Bird, unmanned aerial vehicle and air floating object artificial intelligence identification method based on track continuous checking

Abstract

The present application relates to the technical field of low-altitude target recognition, and particularly relates to a bird, unmanned aerial vehicle and airborne object artificial intelligence recognition method based on track continuous checking, comprising: dividing continuous track segments and determining broken end points through time interval, space displacement and speed change relationship, which can locate the track break position and provide clear objects for subsequent reconnection; through forward extrapolation and reverse backtracking of track segments on both sides of the broken end points, candidate reconnection relationships can be constructed, and track breaks caused by occlusion and weak echoes can be reduced; through endpoint prediction track alignment and motion residual spectrum decomposition, reconnection relationships with abnormal motion differences can be screened out, and the influence of wrong connections on recognition results can be reduced; through category matching of drift components, maneuvering components and disturbance components, airborne objects, unmanned aerial vehicles and birds can be distinguished, and the continuous recognition stability and misjudgment suppression capability of low-altitude targets are improved.

Description

Technical Field

[0001] This invention relates to the technical field of low-altitude target recognition, and in particular to an artificial intelligence recognition method for birds, drones and airborne objects based on continuous point verification. Background Technology

[0002] In the scenario of continuous monitoring of low-altitude, slow-moving, and small targets in airport airspace, low-altitude surveillance radar continuously outputs target traces along the airport perimeter, runway extension area, and approach passage. Target traces typically include sampling time, spatial location, velocity value, and altitude value. Due to factors such as obstruction by surrounding buildings, trees, ground clutter, weather disturbances, bird flocks, and low-speed drift of lightweight airborne objects, the same target is prone to short-term loss, trace breakage, track misconnection, or target category jump during continuous scanning. Birds, UAVs, and airborne objects may all exhibit low-altitude, low-speed, small radar cross-section, and short track characteristics. Traditional low-altitude, slow-moving, and small target identification methods mostly classify based on the speed, heading, altitude, or echo characteristics of a single track segment. It is difficult to determine whether the traces before and after the break belong to the same target, and it is also difficult to distinguish the motion differences caused by UAV active maneuvers, local bird disturbances, and airborne object drift. Therefore, in actual monitoring of airport airspace, there are still problems such as insufficient trace continuity, low reliability of breakage reconnection, and poor stability of classification and identification.

[0003] For example, CN115761421A discloses a multi-source information fusion method for detecting low-altitude, slow-moving, and small targets and an unmanned air defense system. This method uses photoelectric, radar, and radio detection equipment to detect foreign targets such as low-altitude floating balloons, drones, and flocks of birds, and obtains target identification results through multi-source information fusion and a target classification model. However, the key technical aspect of this method lies in the data fusion of multi-source detection equipment and the construction of the target classification model. It mainly relies on the corroboration relationship between multiple types of sensors to improve identification reliability. It does not establish a continuous verification mechanism for the broken endpoints of single radar tracks caused by obstruction, weak echoes, or clutter interference, nor does it perform forward extrapolation, reverse extrapolation, or residual spectrum analysis on the tracks on both sides of the broken endpoints. Therefore, the method disclosed in CN115761421A still suffers from problems such as unreliable reconnection of broken tracks and easy misjudgment between birds and floating objects or low-speed drones when only intermittent radar tracks are available in airport airspace or when multi-source information is temporarily missing.

[0004] For example, CN113537347A discloses a method for classifying UAVs and birds based on track motion features. It extracts average speed, speed standard deviation, heading standard deviation, maneuvering factor, and track oscillation factor from the target track and classifies UAVs and birds using a joint probability matrix. However, this method takes the already formed target track as the processing object and focuses on extracting classification parameters from the track motion features. It does not check the point break position, break endpoint, and candidate reconnection relationship before the track is formed, nor does it decompose the motion residual between the endpoint predicted points into drift component, maneuver component, and disturbance component. Therefore, the method disclosed in CN113537347A still has problems such as discontinuous track input, distorted classification results after erroneous reconnection, and inability to output reliable reconnection status when facing short-term interruption of the track, bird flock crossing, UAV hovering and restarting, or slow drifting of airborne objects in airport airspace.

[0005] In summary, given that existing low-speed, small target recognition technologies suffer from problems such as overemphasizing complete track classification or multi-source fusion verification, lacking continuous verification of broken track points, and difficulty in distinguishing birds, drones, and airborne objects based on the broken and reconnected process, this invention proposes an artificial intelligence recognition method for birds, drones, and airborne objects based on continuous track point verification. This method divides continuous track segments and determines the break endpoints by considering the time interval, spatial displacement, and velocity change relationships of adjacent target tracks. It then performs forward extrapolation and reverse extrapolation based on the continuous track segments on both sides of the break endpoint, uses the motion residual spectrum of the predicted track points at the endpoints to filter out abnormal candidate reconnection relationships, and completes category matching based on drift, maneuver, and disturbance components. This solves the problem of how to continuously verify and stably identify birds, drones, and airborne objects under track breakage conditions in airport airspace. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: An AI-based method for identifying birds, drones, and airborne objects based on continuous point verification includes: dividing target point traces into continuous point trace segments according to the time interval, spatial displacement, and velocity change relationship between adjacent target point traces, and determining the end point and beginning point of adjacent continuous point trace segments as breakpoints; Based on the continuity direction, velocity continuity and height change of the continuous point segments on both sides of the fracture endpoint, forward outward and reverse backward extrapolation are performed to obtain the endpoint prediction point traces and form candidate reconnection relationships. Align the endpoint prediction points in the candidate reconnection relationship according to the sampling order, calculate the motion residual and decompose it to obtain the motion residual spectrum, and filter out abnormal candidate reconnection relationships according to the motion residual spectrum to obtain the target continuous point sequence. The drift component, maneuver component, and disturbance component are separated from the motion residual spectrum. The three types of components are matched according to the break and reconnection positions and reconnection confidence states of the target continuous point sequence, and the bird, drone, or airborne object identification result is output.

[0009] The beneficial effects of this invention are as follows: This invention divides continuous track segments and determines breakpoints by considering time intervals, spatial displacement, and velocity changes, enabling the location of track interruptions and providing clear targets for subsequent reconnection; by pushing forward and backward the track segments on both sides of the breakpoint, candidate reconnection relationships can be constructed, reducing track breaks caused by occlusion and weak echoes; by predicting track alignment and decomposing motion residual spectra, reconnection relationships with abnormal motion differences can be screened out, reducing the impact of misconnections on the recognition results; by matching the categories of drift components, maneuver components, and disturbance components, it can distinguish between airborne objects, UAVs, and birds, improving the stability of continuous low-altitude target recognition and the ability to suppress misjudgments. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification as shown in this invention. Detailed Implementation

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0012] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0013] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0014] In a preferred embodiment, the present invention is applied to the continuous monitoring and classification of low-altitude, slow-moving, small targets in airport airspace. The airport airspace includes the runway end extension area, approach protection surface, takeoff and climb surface, and low-altitude warning area outside the airport perimeter. The target traces are derived from a single observation result formed by the airport airspace low-altitude surveillance radar after threshold detection, clutter removal, and centroid aggregation of echo point clouds within the same detection batch. Each target trace includes the sampling time, spatial location, velocity value, altitude value, and trace source number. The spatial location is represented by local rectangular coordinates with the airport runway center point as the origin, including lateral, longitudinal, and vertical coordinates. The velocity value is the ground speed converted from the spatial location change and sampling time difference of adjacent detection batches. The altitude value is the vertical height of the target trace relative to the airport elevation plane.

[0015] For example, the sampling period of the low-altitude surveillance radar is 0.5s, and the sampling times are 12:00:00.000, 12:00:00.500, 12:00:01.000, and 12:00:01.500 respectively. If an observation point with a spatial location of 320m horizontal coordinate, 1850m vertical coordinate, 96m vertical coordinate, a velocity value of 18m / s, and a height value of 96m is formed at 12:00:01.000, then this observation point is a target trace.

[0016] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates an AI-powered method for identifying birds, drones, and airborne objects based on continuous point verification. The method includes the following steps: S1. Based on the time interval, spatial displacement, and velocity variation relationship between adjacent target point traces, the target point traces are divided into continuous point trace segments, and the end and beginning points of adjacent continuous point trace segments are determined as breakpoints. It should be noted that in this step: S1.1 Arrange the target points according to the sampling time to obtain the target point sequence.

[0017] In this embodiment, when arranging target traces according to sampling time, target traces under the same target candidate number are first arranged from earliest to latest according to sampling time. When two target traces have the same sampling time, they are arranged from smallest to largest according to the source number of the traces. Target traces with the same sampling time whose spatial distance is less than 6m and whose height difference is less than 3m are merged into one target trace. The spatial position of the merged target trace is the average of the spatial positions of the merged target traces, the velocity value is the average of the velocity values ​​of the merged target traces, and the height value is the average of the height values ​​of the merged target traces. The resulting target trace sequence is arranged continuously according to sampling time, and each sampling time corresponds to at most one target trace, reducing duplicate traces caused by multipath reflections or radar sidelobe echoes in the same batch.

[0018] As an example, within the approach protection surface on the north side of the airport's airspace, 10 target points are continuously obtained, with sampling times ranging from 12:00:00.000 to 12:00:04.500, and adjacent sampling times differing by 0.5 seconds. These 10 target points are then arranged into a target point sequence according to their sampling times.

[0019] S1.2 Calculate the time interval, spatial displacement, and velocity change of adjacent target points in the target point sequence one by one, and determine the time continuity boundary, displacement continuity boundary, and velocity continuity boundary respectively. Wherein: S1.2.1 Subtract the sampling time of the previous target point from the sampling time of the next target point in the target point sequence to obtain the time interval sequence; For example, the calculation formula is as follows:

[0020] in, For the target point sequence, the first The target point and the first The time interval between each target point; For the first The sampling time of each target point; For the first The sampling time of each target point; The index of the adjacent point in the target point sequence.

[0021] S1.2.2 Calculate the distance between the spatial position of the next target point in the target point sequence and the spatial position of the previous target point to obtain the spatial displacement sequence; For example, the calculation formula is as follows:

[0022] in, For the target point sequence, the first The target point and the first Spatial displacement between target points; , , The first The horizontal, vertical, and longitudinal coordinates of each target point; , , The first The horizontal, vertical, and longitudinal coordinates of each target point; The index of the adjacent point in the target point sequence.

[0023] S1.2.3 Calculate the difference between the velocity value of the next target point in the target point sequence and the velocity value of the previous target point to obtain the velocity change sequence; For example, the calculation formula is as follows:

[0024] in, For the target point sequence, the first The target point and the first The change in velocity between each target point; For the first The velocity value of each target point; For the first The velocity value of each target point; The index of the adjacent point in the target point sequence.

[0025] S1.2.4 Extract the upper quartile and interquartile range of the time interval sequence, spatial displacement sequence and velocity change sequence respectively, and use the sum of the upper quartile and interquartile range as the time continuity boundary, displacement continuity boundary and velocity continuity boundary respectively.

[0026] In a preferred embodiment, the time interval sequence, spatial displacement sequence, and velocity change sequence are arranged in ascending order to determine the upper quartile and lower quartile of each sequence. The lower quartile is then subtracted from the upper quartile to obtain the interquartile range. The sum of the upper quartile and the interquartile range is then used as the corresponding continuous boundary.

[0027] For monitoring low, slow, and small targets in airport airspace, when the sampling period is 0.5s, the time interval of normal continuous points is mostly concentrated between 0.5s and 1.0s, the spatial displacement is mostly concentrated between 4m and 25m, and the velocity change is mostly concentrated between 0.2m / s and 3.5m / s.

[0028] For example, if the upper quartile of the time interval sequence is 0.5s and the interquartile range is 0.5s, then the time continuity boundary is 1.0s; if the upper quartile of the spatial displacement sequence is 18m and the interquartile range is 9m, then the displacement continuity boundary is 27m; if the upper quartile of the velocity change sequence is 2.1m / s and the interquartile range is 1.4m / s, then the velocity continuity boundary is 3.5m / s.

[0029] Preferably, the continuous boundaries of time, displacement, and velocity can adaptively delineate discontinuous positions in the target point sequence, avoiding the mis-segmentation of slow-moving objects or the mis-connection of high-speed UAV maneuver segments due to relying solely on fixed distance thresholds.

[0030] S1.3. The positions between adjacent target points in the target point sequence where the time interval exceeds the time continuity boundary, or the spatial displacement exceeds the displacement continuity boundary and the velocity change exceeds the velocity continuity boundary, are determined as point break locations.

[0031] S1.4. Using the break point of the dot pattern as the boundary, the target dot pattern sequence is divided into continuous dot pattern segments, and the end dot pattern and the beginning dot pattern located on both sides of the break point are determined as the break point.

[0032] Specifically, the target point sequence is split into continuous point segments, including: The fracture locations are sorted according to the order of the target point sequence to obtain the fracture location sequence; The target point traces located before the first break point in the target point trace sequence are designated as the first candidate continuous point trace segment; In the fracture location sequence, target points between adjacent fracture locations are divided into intermediate candidate continuous point segments; The target point traces located after the break point of the last point trace in the target point trace sequence are classified as candidate continuous point trace segments for the last segment. The first, middle, and last candidate continuous point segments, which have no break points between adjacent target points, are identified as continuous point segments.

[0033] For example, if the time interval between the 8th and 9th target points is 1.5s, which is greater than the time continuity boundary of 1.0s, then the position between the 8th and 9th target points is the point break position, the 8th target point is the end point, and the 9th target point is the beginning point.

[0034] It should be noted that step S1 addresses the issue of track interruptions that are easily generated by low-altitude, slow-moving, and small targets in the airport airspace under conditions of low-altitude obstruction, radar blind spots, clutter suppression, and target intersection. It uses the joint boundary of time interval, spatial displacement, and velocity change to determine the location of track breakage, reducing continuous track misinterpretation and misconnection of heterogeneous tracks caused by single distance judgment, thereby improving the basic quality of track identification for subsequent bird, UAV, and airborne object classification.

[0035] S2. Based on the continuity direction, velocity continuity, and height change of the continuous point segments on both sides of the fracture endpoint, perform forward outward extrapolation and reverse back extrapolation to obtain the predicted endpoint points and form candidate reconnection relationships. It should be noted that in this step: S2.1. Using each fracture endpoint as a reference, determine the continuous dot segment at the end of the fracture endpoint and the continuous dot segment at the beginning of the fracture endpoint.

[0036] Specifically, if a break in a point is located between the first and second consecutive point segments, then the first consecutive point segment is the end of the consecutive point segment, and the second consecutive point segment is the beginning of the consecutive point segment. If multiple consecutive point segments are arranged sequentially, then each break position in the break position sequence is used to establish an endpoint combination between the end and beginning of the consecutive point segments. The end of the consecutive point segment contains at least 3 consecutive target points, and the beginning of the consecutive point segment contains at least 3 consecutive target points. When any consecutive point segment contains fewer than 3 consecutive target points, that consecutive point segment does not participate in forward extrapolation or reverse extrapolation.

[0037] For example, if a suspected drone target forms a continuous dot segment within the airport's airspace from 12:00:00.000 to 12:00:04.000, and forms another continuous dot segment from 12:00:05.500 to 12:00:09.000, then the first continuous dot segment is the end continuous dot segment, and the second continuous dot segment is the beginning continuous dot segment.

[0038] S2.2. Based on the continuous target dots near the break point in the continuous dot segment at the end, determine the forward dot direction, forward velocity continuity, and forward height change.

[0039] In this embodiment, "near the break point" refers to the four consecutive target points closest to the end point in the continuous point segment at the end; when the continuous point segment at the end contains only three consecutive target points, these three consecutive target points are taken.

[0040] In a preferred embodiment, the forward continuation direction of the point trace is determined according to the direction of the line connecting the spatial positions of the continuous target point traces, specifically including lateral positive continuation, lateral negative continuation, longitudinal approach continuation, longitudinal departure continuation, vertical upward continuation, and vertical downward continuation; when the difference between the forward and backward directions is less than 15°, it is classified as same-direction continuation; when the difference between the forward and backward directions is between 15° and 45°, it is classified as gradual change continuation; and when the difference between the forward and backward directions is greater than 45°, it is classified as turning continuation.

[0041] In a preferred embodiment, the forward velocity continuity state is determined based on the velocity value changes of the continuous target points, specifically including a velocity increase state, a velocity decrease state, and a velocity stability state; when the average difference between adjacent velocity values ​​is greater than 0.8 m / s, it is a velocity increase state; when the average difference between adjacent velocity values ​​is less than -0.8 m / s, it is a velocity decrease state; and when the average difference between adjacent velocity values ​​is between -0.8 m / s and 0.8 m / s, it is a velocity stability state.

[0042] In a preferred embodiment, the forward altitude change state is determined based on the altitude changes of the continuous target points, specifically including an altitude increase state, an altitude decrease state, and an altitude stability state; when the average value of the adjacent altitude difference is greater than 1.5m, it is an altitude increase state; when the average value of the adjacent altitude difference is less than -1.5m, it is an altitude decrease state; and when the average value of the adjacent altitude difference is between -1.5m and 1.5m, it is an altitude stability state.

[0043] S2.3. Based on the continuous target dots near the break point in the continuous dot segment at the beginning, determine the direction of the reverse dot continuation, the state of the reverse velocity continuation, and the state of the reverse height change.

[0044] In this embodiment, "close to the break point" refers to the four consecutive target points closest to the starting point in the continuous point segment at the beginning; when the continuous point segment at the beginning contains only three consecutive target points, those three consecutive target points are taken.

[0045] In a preferred embodiment, the direction of the reverse point continuation is determined by the direction of the line connecting the spatial positions of the continuous target points arranged in reverse according to the sampling time, specifically including lateral positive pushback, lateral negative pushback, longitudinal approach pushback, longitudinal departure pushback, vertical upward pushback, and vertical downward pushback.

[0046] In a preferred embodiment, the reverse velocity continuation state is determined by the velocity value change after the continuous target points are arranged in reverse according to the sampling time, specifically including the velocity increase back push state, the velocity decrease back push state, and the velocity steady back push state.

[0047] In a preferred embodiment, the reverse height change state is determined by the height change of the continuous target points arranged in reverse order of the sampling time, specifically including the height rise back state, the height fall back state, and the height stable back state.

[0048] For example, the height values ​​of the first four consecutive target points in the first continuous point segment are 103m, 104m, 105m, and 106m respectively. After being arranged in reverse, the height values ​​are 106m, 105m, 104m, and 103m respectively. The reverse height change state is the height decrease and pushback state.

[0049] S2.4. Based on the missing sampling time between the fracture endpoints, generate the forward endpoint prediction point based on the forward point continuation direction, forward velocity continuation state, and forward height change state, and generate the reverse endpoint prediction point based on the reverse point continuation direction, reverse velocity continuation state, and reverse height change state. In this embodiment, the missing sampling times between the terminal fracture endpoint and the head fracture endpoint are first determined. If the sampling time of the terminal fracture endpoint is 12:00:04.000 and the sampling time of the head fracture endpoint is 12:00:05.500, with a sampling period of 0.5s, then the missing sampling times are 12:00:04.500 and 12:00:05.000. When generating the forward endpoint prediction point, the spatial position, height, and velocity values ​​of the terminal fracture endpoint are used as the starting point. Based on the unit direction corresponding to the forward point continuation direction, the velocity change trend corresponding to the forward velocity continuation state, and the height change trend corresponding to the forward height change state, the missing sampling times are gradually extrapolated. When generating the reverse endpoint prediction point, the spatial position, height, and velocity values ​​of the head fracture endpoint are used as the starting point. Based on the unit direction corresponding to the reverse point continuation direction, the velocity change trend corresponding to the reverse velocity continuation state, and the height change trend corresponding to the reverse height change state, the missing sampling times are gradually extrapolated.

[0050] As an example, the spatial location of the terminal fracture endpoint is 400m horizontally, 2400m vertically, and 120m vertically, with a velocity of 16m / s. The forward trajectory continues in the longitudinal approach direction, the forward velocity continues in a stable velocity state, and the forward height changes in a stable height state. Therefore, the predicted forward endpoint trajectory obtained after 0.5s can be located near 402m horizontally, 2408m vertically, and 120m vertically. If the predicted reverse endpoint trajectory obtained by reverse tracing back from the terminal fracture endpoint at the same missing sampling time is located near 403m horizontally, 2407m vertically, and 121m vertically, then the two have a basis for reconnection and verification.

[0051] S2.5 Calculate the spatial deviation, velocity deviation, and height deviation of adjacent target points within the continuous point segments at the end and the continuous point segments at the beginning, respectively, to obtain the spatial deviation sequence, velocity deviation sequence, and height deviation sequence, and determine the spatial deviation boundary, velocity deviation boundary, and height deviation boundary based on the spatial deviation sequence, velocity deviation sequence, and height deviation sequence.

[0052] Specifically, spatial deviation is the absolute difference between the actual spatial displacement of adjacent target points and the median of adjacent spatial displacements within the continuous point segment; velocity deviation is the absolute difference between the velocity change of adjacent target points and the median of adjacent velocity changes within the continuous point segment; height deviation is the absolute difference between the height change of adjacent target points and the median of adjacent height changes within the continuous point segment. The calculation formulas are as follows:

[0053]

[0054]

[0055] in, This refers to the spatial deviation. For the first continuous point segment The target point and the first Spatial displacement between target points; It is the set of spatial displacements of all adjacent target points within a continuous point segment; The median of the set of spatial displacements; This refers to the speed deviation. For the first continuous point segment The target point and the first The change in velocity between each target point; It is the set of velocity changes of all adjacent target points within a continuous point segment; The median of the set of velocity changes; This represents the degree of deviation. For the first continuous point segment The target point and the first The change in height between individual target points; It is the set of height changes of all adjacent target points within a continuous point segment; The median of the set of height variations; It represents the sequence number of adjacent target points within a continuous point segment.

[0056] Furthermore, a spatial deviation sequence is formed from each spatial deviation, a velocity deviation sequence is formed from each velocity deviation, and a height deviation sequence is formed from each height deviation. The upper quartile and interquartile range of each of the three types of deviation sequences are summed to obtain the spatial deviation boundary, velocity deviation boundary, and height deviation boundary.

[0057] Preferably, the spatial deviation boundary is 8m, the velocity deviation boundary is 2.5m / s, and the height deviation boundary is 4m. If bird targets within the airport's airspace are affected by gusts of wind and the height deviation sequence within a continuous point segment is large, then the height deviation boundary increases as the actual fluctuations within the continuous point segment increase, thus avoiding misjudging the actual wing flapping of birds as unrepeatable.

[0058] S2.6 Pair the forward endpoint prediction points and reverse endpoint prediction points at the same missing sampling time. If the spatial deviation, velocity deviation and height deviation of the paired points do not exceed the spatial deviation boundary, velocity deviation boundary and height deviation boundary respectively, then the corresponding end break endpoint and beginning break endpoint are determined as candidate reconnection relationships.

[0059] Specifically, a forward endpoint prediction point and a reverse endpoint prediction point form a paired point. If there are multiple forward endpoint prediction points or multiple reverse endpoint prediction points at the same missing sampling time, the combination with the smallest spatial distance is selected. If the spatial distances are the same, the combination with the smaller velocity difference is selected. If the velocity differences are the same, the combination with the smaller height difference is selected. For each paired point, the spatial deviation, velocity deviation, and height deviation are calculated. When the spatial deviation, velocity deviation, and height deviation of the paired point do not exceed the spatial deviation boundary, the velocity deviation does not exceed the velocity deviation boundary, and the height deviation does not exceed the height deviation boundary at all missing sampling times, the corresponding end break endpoint and the beginning break endpoint are determined as candidate reconnection relationships.

[0060] Among them, the candidate reconnection relationship includes the end continuous point trace number, the beginning continuous point trace number, the end break endpoint, the beginning break endpoint, the set of missing sampling times, the set of predicted points for the forward endpoint, and the set of predicted points for the reverse endpoint.

[0061] It should be noted that step S2 addresses the short-term breakage problem caused by targets being obscured by buildings, radar scanning holes, or bird flocks crossing and obstructing within the airport's airspace. It adopts the motion states on both sides of the breakage to jointly constrain the reconnection possibility, reducing one-way misconnection caused by only end extrapolation, thereby improving the reliability of continuous point sequence recovery.

[0062] S3. Align the predicted endpoints in the candidate reconnection relationships according to the sampling order, calculate the motion residuals and decompose them to obtain the motion residual spectrum. Based on the motion residual spectrum, eliminate abnormal candidate reconnection relationships to obtain the target continuous point sequence. Note that the following should be noted in this step: S3.1 Extract the forward endpoint prediction trace and the reverse endpoint prediction trace at the same missing sampling time in the candidate reconnection relationship to obtain the paired endpoint prediction trace.

[0063] Specifically, a sampling time index is first formed according to the set of missing sampling times in the candidate reconnection relationship. Then, the same sampling time is found in the set of forward endpoint prediction points and the set of reverse endpoint prediction points, and the two are combined into a pair of endpoint prediction points. The pair of endpoint prediction points includes the missing sampling time, forward prediction spatial position, reverse prediction spatial position, forward prediction velocity value, reverse prediction velocity value, forward prediction height value, and reverse prediction height value.

[0064] For example, at the missing sampling time of 12:00:04.500, the predicted points of the forward endpoints are spatial positions of 402m, 2408m, and 120m, with a velocity value of 16.0m / s and a height value of 120m; the predicted points of the reverse endpoints are spatial positions of 403m, 2407m, and 121m, with a velocity value of 16.4m / s and a height value of 121m. The two constitute the paired endpoint predicted points at the missing sampling time.

[0065] S3.2 Arrange the paired endpoint prediction points according to the missing sampling time to obtain the endpoint prediction point alignment sequence.

[0066] Specifically, the paired endpoint prediction points are arranged from earliest to latest according to the missing sampling time to obtain the endpoint prediction point alignment sequence. If there are more than two missing sampling times, the endpoint prediction point alignment sequence retains the forward endpoint prediction points and the reverse endpoint prediction points at each missing sampling time in chronological order. If there is only one missing sampling time, the endpoint prediction point alignment sequence contains one set of paired endpoint prediction points and enters the subsequent residual calculation.

[0067] Preferably, this arrangement ensures a consistent time correspondence between the subsequent motion residual sequence and the missing sampling time.

[0068] S3.3 Calculate the spatial residual, velocity residual and height residual of the paired endpoint prediction traces in the endpoint prediction trace alignment sequence to obtain the motion residual sequence.

[0069] As an example, the calculation formula is as follows:

[0070]

[0071]

[0072] in, For the first Spatial residuals corresponding to each missing sampling time; For the first The lateral coordinates of the predicted points at the forward endpoints at each missing sampling time; For the first The longitudinal coordinates of the predicted point trace at the forward endpoint at each missing sampling time; For the first Vertical coordinates of the predicted point trace of the forward endpoint at each missing sampling time; For the first The lateral coordinates of the predicted points at the reverse endpoints at each missing sampling time. For the first The longitudinal coordinates of the predicted point trace at the reverse endpoint at each missing sampling time; For the first The vertical coordinates of the predicted point trace at the reverse endpoint at each missing sampling time; For the first The velocity residuals corresponding to the missing sampling moments; For the first The velocity value of the predicted point trace at the forward endpoint at each missing sampling time; For the first The velocity value of the predicted point trace at the reverse endpoint at the missing sampling time; For the first The height residual corresponding to each missing sampling time; For the first The height value of the predicted point trace at the forward endpoint at each missing sampling time; For the first The height value of the predicted point trace at the reverse endpoint at the missing sampling time; This represents the sequence number of the missing sampling time.

[0073] The spatial residual, velocity residual, and height residual corresponding to each missing sampling moment are arranged in chronological order to form a motion residual sequence.

[0074] S3.4 Perform frequency band decomposition on the motion residual sequence to obtain the motion residual spectrum containing low-frequency drift components, mid-frequency maneuver components, and high-frequency disturbance components.

[0075] In this embodiment, when performing frequency band decomposition on the motion residual sequence, the spatial residual, velocity residual, and altitude residual are first normalized to the 0 to 1 interval according to the same missing sampling time. Then, the three types of normalized residuals are averaged according to the same missing sampling time to obtain the comprehensive residual sequence. Then, the comprehensive residual sequence is decomposed using a three-level discrete wavelet decomposition. The low-frequency approximation component is selected as the low-frequency drift component, the second-level detail component is selected as the mid-frequency maneuver component, and the first-level detail component is selected as the high-frequency perturbation component to obtain the motion residual spectrum.

[0076] As an example, for airport airspace monitoring data with a sampling period of 0.5s, the low-frequency drift component corresponds to slow wind drift or radar systemic deviation, the medium-frequency maneuver component corresponds to the UAV turning, accelerating or climbing process, and the high-frequency disturbance component corresponds to bird wing flapping, short-term attitude changes or echo flicker.

[0077] S3.5 Calculate the energy proportions of the low-frequency drift component, the mid-frequency maneuver component, and the high-frequency disturbance component in the motion residual spectrum, respectively, to obtain the proportions of the drift component, the maneuver component, and the disturbance component.

[0078] As an example, the calculation formula is as follows:

[0079]

[0080]

[0081] in, The percentage of drift components; The proportion of the motor component; The percentage of the disturbance component; For the first The amplitude of the low-frequency drift component corresponding to each missing sampling moment; For the first The amplitude of the intermediate frequency maneuver component corresponding to each missing sampling moment; For the first The amplitude of the high-frequency disturbance component corresponding to each missing sampling moment; This represents the number of missing sampling moments.

[0082] S3.6. Calculate the adjacent amplitude differences of the low-frequency drift component, the mid-frequency maneuver component, and the high-frequency disturbance component according to the order of the missing sampling times, and obtain the continuous state of the drift component, the continuous state of the maneuver component, and the continuous state of the disturbance component.

[0083] As an example, the calculation formula is as follows:

[0084]

[0085]

[0086] in, For the low-frequency drift component in the first The missing sampling time and the first The difference in amplitude between adjacent missing sampling times; For the mid-frequency maneuvering component in the th The missing sampling time and the first The difference in amplitude between adjacent missing sampling times; For the high-frequency disturbance component in the first... The missing sampling time and the first The difference in amplitude between adjacent missing sampling times; For the first The amplitude of the low-frequency drift component corresponding to each missing sampling moment; For the first The amplitude of the intermediate frequency maneuver component corresponding to each missing sampling moment; For the first The amplitude of the high-frequency disturbance component corresponding to the missing sampling time.

[0087] Furthermore, based on the sign and magnitude of adjacent amplitude differences, the continuous states of the drift component, the continuous state of the maneuver component, and the continuous state of the disturbance component include continuous increase, continuous decrease, stable continuation, and abrupt change, respectively. Among them, continuous increase means that two or more consecutive adjacent amplitude differences are all greater than 0, and the subsequent adjacent amplitude difference is not less than 70% of the previous adjacent amplitude difference; continuous decrease means that two or more consecutive adjacent amplitude differences are all less than 0, and the absolute value of the subsequent adjacent amplitude difference is not less than 70% of the absolute value of the previous adjacent amplitude difference; stable continuation means that the absolute value of all adjacent amplitude differences is less than 0.08; abrupt change means that the absolute value of any adjacent amplitude difference is greater than 0.25, and the absolute value of this adjacent amplitude difference is greater than twice the average of the absolute values ​​of the preceding and following adjacent amplitude differences.

[0088] S3.7 If the proportion of the maneuver component is higher than that of the drift component and the maneuver component continuously increases, or the proportion of the disturbance component is higher than that of the maneuver component and the disturbance component continuously changes abruptly, then the corresponding candidate reconnection relationship will be determined as an abnormal candidate reconnection relationship.

[0089] In this embodiment, when the proportion of the maneuver component is higher than that of the drift component, and the maneuver component continuously increases, it indicates that the predicted points on both sides of the break exhibit a continuously amplified mid-frequency motion difference within the reconnection interval. This difference usually does not conform to the natural motion continuation of a target after a short-term blockage. Therefore, this candidate reconnection relationship is determined as an abnormal candidate reconnection relationship. When the proportion of the disturbance component is higher than that of the maneuver component, and the disturbance component undergoes a sudden change, it indicates that the predicted points on both sides of the break exhibit a high-frequency residual jump within the reconnection interval. This jump usually corresponds to radar scintillation, bird flock splitting, or misconnection of different targets. This candidate reconnection relationship is also determined as an abnormal candidate reconnection relationship.

[0090] For example, if the proportion of the maneuver component in a candidate reconnection is 0.58 and the proportion of the drift component is 0.21, and the amplitude of the mid-frequency maneuver component is 0.18, 0.31, 0.49 and 0.70 in sequence according to the missing sampling time, and the difference between adjacent amplitudes is positive and there is no obvious drop, then the candidate reconnection is an abnormal candidate reconnection.

[0091] S3.8. Filter out abnormal candidate reconnection relationships, determine the remaining candidate reconnection relationships as target reconnection relationships, and connect the corresponding continuous point trace segments according to the target reconnection relationships to obtain the target continuous point trace sequence.

[0092] In an optional implementation, if a continuous point segment simultaneously conflicts with two or more target reconnection relationships, the target reconnection relationship with the smallest sum of the average spatial residual, the average velocity residual, and the average height residual is selected; the discarded target reconnection relationship does not participate in the splicing.

[0093] S3.9 Mark the motion residual spectrum corresponding to the target reconnection relationship as reliable, and mark the motion residual spectrum corresponding to the abnormal candidate reconnection relationship as pending verification, to obtain the reliable reconnection status.

[0094] It should be noted that this embodiment solves the problem of false reconnection caused by multiple target intersections, bird flock splits, UAV sharp turns, and radar flashing in airport airspace by performing residual spectrum decomposition on the endpoint prediction points in the candidate reconnection relationship and filtering out abnormal candidate reconnection relationships based on the proportion of drift, maneuver, and disturbance components and the continuous state. This makes the target continuous point sequence no longer determined solely by geometric proximity, but by introducing the frequency band characteristics of the motion residual spectrum for verification, thereby improving the consistency between the broken reconnection result and the actual target motion process.

[0095] S4. Separate the drift component, maneuver component, and disturbance component from the motion residual spectrum. Perform category matching on the three types of components according to the break and reconnection positions and reconnection confidence states of the target continuous point sequence, and output the identification results of birds, drones, or airborne objects. Note that the following should be noted in this step: S4.1. Extract the trace segments located before and after the reconnection in the target continuous trace sequence according to the break and reconnection position to obtain the reconnection check trace segment.

[0096] Specifically, taking the break-and-reconnection location as the center, six target points are captured forward and six target points are captured backward. If the number of target points before or after reconnection is less than six, all target points on that side are captured. If the total number of target points on both sides is less than six, the identification status corresponding to the break-and-reconnection location enters the pending verification state.

[0097] Preferably, the extracted reconnection verification point segments include the point segment before reconnection, the endpoint prediction point segment at the missing sampling time, and the point segment after reconnection.

[0098] S4.2 Separate the low-frequency drift component, mid-frequency maneuver component, and high-frequency disturbance component from the motion residual spectrum, and calculate the component proportion and amplitude continuity of the three types of components in the reconnection check point trace segment respectively.

[0099] Specifically, step S4.2 uses the same frequency band decomposition results as step S3.4, and extracts the amplitude of each component according to the sampling time range corresponding to the reconnection check point trace segment; the component proportion and amplitude continuity of the three types of components in the reconnection check point trace segment are calculated respectively, and the calculation formula is as follows:

[0100] in, For the first The proportion of class components within the trace segment of the reconnection verification point; The component type is L, M, or H; L represents low-frequency drift component; M represents mid-frequency maneuver component; H represents high-frequency disturbance component. To reconnect the verification point within the segment The sampling time corresponding to the first sampling time Class component amplitude; To reconnect the verification point within the segment The amplitude of the low-frequency drift component corresponding to each sampling time; To reconnect the verification point within the segment The amplitude of the intermediate frequency maneuver component corresponding to each sampling time; To reconnect the verification point within the segment The amplitude of the high-frequency disturbance component corresponding to each sampling time; The number of sampling times within the reconnection and verification point trace segment; This is the sampling time sequence number within the reconnection and verification point trace segment.

[0101] Furthermore, the amplitude continuity state is determined based on the direction of component amplitude change before and after reconnection, including no direction reversal, continuation in the same direction, alternating fluctuations, and amplitude jumps. Among them, no direction reversal means that the direction of component amplitude change before and after reconnection is either increasing, decreasing, or stable. Continuation in the same direction means that the proportion of the signs of the last three adjacent amplitude differences before reconnection and the first three adjacent amplitude differences after reconnection is consistent is not less than 2 / 3. Alternating fluctuations means that the signs of adjacent amplitude differences within the reconnection check point trace segment alternate no less than 4 times, and the difference between the largest and smallest amplitudes is greater than 0.18. Amplitude jumps mean that the absolute value of the amplitude difference between adjacent components before and after reconnection is greater than 0.25.

[0102] S4.3 When the low-frequency drift component has the highest proportion and the amplitude of the low-frequency drift component does not reverse direction before and after the break and reconnection position, the airborne object matching result is generated.

[0103] The airborne object matching results include the target category as airborne object, the dominant component type as low-frequency drift component, the proportion of low-frequency drift component, the amplitude continuity status, the break and reconnection location, and the reconnection credibility status.

[0104] For example, if a target moves slowly at a height of 80m to 95m outside the western boundary of an airport, with a speed between 3m / s and 6m / s, and the low-frequency drift component accounts for 0.69, the mid-frequency maneuver component accounts for 0.18, and the high-frequency disturbance component accounts for 0.13, and the low-frequency drift component continues to increase slowly before and after the reconnection, then the airborne object matching result is generated.

[0105] S4.4 When the intermediate frequency maneuvering component has the highest proportion and the amplitude of the intermediate frequency maneuvering component continues in the same direction before and after the break-and-reconnection position, the UAV matching result is generated.

[0106] The drone matching results include the target category being drone, the dominant component type being mid-frequency maneuvering component, the proportion of mid-frequency maneuvering component, the same-direction continuity state, the break-and-reconnection location, and the reconnection reliability state.

[0107] For example, if a target moves along the longitudinal approach direction at an altitude of about 120m in the outer region of the runway end, and its speed increases from 12m / s to 18m / s, with the mid-frequency maneuver component accounting for 0.62, the low-frequency drift component accounting for 0.21, and the high-frequency disturbance component accounting for 0.17, and the mid-frequency maneuver component shows an increasing trend before and after the reconnection, then the UAV matching result is generated.

[0108] S4.5 When the proportion of high-frequency disturbance components is the highest, and the amplitude of high-frequency disturbance components fluctuates alternately before and after the break and reconnection position, bird matching results are generated.

[0109] Among them, the bird matching results include the target category as bird, the dominant component type as high-frequency perturbation component, the proportion of high-frequency perturbation component, the alternating fluctuation state, the break and reconnection location, and the reconnection credibility state.

[0110] For example, if a target makes short-term fluctuations within a height range of 60m to 85m to the side of the airport approach protection surface, with a speed value between 8m / s and 14m / s, and the proportion of high-frequency disturbance component is 0.57, the proportion of low-frequency drift component is 0.19, the proportion of mid-frequency maneuver component is 0.24, and the high-frequency disturbance component shows alternating positive and negative fluctuations, then a bird matching result is generated.

[0111] S4.6 Select the matching result with the highest component ratio from the matching results of airborne objects, drones, and birds as the final matching result; when the reconnection confidence state corresponding to the final matching result is confidence, output the bird, drone, or airborne object identification result according to the final matching result; when the reconnection confidence state corresponding to the final matching result is pending verification, output the pending verification mark.

[0112] Specifically, the bird, drone, or airborne object identification results include target category, identification status, breakage and reconnection location, reconnection reliability status, and residual spectrum discrimination information; among which: The target category is one of the following: bird, drone, or airborne object; The identification status is either a trusted output status or a pending verification status; The break-and-reconnection position is the position between adjacent continuous point segments connected by the target reconnection relationship when forming a continuous point sequence of the target; The reconnection confidence state is the confidence state formed when the motion residual spectrum corresponding to the target reconnection relationship meets the reconnection screening rules, or the state to be verified formed when the motion residual spectrum corresponding to the abnormal candidate reconnection relationship does not meet the reconnection screening rules. The residual spectrum discrimination information includes the component type, component proportion, and amplitude continuity of the low-frequency drift component, mid-frequency maneuver component, and high-frequency disturbance component that correspond to the target category.

[0113] It should be noted that the reconnection screening rules in this embodiment specifically include: the candidate reconnection relationship is not determined as an abnormal candidate reconnection relationship by step S3.7; when the proportion of the maneuver component is higher than the proportion of the drift component, the continuous state of the maneuver component does not show a continuous increase; when the proportion of the disturbance component is higher than the proportion of the maneuver component, the continuous state of the disturbance component does not show an abrupt change; the target reconnection relationship is not discarded in the connection conflict; the total number of target points on both sides of the reconnection verification point trace segment is not less than 6; when the above rules are met, the reconnection trust status is trustworthy; when any rule is not met, the reconnection trust status is pending verification.

[0114] In a specific example, a low-speed, small target was detected 3.2 km outside the runway end of the airport's clear zone. The continuous target track sequence had a break and reconnection point between 12:14:18.000 and 12:14:26.500. A total of 12 target tracks were captured before and after the reconnection. Residual spectrum analysis of the reconnected check track segment revealed that the low-frequency drift component accounted for 0.16%, the mid-frequency maneuvering component for 0.67%, and the high-frequency disturbance component for 0.17%. The amplitude differences of the last three adjacent mid-frequency maneuvering components before the reconnection were 0.05, 0.06, and 0.07, while the differences of the first three adjacent mid-frequency maneuvering components after the reconnection were 0.06, 0.08, and 0.07. (Symbol 1) The ratio is 1; the reconnection confidence status corresponding to the break-reconnection position is confidence; the UAV matching result is selected as the final matching result, and the target category is UAV, the identification status is confidence output status, the break-reconnection position is between 12:14:22.000 and 12:14:23.500, the reconnection confidence status is confidence, the residual spectrum discrimination information is mid-frequency maneuvering component, the component proportion is 0.67, and the amplitude continuity status is continuous in the same direction; if the reconnection confidence status of the same target is pending verification, the target category is UAV, the identification status is pending verification status, and the break-reconnection position and residual spectrum discrimination information are output together for airport airspace personnel to verify.

[0115] Preferably, step S4 addresses the problem that low-speed, small targets in airport airspace are difficult to distinguish reliably based solely on speed, altitude, or radar scattering intensity. It introduces the residual spectral component structure and amplitude continuity state before and after the break and reconnection location, so that the slow drift of airborne objects, the mid-frequency maneuver of UAVs, and the high-frequency disturbance of birds have calculable and verifiable discrimination criteria, thereby improving the reliability of continuous monitoring and classification results of low-speed, small targets in airport airspace.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification, characterized in that, include: Based on the time interval, spatial displacement, and velocity variation relationship between adjacent target points, the target points are divided into continuous point segments, and the end points and beginning points of adjacent continuous point segments are determined as break points. Based on the continuity direction, velocity continuity and height change of the continuous point segments on both sides of the fracture endpoint, forward outward and reverse backward extrapolation are performed to obtain the endpoint prediction point traces and form candidate reconnection relationships. Align the endpoint prediction points in the candidate reconnection relationship according to the sampling order, calculate the motion residual and decompose it to obtain the motion residual spectrum, and filter out abnormal candidate reconnection relationships according to the motion residual spectrum to obtain the target continuous point sequence. The drift component, maneuver component, and disturbance component are separated from the motion residual spectrum. The three types of components are matched according to the break and reconnection positions and reconnection confidence states of the target continuous point sequence, and the bird, drone, or airborne object identification result is output.

2. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 1, characterized in that, The determination of the fracture endpoint includes: Arrange the target points according to the sampling time to obtain the target point sequence; Calculate the time interval, spatial displacement, and velocity change of adjacent target points in the target point sequence one by one, and determine the time continuity boundary, displacement continuity boundary, and velocity continuity boundary respectively; The location between adjacent target points in the target point sequence where the time interval exceeds the time continuity boundary, or the spatial displacement exceeds the displacement continuity boundary and the velocity change exceeds the velocity continuity boundary, is determined as the point break location. Using the break point of the dot pattern as the boundary, the target dot pattern sequence is divided into continuous dot pattern segments, and the end dot pattern and the beginning dot pattern located on both sides of the break point are determined as the break point.

3. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 2, characterized in that, The determination of the time continuity boundary, displacement continuity boundary, and velocity continuity boundary includes: Subtract the sampling time of the previous target point from the sampling time of the next target point in the target point sequence to obtain the time interval sequence; The spatial displacement sequence is obtained by calculating the distance between the spatial position of the next target point in the target point sequence and the spatial position of the previous target point. The velocity change sequence is obtained by calculating the difference between the velocity value of the next target point in the target point sequence and the velocity value of the previous target point. The upper quartile and interquartile range of the time interval sequence, the spatial displacement sequence, and the velocity change sequence are extracted respectively, and the sum of the upper quartile and the interquartile range is used as the time continuity boundary, the displacement continuity boundary, and the velocity continuity boundary, respectively.

4. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 2, characterized in that, The target point sequence is split into continuous point segments, including: The break positions of the dots are sorted according to the order of the target dot sequence to obtain the break position sequence; The target point traces located before the first break point in the target point trace sequence are designated as the first candidate continuous point trace segment; The target point traces between adjacent point trace breakage locations in the fracture location sequence are divided into intermediate candidate continuous point trace segments; The target point traces located after the last point trace break point in the target point trace sequence are classified as candidate continuous point trace segments for the last segment; The first, middle, and last candidate continuous point segments, which do not have the break points between adjacent target points, are identified as continuous point segments.

5. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 1, characterized in that, The process of obtaining the endpoint prediction point trace includes: Using each fracture endpoint as a reference, determine the continuous dot segment at the end of the fracture endpoint and the continuous dot segment at the beginning of the fracture endpoint. Based on the continuous target dots near the break point in the continuous dot segment at the end, determine the forward dot continuation direction, forward velocity continuation state, and forward height change state. Based on the continuous target dots near the break point in the continuous dot segment at the beginning, determine the direction of continuation of the reverse dots, the state of continuation of the reverse velocity, and the state of change of the reverse height. Based on the missing sampling time between the fracture endpoints, a forward endpoint prediction point is generated according to the forward point continuation direction, the forward velocity continuation state, and the forward height change state; and a reverse endpoint prediction point is generated according to the reverse point continuation direction, the reverse velocity continuation state, and the reverse height change state. The endpoint prediction traces include the forward endpoint prediction traces and the reverse endpoint prediction traces.

6. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 5, characterized in that, The formation of candidate reconnection relationships includes: The spatial deviation, velocity deviation, and height deviation of adjacent target points within the continuous point trace segments at the end and the continuous point trace segments at the beginning are calculated one by one to obtain spatial deviation sequences, velocity deviation sequences, and height deviation sequences, respectively. The spatial deviation boundaries, velocity deviation boundaries, and height deviation boundaries are determined based on the spatial deviation sequences, velocity deviation sequences, and height deviation sequences, respectively. Pair the forward endpoint prediction traces and the reverse endpoint prediction traces at the same missing sampling time. If the spatial deviation, velocity deviation, and height deviation of the paired traces do not exceed the spatial deviation boundary, the velocity deviation boundary, and the height deviation boundary, respectively, then the corresponding end break endpoint and the beginning break endpoint are determined as candidate reconnection relationships.

7. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 1, characterized in that, The obtained motion residual spectrum includes: Extract the forward endpoint prediction trace and the reverse endpoint prediction trace at the same missing sampling time in the candidate reconnection relationship to obtain the paired endpoint prediction trace; Arrange the paired endpoint prediction points according to the missing sampling time to obtain the endpoint prediction point alignment sequence; The spatial residual, velocity residual, and height residual of the paired endpoint prediction traces in the endpoint prediction trace alignment sequence are calculated respectively to obtain the motion residual sequence; The motion residual sequence is decomposed into frequency bands to obtain a motion residual spectrum containing low-frequency drift components, mid-frequency maneuver components, and high-frequency disturbance components.

8. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 7, characterized in that, The process of obtaining the target continuous point sequence includes: The energy proportions of the low-frequency drift component, the mid-frequency maneuver component, and the high-frequency disturbance component in the motion residual spectrum are calculated respectively to obtain the proportions of the drift component, the maneuver component, and the disturbance component. According to the order of the missing sampling times, the adjacent amplitude difference of the low-frequency drift component, the mid-frequency maneuver component and the high-frequency disturbance component are calculated respectively to obtain the continuous state of the drift component, the continuous state of the maneuver component and the continuous state of the disturbance component. If the proportion of the maneuver component is higher than the proportion of the drift component and the maneuver component continuously increases, or the proportion of the disturbance component is higher than the proportion of the maneuver component and the disturbance component continuously changes abruptly, then the corresponding candidate reconnection relationship is determined as an abnormal candidate reconnection relationship. The abnormal candidate reconnection relationships are screened out, the remaining candidate reconnection relationships are determined as target reconnection relationships, and the corresponding continuous point trace segments are connected according to the target reconnection relationships to obtain the target continuous point trace sequence. The motion residual spectrum corresponding to the target reconnection relationship is marked as reliable, and the motion residual spectrum corresponding to the abnormal candidate reconnection relationship is marked as pending verification, thus obtaining the reliable reconnection status.

9. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 1, characterized in that, The output bird, drone, or airborne object identification results include: According to the break and reconnection position, extract the trace segments in the target continuous trace sequence located before and after the reconnection to obtain the reconnection check trace segment; The low-frequency drift component, the mid-frequency maneuver component, and the high-frequency disturbance component are separated from the motion residual spectrum, and the component proportion and amplitude continuity of the three types of components in the reconnection verification point trace segment are calculated respectively. When the low-frequency drift component has the highest proportion, and the amplitude of the low-frequency drift component does not reverse direction in the continuous state before and after the break-and-reconnection position, an airborne object matching result is generated. When the proportion of the intermediate frequency maneuvering component is the highest, and the amplitude of the intermediate frequency maneuvering component continues in the same direction before and after the break-and-reconnection position, the UAV matching result is generated. When the proportion of the high-frequency disturbance component is the highest, and the amplitude of the high-frequency disturbance component fluctuates alternately before and after the break-and-reconnection position, a bird matching result is generated. The matching result with the highest component ratio among the airborne object matching result, the drone matching result, and the bird matching result is selected as the final matching result; when the reconnection confidence status corresponding to the final matching result is confidence, the bird, drone, or airborne object identification result is output according to the final matching result; when the reconnection confidence status corresponding to the final matching result is pending verification, a pending verification mark is output.

10. The artificial intelligence recognition method for birds, drones, and airborne objects based on continuous point verification according to claim 9, characterized in that, The bird, drone, or airborne object identification results include target category, identification status, breakage and reconnection location, reconnection reliability status, and residual spectrum discrimination information; wherein: The target category is one of birds, drones, or airborne objects; The identification status is either a trusted output status or a status pending verification. The break-and-reconnection position is the position between adjacent continuous point segments connected by the target reconnection relationship when the target continuous point sequence is formed; The reconnection confidence state is a confidence state formed when the motion residual spectrum corresponding to the target reconnection relationship meets the reconnection screening rules, or a state to be reviewed formed when the motion residual spectrum corresponding to the abnormal candidate reconnection relationship does not meet the reconnection screening rules. The residual spectrum discrimination information includes the component type, component proportion, and amplitude continuity of the low-frequency drift component, mid-frequency maneuver component, and high-frequency disturbance component that correspond to the target category.

Images