A digital signal processing method and apparatus for bird flock detection

Abstract

The application discloses a kind of digital signal processing method and device for bird detection, it is related to bird situation monitoring and low-altitude target digital signal processing technical field, the application is unified modeling to runway end outside boundary, approach-departure channel projection area and bird attractor area, and division attractor association area, connection transition area and channel area under unified coordinate, the application directly writes into processing chain between the space relation of bird flock activity and runway end key channel.Subsequently, with candidate return unit forming bird flock envelope, with adjacent time bird flock envelope constructing continuous relationship diagram, and when multiple previous candidates exist simultaneously, determine main inheritance edge, retain the rest of connection relationship for split relationship or merge relationship determination, so that the crowd in continuous time, split group, compression and rotation is no longer only dependent on single target trajectory continuation expression, but is transformed into the continuous evolution expression of the same bird flock state object.

Description

Technical Field

[0001] This invention relates to the field of bird monitoring and low-altitude target digital signal processing technology, and in particular to a digital signal processing method and apparatus for bird flock detection. Background Technology

[0002] If the outer edge of an airport runway is adjacent to wetlands, aquaculture ponds, or garbage transfer points, low-altitude bird flocks are likely to form during dawn and dusk. Existing airport bird monitoring systems typically employ detection radar, video surveillance, and alarm linkage devices to perform target detection, trajectory tracking, area alarms, and result output on the detection data of the low-altitude area outside the runway edge, in order to support bird monitoring and response.

[0003] Existing processing chains typically rely on the correlation of adjacent time points and trajectory continuation to merge continuous detection results into the target trajectory and combine this with regional partitioning to generate alarm outputs. This approach assumes the front end can form relatively stable continuous objects; however, in low-altitude bird activity outside the runway, flocks often merge, split, compress, turn, and rapidly change boundaries, causing detection results from consecutive time points to be repeatedly split or reassembled. It's difficult to maintain consistent object identification for the same group of birds. Consequently, the system output is prone to issues such as the same bird activity being triggered repeatedly, different bird activities being incorrectly pieced together, and inconsistencies in historical records.

[0004] Meanwhile, the outer edge of the runway also suffers from near-ground clutter, boundary jitter, and short-term missing data, making it more difficult to reliably transmit detection results across consecutive moments. If the output only indicates bird activity in a certain area or risk at a certain moment, it is difficult to continuously express the migration status of the same flock of birds relative to the critical entry and exit channels. Consequently, it is difficult to distinguish in a timely manner whether they are approaching, cutting in, traversing, or occupying the critical channels, increasing the burden of on-site handling and subsequent tracing.

[0005] Therefore, how to maintain the continuous object representation of the same flock of birds in low-altitude flocking scenarios outside the runway end, under conditions of object splitting, merging, and boundary jitter, and how to continuously determine its entry, cutting in, traversing, and occupying status on key channels, has become a technical problem that needs to be solved. Summary of the Invention

[0006] This application provides a digital signal processing method and apparatus for bird flock detection, which solves the problems of unstable bird flock object identification and difficulty in continuously determining the status of key channels in the scenario of flocks merging outside the runway end.

[0007] In a first aspect, embodiments of the present invention provide a digital signal processing method for bird flock detection, comprising:

[0008] Acquire multiple scan echo frames and scene constraint data arranged in chronological order. The scene constraint data includes at least the outer boundary of the runway end, the projection area of ​​the approach and departure passage, and the bird attraction source area.

[0009] Based on the scene constraint data, a scene constraint model under a unified coordinate system is established, and the attraction source association area, connection transition area, and channel area are divided.

[0010] The multi-frame scan echo is time-series registered and background stripped to obtain candidate echo units at each time point.

[0011] Based on spatial adjacency, candidate echo units at each time step are aggregated into flock envelopes. With flock envelopes at adjacent time steps as nodes, candidate node pairs are first screened based on regional inheritance relationships. Inheritance edges are first established based on envelope overlap relationships. For candidate node pairs without established inheritance edges, inheritance edges are then established based on displacement continuity relationships. When the same subsequent node corresponds to multiple preceding nodes, the main inheritance edge is determined and the remaining connection relationships are retained for splitting or merging relationship determination, thereby constructing a continuous relationship graph.

[0012] Generate a bird flock state object that spans multiple frames based on the continuous relationship graph;

[0013] Map the bird flock state object to the scene constraint model to generate a region occupation sequence;

[0014] Based on the region occupation sequence and the continuous relationship graph, entry events, cutting events, traversal events, road occupation events, grouping events, and grouping events are extracted to generate an event sequence;

[0015] Based on the bird flock state object, the region occupation sequence, and the event sequence, a structured detection result is output.

[0016] In some embodiments, establishing the scenario constraint model includes:

[0017] A scene reference coordinate system is established using the outer boundary of the runway end;

[0018] A channel area is generated based on the projection area of ​​the entry and exit channels;

[0019] An attraction source association zone is generated based on the described bird attraction source region;

[0020] A connection transition zone is generated based on the connectivity range between the attraction source association zone and the channel zone;

[0021] Fixed area identifiers are written for the attraction source association area, the connection transition area, and the channel area, respectively.

[0022] In some embodiments, the background stripping includes:

[0023] Perform uniform coordinate registration on scan echo frames from adjacent time points;

[0024] A static background set is established for echo cells with stable positions in multiple consecutive frames;

[0025] When the overlap ratio between the echo unit and the static background set reaches the background threshold, the echo unit is marked as a background echo.

[0026] The candidate echo cells are generated based on the connectivity of echo cells that are not marked as background echoes.

[0027] In some embodiments, when generating the flock state object, an object identifier, a start time, an end time, and a predecessor object identifier are written for each flock state object.

[0028] When a split relationship occurs in the continuous relationship graph, the same preceding object identifier is written to the multiple bird flock state objects after the split.

[0029] When a merging relationship occurs in the continuous relationship graph, multiple predecessor object identifiers are written for the merged flock state object.

[0030] In some embodiments, the extracted events include: recording the first state change of the region occupancy sequence from the attraction source association zone to the connection transition zone as an entry event; recording the first state change of the region occupancy sequence from the connection transition zone to the channel zone as a cutting event; and recording the movement of the bird flock state object from one side of the channel centerline to the other side within the channel zone as a traverse event, based on the channel centerline, where the channel centerline is a reference line generated by the central axis of the channel zone.

[0031] In some embodiments, the event extraction further includes: determining that when a flock of birds enters the channel area based on the continuous occupation status and remains in the channel area for a continuous period of time that reaches a continuous frame count threshold, recording it as an occupation event; and continuously recording it as the same occupation event before the release frame count threshold is reached when the bird enters the channel area continuously in the area occupation sequence under boundary jitter conditions; recording the merging relationship in the continuous relationship graph as a grouping event, and recording the splitting relationship in the continuous relationship graph as a grouping event.

[0032] In some embodiments, the structured detection results further include timestamps, region identifiers, preceding object identifiers, and subsequent object identifiers;

[0033] The region occupation sequence and the event sequence of the same flock state object are associated and written into the result record in chronological order;

[0034] When the event sequence contains both an entry event and a road occupation event, a critical channel handling record is generated. The critical channel handling record is a handling result record associated with the corresponding bird flock state object.

[0035] Secondly, embodiments of the present invention provide a digital signal processing device for bird flock detection, comprising,

[0036] The data acquisition module is used to acquire multi-frame scan echo frames and scene constraint data arranged in chronological order;

[0037] The scene modeling module is used to establish a scene constraint model based on the scene constraint data, and to divide the attraction source association area, the connection transition area, and the channel area.

[0038] The background processing module is used to perform time-series registration and background stripping on the multi-frame scan echo frames to obtain candidate echo units.

[0039] An envelope generation module is used to aggregate the candidate echo units into a flock envelope based on spatial adjacency relationships;

[0040] The continuous relationship module is used to select candidate node pairs based on the regional inheritance relationship, and then establish inheritance edges based on the envelope overlap relationship and displacement continuation relationship. When the same subsequent node corresponds to multiple predecessor nodes, the main inheritance edge is determined and the remaining connection relationship is retained for the determination of split relationship or merge relationship. The split relationship and merge relationship are recorded to construct a continuous relationship graph.

[0041] The state object module is used to generate a bird flock state object that continues across frames based on the inheritance, splitting, and merging relationships in the continuous relationship graph.

[0042] The region mapping module is used to map the bird flock state object to the scene constraint model and generate a region occupation sequence.

[0043] The event extraction module is used to extract entry events, cutting events, traversal events, lane occupation events, grouping events, and grouping events based on the region occupation sequence and the continuous relationship graph.

[0044] The result output module is used to output the structured detection results corresponding to the bird flock state object.

[0045] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the digital signal processing method for bird flock detection as described in the first aspect of the present invention.

[0046] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the digital signal processing method for bird flock detection as described in the first aspect of the present invention.

[0047] Through the above technical solution, the present invention can achieve at least the following beneficial effects:

[0048] This invention unifies the modeling of the outer boundary of the runway end, the projection area of ​​the approach and departure channels, and the bird attraction source area, and divides the attraction source association area, the connecting transition area, and the channel area under a unified coordinate system. This invention directly incorporates the spatial relationship between bird flock activity and key channels at the runway end into the processing chain. Subsequently, a bird flock envelope is formed using candidate echo units, and a continuous relationship graph is constructed using bird flock envelopes from adjacent time points. When multiple preceding candidates exist simultaneously, the primary inheritance edge is determined, and the remaining connection relationships are retained for determining splitting or merging relationships. This transforms merging, splitting, compression, and turning in continuous time points from relying solely on the continuation of a single target trajectory to a continuous evolutionary expression of the same bird flock state object. This enables the same batch of bird activity to be stably merged into the same object chain within continuous time periods, reducing the probability of repeated triggering of the same batch of bird activity and incorrect splicing of different batches of bird activity. This directly addresses the problems of unstable object identification and frequent alarm state switching in low-altitude merging scenarios at dawn and dusk outside the runway end.

[0049] By mapping bird flock state objects to a region occupation sequence and extracting entry, cutting, and traversing events from the region occupation sequence, this invention outputs not the presence or absence of birds within a static region, but rather the migration path and crossing status of bird flock activities relative to a key passage. An entry event corresponds to the flock moving from the attraction source association zone to the connection transition zone; a cutting event corresponds to the flock further entering the passage zone; and a traversing event corresponds to the flock crossing the passage centerline. This allows for the differentiation of the processes of approaching, cutting into, and traversing the key passage. In this way, the system can continuously express the state changes of the same bird flock—approaching, entering, and crossing the key passage—around the same bird flock state object in a runway scenario, reducing processing lag caused by relying solely on single-frame results.

[0050] By determining continuous occupancy states and extracting occupancy events, this invention further separates the continuous stay and short-term crossing of bird flocks within the passage area. The bird flock state object corresponding to the occupancy event remains located within the passage area for a period continuously reaching a consecutive frame count threshold. Even under boundary jitter conditions, the same occupancy event is continuously recorded until the release frame count threshold is reached. Therefore, it can suppress frequent opening and closing of occupancy states caused by slight boundary fluctuations. In this way, the structured detection results can not only indicate whether a bird flock has entered a critical passage, but also whether the bird flock continuously occupies a critical passage, thus providing direct evidence for generating critical passage handling records.

[0051] By writing object identifiers, preceding object identifiers, and subsequent object identifiers to bird flock status objects, and linking the area occupation sequence and event sequence in chronological order and writing them into the result record, this invention forms an object-oriented structured detection result. This result simultaneously preserves object continuity, area migration relationships, and event occurrence order, providing a unified object index and event chain for bird control, release decisions, and post-event tracing. In this way, the entire process of the same bird flock activity, from discovery and migration to road occupation, can be saved as a single object chain, reducing the manual splicing burden on personnel when faced with multiple short-term alarm fragments. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0053] Figure 1 This is a flowchart of the digital signal processing method for bird flock detection in the embodiment;

[0054] Figure 2 This is a frame diagram of the digital signal processing device for bird flock detection in the embodiment.

[0055] Figure 3 This is a flowchart illustrating the construction and ambiguity resolution of the bird flock envelope in the embodiment.

[0056] Figure 4 This is a flowchart illustrating the determination and triggering of road occupancy events based on the area occupancy ratio in this embodiment. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0059] To facilitate understanding, the relevant terms and concepts involved in the embodiments of this application will be introduced below:

[0060] A scan echo frame refers to the set of echo data collected by the detection equipment in a scan cycle of the low-altitude region outside the runway end and expressed in a unified coordinate system. A scan echo frame includes at least a time stamp, spatial location, and echo intensity information. Scene constraint data refers to the basic data set used to define the spatial relationships outside the runway end. Scene constraint data includes at least the runway end outer boundary, the projection area of ​​the approach / departure passage, the bird attraction source area, data source markers, and update time stamps. A scene constraint model is a unified spatial model generated from scene constraint data and used for region mapping. An attraction source association zone is an activity association area established around the bird attraction source area. A passage zone is a key passage area generated from the projection area of ​​the approach / departure passage. A connecting transition zone is a connecting area located between the attraction source association zone and the passage zone, used to characterize the migration process. A candidate echo unit is a dynamic echo set with spatial connectivity retained after background stripping. A flock envelope is the outer spatial range formed by aggregating multiple candidate echo units at the same time. A continuity graph is a temporal correlation structure used to record the inheritance, splitting, and merging relationships between flock envelopes at adjacent times. A flock state object refers to a continuous object spanning multiple frames, generated based on a continuous relationship graph. A region occupation sequence refers to a serialized record that records the correspondence between flock state objects and region identifiers in chronological order. Structured detection results refer to the output records indexed by flock state object identifiers; these records include at least a timestamp, region identifier, event type, preceding object identifier, and subsequent object identifier.

[0061] Furthermore, the comprehensive association result refers to the association judgment result obtained after uniformly adjudicating the envelope overlap result, displacement continuation result, and region inheritance result under the same parameter version identifier. The ambiguity status flag refers to the status flag written when the same subsequent node corresponds to multiple candidate preceding nodes and the differences between the candidate evidence do not meet the stable adjudication requirements. The fallback window refers to the maximum frame span allowed for supplementing inheritance edges in the case of a missing single frame or a single frame crossing boundaries; the default is 1 frame, and the adjustable range is 1~2 frames. The parameter version identifier refers to the record field bound to the inheritance relaxation coefficient, overlap score weight, displacement continuation score weight, region inheritance score weight, overlap priority threshold, comprehensive association threshold, score tolerance, dominant proportion threshold, confirmation frame count, entry hold threshold, hold threshold, occupation duration benchmark, traverse upper limit frame count, and scan update cycle. The conservative parameter group refers to the parameter combination activated when there are insufficient offline tuning samples; it corresponds to a higher overlap priority threshold and entry hold threshold, and a lower inheritance relaxation coefficient. To maintain consistency in the recorded results, the structured detection results can also include a frame index field, an object index field, a parameter version field, an ambiguity status field, and a missing detection status field. The frame index field is used to mark the order of the scan echo frames, the object index field is used to mark the order of bird flock status objects, the parameter version field is used to align the association determination criteria, and the ambiguity status field and the missing detection status field are used to record the ambiguity resolution and missing detection retention status, respectively.

[0062] In some embodiments, scan echo frames are generated by the detection radar according to a continuous scanning cycle, and each scan echo frame corresponds to a unique frame number and a unique time stamp. Scene constraint data is formed by airport electronic map data, runway end outer calibration data, arrival and departure channel projection data, and bird attraction source annotation data, and spatial alignment is completed under a unified coordinate system. The structured detection results are organized with bird flock status object identifiers as the main index. The area occupation sequence, event sequence, start time, end time, preceding object identifier, and subsequent object identifier corresponding to the same bird flock status object are written into the same result record, thereby maintaining a traceable expression of the same bird flock activity within a continuous time period.

[0063] Example 1:

[0064] To facilitate the explanation of the technical content of the embodiments of this application, the following implementation focuses on the scenario outside the airport runway end. In this scenario, the detection equipment continuously scans the low-altitude area outside the runway end and establishes a scenario constraint model in conjunction with scenario constraint data. Subsequent processing sequentially completes background stripping, candidate echo unit generation, bird flock envelope generation, continuous relationship graph construction, bird flock state object generation, region mapping, event extraction, and structured detection result output, thereby expressing the activity of the same bird flock within a continuous time period as a traceable object-level result.

[0065] like Figure 1As shown, this embodiment provides a digital signal processing method for bird flock detection, including:

[0066] Step S1: Obtain multiple scan echo frames and scene constraint data arranged in chronological order. The scene constraint data includes at least the outer boundary of the runway end, the projection area of ​​the approach and departure passage, and the bird attraction source area.

[0067] Step S2: Establish a scene constraint model under a unified coordinate system based on the scene constraint data, and divide the attraction source association area, the connection transition area, and the channel area.

[0068] Step S3: Perform time-series registration and background stripping on multiple scan echo frames to obtain candidate echo units at each time point;

[0069] Step S4: Based on spatial adjacency, candidate echo cells at each time step are aggregated into a flock envelope. Using flock envelopes at adjacent time steps as nodes, candidate node pairs are first screened based on regional inheritance relationships. For candidate node pairs that meet the screening criteria, inheritance edges are first established based on envelope overlap relationships. For candidate node pairs that do not have inheritance edges established, inheritance edges are then established based on displacement continuity relationships. A predecessor node connecting multiple successor nodes is recorded as a splitting relationship, and multiple predecessor nodes connecting one successor node is recorded as a merging relationship, thereby constructing a continuous relationship graph.

[0070] Step S5: Generate a bird flock state object that continues across frames based on the continuous relationship graph;

[0071] Step S6: Map the bird flock state object to the scene constraint model to generate the region occupation sequence;

[0072] Step S7: Extract entry events, cutting events, traversal events, road occupation events, grouping events, and grouping events based on the regional occupation sequence and continuous relationship diagram to generate an event sequence;

[0073] Step S8: Based on the bird flock status object, the region occupation sequence, and the event sequence, output the structured detection results;

[0074] In one embodiment of Example 1, the continuous relationship graph uses frame number and flock envelope identifier as node indices, and preceding node index and subsequent node index as edge indices. Each node corresponds to the flock envelope at a given time, and each edge records at least the relationship type, envelope overlap determination result, displacement continuation determination result, and region inheritance determination result. For two nodes at adjacent time points, it is first determined whether to establish a candidate inheritance edge based on the envelope overlap relationship. For node pairs that do not establish a candidate inheritance edge, it is then determined whether to establish a candidate inheritance edge based on the displacement continuation relationship. When the region inheritance relationship is established and the envelope overlap relationship meets the inheritance conditions, or when the region inheritance relationship is established but the envelope overlap relationship does not meet the inheritance conditions, and the comprehensive association result meets the inheritance conditions, an inheritance edge is established between the two nodes. When a preceding node maintains continuous connection with multiple subsequent nodes, it is recorded as a split relationship; when multiple preceding nodes maintain continuous connection with one subsequent node, it is recorded as a merge relationship. In cases where the same subsequent node corresponds to multiple candidate preceding nodes, the envelope overlap results are compared first, followed by the displacement continuation results, and finally the region inheritance results to determine the primary inheritance relationship. Other connection relationships not retained as primary inheritance relationships are continued to be written into the continuous relationship graph for subsequent determination of splitting or merging relationships. The continuous relationship graph also records the corresponding times for each node, and sequentially records the start and end times of the flock state object according to the times corresponding to the first and last nodes associated with the same flock state object.

[0075] In one embodiment of Example 1, the scene constraint model is generated by performing a unified coordinate transformation on the scene constraint data.

[0076] The outer boundary of the runway end is represented by a boundary line or a boundary polygon. The projection area of ​​the arrival and departure passages is represented by a passage polygon aligned with the runway direction. The bird attraction source area is represented by an attraction source outline polygon. The attraction source association area is generated directly from the bird attraction source area or by expanding outward from the bird attraction source area. The passage area is generated directly from the projection area of ​​the arrival and departure passages. The connecting transition area is generated by the connectivity range between the attraction source association area and the passage area. The scene constraint model stores area identifiers, area boundaries, area types, data source markers, and update time markers to maintain consistency between different batches of data during subsequent area mapping. Bird attraction source areas can originate from wetland area data, aquaculture pond area data, waste transfer point operation area data, manually labeled data, or historical bird activity hotspot labeling data.

[0077] Specifically, when aligning scene constraint data with scan echo frames in time, the time stamp of each scan echo frame is used as the primary time stamp, and the most recent valid scene constraint data no later than that time stamp is called. When scene constraint data is updated between two adjacent frames, the updated region boundary and region identifier are used from the next newly created bird flock state object. Existing bird flock state objects maintain their original region mapping until the object ends, and the corresponding version is written in the parameter version field. The channel axis refers to the direction reference line determined in the scene reference coordinate system based on the long side direction of the projection area of ​​the entry and exit channel. The channel centerline refers to the determination reference line located in the middle of the channel area and consistent with the channel axis. The legal region transfer set refers to the set of region transfers that are allowed to be inherited continuously in adjacent moments, including at least the attraction source association area to the connection transition area, the connection transition area to the attraction source association area, the connection transition area to the channel area, the channel area to the connection transition area, and continuation within the channel area. The direction consistency mark refers to the number of records of whether the displacement direction of the bird flock state object is consistent with the direction of the channel axis. Subsequent region inheritance results, orientation consistency marking, and cross-cutting event determination are all performed based on the same channel axis and the same channel centerline to maintain consistent region mapping apertures between different frames.

[0078] In one embodiment of Example 1, background stripping is performed according to the spatial stability of continuous time intervals.

[0079] Echo cells whose positions remain stable across multiple consecutive frames and whose boundary changes are within a predetermined range are written into a static background set. The static background set includes at least the background cell identifier, spatial boundary, duration, and echo strength statistics. For the echo cell at the current moment, an overlap comparison is first performed with the static background set based on uniform coordinates, followed by dynamic discrimination based on boundary changes. Echo cells that overlap with the static background set and whose boundary changes remain stable are marked as background echoes; echo cells not marked as background echoes are treated as dynamic echoes and participate in subsequent processing. Candidate echo cells include at least the cell identifier, frame number, time stamp, spatial boundary, center position, area information, and echo strength statistics. Echo cells close to the ground boundary but whose positions are unstable across consecutive frames are retained as candidate echo cells to avoid being mistakenly included in the static background set due to the near-ground bird flocks moving close to the edge.

[0080] For example, a scanning range resolution unit (SMR) refers to the smallest discrete unit in the range direction of a scanning echo frame. The position of a stable echo unit across multiple consecutive frames can be determined by combining center position drift and boundary changes within 3 to 10 consecutive frames. The center position drift is assumed to be no more than one SMR by default, with an adjustable range of 1 to 2 SMRs. Boundary changes are measured as a percentage of the change in the envelope area of ​​adjacent frames, with a default value no more than 15% and an adjustable range of 5% to 20%. The background threshold, as an implementation parameter, can be set based on historical sample quantiles or validation set adjustments, with a default value of 0.70 and an adjustable range of 0.60 to 0.85. Its value is neither set to 1 to prevent the static background set from failing to converge stably when the boundary fluctuates slightly, nor is it set to 0 to prevent dynamic echoes from being incorporated into the background echo. Each record in the static background set can also be written with the most recent update time stamp, the number of consecutive stable frames, and the data source stamp. When a single frame of the scan echo frame is missing, the timestamp is out of order, or the scene constraint data has not yet been aligned with the current frame, the previous valid static background set of 1 to 2 frames is maintained, and the corresponding stamp is written in the missing status field. The update will continue after the subsequent valid scan echo frames are restored.

[0081] In one embodiment of Example 1, flock state objects are generated and maintained according to temporal continuity rules. A flock envelope that appears for the first time and has no valid predecessor node generates a new flock state object identifier; flock envelopes with a unique predecessor node and satisfying inheritance relationships inherit the predecessor object identifier; when a splitting relationship exists, multiple new flock state objects are derived from the same predecessor object, and the same predecessor object identifier is written into each new flock state object; when a merging relationship exists, multiple predecessor objects converge to form a new flock state object, and multiple predecessor object identifiers are written into the new flock state object. Each flock state object includes at least an object identifier, a start time, an end time, the current envelope boundary, a set of predecessor object identifiers, and a set of subsequent object identifiers. When there is no valid subsequent node at the current time, the corresponding flock state object is marked as terminated, and the termination time is written.

[0082] In one embodiment of Example 1, the structured detection results are saved using an object-level recording method. Each result record includes at least a flock status object identifier, a timestamp, a current area identifier, an event type, a start time, an end time, a preceding object identifier, a subsequent object identifier, a corresponding frame number, and a recording status. The area occupation sequence saves area entry, area stay, and area departure information in chronological order, while the event sequence saves entry events, cutting events, traversing events, path occupation events, merging events, and splitting events in chronological order. The critical channel handling record is associated with the flock status object identifier that triggered the record and is written with the trigger time, the event sequence before the trigger, and the status change after the trigger, thereby maintaining a consistent correspondence between the handling chain and the flock activity chain.

[0083] Similarly, structured detection results can be stored in result record tables and disposition record tables. The result record table is a collection of object-level records organized by the object index field. Each record includes at least the flock status object identifier, frame index field, timestamp, current area identifier, event type, start time, end time, previous object identifier set, subsequent object identifier set, parameter version field, ambiguous status field, and missing status field. The disposition record table is a collection of disposition records indexed by the object index field and trigger time. Each record includes at least the object identifier, trigger time, pre-trigger event sequence, post-trigger status change, disposition priority, disposition time window, and record status. Both the result record table and the disposition record table are appended chronologically and aligned using the object index field and parameter version field. When a flock status object extends its object chain due to splitting or merging relationships, the association is synchronously written to the previous and subsequent object identifier sets to maintain the continuity of the same flock activity chain during retrieval, playback, and disposition tracking.

[0084] like Figure 3 As shown, in an optional implementation of Example 1, when constructing the continuous relationship graph, the bird flock envelopes at adjacent times are not directly connected pairwise. Instead, edge relationships are established in the order of region inheritance constraint outer layer screening, envelope overlap priority, displacement continuation supplementation, and local conflict adjudication.

[0085] Furthermore, the envelope overlap priority threshold, comprehensive association threshold, inheritance relaxation coefficient, and scoring tolerance are all tuned offline as implementation parameters. The default value for the envelope overlap priority threshold is 0.40, with an adjustable range of 0.35–0.50; the default value for the comprehensive association threshold is 0.60, with an adjustable range of 0.55–0.70; the default value for the inheritance relaxation coefficient is 0.75, with an adjustable range of 0.60–0.85; and the default value for the scoring tolerance is 0.08, with an adjustable range of 0.05–0.10. The sum of the overlap scoring weight, displacement continuation scoring weight, and region inheritance scoring weight is kept at 1, with default values ​​of 0.60, 0.25, and 0.15 respectively, and adjustable ranges of 0.55–0.65, 0.20–0.30, and 0.15–0.25 respectively. None of the weights are set to 0 or 1 to avoid a single criterion being permanently masked or completely dominating the comprehensive association result. If the predicted position of the preceding flock envelope is unavailable, no valid primary inherited edge was formed in the previous time step, or an ambiguous state flag was written in the previous time step, the displacement continuation result is not used in the comprehensive association result. Instead, the remaining valid judgment results are normalized and adjudicated. If there are fewer than two valid judgment results, only candidate inherited edges that meet the envelope overlap priority threshold and the legal region transfer set are retained. If there are multiple candidate preceding nodes for the same subsequent node and the comprehensive association results are close, an ambiguous state flag is written and confirmation is delayed by one frame. If a new valid scan echo frame is obtained within the backtracking window and the local relationship at the new time step still meets the legal region transfer set, the splitting or merging relationship is confirmed based on the relationship at the new time step. Otherwise, the primary inherited edge is retained and the remaining candidate edges are terminated.

[0086] The envelope of each flock in the previous time step is recorded as the envelope of the preceding flock. Each flock envelope at a later time step is recorded as the subsequent flock envelope. , For the envelope index of preceding flocks, This serves as the envelope index for subsequent bird flocks. For any pair of preceding and following envelopes, first calculate the envelope overlap score:

[0087] ,

[0088] in, Enveloping the preceding flock of birds Envelope of subsequent flocks Overlapping scores For the previous moment A flock of birds encircles, For the next moment The overlap score is normalized using a smaller envelope area to suppress bias caused by envelope expansion or contraction; the overlap threshold is preferably calibrated to 0.35 to 0.50, with the lower edge of the interval taken when the deformation is strong and the upper edge of the interval taken when the background is relatively stable, and the value is not lower than 0.30.

[0089] For envelope pairs with insufficient overlap but exhibiting a continuous motion trend, the displacement continuity score is then calculated:

[0090] ,

[0091] in, Enveloping the preceding flock of birds Envelope of subsequent flocks Displacement duration score, Enveloping the preceding flock of birds The position of the center of mass, For subsequent flocks of birds to encircle The position of the center of mass, Enveloping the preceding flock of birds The estimated displacement vector, The frame interval between adjacent time points. The displacement duration limit is calculated by multiplying the upper limit velocity of the bird flock by the frame interval, and then adding the caliber of one or two distance unit positioning errors, and limiting it to within half of the local channel width; it increases synchronously when the frame interval increases, and tightens towards the lower edge of the interval when clutter tails increase.

[0092] When writing runway-end scene information into the region inheritance relationship, the dominant region identifier of each envelope in the scene constraint model is used as the inheritance criterion. The region inheritance score can be written as:

[0093] ,

[0094] in, Enveloping the preceding flock of birds Envelope of subsequent flocks Regional inheritance score, Enveloping the preceding flock of birds The dominant area identifier, For subsequent flocks of birds to encircle The dominant area identifier, This is the inheritance relaxation factor for transfers between adjacent legal regions. Marking for consistent direction, This is a set of legal region transfers. A legal region transfer set must include at least the following: from the attractor-associated area to the connecting transition area, from the connecting transition area to the attractor-associated area, from the connecting transition area to the channel area, from the channel area to the connecting transition area, and continuations within the channel area. It does not include direct jumps from the attractor-associated area to the channel area; a consistent direction marker is used. In channel-related transfers, the direction of the displacement vector's projection onto the channel axis is used, with 1 for the same direction and 0 for the opposite direction; in legal region transfers outside the channel region... The inheritance relaxation coefficient is preferably between 0.60 and 0.85, and is only used for inheritance supplementation between adjacent legal regions.

[0095] Calculate the comprehensive association score for envelope pairs constrained by region inheritance:

[0096] ,

[0097] in, Enveloping the preceding flock of birds Envelope of subsequent flocks The overall correlation score, For overlapping scoring weights, As a weight for the displacement continuity score, For regional inheritance scoring weights, , and The definitions mentioned above shall be used respectively. , and During the deployment and calibration phase, the values ​​are preferably set to 0.55 to 0.65, 0.20 to 0.30, and 0.15 to 0.25, respectively, with the sum of the three being 1. When there is significant runway boundary jitter, a higher weight is used for the region inheritance score; when the envelope deformation is large, a higher weight is used for the displacement continuation score, but all weights are kept within the range of 0 to 1. If the estimated displacement vector of the preceding flock envelope is unavailable, the predicted position extrapolated from the preceding flock envelope is unavailable, or the preceding flock envelope has been written with an ambiguous state marker, the displacement continuation score is not included in the comprehensive association score with a zero value. Instead, the displacement continuation component is temporarily not enabled, and the association score weights are redistributed according to the original relative proportions of the remaining effective components. When there are fewer than two effective components, candidate inheritance edges are not established based on the comprehensive association score; only subsequent candidate inheritance edges that meet the overlap priority threshold are retained.

[0098] Based on this, the condition for establishing the edge is written as:

[0099] ,

[0100] in, Enveloping the preceding flock of birds Envelope of subsequent flocks Mark the inherited edges between them. This is an indicator function; it takes the value 1 when the condition is true and 0 when the condition is false. , and Using the definitions mentioned above, The overlap priority threshold, This is a comprehensive correlation threshold. Overlap priority threshold. The preferred calibration value is 0.35 to 0.50, with the lower edge of the interval used when the deformation is strong and the upper edge used when the background is relatively stable, and the value should not be lower than 0.30; the comprehensive correlation threshold is preferably 0.55 to 0.70; when At that time, in the encirclement of the preceding flock of birds Envelope of subsequent flocks Candidate inheritance edges are established between them. After this processing, overlapping relationships are responsible for determining the main inheritance, displacement continuation relationships are responsible for determining the supplementation in low-overlap scenarios, and region inheritance relationships are responsible for gating the legality of the runway-end scenario.

[0101] After candidate inheritance edges are established, each predecessor node and each successor node are sorted from highest to lowest based on their comprehensive association score. The edge with the highest score is retained as the primary inheritance edge, and the predecessor object identifier is retained along this primary inheritance edge. When a predecessor node corresponds to multiple successor nodes, if the primary inheritance edge is already established, and the remaining edges still satisfy the legal region transition, and the mutual spacing between the envelopes corresponding to multiple successor nodes continues to increase in the next time step, then the relationship from the predecessor node to multiple successor nodes is recorded as a split relationship. In this case, the successor object corresponding to the primary inheritance edge retains the predecessor object identifier, and the other successor objects are written with the same predecessor object identifier. When multiple predecessor nodes correspond to one successor node, if the successor envelope covers the predicted centroids of multiple predecessor envelopes, and the area of ​​the successor envelope is closer to the sum of the areas of multiple predecessor envelopes rather than any single predecessor envelope area, then it is recorded as a merge relationship, and multiple predecessor object identifiers are written into the successor object.

[0102] When one-to-many and many-to-one relationships appear simultaneously within the local candidate relationship subgraph formed by two consecutive frames, the object relationships are not directly rewritten based on the single-frame connection results. Instead, the primary inheritance edges with the highest mutual selection and continuous region inheritance are retained first, and then the area convergence trend and centroid divergence trend reflected by the remaining edges are compared. When the subsequent envelope has a higher coverage of multiple preceding predicted centroids and its area is closer to the sum of the areas of multiple preceding envelopes, it is processed with priority for merging. When the spacing between multiple subsequent envelopes continues to increase, and each subsequent envelope maintains a valid region inheritance from the same preceding envelope, it is processed with priority for splitting. When the two types of evidence are close, an ambiguous state marker is written, only the primary inheritance edges are retained, and the splitting or merging relationship is postponed to the next frame for confirmation. After this processing, when splitting and merging occur simultaneously, the object identifier will not repeatedly jump due to single-frame echo compression or envelope adhesion. When the difference in the comprehensive correlation score between the primary candidate inheritance edge and the secondary candidate inheritance edge is not greater than the scoring tolerance, the two types of evidence are considered to be close. The scoring tolerance is only used for ambiguous state marking, delayed confirmation, and backtracking binding, and is preferably set to 0.05 to 0.10. When the background fluctuation is large, the upper edge of the interval is taken. If the corresponding envelope is missing in the next time step, or if the corresponding envelope is still in an ambiguous state in the next time step, only the primary inheritance edge is retained and the ambiguous state marking is maintained. Split relations or merging relations are not written in advance.

[0103] For jittery scenes near region boundaries, jitter buffers are set on both sides of each region boundary. The width of the jitter buffer is calibrated to be 0.5 to 1.0 times the larger of the distance resolution and the median equivalent width of the bird flock envelope. When the centroid of the bird flock envelope only enters the jitter buffer, or when the dominant region identifier changes back and forth within two consecutive frames, the identifier of the preceding object remains unchanged, and the dominant region identifier of the preceding bird flock envelope remains unchanged. Only an ambiguous state flag is written, and the region inheritance result is not immediately rewritten. Only when the proportion of the area of ​​the current bird flock envelope falling into the new region to the total area of ​​the bird flock envelope continuously reaches the dominant proportion threshold and continues to reach the confirmation frame number, and the region transfer belongs to the set of legal region transfers, is the candidate edge confirmed as a region inheritance edge. The dominant proportion threshold is preferably not less than 0.60, and the confirmation frame number is preferably 2 to 3 frames. For inheritance breaks caused by single-frame missing or single-frame crossover, if the same path trend exists in both frames before and after, the predicted trajectory passes through the connection transition area, the envelope of the preceding flock and the envelope of the subsequent flock are under the same parameter version identifier, the region transfer still belongs to the legal region transfer set and does not conflict with the already retained main inheritance edge after reconstruction, it is allowed to reconstruct the inheritance edge within a 1-frame rollback window and retain the previous object identifier.

[0104] The inheritance relaxation coefficient, overlap scoring weight, displacement continuation scoring weight, region inheritance scoring weight, overlap priority threshold, comprehensive correlation threshold, scoring tolerance, dominant proportion threshold, confirmation frame count, entry hold threshold, hold threshold, lane occupation duration baseline, traverse upper limit frame count, and scan update cycle are all bound using the same parameter version identifier. The parameter version identifier is only used for recording, alignment, rollback, and version binding. Online rewriting is not performed under the same parameter version identifier; the parameter version identifier is only switched when the scene constraint model is adjusted, the channel axis definition is changed, the scan system is switched, or the offline tuning version is changed. Parameter version switching takes effect from the creation of a new flock state object; existing flock state objects retain their original parameter version identifier until the object ends. When there are insufficient offline tuning samples, the previous valid parameter version is frozen; when there is no previous valid parameter version, a conservative parameter group is used, which takes a higher overlap priority threshold and entry hold threshold, a lower inheritance relaxation coefficient, and prohibits increasing the displacement continuation scoring weight.

[0105] If no supplementary evidence is obtained after the rollback window expires or any condition is not met, the inheritance edge is not constructed, the inheritance is terminated, and a new object is generated. Thus, each inheritance edge in the continuous relationship graph reflects both geometric continuity and the migration semantics between the attraction source association area, the connecting transition area, and the channel area, thereby providing a continuous and traceable object foundation for subsequent entry events, cutting events, traversing events, occupancy events, grouping events, and grouping events.

[0106] Example 2:

[0107] Based on Example 1, this example further explains the event extraction process. This implementation method, based on the region occupation sequence and continuous relationship graph, continuously determines entry events, cutting events, traversal events, lane occupation events, grouping events, and grouping events, and specifically explains the formation and maintenance methods of lane occupation events, as follows:

[0108] Event extraction is performed based on the combined results of the region occupation sequence and the continuous relationship graph. An entry event corresponds to the first state change of a flock of birds transitioning from the attraction source association zone to the connection transition zone; a cutting event corresponds to the first state change of a flock of birds transitioning from the connection transition zone to the channel zone. Cross-cutting events are determined based on the aforementioned channel centerline; when a flock of birds enters the channel zone and its centroid moves from one side of the channel centerline to the other, it is recorded as a cross-cutting event. Encroachment events are determined based on continuous occupation status; when a flock of birds enters the channel zone and remains within the channel zone for a predetermined number of consecutive frames, it is recorded as an encroachment event; when a flock of birds briefly leaves the channel zone under boundary jitter conditions and then returns, it is continuously recorded as the same encroachment event until the exit condition corresponding to the release frame threshold is met, until the exit condition corresponding to the release frame threshold is met. The event sequence includes at least the event type, occurrence time, corresponding region identifier, and associated object identifier.

[0109] Specifically, the entry-hold threshold refers to the required occupation ratio threshold when a flock of birds first stably enters the passage area; the hold threshold refers to the required occupation ratio threshold to maintain the occupied state after it has been formed; the occupation duration baseline refers to the minimum duration used to convert a continuous occupation state into an occupation event; and the maximum traverse frame count refers to the maximum number of frames required to distinguish between rapid crossing of the passage centerline and continuous occupation of the passage area. All of the above parameters are tuned offline along with the scan update cycle. The entry-hold threshold defaults to 0.60, with an adjustable range of 0.55~0.70; the hold threshold defaults to 0.40, with an adjustable range of 0.35~0.50; the occupation duration baseline defaults to 3s, with an adjustable range of 2s~5s; and the maximum traverse frame count defaults to 0.5 times the threshold of consecutive frames required to trigger an occupation event, rounded down to at least 2 frames. The entry-hold threshold is higher than the hold threshold to ensure higher stability upon first entering the passage area, allowing slight boundary oscillations after the hold state is formed without immediate reversal. The release frame threshold refers to the number of frames required to end a lane occupation event after the lane occupation state has been formed, when the number of consecutive departures of the area occupation sequence from the channel area reaches this threshold. The default release frame threshold is 2 frames, and the adjustable range is 2 to 4 frames. The handling priority refers to the handling level field written to the structured detection results. The handling time window refers to the continuous handling time range corresponding to the lane occupation event. The handling priority can be recorded in levels 1 to 3, and the handling time window must cover at least one lane occupation duration baseline and extend as the lane occupation count continues. For cases where timestamps are missing, out of order, or the interval between adjacent timestamps exceeds twice the scan update cycle corresponding to the current parameter version, the current duration, handling priority, and handling time window output are frozen, the previous valid record is retained, and written to the missing detection status field.

[0110] like Figure 4 As shown, in an optional implementation of Embodiment 2, the region occupation sequence that remains in the channel area for a predetermined number of consecutive frames is recorded as a channel occupation event. The determination is not only based on the single-frame region identifier, but also on the degree of continuous occupation of the channel area by the bird flock state object in adjacent time moments, the boundary jitter maintenance state, and the channel centerline crossing process.

[0111] For any given moment, calculate the proportion of the bird flock's state in the channel area and generate a single-frame channel occupancy criterion:

[0112] ,

[0113] ,

[0114] in, For the first The proportion of the channel area occupied by the frame bird flock state object. For the first The flock envelope corresponding to the frame flock state object. The area of ​​the passageway. For the first Single-frame occupancy marker for a frame. This is an indicator function; it takes the value 1 when the condition is true and 0 when the condition is false. To reach the holding threshold, For the first Frame occupancy retention count, when This indicates that the previous frame was in a hold-alive state. If a valid primary inheritance edge has not been formed in the previous frame, or the hold-alive count for the previous frame is unavailable, then press [the button]. Processing, disabling branch retention, To maintain the threshold, the entry holding threshold is jointly calibrated based on the effective width of the channel area and the scanning resolution unit, preferably between 0.55 and 0.70; the holding threshold is lower than the entry holding threshold, preferably between 0.35 and 0.50, used to maintain the already entered occupancy state during boundary jitter. With this setting, the bird flock envelope only raises the trigger threshold when it first stably enters the channel area, while allowing small-amplitude boundary swings in subsequent consecutive frames, avoiding frequent opening and closing of the occupancy state caused by single-frame jitter near the area boundary.

[0115] After the single-frame occupancy marker is generated, it is continuously updated for both entry hold and exit hold. The occupancy hold count can be written as:

[0116] ,

[0117] in, For the first Frame occupancy count, For the first Frame occupancy count, To maintain the maximum count for road occupancy, The exit hold step size is set to 1 to 2. A step size of 1 is used when the scan echo boundary fluctuates significantly, and a step size of 2 is used when there is a lot of clutter at the channel edge. The upper limit of the occupancy hold count is set according to the ratio of the predetermined longest occupancy observation time to the scan update cycle. This update method achieves exit hold by incrementing one frame at a time and releasing slowly. The occupancy state is not immediately canceled when a single frame leaves the channel area; it only ends when consecutive departures meet the exit release requirement, thus covering the situation where the bird flock envelope swings back and forth near the channel boundary. If two consecutive frames show a sudden decrease in envelope area, and the centroid position deviates from the predicted position extrapolated from the previous bird flock envelope by a distance exceeding the aforementioned displacement continuation limit, the exit hold will be cancelled. If an abnormal condition is encountered, the occupancy count of the previous frame remains unchanged at 1 frame, and increasing the occupancy count is prohibited. Only maintaining or releasing the count is allowed to reduce the risk of false triggering. Anomaly detection is only performed when there is a valid primary inherited edge in the previous frame and the predicted position obtained by extrapolating the envelope of the preceding flock is available. If the predicted position obtained by extrapolating the envelope of the preceding flock is unavailable, no valid primary inherited edge was formed in the previous frame, or an ambiguous state flag was written in the previous frame, anomaly detection based on the predicted position is not performed. Instead, the occupancy count is maintained or released based on the current single-frame occupancy flag and occupancy ratio, and no increase is performed.

[0118] When setting the predetermined number of consecutive frames in conjunction with the scan update cycle, the number of consecutive frames is not hardcoded directly. Instead, the occupancy duration baseline is first set, and then converted into a frame number threshold.

[0119] ,

[0120] in, The predetermined threshold of consecutive frames required to trigger a road occupancy event. As a benchmark for the duration of road occupation, For the scan update cycle, The calculation is rounded up. The runway occupancy duration is calibrated based on runway-side response requirements, preferably between 2 and 5 seconds. The scan update cycle is determined by the equipment's scan rate; as the scan speed increases, the consecutive frame count threshold increases accordingly, and as the scan speed decreases, the consecutive frame count threshold decreases accordingly, but the consecutive frame count threshold is not less than 3 frames. Using this conversion method ensures that the actual duration of occupancy remains consistent across different scanning systems, avoiding inconsistencies in the judgment criteria for the same bird flock trajectory at different deployment points simply due to differences in equipment refresh rates.

[0121] When a road occupancy event is triggered, it is not categorized as a traverse event or a road occupancy event. Instead, it is differentiated into short-term crossing and continuous occupation based on changes in position on both sides of the passage's centerline. The directed distance from the centroid of the bird flock state object to the channel centerline is: When adjacent times satisfy and A cross-cutting event is recorded when the sign of the signal changes, and the sum of the number of consecutive frames in the channel area before and after the sign change is less than the cross-cutting limit. A cross-cutting event is recorded when the lane occupancy count meets the trigger condition but not the aforementioned cross-cutting condition. The trigger condition for a lane occupancy event can be written as:

[0122] ,

[0123] in, For the first The occupancy event trigger flag of the frame, For the first Frame occupancy count, The predetermined threshold of consecutive frames required to trigger a road occupancy event. For the first Frame transverse exclusive marker, This indicates that the object currently meets the cross-cutting event conditions. This indicates that the object does not currently meet the cross-cut event conditions. The preferred maximum number of frames for a cross-cut is set to [value missing]. The value is 0.4 to 0.6 times the value and rounded down, for a minimum of 2 frames, used to distinguish between the process of quickly crossing the centerline of the channel and the process of continuously staying in the channel area. After this processing, objects that cross and then leave are classified as traversing events, while objects that stay or continue moving along the channel are classified as occupying events, avoiding duplicate coverage of the two types of events in the result record.

[0124] When a lane occupancy event is triggered, the start time of the occupancy, the current duration, the handling priority, and the handling time window are written into the structured detection results. The handling priority increases as the lane occupancy holding count increases, the object envelope width approaches the upper edge of the effective channel width, and the object's movement speed along the channel axis decreases. When the occupancy ratio continuously falls below the holding threshold and enters the exit and release phase, the handling priority is downgraded or the handling flag is revoked.

[0125] The current duration is calculated based on the time difference between the current moment and the start time of the occupation. If the timestamp is missing, out of order, or the interval between adjacent timestamps exceeds twice the scan update cycle corresponding to the current parameter version, the current duration and the processing time window output are frozen, the previous valid result is retained, and no extrapolation is performed. The processing priority and processing time window are only used for writing structured detection results and are not used to write back the continuous relationship score, region inheritance score, or occupation event triggering condition. The processing time window starts from the first triggering moment of the occupation event, covers at least one occupation duration benchmark, and is extended frame by frame as the occupation holding count continues to increase. For objects located at the boundary between the connection transition area and the channel area, an occupation event is only allowed to be generated when the entry holding threshold is continuously met and the occupation holding count is greater than 0; the continuous meeting of the entry holding threshold is preferably met for 2 to 3 consecutive frames; 2 frames are taken when the scan update cycle is long, and 3 frames are taken when the boundary jitter is obvious. For objects that have already triggered an occupation event, the occupation end time is only written after the occupation ratio is continuously lower than the holding threshold and the exit release is completed. Therefore, the lane occupancy event is no longer an instantaneous result of a single-frame region label, but a stable expression of the continuous occupancy state of the channel area, which can directly support the generation of critical channel handling records.

[0126] Example 3:

[0127] like Figure 2As shown, based on the methods described in Embodiments 1 and 2, this embodiment provides a digital signal processing system for bird flock detection. The system organizes its processing units according to the following sequence: scan echo frame acquisition, scene modeling, background processing, bird flock envelope generation, continuous relationship graph construction, bird flock state object generation, region mapping, event extraction, and result output, to output structured detection results corresponding to the bird flock state objects.

[0128] The data acquisition module is used to acquire multi-frame scan echo frames and scene constraint data arranged in chronological order;

[0129] The scene modeling module is used to build a scene constraint model based on scene constraint data and to divide the attraction source association area, connection transition area and channel area.

[0130] The background processing module is used to perform time-series registration and background stripping on multiple scan echo frames to obtain candidate echo units.

[0131] The envelope generation module is used to aggregate candidate echo cells into a flock envelope based on spatial adjacency.

[0132] The continuous relationship module is used to construct a continuous relationship graph by first filtering candidate node pairs based on the regional inheritance relationship, then establishing inheritance edges based on the envelope overlap relationship and displacement continuation relationship, and recording the split relationship and merging relationship.

[0133] The State Object module is used to generate bird flock state objects that continue across frames based on inheritance, splitting, and merging relationships in the continuous relationship graph;

[0134] The region mapping module is used to map bird flock state objects to scene constraint models and generate region occupation sequences;

[0135] The event extraction module is used to extract entry events, cutting events, traversal events, road occupation events, grouping events, and grouping events based on the regional occupation sequence and continuous relationship diagram.

[0136] The results output module is used to output the structured detection results corresponding to the bird flock state object;

[0137] In one embodiment of Example 3, the various processing steps in the system are executed in a unified data link sequence. The data acquisition module outputs scan echo frames and scene constraint data; the scene modeling module outputs a scene constraint model based on the scene constraint data; the background processing module outputs candidate echo cells based on the scan echo frames; the envelope generation module outputs a flock envelope based on the candidate echo cells; the continuity relationship module outputs a continuity relationship graph based on the flock envelopes at adjacent time points; the state object module outputs a flock state object based on the continuity relationship graph; the region mapping module outputs a region occupation sequence based on the flock state object and the scene constraint model; the event extraction module outputs an event sequence based on the region occupation sequence and the continuity relationship graph; and the result output module outputs structured detection results based on the flock state object, the region occupation sequence, and the event sequence. The system can be deployed on a single processing device or on multiple processing nodes sharing the same result storage area; regardless of the deployment method, the scan echo frames, scene constraint models, continuity relationship graphs, and structured detection results are all associated and stored using a unified object identifier.

[0138] This embodiment also provides an electronic device, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the digital signal processing method for bird flock detection proposed in the above embodiment.

[0139] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the digital signal processing method for bird flock detection as proposed in the above embodiment; the computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the aforementioned digital signal processing method for bird flock detection.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0141] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A digital signal processing method for bird flock detection, characterized in that, include: Acquire multiple scan echo frames and scene constraint data arranged in chronological order. The scene constraint data includes at least the outer boundary of the runway end, the projection area of ​​the approach and departure passage, and the bird attraction source area. Based on the scene constraint data, a scene constraint model under a unified coordinate system is established, and the attraction source association area, connection transition area, and channel area are divided. The multi-frame scan echo is time-series registered and background stripped to obtain candidate echo units at each time point. Based on spatial adjacency, candidate echo units at each time step are aggregated into flock envelopes. With flock envelopes at adjacent time steps as nodes, candidate node pairs are first screened based on regional inheritance relationships. Inheritance edges are first established based on envelope overlap relationships. For candidate node pairs without established inheritance edges, inheritance edges are then established based on displacement continuity relationships. When the same subsequent node corresponds to multiple preceding nodes, the main inheritance edge is determined and the remaining connection relationships are retained for splitting or merging relationship determination, thereby constructing a continuous relationship graph. Generate a bird flock state object that spans multiple frames based on the continuous relationship graph; Map the bird flock state object to the scene constraint model to generate a region occupation sequence; Based on the region occupation sequence and the continuous relationship graph, entry events, cutting events, traversal events, road occupation events, grouping events, and grouping events are extracted to generate an event sequence; Based on the bird flock state object, the region occupation sequence, and the event sequence, a structured detection result is output.

2. The digital signal processing method for bird flock detection according to claim 1, characterized in that, Establishing the scenario constraint model includes: A scene reference coordinate system is established using the outer boundary of the runway end; A channel area is generated based on the projection area of ​​the entry and exit channels; An attraction source association zone is generated based on the described bird attraction source region; A connection transition zone is generated based on the connectivity range between the attraction source association zone and the channel zone; Fixed area identifiers are written for the attraction source association area, the connection transition area, and the channel area, respectively.

3. The digital signal processing method for bird flock detection according to claim 1, characterized in that, The background stripping includes: Perform uniform coordinate registration on scan echo frames from adjacent time points; A static background set is established for echo cells with stable positions in multiple consecutive frames; When the overlap ratio between the echo unit and the static background set reaches the background threshold, the echo unit is marked as a background echo. The candidate echo cells are generated based on the connectivity of echo cells that are not marked as background echoes.

4. The digital signal processing method for bird flock detection according to claim 1, characterized in that, When generating the flock state object, write an object identifier, start time, end time and previous object identifier for each flock state object; When a split relationship occurs in the continuous relationship graph, the same preceding object identifier is written to the multiple bird flock state objects after the split. When a merging relationship occurs in the continuous relationship graph, multiple predecessor object identifiers are written for the merged flock state object.

5. The digital signal processing method for bird flock detection according to claim 1, characterized in that, The extracted events include: recording the first state change of the region occupancy sequence from the attraction source association zone to the connection transition zone as an entry event; recording the first state change of the region occupancy sequence from the connection transition zone to the channel zone as a cutting event; and recording the movement of the bird flock state object from one side of the channel centerline to the other side within the channel zone as a traverse event, based on the channel centerline. The channel centerline is a reference line generated by the central axis of the channel zone.

6. The digital signal processing method for bird flock detection according to claim 5, characterized in that, The extracted events also include: when a flock of birds enters the channel area based on the continuous occupation status and remains in the channel area for a continuous period of time that reaches the continuous frame number threshold, it is recorded as an occupation event; and under the boundary jitter condition, if the bird enters the channel area continuously before the release frame number threshold is reached, it is continuously recorded as the same occupation event; the merging relationship in the continuous relationship graph is recorded as a grouping event, and the splitting relationship in the continuous relationship graph is recorded as a grouping event.

7. The digital signal processing method for bird flock detection according to claim 1, characterized in that, The structured detection results also include timestamps, region identifiers, preceding object identifiers, and subsequent object identifiers; The region occupation sequence and the event sequence of the same flock state object are associated and written into the result record in chronological order; When the event sequence contains both an entry event and a road occupation event, a critical channel handling record is generated. The critical channel handling record is a handling result record associated with the corresponding bird flock state object.

8. A digital signal processing apparatus for bird flock detection, based on the digital signal processing method for bird flock detection according to any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire multi-frame scan echo frames and scene constraint data arranged in chronological order; The scene modeling module is used to establish a scene constraint model based on the scene constraint data, and to divide the attraction source association area, the connection transition area, and the channel area. The background processing module is used to perform time-series registration and background stripping on the multi-frame scan echo frames to obtain candidate echo units. An envelope generation module is used to aggregate the candidate echo units into a flock envelope based on spatial adjacency relationships; The continuous relationship module is used to select candidate node pairs based on the regional inheritance relationship, and then establish inheritance edges based on the envelope overlap relationship and displacement continuation relationship. When the same subsequent node corresponds to multiple predecessor nodes, the main inheritance edge is determined and the remaining connection relationship is retained for the determination of split relationship or merge relationship. The split relationship and merge relationship are recorded to construct a continuous relationship graph. The state object module is used to generate a bird flock state object that continues across frames based on the inheritance, splitting, and merging relationships in the continuous relationship graph. The region mapping module is used to map the bird flock state object to the scene constraint model and generate a region occupation sequence. The event extraction module is used to extract entry events, cutting events, traversal events, lane occupation events, grouping events, and grouping events based on the region occupation sequence and the continuous relationship graph. The result output module is used to output the structured detection results corresponding to the bird flock state object.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the digital signal processing method for bird flock detection as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the digital signal processing method for bird flock detection as described in any one of claims 1 to 7.

Images