Non-cooperative multi-dimensional data real-time identification method and system for unmanned aerial vehicle flight operation

Abstract

The application provides a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification method and system, relates to the unmanned aerial vehicle flight identification technical field, and the method comprises the following steps: constructing a typical behavior mode library based on cooperative unmanned aerial vehicles, including a typical motion mode, a typical infrared mode and a typical load mode; collecting multi-dimensional data of a target airspace in real time through a multi-dimensional sensor group, and analyzing and obtaining flight state data, infrared feature data and image feature data of a plurality of unmanned aerial vehicles; based on the typical behavior mode library, the flight state data, the infrared feature data and the image feature data are iterated to perform multi-dimensional parallel matching screening, and a multi-dimensional candidate object set is obtained; taking the multi-dimensional candidate object set as an analysis target, performing multi-dimensional fusion identification, and determining a non-cooperative flight operation object corresponding to the non-cooperative unmanned aerial vehicle. The technical problems of low recognition accuracy and high false alarm and missed alarm rates in the prior art non-cooperative unmanned aerial vehicle flight operation identification method are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned aerial vehicle flight identification, and in particular to a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification method and system. BACKGROUND

[0002] With the rapid popularization of unmanned aerial vehicle technology, cooperative unmanned aerial vehicles and non-cooperative unmanned aerial vehicles coexist in low-altitude airspace, posing a serious challenge to airspace safety management. However, traditional non-cooperative unmanned aerial vehicle flight operation identification methods rely on single-dimensional data such as optical images, and have the problems of low identification accuracy and high false alarm and missed alarm rates. Therefore, there is an urgent need for a non-cooperative unmanned aerial vehicle flight operation identification technology that integrates multi-dimensional features and takes into account real-time and accuracy to meet the actual needs of low-altitude safety prevention and control. SUMMARY

[0003] The present application provides a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification method and system to solve the technical problems of low identification accuracy and high false alarm and missed alarm rates in the prior art.

[0004] The technical solution of the present application to solve the above technical problems is as follows:

[0005] In a first aspect, the present application provides a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification method, comprising:

[0006] constructing a typical behavior pattern library based on cooperative unmanned aerial vehicles, the typical behavior pattern library including typical motion patterns, typical infrared patterns, and typical load patterns;

[0007] real-time acquisition of multi-dimensional data of a target airspace by a multi-dimensional sensor group, and parsing to obtain flight state data, infrared feature data, and image feature data of multiple unmanned aerial vehicles;

[0008] based on the typical behavior pattern library, performing multi-dimensional parallel matching and screening on the flight state data, the infrared feature data, and the image feature data to obtain a multi-dimensional candidate object set;

[0009] taking the multi-dimensional candidate object set as an analysis target, performing multi-dimensional fusion identification based on the flight state data, the infrared feature data, and the image feature data to determine a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle.

[0010] In a second aspect, the present application provides a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification system, comprising:

[0011] a pattern library construction module for constructing a typical behavior pattern library based on cooperative unmanned aerial vehicles, the typical behavior pattern library including typical motion patterns, typical infrared patterns, and typical load patterns;

[0012] The data acquisition module is used for acquiring multi-dimensional data of the target airspace in real time through a multi-dimensional sensor group, and analyzing and obtaining flight state data, infrared feature data and image feature data of a plurality of unmanned aerial vehicles;

[0013] The matching and screening module is used for performing multi-dimensional parallel matching and screening on the flight state data, the infrared feature data and the image feature data based on the typical behavior mode library, and obtaining a multi-dimensional candidate object set.

[0014] The fusion recognition module is used for taking the multi-dimensional candidate object set as an analysis target, performing multi-dimensional fusion recognition based on the flight state data, the infrared feature data and the image feature data, and determining a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle.

[0015] The present application has the following advantages:

[0016] Compared with the prior art, the present application firstly constructs a typical behavior mode library based on cooperative unmanned aerial vehicles, the typical behavior mode library including a typical motion mode, a typical infrared mode and a typical load mode, and the features of the cooperative unmanned aerial vehicles are extracted into the typical behavior mode library through clustering, thereby providing a precise comparison basis for subsequent real-time recognition of non-cooperative unmanned aerial vehicles. Secondly, multi-dimensional data of the target airspace are acquired in real time through a multi-dimensional sensor group, and flight state data, infrared feature data and image feature data of a plurality of unmanned aerial vehicles are analyzed and obtained, the target airspace data are synchronously acquired through a plurality of sensors, and are converted into structured flight state data, infrared feature data and image feature data through analysis, thereby providing a reliable data basis for subsequent matching and screening. Thirdly, based on the typical behavior mode library, multi-dimensional parallel matching and screening are performed on the flight state data, the infrared feature data and the image feature data, a multi-dimensional candidate object set is obtained, and unmanned aerial vehicles not conforming to the typical behavior mode library are screened out from all unmanned aerial vehicles in the target airspace to form the candidate object set, thereby providing a reliable data basis for subsequent recognition of non-cooperative unmanned aerial vehicles. Finally, taking the multi-dimensional candidate object set as an analysis target, multi-dimensional fusion recognition is performed based on the flight state data, the infrared feature data and the image feature data, a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle is determined, and a high-confidence non-cooperative flight operation object recognition result is output, thereby providing a reliable decision basis for low-altitude safety control.

[0017] By the technical solution, the application constructs a typical behavior mode library of cooperative unmanned aerial vehicles, as a comparison benchmark, collects motion, infrared and load characteristic data of a target airspace in real time through a multi-dimensional sensor group such as a radar, an infrared sensor and an optical sensor, filters out unmanned aerial vehicles with non-cooperative suspicion through parallel matching of multi-dimensional features and the typical behavior mode library, obtains a multi-dimensional candidate object set, and finally performs multi-dimensional fusion recognition on the multi-dimensional candidate object set to determine a non-cooperative flight operation object of the non-cooperative unmanned aerial vehicle. In this way, the risk of misjudgment of a single dimension is reduced, and real-time and accurate recognition of the non-cooperative flight operation object of the non-cooperative unmanned aerial vehicle is realized. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification method provided by the application is shown.

[0019] Figure 2 A structural diagram of a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification system provided by the application is shown.

[0020] In the drawings, the components represented by the numbers are as follows:

[0021] The mode library construction module 11, the data acquisition module 12, the matching and filtering module 13 and the fusion recognition module 14. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the application.

[0023] In the description of the application, the terms "first" and "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0024] In the description of the present application, the term "for example" is used to mean "serving as an example, instance, or illustration." Any embodiment described as "for example" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the present application. In the following description, for the purpose of explanation, details are set forth in order to provide a thorough understanding of the present application. It will be appreciated that one skilled in the art will be able to practice the present application without these specific details. In other instances, well known structures and processes are not elaborated upon in order to avoid obscuring the description of the present application. Thus, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features presented herein.

[0025] Embodiment one, as shown, the present application embodiment provides a multi-dimensional data non-cooperative unmanned aerial vehicle flight operation real-time identification method, comprising: Figure 1

[0026] S10: Construct a typical behavior pattern library based on cooperative unmanned aerial vehicles, the typical behavior pattern library includes typical motion patterns, typical infrared patterns and typical load patterns;

[0027] Cooperative unmanned aerial vehicles refer to authorized and recorded unmanned aerial vehicles with legal flight qualifications, whose flight tasks, airspace routes, equipment parameters and other information have been included in the supervision system, such as logistics distribution unmanned aerial vehicles, surveying and mapping operation unmanned aerial vehicles, etc., and the flight behavior follows the preset specification and is traceable. Non-cooperative unmanned aerial vehicles refer to unmanned aerial vehicles that have not been authorized and recorded, without legal flight qualifications, whose flight plans, equipment characteristics, task purposes and other information are not included in the supervision, such as black flight unmanned aerial vehicles, illegal modification unmanned aerial vehicles, etc., which may have risks such as airspace occupation and safety threat.

[0028] Traditional non-cooperative unmanned aerial vehicle identification methods mostly rely on single-dimensional data, such as judgment by radar trajectory or optical image only, which has the problems of weak anti-interference ability and one-sided feature matching, resulting in high false positive rate and false negative rate, and cannot accurately distinguish between cooperative and non-cooperative targets. Therefore, by constructing a typical behavior pattern library based on cooperative unmanned aerial vehicles, the common characteristics of motion, infrared, load and other multi-dimensional data can be extracted, providing an accurate benchmark for the identification of non-cooperative unmanned aerial vehicles.

[0029] To solve the above problems, the present application constructs a typical behavior pattern library based on cooperative unmanned aerial vehicles, which includes typical motion patterns, typical infrared patterns and typical load patterns.

[0030] Specifically, step S10 in the method comprises:

[0031] ​According to the unmanned aerial vehicle list of the cooperative unmanned aerial vehicle, the historical task data and the planned task data are extracted and data deduplication processing is performed to obtain standard task data;

[0032] Based on the preset behavior index set, the standard task data is traversed to perform index value extraction, and a standard behavior feature set is correspondingly generated;

[0033] The standard behavior feature set is mapped to a multi-dimensional feature space, density-based clustering analysis is performed, and typical mode calibration is performed according to a plurality of clustering clusters of the distance analysis result, and the output is the typical behavior mode library.

[0034] In the embodiment of the application, first, according to the unmanned aerial vehicle list of the cooperative unmanned aerial vehicle, the historical task data and the planned task data are extracted and data deduplication processing is performed to obtain standard task data, wherein the unmanned aerial vehicle list of the cooperative unmanned aerial vehicle refers to the unmanned aerial vehicle ID list, the model list and the like registered and recorded, so as to define the data collection range, the historical task data refers to the historical record data of the completed flight task, including the historical flight height, the infrared frequency, the load size and the like, and the planned task data refers to the data of the future flight task reported, including the planned flight height, the infrared frequency, the load size and the like. For example, according to the unmanned aerial vehicle list of the cooperative unmanned aerial vehicle, such as the registered and recorded cooperative unmanned aerial vehicle ID list, the historical task data of a plurality of cooperative unmanned aerial vehicles is extracted, for example, the historical task data of a certain logistics unmanned aerial vehicle extracted from the historical flight log is flight height 50m, infrared frequency 3μm, load size 0.6m*0.4m and the like, and the planned task data is extracted from the reported future flight plan, for example, the planned task data reported by a certain surveying and mapping unmanned aerial vehicle is flight height 45m, infrared frequency 4μm, load size 0.5m*0.6m and the like, and then through the unique identification of the cooperative unmanned aerial vehicle (such as the unmanned aerial vehicle number), the task timestamp and the like, the repeated or redundant data is removed, for example, only one valid data is retained for the plurality of historical task data of the same unmanned aerial vehicle and the same task, and finally the standard task data is formed, which is the basis for subsequent feature extraction, ensuring the uniqueness and reliability of the data.

[0035] Secondly, based on the preset behavior index set, the standard task data is traversed to perform index value extraction, and a standard behavior feature set is correspondingly generated, wherein the preset behavior index set is the screening standard for behavior feature extraction, including the motion index set, the infrared index set and the load index set, the motion index set includes the flight height, the velocity vector and the acceleration vector, the infrared index set includes the infrared distribution and the infrared frequency, and the load index set includes the load-flight directionality, the load size, the load shape and the load form.

[0036] Finally, the standard behavior feature set is mapped to a multi-dimensional feature space, density-based clustering analysis is performed, and according to the distance analysis result of multiple cluster, a typical mode is labeled, and the output is the typical behavior mode library. Exemplarily, each standard behavior feature in the standard behavior feature set corresponds to a data point in the multi-dimensional feature space, each component of the vector corresponds to a coordinate axis in the space, the component value size corresponds to the specific position on the coordinate axis, the dimension of the vector determines the dimension of the feature space, for example, a standard behavior feature vector [20, 5, 0.87, 0.3, 0.87, 0.4, 0.3, 0.3, 2.8, 0.98, 0.5, 0.3, 0.6, 1, 1] containing 15 components, that is, a point in the 15-dimensional feature space. Through this mapping, the abstract behavior feature is converted into a quantifiable space coordinate for subsequent clustering analysis.

[0037] Exemplarily, the density-based clustering analysis can adopt a density clustering algorithm such as DBSCAN, and the similar samples are aggregated by analyzing the distribution density of the points in the feature space to form multiple cluster clusters, wherein each cluster represents a common behavior mode, and the core logic is to set a neighborhood radius (ε) and a minimum point number (minPts) with any point in the space as the center, if the number of points contained in the neighborhood is ≥ the minimum point number, the point is defined as a core point, and the potential cluster cluster is formed by the core point and its reachable high-density area, wherein the minimum point number is dynamically determined according to the actual data amount. The specific clustering process is: traverse all points in the feature space, identify the core points and other points connected with the density (i.e. indirectly associated points through the core points), and classify these density-connected points into the same cluster, for example, multiple surveying and mapping unmanned aerial vehicles highly coincide in the characteristics of “flight height 500-800 meters, speed size 8-12 m / s, infrared main frequency 3-4 μm”, etc. will be aggregated into the same cluster. Compared with traditional clustering algorithms such as K-means, density clustering does not need to pre-set the number of clusters, can adaptively identify irregular-shaped clusters, and has stronger fault tolerance to noise data (such as occasional abnormal flight records), and the clustering result is more in line with the actual behavior rule.

[0038] Exemplarily, "calibrating a typical mode according to a plurality of clustering clusters of distance analysis results, and outputting a typical behavior mode library" refers to extracting a representative core feature as a typical mode for each clustering cluster, wherein distance analysis verifies the consistency of features within the cluster by calculating the distance (such as Euclidean distance, cosine distance) between samples within the cluster. The smaller the distance, the higher the similarity of the samples, and the better the stability of the cluster. On this basis, the mean vector (i.e. centroid) of all feature vectors within the cluster is calculated as the typical mode of the cluster. For example, the mean motion feature vector of a surveying and mapping unmanned aerial vehicle cluster is [650, 10, 0.9] (corresponding to a height of 650 meters, a speed of 10 m / s, and a direction angle cosine of 0.9), and the mean vector is calibrated as the typical motion mode of the surveying and mapping unmanned aerial vehicle. According to the same method, the mean infrared feature vector and the mean load feature vector of the cluster are extracted as the corresponding typical infrared mode and typical load mode. Finally, the typical modes of all clustering clusters are classified and integrated according to the motion→infrared→load dimension to form a typical behavior mode library covering various common behaviors of cooperative unmanned aerial vehicles, providing a standardized benchmark for subsequent multi-dimensional recognition of non-cooperative unmanned aerial vehicles.

[0039] Specifically, the "extracting and generating a behavior feature set based on a preset behavior index set and traversing the standard task data execution index values" includes:

[0040] Taking the motion index set in the behavior index set as the target, the motion index extraction and vectorization of the standard task data are traversed to generate a motion feature vector set;

[0041] Taking the infrared index set in the behavior index set as the target, the infrared index extraction and vectorization of the standard task data are traversed to generate an infrared feature vector set;

[0042] Taking the load index set in the behavior index set as the target, the load index extraction and vectorization of the standard task data are traversed to generate a load feature vector set;

[0043] According to the unique identification mark of the cooperative unmanned aerial vehicle, the mapping relationship between the motion feature vector set, the infrared feature vector set, and the load feature vector set is established, and is combined and output as the behavior feature set.

[0044] In the embodiment of the application, first, taking the motion index set in the behavior index set as the target, the motion index extraction and vectorization of the standard task data are traversed to generate a motion feature vector set. Exemplarily, taking the motion index set (including flight height, speed vector, acceleration vector, etc.) in the behavior index set as the target, the specific data of each motion index is extracted by traversing the standard task data, for example, the flight height is 20 meters, the speed vector is 5 m / s (east by north 30°), and the acceleration vector is 0.3 m / s2 (30° east of north), and performs vectorization processing, for example, for numerical indicators such as flight altitude, speed size, acceleration size, etc., directly retaining the original numerical values as feature components, for angle indicators such as speed direction angle, acceleration direction angle, etc., converting them into dimensionless numerical features through trigonometric functions, such as taking the cosine value cos30° = 0.87, and then concatenating in a fixed order of flight altitude, speed size, speed direction cosine value, acceleration size, and acceleration direction cosine value to generate the motion feature vector of this data, such as [20, 5, 0.87, 0.3, 0.87]. In this way, each standard task data is converted into a motion feature vector with consistent structure according to the same method, and finally the motion feature vector set is formed. Second, taking the infrared indicator set in the behavior indicator set as the target, the infrared indicator extraction and vectorization are performed on the standard task data to generate the infrared feature vector set. Exemplarily, taking the infrared indicator set (including infrared distribution, infrared frequency, etc.) in the behavior indicator set as the target, the specific data of each infrared indicator is extracted from the standard task data, for example, the front body infrared intensity proportion of 40%, the middle body infrared intensity proportion of 30%, the rear body infrared intensity proportion of 30%, and the infrared main frequency of 2.8 μm are extracted from a standard task data, and vectorization processing is performed, for example, for numerical indicators such as infrared distribution, infrared frequency, etc., directly retaining the original numerical values as feature components, and then concatenating in a fixed order of front body infrared intensity proportion, middle body infrared intensity proportion, rear body infrared intensity proportion, and infrared main frequency to generate the infrared feature vector of this data, such as [0.4, 0.3, 0.3, 2.8]. In this way, each standard task data is converted into an infrared feature vector with consistent structure according to the same method, and finally the infrared feature vector set is formed.

[0045] Again, targeting the load indicator set in the behavior indicator set, the standard task data is traversed to extract and vectorize the load indicators, generating a load feature vector set. Illustratively, targeting the load indicator set (including load-flight directionality, load size, load shape, load form, etc.) in the behavior indicator set, the standard task data is traversed to extract the specific data of each load indicator, for example, from a piece of standard task data, the load-flight direction angle is 10°, the load size is 0.5m x 0.3m x 0.6m (length x width x height), the load shape is a cube, and the load form is ordinary goods, and vectorization processing is performed, for example, for angle-type indicators such as load-flight direction angle, through trigonometric function conversion to dimensionless numerical features, such as taking the cosine value cos40° = 0.98, for numerical indicators such as load size, the original numerical value is directly retained as a feature component, for classification indicators such as load shape and load form, based on a preset coding rule, such as presetting cube = 1, cylinder = 2, irregular shape = 3, ordinary goods = 1, liquid = 2, communication equipment = 3, then according to the fixed order of load-flight direction angle cosine value, load length, load width, load height, load shape code, and load form code, the load feature vector of this piece of data is generated, such as [0.98, 0.5, 0.3, 0.6, 1, 1], in this way, according to the same method, the standard task data is traversed to convert each piece of standard task data into a load feature vector with consistent structure, and finally the load feature vector set is formed by aggregating.

[0046] Finally, a mapping relationship between the motion feature vector set, the infrared feature vector set and the load feature vector set is established according to the unique identification mark of the cooperative UAV, and the mapping relationship is merged and output as the behavior feature set. For example, the multi-dimensional features of the same UAV are accurately associated through the unique identification mark of the cooperative UAV. Taking the cooperative UAV UAV-001 as an example, the motion feature vector (such as [20, 5, 0.87, 0.3, 0.87]), the infrared feature vector (such as [0.4, 0.3, 0.3, 2.8]) and the load feature vector (such as [0.98, 0.5, 0.3, 0.6, 1, 1]) of the cooperative UAV UAV-001 are bound through the registration ID “UAV-001” of the cooperative UAV UAV-001, so as to ensure that the three types of feature vectors correspond to the same subject, and then the associated motion, infrared and load feature vectors are merged and output to obtain the behavior feature of the cooperative UAV UAV-001: [20, 5, 0.87, 0.3, 0.87, 0.4, 0.3, 0.3, 2.8, 0.98, 0.5, 0.3, 0.6, 1, 1]. In this way, the multi-dimensional features of all cooperative UAVs are associated and merged according to the same method, and finally the behavior feature set containing all cooperative UAVs is formed, which provides complete samples for subsequent clustering analysis of the typical behavior mode library, and ensures that the clustering result can accurately reflect the comprehensive behavior rule of the same UAV.

[0047] Further, the typical motion mode at least includes a flight height item, a speed vector item and an acceleration vector item; the typical infrared mode at least includes an infrared distribution item and an infrared frequency item; and the typical load mode at least includes a load-flight directionality item, a load size item, a load shape item and a load form item.

[0048] In the embodiment of the application, the typical motion mode can reflect the flight behavior rule of the cooperative UAV, and at least includes a flight height item, a speed vector item and an acceleration vector item. The flight height item refers to the commonly used height range of the cooperative UAV in the conventional task. For example, the working height of the surveying and mapping UAV is usually stable at 500-800 meters. If it is detected that the UAV flies beyond the range for a long time, it may belong to non-cooperative behavior. The speed vector item includes the speed size and the direction angle. The speed size needs to comply with the general restriction of civil UAV (such as ≤15 m / s), and the direction angle needs to be regular, such as gently changing along the preset route. If there is high-speed flight or irregular turning, such as the direction angle suddenly changes by ≥60° in a short time, it may be non-cooperative behavior. The acceleration vector item includes the acceleration size and the direction angle. The acceleration size and direction of the cooperative UAV are usually stable. If there is a sharp acceleration or deceleration, such as the acceleration suddenly exceeds 3 m / s 2Or irregular variable speed, which may be non-cooperative behavior. The typical infrared mode can reflect the physical characteristics of the fuselage, at least including the infrared distribution term and the infrared frequency term, wherein the infrared distribution term is fixed due to the location of the heating components such as motors and batteries of the cooperative unmanned aerial vehicle, and the infrared intensity distribution has significant regularity, such as the infrared energy proportion of the motor part ≥40%, the load area in the middle of the fuselage ≤20%, if the infrared distribution proportion is detected to be abnormal, such as the infrared proportion of the load area suddenly increases, it may be a modified model or a non-cooperative unmanned aerial vehicle; the infrared frequency term refers to the main frequency of the infrared radiation of the fuselage heating component, for example, the infrared main frequency of the motor of the civilian unmanned aerial vehicle is usually concentrated in 2-4 μm, if non-typical high-frequency infrared radiation of 5-7 μm is detected, it may be equipped with special equipment, which is non-cooperative behavior. The typical load mode can reflect the mounting feature regularity, at least including the load-flight directionality term, the load size term, the load shape term, and the load form term, wherein the load-flight directionality term is used to judge the consistency of the load direction and the flight direction, the load direction angle and the speed vector angle of the cooperative unmanned aerial vehicle are usually ≤15°, if the angle ≥45°, it may be temporarily modified or abnormally operated; the load size term needs to meet the size range corresponding to the load bearing design of the unmanned aerial vehicle, such as the maximum load size of a certain type of unmanned aerial vehicle is 1m×0.8m×0.5m, if the oversized load is detected, it may be non-cooperative behavior; the load shape term refers to the legal load which is usually regular geometric shape, such as cube, cylinder, etc., if irregular shapes such as long strip and multiple protrusions appear and are not within the reported range, it may be illegal equipment; the load form term needs to be consistent with the legal type reported, such as ordinary goods, surveying and mapping cameras, communication repeaters, etc., if the unreported type is detected, it may be non-cooperative behavior. In summary, compared with the prior art, the application constructs a typical behavior mode library based on cooperative unmanned aerial vehicles, which includes typical motion mode, typical infrared mode and typical load mode. In this way, the characteristics of cooperative unmanned aerial vehicles are extracted into a typical behavior mode library through clustering, which provides accurate comparison basis for subsequent real-time identification of non-cooperative unmanned aerial vehicles.

[0049] S20: Real-time collection of multi-dimensional data of the target airspace by the multi-dimensional sensor group, and analysis and acquisition of flight state data, infrared feature data and image feature data of multiple unmanned aerial vehicles;

[0050] The flight state data (reflecting the motion regularity), the infrared feature data (reflecting the physical characteristics of the fuselage) and the image feature data (reflecting the load mounting characteristics) of the unmanned aerial vehicle can accurately depict the individual feature differences of the unmanned aerial vehicle from different dimensions, for example, the height range, infrared radiation regularity and load form of different types of unmanned aerial vehicles have significant differences. Therefore, by real-time collection of multi-dimensional data of the target airspace, the difference information of the three types of features can be comprehensively analyzed, which provides a basis for distinguishing cooperative and non-cooperative unmanned aerial vehicles.

[0051] To solve the above problems, the multi-dimensional sensor group is used to collect multi-dimensional data of the target airspace in real time, and flight state data, infrared feature data and image feature data of multiple unmanned aerial vehicles are obtained.

[0052] Specifically, step S20 in the method comprises:

[0053] Based on the preset digital fence, radar data, infrared data and optical data of the target airspace are synchronously collected;

[0054] According to the radar data, flight height value, flight speed vector and acceleration vector are calculated and output as the flight state data;

[0055] The infrared data are analyzed to obtain infrared distribution and infrared frequency, and the infrared feature data are output;

[0056] Based on the optical data, edge-target joint detection is performed, load size, load shape and load form are extracted according to the edge-target joint detection result, load-flight directionality is defined in combination with the flight speed vector, and the load size, the load shape, the load form and the load-flight directionality are combined to obtain the image feature data.

[0057] In the embodiment of the application, first, based on the preset digital fence, radar data, infrared data and optical data of the target airspace are synchronously collected, wherein the preset digital fence refers to a preset airspace boundary, which is used to limit the spatial range of data collection to avoid irrelevant airspace data interference. The spatial range can be determined by geographic coordinates, for example, a low-altitude airspace of 30°-30.5° north latitude and 120°-120.5° east longitude. Illustratively, the radar data can be used to obtain the spatial position and motion trajectory of the unmanned aerial vehicle through radar echo, the infrared data can be used to collect the thermal imaging image and infrared radiation signal of the unmanned aerial vehicle through an infrared sensor, and the optical data can be used to obtain the visible light image of the unmanned aerial vehicle through a high-definition camera or an optical telescope. The three types of sensors need to collect data according to a unified time stamp (time error ≤0.1 second) to ensure that the radar, infrared and optical data of the same unmanned aerial vehicle can be accurately associated, and to avoid confusing different unmanned aerial vehicle data at different times.

[0058] Secondly, the flight height value, flight speed vector, acceleration vector are calculated according to the radar data, and the output is spliced into the flight state data. Exemplarily, the motion core indicators of the unmanned aerial vehicle are extracted and calculated from the radar data: the flight height value can be calculated by the elevation angle and slant range in the radar data, combined with the trigonometric function to calculate the absolute height, for example, the slant range is 1000 meters, the elevation angle is 30°, then the flight height value = 1000 x sin 30° = 500 meters; the flight speed vector can be calculated by the position change rate of the continuous frame radar data, including the speed size (such as 8 m / s) and the speed direction (such as east by north 45°); the acceleration vector can be calculated by the change rate of the speed vector, including the acceleration size (such as 0.5 m / s 2 ) and the acceleration direction (such as consistent or opposite to the speed direction); finally, the flight height value, flight speed vector, acceleration vector are spliced in a fixed format to form structured data, such as vector form: [flight height value, flight speed size, flight speed direction angle cosine value, acceleration size, acceleration direction angle cosine value]. Thirdly, the infrared data is analyzed to obtain the infrared distribution and infrared frequency, and the output is spliced into the infrared feature data. Exemplarily, the infrared distribution can be analyzed by analyzing the thermal imaging map collected by the infrared sensor, and the infrared intensity distribution of different regions of the fuselage is analyzed according to the gray value, for example, the motor part is high in infrared intensity (high in gray value) due to continuous heating, the middle part of the fuselage is low in infrared intensity (small in gray value) without heating components, and finally the infrared distribution characteristics are output, such as the infrared intensity of the front part of the fuselage accounts for 30%, the infrared intensity of the middle part of the fuselage accounts for 20%, and the infrared intensity of the rear part of the fuselage accounts for 50%; the infrared frequency can be obtained by frequency spectrum analysis of the infrared radiation signal, such as the peak frequency obtained by Fourier transform, for example, the infrared main frequency is 4 μm; finally, the infrared distribution and the infrared frequency are spliced to form structured data, such as vector form: [infrared intensity of the front part of the fuselage accounts for 30%, infrared intensity of the middle part of the fuselage accounts for 20%, infrared intensity of the rear part of the fuselage accounts for 50%, infrared main frequency is 4 μm].

[0059] Finally, edge-target joint detection is performed based on the optical data, load size, load shape, and load form are extracted according to the edge-target joint detection result, and load-flight directionality is defined in combination with the flight speed vector, and the load size, the load shape, the load form, and the load-flight directionality are merged to obtain the image feature data. For example, for the collected visible light image, edge-target joint detection is performed, wherein edge detection extracts the edge profile in the image through a Canny operator or the like, such as the gray difference boundary between the unmanned aerial vehicle body and the background, and complex background interference is excluded, target positioning is performed on the basis of the edge profile, and the unmanned aerial vehicle body and the mounted load area are accurately positioned through a target detection model (such as YOLO), so as to ensure that the feature extraction focuses on the target itself. For example, the load size can be converted into the actual physical size through pixel proportion according to the target positioning result, and the length, width, and height of the load are calculated; the load shape can be judged by fitting the edge profile, such as a rectangle→a cube, a circle→a cylinder, and an irregular polygon→a non-standard shape; the load form can be identified in combination with the shape and texture features, for example, a cube with a lens→a surveying camera, a sealed box→cargo, and an antenna-like protrusion→a communication device; the load-flight directionality can be calculated by combining the flight speed vector, the included angle between the load orientation and the flight direction, the load orientation is judged by the geometric center line of the load in the optical image, the flight direction is the direction angle of the speed vector, and the larger the included angle, the more abnormal the load mounting is; and finally, the load size, the load shape, the load form, and the load-flight directionality are merged to obtain the image feature data, such as [length, width, height, shape code, form code, and direction included angle cosine value].

[0060] In summary, compared with the prior art, the present application acquires multi-dimensional data of a target airspace in real time through a multi-dimensional sensor group, and analyzes and obtains flight state data, infrared feature data, and image feature data of multiple unmanned aerial vehicles. In this way, the target airspace data is synchronously collected through multiple sensors, and is converted into structured flight state, infrared feature, and image feature data through analysis, thereby providing a reliable data basis for subsequent matching and screening.

[0061] S30: Based on the typical behavior mode library, multi-dimensional parallel matching and screening are performed on the flight state data, the infrared feature data, and the image feature data to obtain a multi-dimensional candidate object set.

[0062] The typical behavior mode library is established based on cooperative unmanned aerial vehicles, and contains typical motion modes, infrared modes, and load modes. By comparing the real-time collected flight state data with the typical motion modes, the infrared feature data with the typical infrared modes, and the image feature data with the typical load modes, three-dimensional parallel matching and screening are performed, and unmanned aerial vehicles with non-cooperative suspicion can be screened out.

[0063] To solve the above problems, the application performs multi-dimensional parallel matching screening on the flight state data, the infrared feature data and the image feature data based on the typical behavior mode library, and obtains a multi-dimensional candidate object set.

[0064] Specifically, step S30 in the method comprises:

[0065] The typical motion mode is compared with the flight state data for first matching screening, and the unmanned aerial vehicle that does not match the typical motion mode is output as a first candidate object set;

[0066] The typical infrared mode is compared with the infrared feature data for second matching screening, and the unmanned aerial vehicle that does not match the typical infrared mode is output as a second candidate object set;

[0067] The typical load mode is compared with the image feature data for third matching screening, and the unmanned aerial vehicle that does not match the typical load mode is output as a third candidate object set;

[0068] The union of the first candidate object set, the second candidate object set and the third candidate object set is taken as the multi-dimensional candidate object set.

[0069] In the embodiment of the application, first, the typical motion mode is compared with the flight state data for first matching screening, and the unmanned aerial vehicle that does not match the typical motion mode is output as a first candidate object set. As long as any one of the flight height value, the flight speed vector and the acceleration vector in the flight state data does not match the typical motion mode, the unmanned aerial vehicle is classified into the first candidate object set. Exemplarily, the similarity algorithm (such as Euclidean distance, dynamic time warping, etc.) is used to calculate the similarity between the flight state data vector of the target unmanned aerial vehicle and the typical motion mode vector in the typical behavior mode library, and the unmanned aerial vehicle whose flight state data does not match the typical motion mode is screened out through a preset distance threshold. The preset distance threshold can be set according to the behavior fluctuation range of the cooperative unmanned aerial vehicle, for example, the height dimension threshold is set to 100 meters. For example, if the real-time flight height of the unmanned aerial vehicle is 200 meters, and the corresponding typical motion mode of the cooperative unmanned aerial vehicle has a flight height range of 500-800 meters, the Euclidean distance between the flight height and the typical mode is 300 meters, which is greater than the preset distance threshold of 100 meters, so it is determined that the unmanned aerial vehicle does not match in the height dimension, and the unmanned aerial vehicle is classified into the first candidate object set. The unmanned aerial vehicles in the first candidate object set do not match the typical motion mode of the cooperative unmanned aerial vehicle in the motion mode, and there is a suspicion of non-cooperation.

[0070] Secondly, the second matching screening is performed by comparing the typical infrared mode with the infrared feature data, and a UAV that does not match the typical infrared mode is output as a second candidate object set. As long as any one of the infrared distribution and the infrared frequency in the infrared feature data does not match the typical motion mode, the UAV is included in the second candidate object set. For example, the typical infrared mode is compared with the infrared feature data. If the proportion of the middle body infrared in the infrared feature data is 40%, which contradicts the proportion of 30% of the middle body infrared in the typical infrared mode, it is determined that the UAV does not match the typical motion mode in the infrared distribution, and the UAV is included in the second candidate object set. The UAV in the second candidate object set does not match the typical motion mode of the cooperative UAV in the infrared mode, and is suspected of being non-cooperative.

[0071] Thirdly, the third matching screening is performed by comparing the typical load mode with the image feature data, and a UAV that does not match the typical load mode is output as a third candidate object set. As long as any one of the load size, the load shape, the load form, and the load-flight directionality in the image feature data does not match the typical motion mode, the UAV is included in the third candidate object set. For example, the typical load mode is compared with the image feature data. If the load size is 1.5 m x 1.2 m, which exceeds the corresponding load size 1 m x 0.8 m of the typical load mode, it is determined that the UAV does not match the typical motion mode in the load size, and the UAV is included in the third candidate object set. The UAV in the third candidate object set does not match the typical motion mode of the cooperative UAV in the load mode, and is suspected of being non-cooperative.

[0072] Finally, the union of the first candidate object set, the second candidate object set, and the third candidate object set is taken as the multi-dimensional candidate object set. For example, the union of the first candidate object set, the second candidate object set, and the third candidate object set is taken as the final multi-dimensional candidate object set. As long as a UAV does not match the typical mode in any dimension of motion, infrared, and load, the UAV is included in the candidate object set. Through multi-dimensional parallel screening and union, the coverage of potential non-cooperative objects is maximized, and single-dimensional missed detection is avoided. For example, a UAV with normal motion mode but abnormal load form can still be included in the candidate object set through the third screening, thereby providing comprehensive candidate samples for subsequent accurate identification.

[0073] In summary, compared with the prior art, the multi-dimensional parallel matching screening is performed on the flight state data, the infrared feature data, and the image feature data based on the typical behavior mode library, and the multi-dimensional candidate object set is obtained. In this way, the UAVs that do not conform to the typical behavior mode library are screened from all the UAVs in the target airspace, and the candidate object set is formed, thereby providing a reliable data basis for subsequent identification of non-cooperative UAVs.

[0074] S40: taking the multi-dimensional candidate object set as an analysis target, performing multi-dimensional fusion recognition based on the flight state data, the infrared feature data, and the image feature data, and determining a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle.

[0075] The foregoing steps accurately obtain a multi-dimensional candidate object set suspected of being non-cooperative through multi-dimensional parallel matching and screening of the motion, infrared, load features, and typical behavior mode library of the target airspace unmanned aerial vehicle, and further non-cooperative flight operation object recognition can be performed based on the multi-dimensional candidate object set.

[0076] To solve the foregoing problem, the present application takes the multi-dimensional candidate object set as an analysis target, performs multi-dimensional fusion recognition based on the flight state data, the infrared feature data, and the image feature data, and determines a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle.

[0077] Specifically, step S40 in the method includes:

[0078] Taking the multi-dimensional candidate object set as an index, data cleaning is performed on the flight state data, the infrared feature data, and the image feature data.

[0079] In combination with a fusion evaluation model pre-trained based on sample labeled data, multi-dimensional fusion analysis is performed on the data cleaning result to obtain a flight operation fusion evaluation value.

[0080] According to the flight operation fusion evaluation value, confidence screening is performed on the multi-dimensional candidate object set, and an object whose flight operation fusion evaluation value does not satisfy a pre-set confidence constraint is determined as the non-cooperative flight operation object.

[0081] In the embodiment of the present application, first, taking the multi-dimensional candidate object set as an index, data cleaning is performed on the flight state data, the infrared feature data, and the image feature data. Exemplarily, taking the multi-dimensional candidate object set as an index, for the unmanned aerial vehicle pointed to by the index, noise removal, standardization, and other data cleaning are performed on the flight state data, the infrared feature data, and the image feature data of the unmanned aerial vehicle, for example, abnormal values caused by sensor errors are removed, such as a speed of 50 m / s instantaneously misestimated by a radar, data is smoothed through a sliding window mean method, short-time data loss, such as 1-2 frames of data lost due to infrared sensor blocking, is supplemented with reasonable values through an interpolation method, different dimensional data is unified in dimension to ensure fairness of feature weights in subsequent fusion analysis, and through data cleaning, data input into the fusion evaluation model is ensured to be accurate, complete, and standardized.

[0082] Secondly, the data cleaning result is subjected to multi-dimensional fusion analysis combined with a fusion evaluation model pre-trained based on sample labeled data to obtain a flight operation fusion evaluation value, wherein the sample labeled data contains cooperative unmanned aerial vehicle data (labeled as cooperative) and non-cooperative unmanned aerial vehicle data (labeled as non-cooperative), the fusion evaluation model is constructed based on machine learning, the correlation between multi-dimensional feature data and cooperative / non-cooperative labels is learned, and the flight state data, infrared feature data and image feature data subjected to data cleaning are input to output the flight operation fusion evaluation value.

[0083] Exemplarily, the fusion evaluation model can be implemented through the following technical path: 1. Data preparation: collect multi-dimensional feature data containing flight state data, infrared feature data and load feature data, label the data as cooperative or non-cooperative by manual operation combined with airspace supervision records to form sample labeled data, clean the features of the sample labeled data, and then divide the sample labeled data into a training set, a validation set and a test set according to a ratio of 7:1.5:1.5. 2. Model construction: an integrated learning random forest model is used to construct the fusion evaluation model, which is composed of 100 independent decision trees. Each decision tree is independently trained by randomly sampling samples from the training set through Bootstrap sampling to ensure the diversity between trees. The flight state data, infrared feature data and load feature data are spliced in a fixed dimension order to form an N-dimensional comprehensive feature vector, ensuring that all relevant features are included in the analysis. The maximum depth of the decision tree is set to 10 to limit the complexity of the tree and avoid overfitting to the training data. The maximum number of features for each tree split is the square root of the total dimension of the comprehensive feature vector, i.e. The minimum number of split samples is set to 20, i.e. stop splitting when the number of node samples is less than 20 to ensure that the branches have statistical significance. The Gini coefficient is used as the feature splitting criterion, and the output layer generates a fusion evaluation value through a probability voting mechanism. Each decision tree outputs a prediction result of cooperative or non-cooperative for the input sample. The model aggregates the votes of all decision trees and converts the cooperative vote proportion to a quantitative score of 0-1. For example, if 80 out of 100 trees predict cooperation, the flight operation fusion evaluation value is 0.8. 3. Model training: the multi-dimensional feature vector in the training set is input, and the corresponding quantitative target (e.g. flight operation fusion evaluation value 1 for cooperation and flight operation fusion evaluation value 0 for non-cooperation) is used as the learning target to train the model to learn the mapping rule between features and labels. During the training process, the model performance (e.g. accuracy and F1 score) is monitored in real time through the validation set to dynamically adjust hyperparameters such as the number of decision trees and the maximum depth. When the accuracy on the test set is ≥95%, it is considered to have converged, and the fusion evaluation model is obtained. The model output is the flight operation fusion evaluation value (quantitative score of 0-1), and the higher the score, the higher the comprehensive similarity between the target unmanned aerial vehicle and the typical cooperative behavior pattern, and the lower the score, the more significant the comprehensive difference, i.e. the more likely it is a non-cooperative unmanned aerial vehicle.

[0084] Finally, the multi-dimensional candidate object set is confidence filtered according to the flight operation fusion evaluation value, and the candidate object whose flight operation fusion evaluation value does not satisfy a preset confidence constraint is determined as the non-cooperative flight operation object, wherein the preset confidence constraint is an evaluation threshold of the flight operation fusion evaluation value, which can be dynamically set by a person skilled in the art according to actual conditions, for example, the preset confidence constraint is set to 0.6, if the flight operation fusion evaluation value is lower than the preset confidence constraint, the UAV is determined as the non-cooperative flight operation object, and if the flight operation fusion evaluation value is greater than or equal to the preset confidence constraint, the UAV is determined as the cooperative flight operation object. Exemplarily, if the preset confidence constraint is 0.6, for each UAV in the multi-dimensional candidate object set, the flight operation fusion evaluation value thereof is compared with the preset confidence constraint, if the flight operation fusion evaluation value is greater than or equal to 0.6, the non-cooperative suspicion thereof is excluded, if the flight operation fusion evaluation value is less than 0.6, the UAV is determined as the non-cooperative flight operation object, all candidate objects not satisfying the preset confidence constraint are determined as the non-cooperative flight operation object by traversing the multi-dimensional candidate object set, and accurate determination of the non-cooperative flight operation object is realized.

[0085] To sum up, compared with the prior art, the multi-dimensional candidate object set is taken as an analysis target, multi-dimensional fusion recognition is performed based on the flight state data, the infrared feature data and the image feature data, and the non-cooperative flight operation object corresponding to the non-cooperative UAV is determined. In this way, a high-confidence non-cooperative flight operation object recognition result is output, and a reliable decision basis is provided for low-altitude safety control.

[0086] To sum up, the embodiments of the present application have at least the following technical effects:

[0087] Compared with the prior art, the present application first constructs a typical behavior mode library based on cooperative UAVs, which includes a typical motion mode, a typical infrared mode and a typical load mode. In this way, the features of cooperative UAVs are extracted as a typical behavior mode library through clustering, providing accurate comparison basis for subsequent real-time recognition of non-cooperative UAVs.

[0088] Secondly, the present application realizes real-time collection of multi-dimensional data of the target airspace by a multi-dimensional sensor group, and obtains flight state data, infrared feature data and image feature data of multiple UAVs through analysis. In this way, the target airspace data is synchronously collected by multiple sensors, and is converted into structured flight state, infrared feature and image feature data through analysis, providing a reliable data basis for subsequent matching and screening.

[0089] Furthermore, based on the aforementioned typical behavior pattern library, this application performs multi-dimensional parallel matching and filtering by traversing the flight status data, the infrared feature data, and the image feature data to obtain a multi-dimensional candidate object set. In this way, drones that do not conform to the typical behavior pattern library are selected from all drones in the target airspace, forming a candidate object set, providing a reliable data foundation for the subsequent identification of non-cooperative drones.

[0090] Finally, this application uses the aforementioned multi-dimensional candidate object set as the analysis target, and performs multi-dimensional fusion recognition based on the flight status data, the infrared feature data, and the image feature data to determine the non-cooperative flight operation objects corresponding to the non-cooperative UAVs. In this way, a high-confidence non-cooperative flight operation object identification result is output, providing a reliable decision-making basis for low-altitude safety management.

[0091] Through the aforementioned technical solution, this application constructs a typical behavior pattern library for cooperative UAVs as a comparison benchmark. Using a multi-dimensional sensor array encompassing radar, infrared, and optical sensors, it collects real-time motion, infrared, and payload characteristic data of the target airspace. By parallel matching of multi-dimensional features with the typical behavior pattern library, UAVs suspected of non-cooperative behavior are screened out, resulting in a multi-dimensional candidate set. Finally, multi-dimensional fusion identification is performed on this candidate set to determine the non-cooperative flight targets of non-cooperative UAVs. This reduces the risk of misjudgment based on a single dimension and achieves real-time and accurate identification of non-cooperative flight targets of non-cooperative UAVs.

[0092] Example 2, as Figure 2 As shown, based on the same inventive concept as the multidimensional data non-cooperative drone flight operation real-time identification method provided in Embodiment 1, this embodiment of the invention also provides a multidimensional data non-cooperative drone flight operation real-time identification system, including:

[0093] The pattern library construction module 11 is used to construct a typical behavior pattern library based on cooperative UAVs. The typical behavior pattern library includes typical motion patterns, typical infrared patterns, and typical payload patterns.

[0094] The data acquisition module 12 is used to acquire multidimensional data of the target airspace in real time through a multidimensional sensor group, and to analyze and obtain flight status data, infrared feature data and image feature data of multiple UAVs.

[0095] Matching and filtering module 13 is used to perform multi-dimensional parallel matching and filtering based on the typical behavior pattern library, traversing the flight status data, the infrared feature data and the image feature data to obtain a multi-dimensional candidate object set;

[0096] The fusion recognition module 14 is configured to take the multi-dimensional candidate object set as an analysis target, perform multi-dimensional fusion recognition based on the flight state data, the infrared feature data, and the image feature data, and determine a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle.

[0097] The mode library construction module 11 is specifically configured to:

[0098] According to the unmanned aerial vehicle list of the cooperative unmanned aerial vehicle, historical task data and planned task data corresponding to the standard task data are extracted and data deduplication processing is performed to obtain the standard task data.

[0099] Based on a preset behavior index set, index value extraction is performed on the standard task data, and a standard behavior feature set is generated accordingly.

[0100] The standard behavior feature set is mapped to a multi-dimensional feature space, density-based clustering analysis is performed, and typical mode calibration is performed according to a plurality of clustering clusters of the distance analysis result, and the output is the typical behavior mode library.

[0101] Specifically, the "based on the preset behavior index set, the index value extraction is performed on the standard task data, and the behavior feature set is generated accordingly", including:

[0102] Taking a motion index set in the behavior index set as a target, motion index extraction and vectorization are performed on the standard task data to generate a motion feature vector set;

[0103] Taking an infrared index set in the behavior index set as a target, infrared index extraction and vectorization are performed on the standard task data to generate an infrared feature vector set;

[0104] Taking a load index set in the behavior index set as a target, load index extraction and vectorization are performed on the standard task data to generate a load feature vector set;

[0105] According to the unique identification mark of the cooperative unmanned aerial vehicle, a mapping relationship of the motion feature vector set, the infrared feature vector set, and the load feature vector set is established, and the behavior feature set is output.

[0106] Further, the typical motion mode at least includes a flight height item, a speed vector item, and an acceleration vector item; the typical infrared mode at least includes an infrared distribution item and an infrared frequency item; and the typical load mode at least includes a load-flight directionality item, a load size item, a load shape item, and a load form item.

[0107] The data acquisition module 12 is specifically configured to:

[0108] Synchronously collect radar data, infrared data and optical data of a target airspace based on a preset digital fence;

[0109] According to the radar data, calculate a flight height value, a flight speed vector and an acceleration vector, and splice and output the flight height value, the flight speed vector and the acceleration vector as the flight state data;

[0110] Analyze the infrared data to obtain infrared distribution and infrared frequency, and splice and output the infrared distribution and the infrared frequency as the infrared feature data;

[0111] Based on the optical data, perform edge-target joint detection, extract load size, load shape and load form according to the edge-target joint detection result, and combine the load size, the load shape, the load form and load-flight directionality to obtain the image feature data.

[0112] The matching and screening module 13 is specifically configured to:

[0113] Perform first matching and screening on the typical motion mode and the flight state data, and output a first candidate object set that does not match the typical motion mode;

[0114] Perform second matching and screening on the typical infrared mode and the infrared feature data, and output a second candidate object set that does not match the typical infrared mode;

[0115] Perform third matching and screening on the typical load mode and the image feature data, and output a third candidate object set that does not match the typical load mode;

[0116] Take the union of the first candidate object set, the second candidate object set and the third candidate object set as the multi-dimensional candidate object set.

[0117] The fusion recognition module 14 is specifically configured to:

[0118] Take the multi-dimensional candidate object set as an index to perform data cleaning on the flight state data, the infrared feature data and the image feature data;

[0119] Combine a fusion evaluation model pre-trained based on sample label data to perform multi-dimensional fusion analysis on the data cleaning result to obtain a flight operation fusion evaluation value;

[0120] According to the flight operation fusion evaluation value, perform confidence screening on the multi-dimensional candidate object set, and determine a candidate object that does not satisfy a preset confidence constraint as the non-cooperative flight operation object.

[0121] To sum up, the embodiments of the present application have at least the following technical effects:

[0122] Compared with the prior art, firstly, the mode library construction module is used to construct a typical behavior mode library based on cooperative unmanned aerial vehicles, the typical behavior mode library includes a typical motion mode, a typical infrared mode and a typical load mode, and the features of the cooperative unmanned aerial vehicles are extracted into the typical behavior mode library through clustering, thereby providing a precise comparison basis for subsequent real-time identification of non-cooperative unmanned aerial vehicles. Secondly, the data acquisition module is used to acquire multi-dimensional data of a target airspace in real time through a multi-dimensional sensor group, and to analyze and obtain flight state data, infrared feature data and image feature data of multiple unmanned aerial vehicles, thereby providing a reliable data basis for subsequent matching and screening through synchronous acquisition of target airspace data by multiple sensors and conversion of the data into structured flight state, infrared feature and image feature data. Thirdly, the matching and screening module is used to perform multi-dimensional parallel matching and screening on the flight state data, the infrared feature data and the image feature data based on the typical behavior mode library, to obtain a multi-dimensional candidate object set, and to screen out unmanned aerial vehicles that do not conform to the typical behavior mode library from all unmanned aerial vehicles in the target airspace, thereby providing a reliable data basis for subsequent identification of non-cooperative unmanned aerial vehicles. Finally, the fusion identification module is used to take the multi-dimensional candidate object set as an analysis target, to perform multi-dimensional fusion identification based on the flight state data, the infrared feature data and the image feature data, to determine a non-cooperative flight operation object corresponding to the non-cooperative unmanned aerial vehicle, and to output a high-confidence non-cooperative flight operation object identification result, thereby providing a reliable decision basis for low-altitude safety control. In this way, the risk of misjudgment in a single dimension is reduced, and real-time and accurate identification of a non-cooperative flight operation object of a non-cooperative unmanned aerial vehicle is achieved.

[0123] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0124] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or blocks of the flowcharts. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or blocks of the flowcharts.

[0126] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or blocks of the flowcharts. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or blocks of the flowcharts.

[0127] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or blocks of the flowcharts. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or blocks of the flowcharts.

[0128] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments described and shown, and it is therefore intended that the application cover any and all modifications and forms derived from the embodiments set forth in the concept, and their equivalents.

Claims

1. A non-cooperative unmanned aerial vehicle flight operation real-time identification method for multi-dimensional data, characterized in that, The application relates to a method for identifying non-cooperative flight operation objects of non-cooperative unmanned aerial vehicles. The method comprises the following steps: a typical behavior mode library based on cooperative unmanned aerial vehicles is constructed, the typical behavior mode library comprising a typical motion mode, a typical infrared mode and a typical load mode; multidimensional data of a target airspace are collected in real time through a multidimensional sensor group, and flight state data, infrared feature data and image feature data of multiple unmanned aerial vehicles are obtained through analysis; based on the typical behavior mode library, multidimensional parallel matching screening is performed on the flight state data, the infrared feature data and the image feature data, and a multidimensional candidate object set is obtained; the multidimensional candidate object set is taken as an analysis target, and multidimensional fusion recognition is performed based on the flight state data, the infrared feature data and the image feature data, so that a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle is determined; wherein the construction of the typical behavior mode library based on cooperative unmanned aerial vehicles comprises: according to a unmanned aerial vehicle list of the cooperative unmanned aerial vehicles, historical task data and planned task data are extracted and data deduplication processing is performed, so that standard task data are obtained; based on a preset behavior index set, index value extraction is performed on the standard task data, and a standard behavior feature set is correspondingly generated; the standard behavior feature set is mapped to a multidimensional feature space, density-based clustering analysis is performed, and typical mode calibration is performed according to multiple clustering clusters of distance analysis results, and the output is the typical behavior mode library; wherein, based on the typical behavior mode library, multidimensional parallel matching screening is performed on the flight state data, the infrared feature data and the image feature data, and a multidimensional candidate object set is obtained, which comprises: first matching screening is performed by comparing the typical motion mode with the flight state data, and unmanned aerial vehicles that do not match the typical motion mode are output as a first candidate object set; second matching screening is performed by comparing the typical infrared mode with the infrared feature data, and unmanned aerial vehicles that do not match the typical infrared mode are output as a second candidate object set; third matching screening is performed by comparing the typical load mode with the image feature data, and unmanned aerial vehicles that do not match the typical load mode are output as a third candidate object set; 2. The non-cooperative UAV flight operation real-time identification method of multi-dimensional data according to claim 1, characterized in that, the union of the first candidate object set, the second candidate object set and the third candidate object set is taken as the multidimensional candidate object set. based on the preset behavior index set, index value extraction is performed on the standard task data, and a behavior feature set is correspondingly generated, which comprises: taking a motion index set in the behavior index set as a target, motion index extraction is performed on the standard task data and vectorization is performed, and a motion feature vector set is generated; taking an infrared index set in the behavior index set as a target, infrared index extraction is performed on the standard task data and vectorization is performed, and an infrared feature vector set is generated; taking a load index set in the behavior index set as a target, load index extraction is performed on the standard task data and vectorization is performed, and a load feature vector set is generated; according to unique identification marks of the cooperative unmanned aerial vehicles, a mapping relationship of the motion feature vector set, the infrared feature vector set and the load feature vector set is established, and the mapping relationship is combined and output as the behavior feature set.

3. The non-cooperative UAV flight operation real-time identification method of multi-dimensional data according to claim 2, characterized in that, Real-time collection of multi-dimensional data of the target airspace by a multi-dimensional sensor group, and analysis and acquisition of flight state data, infrared feature data and image feature data of multiple unmanned aerial vehicles, including: Based on the preset digital fence, radar data, infrared data and optical data of the target airspace are synchronously collected; According to the radar data, flight height value, flight speed vector and acceleration vector are calculated and output as the flight state data; The infrared data are analyzed to obtain infrared distribution and infrared frequency, and the infrared feature data are output by splicing; Based on the optical data, edge-target joint detection is performed, load size, load shape and load form are extracted according to the edge-target joint detection result, and load-flight directionality is defined in combination with the flight speed vector, and the load size, load shape, load form and load-flight directionality are merged to obtain the image feature data.

4. The non-cooperative UAV flight operation real-time identification method of multi-dimensional data according to claim 1, wherein, The typical motion mode at least includes flight height item, speed vector item and acceleration vector item; the typical infrared mode at least includes infrared distribution item and infrared frequency item; and the typical load mode at least includes load-flight directionality item, load size item, load shape item and load form item.

5. The non-cooperative UAV flight operation real-time identification method of multi-dimensional data according to claim 4, characterized in that, Taking the multi-dimensional candidate object set as an analysis target, multi-dimensional fusion recognition is performed based on the flight state data, the infrared feature data and the image feature data to determine a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle, including: Taking the multi-dimensional candidate object set as an index, data cleaning is performed on the flight state data, the infrared feature data and the image feature data; In combination with a fusion evaluation model pre-trained based on sample label data, multi-dimensional fusion analysis is performed on the data cleaning result to obtain a flight operation fusion evaluation value; According to the flight operation fusion evaluation value, confidence screening is performed on the multi-dimensional candidate object set, and an object whose flight operation fusion evaluation value does not satisfy a preset confidence constraint is determined as the non-cooperative flight operation object.

6. A non-cooperative UAV flight operation real-time identification system of multi-dimensional data, characterized in that, For performing the method of any one of claims 1-5, including: A mode library construction module for constructing a typical behavior mode library based on cooperative unmanned aerial vehicles, the typical behavior mode library including a typical motion mode, a typical infrared mode and a typical load mode; A data collection module for real-time collection of multi-dimensional data of the target airspace by a multi-dimensional sensor group, and analysis and acquisition of flight state data, infrared feature data and image feature data of multiple unmanned aerial vehicles; A matching and screening module for multi-dimensional parallel matching and screening of the flight state data, the infrared feature data and the image feature data based on the typical behavior mode library to obtain a multi-dimensional candidate object set; A fusion recognition module for taking the multi-dimensional candidate object set as an analysis target, performing multi-dimensional fusion recognition based on the flight state data, the infrared feature data and the image feature data, and determining a non-cooperative flight operation object corresponding to a non-cooperative unmanned aerial vehicle.

Images