, and the neighborhood information (target neighborhood) corresponding to the target grid cell is determined from the grid map information based on the first feature information and associated values, and then the threat score of the target grid cell is obtained by performing threat scoring on the target grid cell based on the second feature information and the first feature information of other grid cells contained in the neighborhood information, and then the three-dimensional threat situation field is determined based on the target grid cell and the corresponding threat score.

[0198] The calculation of the optimal approach distance includes: constructing a spatiotemporal feature encoder based on a bidirectional long short-term memory network (BiLSTM), inputting the temporal position coordinates of the enemy target and related environmental information, and outputting a fused representation containing kinematic features (displacement velocity, acceleration vector) and radiation features (spectral intensity, energy fluctuation entropy).

[0199] Among them, the temporal location and environmental information of enemy targets are associated with radiation characteristics through the following paths:

[0200] Kinematic physical effects: Velocity / acceleration directly modulates electromagnetic / acoustic radiation characteristics.

[0201] Environmental interactive modulation: The terrain and obstructions dynamically change the signal propagation path loss.

[0202] Data-driven modeling: BiLSTM extracts the complex nonlinear relationship between motion, environment and radiation from historical data, thereby indirectly inferring the radiation characteristics of a target without the need to deploy radiation sensors, using only low-cost trajectory data.

[0203] In the decision modeling phase, the aforementioned features (kinematic and radial features) are input into a deep reinforcement learning framework along with the real-time pose information of the reconnaissance vehicle and 3D threat field data, utilizing a dual-objective optimization mechanism (observation clarity target and exposure risk target):

[0204] Observational sharpness target: Constructing a quantization function R based on an optical physics model c =1 / (1+αd) 2 ), where d is the reconnaissance distance and α is the weight.

[0205] Exposure risk target: Probability value R output by the integrated threat prediction module r ∈[0,1], dynamically corrected by Kalman filtering.

[0206] Simultaneously, according to the adaptive policy network: the Actor network adopts a Gaussian policy distribution and outputs a continuous action space [Δd,Δθ]∈R. 2 The Critic network achieves value function decomposition and dynamically adjusts the weight coefficients λ of the two objectives. t =σ(β·V r (t)), where σ is the sigmoid function, β is the weight, and V r (t) represents the expected cumulative return of the risk target exposed at time step t (i.e., the Critic network's value assessment of the risk target), R 2 The decision coefficient is used; according to the dynamic optimization strategy, the discount factor γ implements a risk-sensitive mechanism.

[0207] γ t =γ0·(1-k·R) r (t))

[0208] γ t The dynamic discount factor at time t, ranging from (0,1), is used to control the decay rate of future rewards, and k represents the risk sensitivity coefficient (adjusting the risk against γ). t Influence intensity), R r (t) represents the instantaneous exposure risk at time t (the probability value output by the threat prediction module), directly reflecting the current environmental threat level. Therefore, a Pareto front solution set is generated online based on the Non-Dominated Sorting Genetic Algorithm (NSGA-II), and the optimal solution is selected according to the tactical stage requirements (e.g., assault mode prioritizes clarity, concealment mode prioritizes security), thereby obtaining the optimal close-range distance in complex battlefield environments.

[0209] Specifically, by constructing a three-dimensional threat situation field, the kinematic and radiation characteristics of enemy targets, real-time pose information of reconnaissance vehicles, and three-dimensional threat field data within the reconnaissance target range are input into a deep reinforcement learning framework. Based on the observation clarity target and exposure risk target optimization model, the observation clarity and exposure risk of each area within the reconnaissance target range are accurately determined, and the optimal close-in distance is accurately determined according to tactical requirements, thereby improving the reconnaissance quality of the reconnaissance aircraft.

[0210] In some optional embodiments, the target control end uses a dynamic programming algorithm to perform path planning using multiple candidate locations corresponding to the optimal approach distance to obtain the fastest travel path for the reconnaissance vehicle; the target control end determines the target arrival location corresponding to the reconnaissance vehicle based on the fastest travel path (that is, the candidate location with the fastest arrival is configured as the deployment location).

[0211] The S600 control and reconnaissance module conducts reconnaissance of enemy targets at its deployment location.

[0212] For example, the target control terminal controls the reconnaissance module to reach the deployment position, and then the reconnaissance aircraft in the reconnaissance module collects information on the enemy target to obtain the first image information, and the reconnaissance aircraft transmits the first image information back to the target control terminal.

[0213] In some embodiments, the unmanned aerial vehicle (UAV) reconnaissance system further includes a fire support module. The reconnaissance module includes a reconnaissance vehicle and reconnaissance aircraft, strike swarms, and ground-based terminals deployed on the reconnaissance vehicle. Controlling the reconnaissance module at its deployment location to conduct reconnaissance of enemy targets includes:

[0214] S610, the target control terminal controls the reconnaissance vehicle to reach the deployment position.

[0215] For example, the target control terminal generates control commands for the reconnaissance vehicle based on the target location information corresponding to the enemy target, and then sends the control commands to the reconnaissance vehicle so that the reconnaissance vehicle can reach the corresponding deployment position according to the control commands. The reconnaissance module includes the reconnaissance vehicle and reconnaissance aircraft, strike swarm, and ground terminal deployed on the reconnaissance vehicle.

[0216] After the S620 reconnaissance aircraft conducts reconnaissance of the enemy target and obtains the first image information, it transmits the first image information back to the ground terminal so that the ground terminal can send the first image information to the target control terminal.

[0217] For example, after the reconnaissance vehicle arrives at its deployment position, it first controls the decoy drones or swarms of decoy drones in the reconnaissance module (carrying corner reflectors to simulate the real radar cross section and emitting radio noise in a preset frequency band in all directions) to approach the enemy target. Specifically, it first obtains the reconnaissance behavior of the real reconnaissance aircraft when it was conducting reconnaissance in an environment most similar to that of the enemy target from the historical database. Then, it controls the decoy drones / swarms (decoy drones or decoy drone swarms) to approach the enemy target based on the reconnaissance behavior, triggering enemy radar scanning and anti-radio suppression. The decoy drones / swarms have a portable satellite guidance and protection module and adopt a radio silent communication strategy, that is, they plan waypoints based on the enemy target coordinates provided by the decoy drones / swarms when they depart. During operation, they do not communicate with the ground end in the reconnaissance module. Thus, after the reconnaissance aircraft conducts reconnaissance of the enemy target, it obtains the first image information and transmits the first image information back to the ground end, so that the ground end can send the first image information to the target control end.

[0218] S630: The target control terminal obtains the first image information and makes a judgment on the attack on the bee swarm based on the first image information to obtain the first judgment result.

[0219] For example, the target control terminal receives first image information transmitted from the ground terminal, thereby obtaining the first image information. The target control terminal then uses a target recognition model to perform target recognition on the first image information to obtain the true target type and the corresponding block diagram information of the enemy target in the first image information. Based on the block diagram information, combined with sensor information and the location information of the first image information collected by the reconnaissance aircraft, information conversion is performed to obtain the target size information of the enemy target. Then, based on the true target type and target size information, combined with expert verification, the target ammunition quantity for striking the enemy target is obtained. Finally, the strike capability corresponding to the reconnaissance module, such as the known payload, is compared with the target ammunition quantity. If the known payload is greater than the target ammunition quantity, the first judgment result is that the strike can be effectively carried out; if the known payload is less than or equal to the target ammunition quantity, the first judgment result is that the strike cannot be effectively carried out.

[0220] S640: If the first judgment result indicates that the strike can be effectively carried out, the target control terminal determines the strike location information corresponding to the strike swarm based on the first image information and sends the strike location information to the ground terminal.

[0221] For example, sensor parameters are read from the reconnaissance aircraft when the first image information is captured. These sensor parameters include intrinsic parameters such as focal length, principal point, and lens distortion coefficient, and extrinsic parameters such as the sensor's installation angle (pitch, roll, yaw) and position offset relative to the reconnaissance aircraft. Simultaneously, the reconnaissance aircraft's aerial position information at the time of capture, such as latitude, longitude, altitude, and attitude angles provided by GPS / INS, is acquired. Then, based on the camera's intrinsic parameters, the position information of the frame in the image is converted to coordinates in the normalized camera coordinate system. Next, the sensor's extrinsic parameters are used to convert the normalized coordinates to coordinates in the reconnaissance aircraft's corresponding body coordinate system. Then, based on the reconnaissance aircraft's attitude angles, a rotation matrix is ​​used to convert the body coordinate system to the Northeast-Upper-Heaven (ENU) coordinate system. Finally, combined with the reconnaissance aircraft's aerial position information, the ENU coordinate system is converted to the true position information of the enemy target in the world coordinate system. This true position information is then determined as the strike position information corresponding to the swarm attack, and the target control terminal sends this strike position information to the ground.

[0222] The S650 ground unit sends the strike location information to the strike swarm so that the strike swarm can reach the strike location indicated by the strike location information to carry out the strike. After the reconnaissance aircraft acquires images of the strike location, the second image information is sent to the target control unit.

[0223] For example, after receiving the strike location information sent by the target control terminal, the ground terminal sends the strike location information to the strike swarm. Upon receiving the strike location information, the strike swarm quickly arrives at the strike location indicated by the strike location information and strikes the enemy target corresponding to the strike location information. After the strike is completed, the reconnaissance aircraft uses a second image to acquire images of the strike location. The reconnaissance aircraft then sends the second image information to the ground terminal, which in turn sends the second image information to the target control terminal.

[0224] S660: The target control terminal obtains the second image information and, based on the target tracking algorithm, performs a destruction judgment on the second image information to obtain the second judgment result.

[0225] For example, after obtaining the second image information, the target control terminal first extracts the position and features of the latest target using a target recognition algorithm (such as a deep learning detection model). Then, it uses a target tracking algorithm (such as Kalman filtering, Hungarian matching, or IoU association) to match the latest recognition result with the target recognition result in the first image. If the two are highly matched in position, appearance, or motion trajectory (such as the overlap rate exceeding a threshold or the feature similarity meeting the standard), it is determined that the same target has not been destroyed, and the second judgment result is output as "destruction incomplete". Conversely, if they cannot match (such as the target disappearing or the feature difference being significant), it is determined that the target has been destroyed, and the second judgment result is output as "destruction completed".

[0226] S670: If the second judgment result indicates that the destruction is complete, the target control end sends a return command to the ground end.

[0227] For example, if the target control terminal determines that the second judgment result is that the destruction is completed, the target control terminal sends a return command to the ground terminal.

[0228] S680: If the second judgment result indicates that the destruction is not completed, the target control terminal controls the strike swarm to strike the enemy target again until the enemy target is completely destroyed.

[0229] For example, if the target control terminal determines that the destruction is incomplete in the second judgment, the target control terminal updates the strike position information based on the second image information and sends the updated strike position information to the ground terminal. The ground terminal then continues to strike the enemy target again based on the updated strike position information until the enemy target is completely destroyed.

[0230] In some optional embodiments, the reconnaissance vehicle is also equipped with a decoy drone, and after the reconnaissance vehicle reaches the deployment position, the following steps are also included:

[0231] S611. Acquire historical reconnaissance behavior, which represents the reconnaissance behavior of reconnaissance aircraft when their environment was most similar to that of enemy targets in history.

[0232] S612. Control a decoy drone to approach an enemy target using historical reconnaissance behavior. The decoy drone navigates based on the target's location information and does not communicate with the ground station while approaching the enemy target.

[0233] For example, the decoy drones or swarms of decoy drones in the reconnaissance module are first controlled to approach the enemy target. Specifically, the reconnaissance behavior of the real reconnaissance aircraft when it was conducting reconnaissance is most similar to that of the enemy target is obtained from the historical database. Then, the decoy drones / swarms (decoy drones or decoy drone swarms) are controlled to approach the enemy target according to the reconnaissance behavior, triggering enemy radar scanning and anti-radio suppression. The decoy drones / swarms have a portable satellite guidance and protection module and adopt a radio silence communication strategy, that is, they plan waypoints according to the enemy target coordinates provided by the decoy drones / swarms when they depart, and do not communicate with the ground end in the reconnaissance module during operation.

[0234] Specifically, by obtaining reconnaissance behavior data from historical databases of real reconnaissance aircraft operating in environments most similar to those of enemy targets, decoy drones / swarms are controlled to approach enemy targets based on this reconnaissance behavior, triggering enemy radar scanning and counter-radio suppression, thereby inducing the enemy to identify the decoy drones as reconnaissance aircraft and engage them, thus improving the safety of reconnaissance aircraft. A radio silence communication strategy is adopted to ensure the security of ground-based communications and reduce the risk of ground-based exposure.

[0235] In some alternative embodiments, the process of the decoy drone approaching the enemy target using historical reconnaissance behavior further includes:

[0236] S6111. When the decoy drone encounters strong electromagnetic suppression in the data link frequency band, the first strike unit, which is a combination of decoy drones and feints, locks onto and strikes the suppression antenna. The first strike unit is equipped with an anti-radiation module.

[0237] S6112. By feigning other drones in the swarm to attract and occupy the detection, interference, suppression, and attack resources of the anti-drone system, the anti-drone system is saturated.

[0238] S6113. Based on the initial strike resources of the decoy drone, the real-time strike resources corresponding to the detection, interference, suppression and strike resources of the anti-drone system, determine the strikeable duration of the decoy drone and the first time when the second strike unit of the decoy drone begins to strike.

[0239] S6114. Electromagnetic characteristic data is obtained by collecting electromagnetic characteristic data emitted by enemy anti-navigation radar and electromagnetic suppression antennas in real time through decoy drones.

[0240] S6115: Using decoy drones, an anti-electromagnetic map of the corresponding area is generated based on electromagnetic characteristic data of the target location information. Based on the anti-electromagnetic map, the drone autonomously avoids strong detection signals and electromagnetic suppression areas to determine the target's communication method. Based on the target's communication method, the anti-electromagnetic map, the first time, and the attack duration are sent to the ground.

[0241] For example, in an anti-drone system designed to intercept enemy targets, if a decoy drone / swarm encounters strong electromagnetic suppression within the data link frequency band, the strike units within the decoy drone / swarm, which are mixed with feints and equipped with anti-radiation modules, will directly lock onto the suppression antenna and attack. Simultaneously, other drones in the feint swarm will attract and occupy the anti-drone system's detection, jamming, suppression, and strike resources, causing system saturation. The strike duration corresponding to the decoy drone / swarm is determined based on the strike resources available to the decoy drone / swarm, and the strike resources attracted and occupied by the anti-drone system. The start time for the strike units of the decoy drone / swarm to engage is then determined and designated as the first time. Simultaneously, the decoy drone / swarm collects electromagnetic signature data in real time from the target's anti-drone radar and electromagnetic suppression antenna.

[0242] For example, the decoy drone / swarm forms an anti-electromagnetic map of the corresponding area under the target location information based on electromagnetic characteristic data. The anti-electromagnetic map is used to characterize the specific anti-electromagnetic map as a map that marks the location of anti-electromagnetic equipment. Then, based on the analysis of the anti-electromagnetic map, the drone autonomously avoids strong detection signals and electromagnetic suppression areas, determines the target's communication mode, and then sends the anti-electromagnetic map, the first time, and the attackable duration to the ground end of the reconnaissance module according to the target's communication mode.

[0243] Specifically, attack units equipped with anti-radiation modules can directly target and attack the suppression antenna, thus avoiding the strong electromagnetic suppression of the decoy drones. By feigning attacks on other drones in the swarm, the anti-drone system's detection, jamming, suppression, and attack resources are diverted, causing system saturation and reducing the anti-drone system's ability to attack decoy drones and reconnaissance aircraft, thereby improving the reconnaissance effectiveness of the reconnaissance aircraft. An anti-drone electromagnetic map is generated using mobile phone electromagnetic signature data, and analysis of this map enables autonomous avoidance of strong detection signals and electromagnetic suppression areas to determine target communication methods, thereby improving the security of communication between the decoy drones and the ground station.

[0244] In some optional embodiments, before determining the impact on the bee swarm based on the first image information to obtain a first determination result, the method further includes:

[0245] S621. Determine the reconnaissance route of the reconnaissance aircraft based on the anti-electromagnetic map at the ground end, and calculate the second time corresponding to the reconnaissance aircraft conducting reconnaissance along the reconnaissance route.

[0246] For example, the ground end performs path planning based on the anti-electromagnetic map to determine the reconnaissance route of the reconnaissance aircraft, and then determines the reconnaissance time corresponding to the reconnaissance aircraft when it conducts reconnaissance according to the reconnaissance route and the corresponding flight parameters of the reconnaissance aircraft, that is, obtains the second time.

[0247] S622: When the time difference between the first and second time points is less than the strikeable duration, the reconnaissance aircraft is controlled to collect information on the enemy target according to the reconnaissance route to obtain the first image information.

[0248] For example, if the ground terminal determines that the time difference between the first time and the second time is less than the strikeable duration, the reconnaissance aircraft will conduct close-range reconnaissance of the enemy target according to the reconnaissance route. Then, the reconnaissance aircraft will collect information on the enemy target under the target location information according to the reconnaissance route to obtain the first image information. The reconnaissance aircraft will transmit the first image information back to the ground terminal corresponding to the reconnaissance module, and then the ground terminal will send the first image information to the target control terminal.

[0249] S623, if the time difference between the first time and the second time is greater than or equal to the strikeable duration, receive the request for strike control command sent by the ground.

[0250] For example, if the ground terminal determines that the time difference between the first time and the second time is greater than or equal to the strikeable duration, then the ground terminal sends a strike control request command to the target control terminal.

[0251] S624. Obtain enemy targets and their corresponding target location information from the request for strike control command.

[0252] For example, the target control terminal receives a request for strike control command sent by the ground terminal, and then, after receiving the request for strike control command, the target control terminal obtains the enemy target and the target location information corresponding to the enemy target from the request for strike control command.

[0253] S625, the fire support module controls the strike position indicated by the target location information.

[0254] For example, control commands are sent to the fire support module based on the target location information corresponding to the enemy target, so that the fire support module can provide fire coverage to the target location information corresponding to the enemy target according to the control commands, thereby destroying the abnormal target. Furthermore, to further improve the reconnaissance efficiency of the reconnaissance aircraft and shorten the reconnaissance time, the reconnaissance aircraft can wait at a preset position before conducting close-range reconnaissance of the enemy target according to the reconnaissance route, reducing the time it takes for the reconnaissance aircraft to reach the reconnaissance position, thus shortening the reconnaissance time; the preset position is related to the coverage area in the anti-electromagnetic map, and the preset position is the outer perimeter of the coverage area in the anti-electromagnetic map.

[0255] Specifically, the strike time of the swarm is determined by the first moment when the strike begins and the second moment when the reconnaissance aircraft follows its reconnaissance route. This ensures that the strike time is sufficient to avoid insufficient strike time and thus unsatisfactory strike results against enemy targets. Furthermore, in the event of insufficient strike time, the fire support module is used to destroy enemy targets, guaranteeing that enemy targets can be destroyed even when the strike time of the swarm is limited.

[0256] In some embodiments, the method further includes:

[0257] S690: If the first judgment result indicates that an effective strike cannot be carried out, the target control terminal controls the fire support module to strike the strike location indicated by the strike location information.

[0258] For example, if the target control terminal determines that the first judgment result is that the strike cannot be effectively carried out, the target control terminal directly controls the fire support module to strike the strike position indicated by the strike position information.

[0259] S691. After the ground end acquires images of the strike location through reconnaissance aircraft, it sends the third image information to the target control end.

[0260] For example, after the fire support module strikes the strike location indicated by the strike location information, the ground end acquires third image information by acquiring images of the strike location through a reconnaissance aircraft, and sends the third image information to the target control end.

[0261] S692. The target control terminal obtains the third judgment result after completing the destruction judgment on the third image information based on the target tracking algorithm.

[0262] For example, after obtaining the third image information, the target control terminal first extracts the position and features of the latest target corresponding to the third image information through a target recognition algorithm (such as a deep learning detection model). Then, it uses a target tracking algorithm (such as Kalman filtering, Hungarian matching, or IoU association) to match the latest recognition result with the target recognition result in the first image. If the two are highly matched in position, appearance, or motion trajectory (such as the overlap rate exceeding the threshold or the feature similarity meeting the standard), it is determined that the same target has not been destroyed, and the third judgment result is output as "destruction not completed". Conversely, if they cannot match (such as the target disappearing or the feature difference being significant), it is determined that the target has been destroyed, and the third judgment result is output as "destruction completed".

[0263] S693. When the third judgment result indicates that the destruction is complete, the target control end sends a return command to the ground end.

[0264] For example, if the target control terminal determines that the third determination result is that the destruction is completed, the target control terminal sends a return command to the ground terminal in the reconnaissance module so that the reconnaissance module returns from the deployment position.

[0265] S694. If the third judgment result indicates that the destruction has not been completed, the target control terminal controls the fire support module to strike the enemy target again.

[0266] For example, if the target control terminal determines that the third determination result is that the destruction is not completed, the target control terminal continues to send control commands to the fire support module to destroy the enemy target again until the third determination result is that the destruction is completed.

[0267] Specifically, when the target control terminal determines that the enemy target is not effectively attackable based on the first judgment result, the enemy target is attacked through the fire support module to destroy it, and a destruction judgment is made. Based on the destruction judgment, it is determined whether to continue to attack through the fire support module; thus, when the attack swarm does not have the conditions to attack, the enemy target can be destroyed in a timely manner.

[0268] In some alternative embodiments, such as Figure 3 As shown, Figure 3 This is a flowchart of the steps of a drone reconnaissance method according to the present invention. The drone reconnaissance steps include, but are not limited to, steps 1 to 33.

[0269] Step 1: The target control terminal receives the reconnaissance mission, obtains the reconnaissance area from the reconnaissance mission, and determines the reconnaissance range based on the reconnaissance rules corresponding to the reconnaissance experience of the reconnaissance experts and the reconnaissance area.

[0270] Step 2: The target control terminal obtains the initial images acquired by visible light remote sensing satellites within the reconnaissance range;

[0271] Step 3: Perform target recognition on the initial image to obtain abnormal targets and their corresponding initial location information;

[0272] Step 4: Using SAR synthetic aperture radar remote sensing to detect the initial position information, distinguish the authenticity of the corresponding abnormal targets to obtain the enemy target and the target position information corresponding to the enemy target;

[0273] Step 5: The target control terminal uses the target location information as the center and a preset length as the radius to determine the reconnaissance target range and obtains the associated environmental information corresponding to the reconnaissance target range;

[0274] Step 6: Analyze the relevant environmental information to obtain the deployment location of the reconnaissance module;

[0275] Step 7: The target control terminal controls the reconnaissance module to reach the deployment location;

[0276] Step 8: Obtain the reconnaissance behavior of a real reconnaissance aircraft when it is most similar to the environment of the enemy target from the historical database. The decoy drone approaches the enemy target based on the reconnaissance behavior, triggering the enemy's radar scan and anti-radio suppression.

[0277] Step 9: If the decoy drone encounters strong electromagnetic suppression within the data link frequency band, the attack units equipped with anti-radiation modules, which are part of the decoy drone swarm, will directly lock onto the suppression antenna and attack. Simultaneously, other drones in the decoy swarm will attract and occupy the detection, jamming, suppression, and attack resources of the anti-drone system, causing system saturation.

[0278] Step 10: Determine the strike duration of the decoy drone and the first time when the decoy drone's strike unit begins to strike, based on the initial strike resources corresponding to the decoy drone, the real-time strike resources corresponding to the attraction and occupation of the anti-drone system's detection, interference, suppression and strike resources.

[0279] Step 11: Simultaneously, the decoy drone collects electromagnetic feature data in real time from the target's anti-radar and electromagnetic suppression antennas to obtain electromagnetic feature data.

[0280] Step 12: The decoy drone generates an anti-electromagnetic map of the corresponding area based on the target location information using electromagnetic feature data;

[0281] Step 13: Based on the anti-electromagnetic map, autonomously avoid strong detection signals and electromagnetic suppression areas to determine the target's communication method;

[0282] Step 14: Send the anti-electromagnetic map, first time, and strike duration to the ground terminal according to the target's communication method;

[0283] Step 15: The ground end determines the reconnaissance route of the reconnaissance aircraft based on the anti-electromagnetic map, and calculates the second time corresponding to the reconnaissance aircraft conducting reconnaissance along the reconnaissance route;

[0284] Step 16: Determine if the time difference between the first time and the second time is less than the strikeable duration. If so, proceed to step 17; otherwise, proceed to step 21.

[0285] Step 17: The reconnaissance aircraft conducts close-range reconnaissance of enemy targets according to the reconnaissance route;

[0286] Step 18: The reconnaissance aircraft collects information on the enemy target under the target location information according to the reconnaissance route and obtains the first image information. The reconnaissance aircraft transmits the first image information back to the ground end, and the ground end sends the first image information to the target control end.

[0287] Step 19: The target control terminal receives the first image information sent by the ground terminal of the reconnaissance module, and performs target judgment or recognition on the first image information according to the target recognition algorithm or target classification algorithm to obtain the target recognition result;

[0288] Step 20: Based on the first image information, use artificial intelligence algorithms to determine whether the reconnaissance module's strike swarm has an effective strike capability, obtain the first judgment result, and jump to step 24;

[0289] Step 21: The ground control sends a request for strike control to the target control station;

[0290] Step 22: After receiving the request for strike control command, the target control terminal obtains the enemy target and the target location information corresponding to the enemy target from the request for strike control command;

[0291] Step 23: Send a control command to the fire support module based on the target location information, so that the fire support module can carry out fire coverage on the target location information corresponding to the enemy target according to the control command, so as to destroy the abnormal target and end the reconnaissance.

[0292] Step 24: Determine if the first judgment result is that the attack can be effectively carried out. If yes, proceed to step 25; otherwise, proceed to step 30.

[0293] Step 25: The target control terminal determines the strike location information corresponding to the bee swarm based on the first image information and sends the strike location information to the ground terminal;

[0294] Step 26: The ground terminal sends the strike location information to the strike swarm, and the strike swarm arrives at the strike location information to carry out the strike;

[0295] Step 27: After the strike is completed, the reconnaissance aircraft acquires images of the strike results to obtain second image information, and sends the second image information to the ground end, which then sends the second image information to the target control end.

[0296] Step 28: The target control terminal performs a destruction completion judgment on the second image information based on the target tracking algorithm and obtains the second judgment result;

[0297] Step 29: Is the second judgment result that the destruction is complete? If yes, end the reconnaissance; otherwise, proceed to step 26.

[0298] Step 30: The target control terminal sends a control command to the fire support module, controlling the fire support module to strike the strike position indicated by the strike position information, so as to destroy the abnormal target.

[0299] Step 31: After the strike is completed, the reconnaissance aircraft acquires images of the strike results to obtain third image information, and sends the third image information to the ground end, which then sends the third image information to the target control end.

[0300] Step 32: The target control terminal performs a destruction judgment on the third image information based on the target tracking algorithm and obtains the third judgment result;

[0301] Step 33: The third judgment result is whether the destruction is complete. If so, the reconnaissance ends; otherwise, proceed to step 30.

[0302] For the technical problems that can be solved and the technical effects that can be obtained by each step or combination of steps 1 to 33 in this embodiment, please refer to the above embodiment, which will not be described in detail here.

[0303] The beneficial effects of this invention include: when conducting reconnaissance of enemy targets, this invention acquires the reconnaissance range corresponding to the reconnaissance mission and the initial image corresponding to the reconnaissance range obtained through remote sensing satellites; after target identification of the initial image, it acquires abnormal targets and their corresponding initial position information; it detects multi-angle imaging information of abnormal targets through synthetic aperture radar remote sensing; after verifying the authenticity of abnormal targets based on the multi-angle imaging information, it obtains the enemy target and its corresponding target position information; it determines the deployment position of the reconnaissance module based on the target position information, a preset reconnaissance radius centered on the target position information, and associated environmental information, where the associated environmental information represents the environmental information corresponding to the reconnaissance target range constructed through the preset reconnaissance radius and target position information; and it controls the reconnaissance module to conduct reconnaissance of enemy targets at the deployment position. By acquiring the initial image within the reconnaissance range through remote sensing satellites, enemy targets are identified and located based on the initial image, thereby accurately identifying enemy targets within the reconnaissance range; simultaneously, the deployment position of the reconnaissance module is calculated to ensure accurate reconnaissance location, thus improving the timeliness and reliability of reconnaissance. Therefore, this application can obtain enemy information in a timely manner, improving the reliability of UAV system reconnaissance.

[0304] like Figure 4 As shown, Figure 4 A structural block diagram of a controller 1000 according to an embodiment of this application is shown. The components of the controller 1000 include, but are not limited to, a memory 1200 and a processor 1100. The processor 1100 is connected to the memory 1200 via a bus, and the memory 1200 is used to store data.

[0305] The controller 1000 also includes an access device that enables the controller 1000 to communicate via one or more networks. Examples of such networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0306] The controller 1000 can be any type of stationary or mobile electronic device, including mobile computers or mobile electronic devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable electronic devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary electronic devices such as desktop computers or PCs. The controller 1000 can also be a mobile or stationary server.

[0307] The processor 1100 is used to execute computer-executable instructions for unmanned aerial vehicle (UAV) reconnaissance methods.

[0308] The above is an illustrative scheme of a controller according to this embodiment. It should be noted that the technical solution of this controller belongs to the same concept as the technical solution of the above-described UAV reconnaissance method. For details not described in detail in the technical solution of the controller, please refer to the description of the technical solution of the above-described UAV reconnaissance method.

[0309] This application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described UAV reconnaissance method.

[0310] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate, and may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0311] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as processors, such as central processing units, digital signal processors, or microprocessors executing software, or as hardware, or as integrated circuits, such as application-specific integrated circuits. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0312] The above provides a detailed description of the preferred embodiments of this application. However, this application is not limited to the above-described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for unmanned aerial vehicle (UAV) reconnaissance, characterized in that, An application to a target control terminal of an unmanned aerial vehicle (UAV) reconnaissance system, the UAV reconnaissance system including at least one reconnaissance module connected to the target control terminal, the UAV reconnaissance method including: Obtain the reconnaissance range corresponding to the reconnaissance mission and acquire the initial image corresponding to the reconnaissance range through remote sensing satellites; After performing target recognition on the initial image, abnormal targets and their corresponding initial location information are obtained; Multi-angle imaging information of the anomalous target was detected by synthetic aperture radar remote sensing. After verifying the authenticity of the abnormal target based on the multi-angle imaging information, the enemy target and the target location information corresponding to the enemy target are obtained. The deployment location of the reconnaissance module is determined based on the target location information, a preset reconnaissance radius centered on the target location information, and associated environmental information. The associated environmental information represents the environmental information corresponding to the reconnaissance target range constructed by the preset reconnaissance radius and the target location information. The reconnaissance module is controlled to conduct reconnaissance of the enemy target at the deployment location.

2. The UAV reconnaissance method according to claim 1, characterized in that, The step of obtaining abnormal targets and their corresponding initial location information after target recognition of the initial image includes: After performing grayscale and filtering processes on the initial image in sequence, a filtered image is obtained. A data template is determined based on the median of the filtered image. The data template is then moved in the filtered image to obtain the template image corresponding to the data template in the filtered image. Obtain the minimum and maximum data values ​​corresponding to the template image; When the target median is between the minimum and maximum data values, a first threshold corresponding to the template image is obtained based on the mean between the minimum and maximum data values, wherein the target median represents the median of the filtered image; If the target value is not located between the minimum and maximum values ​​of the data, the data template is expanded according to a preset rule to obtain an updated data template. The updated data template is then moved in the filtered image to obtain the updated template image corresponding to the updated data template in the filtered image. Obtain the minimum and maximum values ​​of the updated data corresponding to the updated template image, so that the target median value is located between the minimum and maximum values ​​of the updated data, and obtain the second threshold corresponding to the updated template image, wherein both the first threshold and the second threshold are filtering thresholds; The template image is denoised according to the filtering threshold to obtain a denoised image. The denoised image is divided into multiple image regions using initial segmentation lines; After adjusting the position of the initial segmentation line based on the similarity between multiple image regions, the target edge information is obtained; The abnormal target and its corresponding initial position information are obtained based on the target edge information.

3. The UAV reconnaissance method according to claim 2, characterized in that, After adjusting the position of the initial segmentation line based on the similarity between the multiple image regions, the target edge information is obtained, including: The comparison result is obtained by comparing the first similarity between the first region and the third region and the second similarity between the second region and the third region. The third region is the region covered by the initial segmentation line. The first region is located on the first side of the third region, and the second region is located on the second side of the third region. The first region, the second region, and the third region all represent the image region. If the comparison result indicates that the first similarity is greater than or equal to the second similarity, the initial segmentation line is adjusted to the position of the first region. If the comparison result indicates that the first similarity is less than the second similarity, the initial segmentation line is adjusted to the position of the second region. When the difference between the two sides of the initial dividing line is the largest, the target edge information is obtained.

4. The UAV reconnaissance method according to claim 2, characterized in that, The step of verifying the authenticity of the abnormal target based on the multi-angle imaging information to obtain the enemy target and the target location information corresponding to the enemy target includes: Based on the target edge information and affine transformation, the relevant position information of the abnormal target at different times is obtained; The movement information of the abnormal target is determined based on the initial position information and the relevant position information, wherein the movement information includes turning information, speed information and trajectory information; Construct a false target feature library and a false target prediction model; The false target feature library, the movement information, and the false target prediction model are used to predict the true or false probability of the abnormal target to obtain the true or false probability of the target. The abnormal target with the true or false probability greater than the probability threshold is configured as the enemy target, and the initial position information of the abnormal target is configured as the target position information. When the probability of the target being real or fake is less than a probability threshold, the multi-angle imaging information of the abnormal target is detected by synthetic aperture radar remote sensing. After verifying the authenticity of the abnormal target based on the multi-angle imaging information, the enemy target and the target's location information are obtained.

5. The UAV reconnaissance method according to claim 4, characterized in that, The step of predicting the true / false probability of the abnormal target based on the false target feature library, the movement information, and the false target prediction model includes: Obtain the digital elevation model and road vector data corresponding to the reconnaissance range; After mapping the trajectory information to the digital elevation model, the elevation data associated with the trajectory points is obtained. The elevation data indicates the terrain elevation, slope, and land cover type. Obtain real-time map data corresponding to the reconnaissance range; Based on the real-time map data, the road vector data, and the trajectory information, terrain travel cost information and road distance data are determined. The road distance data represents the distance between the trajectory point indicated by the trajectory information and the road. The terrain travel cost information represents the travel cost of the trajectory point indicated by the trajectory information. After the target trajectory of the abnormal target is split into multiple time-series trajectories according to a preset time period, the target trajectory represents the trajectory indicated by the trajectory information; The elevation data, terrain access cost information, and road distance data corresponding to each time segment trajectory are sequentially input into the false target prediction model. The false target prediction model calls the false target feature library and then performs a true / false prediction on the time segment trajectory to obtain the true / false probability of the segment. The false target prediction model integrates the true and false probabilities of all time-series segment trajectories to obtain the first true and false probability. After performing cluster analysis on the target edge information of all the abnormal targets, different sub-clusters with different cluster ranges are obtained. The cluster range represents the regional range of all the abnormal targets contained in the same sub-cluster. The density ratio is determined based on the distribution of all the anomalous targets in the same sub-cluster within the cluster range; The second true / false probability of the anomalous target is determined based on the density ratio; The true / false probability of the abnormal target is determined based on the first true / false probability and the second true / false probability.

6. The UAV reconnaissance method according to claim 1, characterized in that, The step of determining the deployment location of the reconnaissance module based on the target location information, a preset reconnaissance radius centered on the target location information, and associated environmental information includes: Using the location indicated by the target location information as the target center, the preset reconnaissance radius is used to construct the reconnaissance target range around the target center; The associated environmental information corresponding to the range of the reconnaissance target is obtained, and the associated environmental information includes LiDAR terrain data, electromagnetic spectrum information and visible light image information; The associated environmental information is divided into multiple spatial grid cells after being segmented into grids. The elevation and slope information corresponding to the spatial grid cell are determined based on the LiDAR terrain data, the signal strength corresponding to the spatial grid cell is determined based on the electromagnetic spectrum information, and the target type corresponding to the spatial grid cell is determined based on the visible light image information. Calculate the correlation value between any two spatial grid cells based on the LiDAR terrain data; The grid diagram information is obtained by configuring the spatial grid cells as nodes, configuring the associated values ​​as edges of the corresponding nodes, and configuring the elevation information, the slope information, the signal strength, and the target type as node features of the corresponding nodes. Multiple node features corresponding to any target grid cell are obtained from the grid graph information, and the multiple node features are fused to obtain the first feature information; Based on the first feature information and the association value corresponding to the target grid cell, the target neighborhood of the target grid cell is determined in the grid graph information; After assessing the threat of the target grid cell based on the second feature information of other grid cells in the target's neighborhood and the first feature information, a target score is obtained. A three-dimensional threat situation field is determined based on the target grid cells and the target score; Obtain the temporal location coordinates of the enemy target; After inputting the temporal location coordinates and the associated environmental information into the spatiotemporal feature encoder, kinematic features and radiation features are obtained. The abnormal features of the enemy target include the kinematic features and the radiation features. Obtain the real-time pose information of the reconnaissance module; The real-time pose information, the kinematic features, the radiation features, and the three-dimensional threat situation field are input into the deep reinforcement learning framework; The deep reinforcement learning framework is used to dynamically optimize based on the observation clarity target and the exposure risk target, and a Pareto front solution set is generated based on the non-dominated sorting genetic algorithm. The optimal approach distance is obtained by selecting the optimal solution from the Pareto front solution set according to the tactical stage requirements. The deployment location is determined based on the optimal proximity distance.

7. A UAV reconnaissance method according to claim 6, characterized in that, The step of calculating the correlation value between any two spatial grid cells based on the LiDAR terrain data includes: The terrain information corresponding to each spatial grid cell is determined based on the LiDAR terrain data. Based on the terrain information, determine the terrain connectivity information corresponding to any two spatial grid cells, where the terrain connectivity information represents the connectivity coefficient between the two spatial grid cells; Calculate the Euclidean distance between any two of the spatial grid cells; The correlation value between any two spatial grid cells is determined based on the Euclidean distance and the terrain connectivity information.

8. The UAV reconnaissance method according to claim 1, characterized in that, The unmanned aerial vehicle (UAV) reconnaissance system also includes a fire support module. This module comprises a reconnaissance vehicle and reconnaissance aircraft, a strike swarm, and a ground terminal deployed on the vehicle. Controlling the reconnaissance module to conduct reconnaissance of the enemy target at the deployment location includes: The target control terminal controls the reconnaissance vehicle to reach the deployment location; After the reconnaissance aircraft conducts reconnaissance of the enemy target, it obtains first image information and transmits the first image information back to the ground terminal, so that the ground terminal can send the first image information to the target control terminal. The target control terminal obtains the first image information and judges the attack on the bee swarm based on the first image information to obtain a first judgment result; When the first judgment result indicates that the strike can be effectively carried out, the target control terminal determines the strike location information corresponding to the strike bee swarm based on the first image information, and sends the strike location information to the ground terminal. The ground terminal sends the strike location information to the strike swarm so that the strike swarm can reach the strike location indicated by the strike location information to carry out the strike. After the reconnaissance aircraft acquires images of the strike location, it sends the second image information to the target control terminal. The target control terminal obtains the second image information and, based on the target tracking algorithm, performs a destruction judgment on the second image information to obtain a second judgment result. When the second judgment result indicates that the destruction is complete, the target control terminal sends a return command to the ground terminal. If the second judgment result indicates that the destruction is not complete, the target control terminal controls the strike swarm to strike the enemy target again until the enemy target is completely destroyed.

9. A UAV reconnaissance method according to claim 8, characterized in that, The reconnaissance vehicle is also equipped with a decoy drone. After the reconnaissance vehicle reaches the deployment location, the process further includes: Acquire historical reconnaissance behavior, which represents the reconnaissance behavior of the reconnaissance aircraft when its environment was most similar to that of the enemy target in history; The decoy drone is controlled to approach the enemy target using the historical reconnaissance behavior. The decoy drone navigates according to the target location information and does not communicate with the ground terminal while approaching the enemy target.

10. A UAV reconnaissance method according to claim 9, characterized in that, The process of the decoy drone approaching the enemy target using the aforementioned historical reconnaissance behavior also includes: When the decoy drone encounters strong electromagnetic suppression within the data link frequency band, the first strike unit, which is composed of a swarm of decoy drones, locks onto and strikes the suppression antenna. The first strike unit is equipped with an anti-radiation module. By attracting and occupying the detection, interference, suppression, and attack resources of the anti-drone system through other drones in the feint swarm, the anti-drone system can be saturated. The strikeable duration of the decoy drone and the first time when the second strike unit of the decoy drone begins to strike are determined based on the initial strike resources of the decoy drone, the real-time strike resources corresponding to the attraction and occupation of the detection, interference, suppression and strike resources of the anti-drone system. Electromagnetic characteristic data is obtained by collecting electromagnetic characteristic data emitted by the enemy's anti-navigation radar and electromagnetic suppression antenna in real time using the decoy drone. The decoy drone generates an anti-electromagnetic map of the corresponding area under the target location information based on the electromagnetic feature data, and autonomously avoids strong detection signals and electromagnetic suppression areas based on the anti-electromagnetic map to determine the target communication mode. Based on the target communication mode, the anti-electromagnetic map, the first time, and the strikeable duration are sent to the ground terminal.

11. A method for unmanned aerial vehicle (UAV) reconnaissance according to claim 10, characterized in that, Before determining the attack on the bee swarm based on the first image information to obtain a first determination result, the method further includes: The ground terminal determines the reconnaissance route of the reconnaissance aircraft based on the anti-electromagnetic map, and calculates the second time corresponding to the reconnaissance aircraft conducting reconnaissance along the reconnaissance route; If the time difference between the first time and the second time is less than the strikeable duration, the reconnaissance aircraft is controlled to collect information on the enemy target according to the reconnaissance route to obtain the first image information; If the time difference between the first time and the second time is greater than or equal to the strikeable duration, a strike control request command sent by the ground terminal is received. The enemy target and the target location information corresponding to the enemy target are obtained from the request for strike control command; The fire support module is controlled to strike the target location indicated by the target location information.

12. A method for unmanned aerial vehicle (UAV) reconnaissance according to any one of claims 8-11, characterized in that, The method further includes: If the first judgment result indicates that an effective strike cannot be carried out, the target control terminal controls the fire support module to strike the strike location indicated by the strike location information. The ground terminal acquires images of the strike location via the reconnaissance aircraft and then sends the third image information to the target control terminal. The target control terminal obtains a third judgment result after determining whether the third image information has been destroyed based on the target tracking algorithm. When the third judgment result indicates that the destruction is complete, the target control terminal sends a return command to the ground terminal. If the third judgment result indicates that the destruction is not completed, the target control terminal controls the fire support module to strike the enemy target again.

13. A drone reconnaissance system, characterized in that, The system includes a controller, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the UAV reconnaissance method according to any one of claims 1-12.

14. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which are used to execute the UAV reconnaissance method according to any one of claims 1-12.

Images