Airport intelligent bird repelling system radar detection and identification method based on bird flight path

Abstract

The invention belongs to the technical field of image recognition, and provides an airport intelligent bird repelling system radar detection recognition method based on a bird flight trajectory, which comprises the following steps: obtaining sequential flight trajectory data of birds based on bird situation data of an airport area obtained by a bird detection radar in real time; based on an LSTM classifier and a recurrent neural network, classifying the time sequence flight trajectory data to obtain a bird flight trajectory recognition result; and carrying out feature value similarity matching on the bird flight trajectory recognition result and bird flight standard trajectory data in a set bird flight standard trajectory database to obtain a bird type recognition result. According to the invention, through the learning ability of the LSTM classifier and the recurrent neural network, the fine features of the flight trajectory of the birds can be accurately captured, high-precision trajectory classification is realized, and the intelligent level of an airport bird repelling system is improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a radar detection and recognition method for an intelligent bird control system at airports based on bird flight trajectories. Background Technology

[0002] In recent years, with the rapid development of the aviation industry, the probability of bird strikes has increased significantly worldwide, with major bird strikes occurring frequently, causing serious loss of life and property, and drawing high attention from relevant departments in various countries. The frequent occurrence of bird strikes is a key and challenging issue for air force bases and civil aviation airports. Currently, most bird strike solutions employ technologies such as sound waves, light beams, chemicals, and biological methods, including gas cannons, sonic bird deterrents, laser bird deterrents, bird deterrent windmills, bird deterrent agents, drones, and robotic bird deterrents. While these methods can achieve some effect, they fail to meet expectations due to a lack of detection and identification devices, inability to provide early warnings and assess the effectiveness of bird strikes, the absence of intelligent algorithms, simplistic and untargeted methods, and a lack of big data analysis and deep learning.

[0003] Traditional airport bird surveillance relies on manual labor. However, bird strikes are most likely to occur during the difficult times of dawn, dusk, and night when visual observation is challenging. Radar is a crucial technology for bird surveillance, as it is not limited by visibility and can operate automatically around the clock. Radar can be used not only for airport bird strike prevention but also for bird surveillance in estuary wetland bird sanctuaries, major bird migration routes, and planned wind farms. It can accurately statistically analyze bird population size and activity patterns within these areas, assisting relevant departments in conducting scientific research and engineering feasibility assessments.

[0004] Therefore, it is necessary to provide a radar detection and identification method for intelligent bird control systems at airports based on bird flight trajectories. Summary of the Invention

[0005] This invention provides a radar detection and identification method for intelligent bird control systems at airports based on bird flight trajectories. By leveraging the learning capabilities of LSTM classifiers and recurrent neural networks, it can accurately capture subtle features of bird flight trajectories, achieving high-precision trajectory classification and enhancing the intelligence level of airport bird control systems.

[0006] This invention provides a radar detection and identification method for an intelligent bird control system at airports based on bird flight trajectories, including:

[0007] Based on real-time bird data acquired by bird-detecting radar in the airport area, temporal flight trajectory data of birds is obtained.

[0008] Based on LSTM classifier and recurrent neural network, the time-series flight trajectory data is classified to obtain bird flight trajectory recognition results;

[0009] The bird flight trajectory identification results are matched with the bird flight standard trajectory data in the established bird flight standard trajectory database by feature value similarity matching to obtain the bird type identification results.

[0010] Furthermore, based on real-time bird activity data acquired by bird-detecting radar in the airport area, temporal flight trajectory data of birds is obtained, including:

[0011] Based on multiple configured bird detection radars, real-time bird data of the airport area is acquired. The bird data includes flight trajectory, echo intensity, signal characteristics, flight speed and altitude.

[0012] A dynamically updated bird information database is built based on bird information data;

[0013] Based on the bird information database, the temporal flight trajectory data of birds is extracted, and the trajectory feature parameters of the temporal flight trajectory data are calculated and obtained. The trajectory feature parameters include longitude, latitude, altitude, curvature, speed and acceleration.

[0014] Furthermore, based on the real-time bird activity data of the airport area acquired by the bird detection radar, the temporal flight trajectory data of birds is obtained. This also includes processing the signals from the bird detection radar. The processing flow is as follows: signal sampling, digital down-conversion, pulse compression, amplitude and phase correction, target aggregation, and trajectory tracking.

[0015] Furthermore, the trajectory feature parameters for acquiring time-series flight trajectory data are calculated, including:

[0016] Obtain the coordinates of preset points in the time-series flight trajectory data;

[0017] The preset point coordinates are substituted into the set curve equation to calculate the parameter values ​​of the curve equation. Based on the parameter values, the derivative vector of the curve equation at the preset point is obtained. Specifically, for each flight trajectory, a segmented interpolation connection is performed at every three points to obtain the derivative vector set of the three points on the segmented interpolation curve, which constitutes the characteristic value of the segmented curve. The set of characteristic values ​​obtained by solving segment by segment is used to generate the trajectory characteristic parameters of the time-series flight trajectory data.

[0018] Furthermore, temporal flight trajectory data of birds is extracted, including: using an adaptive segmentation algorithm to dynamically adjust the length of trajectory segments based on the rate of change of bird flight speed, and enriching the trajectory points, specifically:

[0019] The rate of change of velocity is defined as the absolute value of the velocity difference between adjacent time points, and a dynamic threshold for the rate of change of velocity is set.

[0020] If the rate of change of velocity is greater than the dynamic threshold, adaptive segmentation is initiated, and the current point in the time-series flight trajectory data corresponding to each segment is used as the segmentation starting point.

[0021] If the rate of change of velocity is less than the dynamic threshold for N consecutive time points, then the adaptive segmentation ends.

[0022] After each segmentation is completed, the threshold is dynamically updated based on the distribution of the rate of change of the current segment using an exponentially weighted moving average algorithm.

[0023] For each trajectory point within a segment, sharp turn features, acceleration features, and direction change angles are calculated. These features are then concatenated with the global trajectory features and used as input to the LSTM classifier.

[0024] Furthermore, based on an LSTM classifier and a recurrent neural network, the temporal flight trajectory data is classified to obtain bird flight trajectory recognition results, including:

[0025] Input the trajectory feature parameters into the LSTM classifier and output the classification label of the bird's flight trajectory;

[0026] Based on the configured recurrent neural network, the flight trajectory is modeled and classified according to the classification labels of the bird flight trajectory to obtain the bird flight trajectory recognition result. The recurrent neural network is configured with a structure with recurrently connected hidden layer units, the hidden layer activation function is the hyperbolic tangent function (tanh), and the output layer activation function is the sigmoid function.

[0027] Furthermore, the trajectory feature parameters are input into an LSTM classifier, and the output is a classification label for the bird's flight trajectory. This also includes: confidence evaluation and manual verification of the classification labels, specifically:

[0028] Obtain the probability distribution vector of the classification labels, which includes the probability distribution of bird types;

[0029] The difference between the highest probability value and the second highest probability value in the probability distribution vector is defined as the confidence level.

[0030] If the confidence level is greater than or equal to the set confidence threshold, the classification label is considered reliable; if the confidence level is less than the set confidence threshold, the classification is considered low-confidence. The confidence threshold is dynamically updated using a sliding window mechanism based on the statistical results of historical classification data.

[0031] For classification labels determined to be of low confidence, the original radar signal data, feature parameters, and classification results of the corresponding bird flight trajectories are extracted and manually verified.

[0032] Furthermore, inputting the trajectory feature parameters into the LSTM classifier also includes: extracting spatiotemporal features of bird flight patterns based on convolutional neural networks through deep learning of historical data, and optimizing the trajectory feature parameters; specifically:

[0033] Historical flight trajectory data is converted into a spatiotemporal grid matrix, and the training dataset is expanded by random time slicing, spatial translation, and noise injection.

[0034] A 3D convolutional neural network is used to extract spatiotemporal features. Through a spatiotemporal attention module, key time nodes and spatial regions are weighted to output weighted features.

[0035] The weighted features are concatenated with the original LSTM input to optimize the trajectory feature parameters.

[0036] Furthermore, the bird flight trajectory identification results are matched with the bird flight standard trajectory data in the constructed bird flight standard trajectory database using feature value similarity to obtain bird type identification results, including:

[0037] A standard flight trajectory database for birds will be constructed to store the standard flight trajectories and their characteristic values ​​for different birds.

[0038] The bird flight trajectory in the bird flight trajectory recognition results is matched with the standard flight trajectory by feature value similarity. If the similarity matching value is greater than the set matching threshold, the bird flight trajectory is determined to match the standard bird flight trajectory, thereby determining the bird type corresponding to the bird flight trajectory.

[0039] Furthermore, the bird flight paths identified in the bird flight path recognition results are matched with standard flight paths using feature value similarity. If the similarity match value is greater than a set matching threshold, the bird flight path is determined to match the standard bird flight path, thereby determining the bird type corresponding to the bird flight path, including:

[0040] The similarity of the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result is calculated by fast Fourier transform. If the similarity of the trajectory spectrum is greater than the set similarity threshold, an improved dynamic time warping algorithm is used to calculate the morphological similarity between the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result. Specifically, based on the basic dynamic time warping algorithm, a constraint window is introduced to limit the search range of the matching path, and the minimum cumulative cost path between the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result is obtained.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] First, it enables real-time monitoring and accurate identification of bird flight trajectories in airport areas, effectively improving the intelligence level of the airport bird control system. Second, by introducing LSTM classifiers and recurrent neural networks, deep learning and classification of time-series flight trajectory data significantly improve the accuracy and stability of bird flight trajectory identification. Third, it constructs a dynamically updated bird information database and a standard bird flight trajectory database, providing rich data support for bird type identification and trajectory matching. Fourth, it adopts adaptive segmentation algorithms and exponentially weighted moving average algorithms to dynamically adjust the trajectory segment length and enrich trajectory points, improving the accuracy and efficiency of trajectory feature parameter extraction. Fifth, through confidence assessment and manual verification of classification labels, as well as deep learning of historical data, it further optimizes trajectory feature parameters and bird flight trajectory identification results, reducing false alarm and false negative rates.

[0043] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0044] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0046] Figure 1 A schematic diagram of the structural steps of a radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories;

[0047] Figure 2 This is a schematic diagram illustrating the steps of a method to obtain temporal flight trajectory data of birds based on real-time bird situation data acquired by bird detection radar in the airport area.

[0048] Figure 3 This diagram illustrates the steps of a method for classifying time-series flight trajectory data and obtaining bird flight trajectory recognition results based on an LSTM classifier and a recurrent neural network. Detailed Implementation

[0049] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0050] This invention provides a radar detection and identification method for an intelligent bird control system at airports based on bird flight trajectories, such as... Figure 1As shown, it includes:

[0051] Based on real-time bird data acquired by bird-detecting radar in the airport area, temporal flight trajectory data of birds is obtained.

[0052] Based on LSTM classifier and recurrent neural network, the time-series flight trajectory data is classified to obtain bird flight trajectory recognition results;

[0053] The bird flight trajectory identification results are matched with the bird flight standard trajectory data in the established bird flight standard trajectory database by feature value similarity matching to obtain the bird type identification results.

[0054] The working principle of the above technical solution is as follows: In order to realize the radar detection and identification method of airport intelligent bird control system based on bird flight trajectory, the present invention first uses bird detection radar to monitor the airport area in real time, captures the flight information of birds, and then generates time-series flight trajectory data. These data record key information such as the flight path, speed and altitude of the birds in detail. Subsequently, LSTM classifier and recurrent neural network are used to perform deep learning and classification processing on these complex time-series data, so as to accurately identify the type of bird flight trajectory. After obtaining the bird flight trajectory identification results, these results are compared with the preset standard bird flight trajectory database. Through feature value similarity matching algorithm, the species of birds are quickly and accurately identified.

[0055] The beneficial effects of the above technical solution are as follows: The solution provided in this embodiment enables real-time monitoring and intelligent identification of birds within the airport area, effectively improving the efficiency and accuracy of the airport bird control system; real-time monitoring by bird-detecting radar can capture bird flight information and generate detailed temporal flight trajectory data, providing a reliable data foundation for subsequent classification and identification; simultaneously, utilizing the deep learning and classification processing capabilities of LSTM classifiers and recurrent neural networks, the types of bird flight trajectories can be accurately identified; by comparing with a preset standard bird flight trajectory database, the species of birds can be quickly and accurately identified, providing targeted bird control strategies for the airport bird control system and reducing the impact of birds on flight safety.

[0056] In one embodiment, such as Figure 2 As shown, based on real-time bird activity data acquired by bird-detecting radar in the airport area, temporal flight trajectory data of birds is obtained, including:

[0057] Based on multiple configured bird detection radars, real-time bird data of the airport area is acquired. The bird data includes flight trajectory, echo intensity, signal characteristics, flight speed and altitude.

[0058] A dynamically updated bird information database is built based on bird information data;

[0059] Based on the bird information database, the temporal flight trajectory data of birds is extracted, and the trajectory feature parameters of the temporal flight trajectory data are calculated and obtained. The trajectory feature parameters include longitude, latitude, altitude, curvature, speed and acceleration.

[0060] The working principle of the above technical solution is as follows: Multiple bird-detecting radars are configured to achieve comprehensive real-time monitoring of the airport area, ensuring the capture of all possible bird flight information. These radars not only acquire bird flight trajectories but also record detailed information such as echo intensity and signal characteristics, providing rich data support for subsequent bird identification and classification. The constructed bird database is a dynamically updated system capable of receiving and storing new bird data in real time. When extracting temporal flight trajectory data of birds, the flight trajectories are recorded and analyzed in detail based on the information in the bird database. Simultaneously, by calculating trajectory characteristic parameters such as longitude, latitude, altitude, curvature, speed, and acceleration, the flight characteristics and patterns of birds can be further understood.

[0061] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, real-time monitoring and accurate identification of birds in the airport area can be achieved, which effectively improves the efficiency and accuracy of the airport bird control system.

[0062] In one embodiment, obtaining temporal flight trajectory data of birds based on real-time bird situation data of the airport area acquired by bird detection radar further includes: processing the signal of bird detection radar, the processing flow being: signal sampling, digital down-conversion, pulse compression, amplitude and phase correction, target aggregation and trajectory tracking.

[0063] The working principle of the above technical solution is as follows: In the signal sampling stage, the raw signal received by the bird detection radar is first acquired at high speed to ensure complete bird flight information is obtained; then, in the digital down-conversion step, the sampled signal is converted to baseband for subsequent processing; pulse compression technology is used to improve the radar's resolution and signal-to-noise ratio, thereby more accurately capturing the bird's flight trajectory; the amplitude and phase correction stage adjusts the amplitude and phase of the signal to eliminate errors introduced by the radar system itself or environmental factors; the target agglomeration step merges multiple scattering points of the same target into one target point, simplifying the complexity of data processing; finally, the trajectory tracking module constructs the bird's flight trajectory based on continuous target point information, providing a basis for subsequent classification and identification.

[0064] The beneficial effects of the above technical solution are as follows: the solution provided in this embodiment ensures the high quality and reliability of bird detection radar data, providing strong support for the decision-making of the airport's intelligent bird control system.

[0065] In one embodiment, calculating trajectory feature parameters for acquiring time-series flight trajectory data includes:

[0066] Obtain the coordinates of preset points in the time-series flight trajectory data;

[0067] The preset point coordinates are substituted into the set curve equation to calculate the parameter values ​​of the curve equation. Based on the parameter values, the derivative vector of the curve equation at the preset point is obtained. Specifically, for each flight trajectory, a segmented interpolation connection is performed at every three points to obtain the derivative vector set of the three points on the segmented interpolation curve, which constitutes the characteristic value of the segmented curve. The set of characteristic values ​​obtained by solving segment by segment is used to generate the trajectory characteristic parameters of the time-series flight trajectory data.

[0068] The working principle of the above technical solution is as follows: by calculating the derivative vector at preset points, the local changing trend of the flight trajectory can be captured, such as key information like speed and acceleration; the method of segmented interpolation connecting every three points not only simplifies the computational complexity but also effectively smooths the flight trajectory data and reduces noise interference; these trajectory feature parameters, such as speed, acceleration, and their rate of change, provide an important data foundation for subsequent bird behavior analysis, species identification, and the formulation of intelligent bird control strategies; in addition, by solving and integrating the feature values ​​segment by segment, the comprehensiveness and accuracy of the trajectory feature parameters are ensured, providing a strong guarantee for the efficient operation of the airport's intelligent bird control system.

[0069] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, it is possible to accurately capture and efficiently process the flight trajectory of birds; by deeply analyzing the trajectory feature parameters, it is possible to accurately determine the flight status and behavior pattern of birds, and also to provide a scientific basis for the formulation of intelligent bird control strategies, making bird control measures more precise and effective, which not only ensures the flight safety of the airport, but also reduces the interference with the ecological environment of birds.

[0070] In one embodiment, extracting temporal flight trajectory data of birds includes: employing an adaptive segmentation algorithm to dynamically adjust the trajectory segment length based on the rate of change of the bird's flight speed, and enriching the trajectory points, specifically:

[0071] The rate of change of velocity is defined as the absolute value of the velocity difference between adjacent time points, and a dynamic threshold for the rate of change of velocity is set.

[0072] If the rate of change of velocity is greater than the dynamic threshold, adaptive segmentation is initiated, and the current point in the time-series flight trajectory data corresponding to each segment is used as the segmentation starting point.

[0073] If the rate of change of velocity is less than the dynamic threshold for N consecutive time points, then the adaptive segmentation ends.

[0074] After each segmentation is completed, the threshold is dynamically updated based on the distribution of the rate of change of the current segment using an exponentially weighted moving average algorithm.

[0075] For each trajectory point within a segment, sharp turn features, acceleration features, and direction change angles are calculated. These features are then concatenated with the global trajectory features and used as input to the LSTM classifier.

[0076] The working principle of the above technical solution is as follows: Real-time capture of bird flight information, including key parameters such as position and speed, is achieved using radar and other equipment. The extracted temporal flight trajectory data, after being processed by an adaptive segmentation algorithm, can more accurately reflect the dynamic changes in bird flight. When a significant change occurs in the bird's flight speed, the algorithm automatically initiates a segmentation, using that time point as the starting point for a new segment. If the bird's flight speed is relatively stable, and the rate of change of speed at N consecutive time points is lower than a set dynamic threshold (where N represents the number of consecutive stable time points used to determine whether the bird's flight state has entered a stable phase), then the current segmentation ends. The method of adjusting the segment length makes the trajectory data more detailed and consistent with the actual flight behavior of birds. After segmentation, the trajectory points within each segment are analyzed in depth to calculate key indicators such as sharp turn features, acceleration features, and abrupt change angles of direction. These features can intuitively reflect the behavioral characteristics of birds during flight, such as sudden changes in direction and acceleration. Subsequently, these local features are combined with global trajectory features and used as input to an LSTM classifier. The LSTM classifier has powerful time series processing capabilities and can efficiently learn and classify the input trajectory features, thereby achieving accurate identification of bird flight trajectories.

[0077] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, the flight trajectory and behavioral characteristics of birds can be grasped in real time, providing strong data support for subsequent intelligent bird control strategies.

[0078] In one embodiment, such as Figure 3 As shown, based on an LSTM classifier and a recurrent neural network, time-series flight trajectory data is classified to obtain bird flight trajectory recognition results, including:

[0079] Input the trajectory feature parameters into the LSTM classifier and output the classification label of the bird's flight trajectory;

[0080] Based on the configured recurrent neural network, the flight trajectory is modeled and classified according to the classification labels of the bird flight trajectory to obtain the bird flight trajectory recognition result. The recurrent neural network is configured with a structure with recurrently connected hidden layer units, the hidden layer activation function is the hyperbolic tangent function (tanh), and the output layer activation function is the sigmoid function.

[0081] The working principle of the above technical solution is as follows: The LSTM classifier first receives trajectory feature parameters, which cover key information such as the bird's flight speed, direction change, and acceleration. Through its complex internal network structure and weight adjustment, the LSTM classifier can learn the intrinsic relationship between trajectory features and time series, and then perform preliminary classification of flight trajectories, outputting corresponding classification labels. These classification labels represent birds with different flight patterns and behavioral characteristics, such as migratory birds, birds of prey, or roosting birds. Subsequently, these classification labels are fed into a recurrent neural network. The recurrent neural network, with its unique recurrent connection of hidden layer units, can capture long-term dependencies in time series data. That is, the current output is not only related to the current input, but also to the previous historical input. The recurrent neural network uses this characteristic to model the time series features of bird flight trajectories, further refining the classification and recognition. The hidden layer activation function adopts the hyperbolic tangent function (tanh), which can map the input value to between -1 and 1, helping the model learn the nonlinear relationship of the data. The output layer activation function is the sigmoid function, which converts the output value into a probability value between 0 and 1, representing the probability that the bird belongs to a certain category.

[0082] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, the recurrent neural network can accurately classify and identify flight trajectories, output the identification results of bird flight trajectories, and provide comprehensive and accurate data support for subsequent intelligent bird deterrence strategies.

[0083] In one embodiment, the trajectory feature parameters are input into an LSTM classifier, and the output is a classification label for the bird's flight trajectory. The method further includes: performing confidence assessment and manual verification of the classification labels, specifically:

[0084] Obtain the probability distribution vector of the classification labels, which includes the probability distribution of bird types;

[0085] The difference between the highest probability value and the second highest probability value in the probability distribution vector is defined as the confidence level.

[0086] If the confidence level is greater than or equal to the set confidence threshold, the classification label is considered reliable; if the confidence level is less than the set confidence threshold, the classification is considered low-confidence. The confidence threshold is dynamically updated using a sliding window mechanism based on the statistical results of historical classification data.

[0087] For classification labels determined to be of low confidence, the original radar signal data, feature parameters, and classification results of the corresponding bird flight trajectories are extracted and manually verified.

[0088] The working principle of the above technical solution is as follows: By introducing a confidence assessment mechanism, high-confidence classification results can be automatically selected, reducing the workload of manual review. When the confidence of the classification results is low, these results will be automatically marked and prompted for manual review. During manual review, experts can use the original radar signal data, feature parameters, and preliminary classification results, combined with their professional knowledge, to confirm or correct the classification results, ensuring the accuracy of the classification. In addition, a sliding window mechanism is used to dynamically update the confidence threshold, enabling the system to continuously optimize and adjust the classification threshold as historical data accumulates, thereby improving the system's adaptability and accuracy.

[0089] The beneficial effects of the above technical solution are as follows: The solution provided in this embodiment not only improves the automation level of bird flight trajectory classification and recognition, but also ensures the reliability of classification results through a manual review mechanism, providing a more accurate data foundation for subsequent intelligent bird deterrence strategies.

[0090] In one embodiment, inputting trajectory feature parameters into an LSTM classifier further includes: extracting spatiotemporal features of bird flight patterns based on a convolutional neural network through deep learning of historical data, and optimizing the trajectory feature parameters; specifically:

[0091] Historical flight trajectory data is converted into a spatiotemporal grid matrix, and the training dataset is expanded by random time slicing, spatial translation, and noise injection.

[0092] A 3D convolutional neural network is used to extract spatiotemporal features. Through a spatiotemporal attention module, key time nodes and spatial regions are weighted to output weighted features.

[0093] The weighted features are concatenated with the original LSTM input to optimize the trajectory feature parameters.

[0094] The working principle of the above technical solution is as follows: By converting historical flight trajectory data into a spatiotemporal grid matrix, the flight trajectory of birds can be represented in matrix form, where each grid represents a specific spatial location and time point. Data augmentation techniques such as random time slicing, spatial translation, and noise injection are used to expand the training dataset, which helps to improve the generalization ability of the model and prevent overfitting. A 3D convolutional neural network is used to extract spatiotemporal features from the expanded training dataset. This network structure can consider information in both time and spatial dimensions, thereby capturing the dynamic changes in bird flight trajectories. The spatiotemporal attention module further weights key time points and spatial regions, which helps the model pay more attention to features that have a greater impact on the classification results. Finally, the weighted features are concatenated with the original LSTM input to form optimized trajectory feature parameters. These optimized feature parameters can more comprehensively reflect the flight patterns of birds, thereby improving the recognition accuracy of the LSTM classifier.

[0095] The beneficial effects of the above technical solution are as follows: The solution provided in this embodiment can significantly improve the accuracy and robustness of the airport intelligent bird control system in recognizing bird flight trajectories. By using a spatiotemporal grid matrix representation method, complex flight trajectory data is effectively transformed into an easily processed matrix form, providing a solid foundation for subsequent feature extraction and classification. The application of data augmentation technology not only increases the diversity of the training dataset but also effectively avoids overfitting during model training, improving the model's generalization ability. The introduction of a 3D convolutional neural network fully considers the temporal and spatial continuity of bird flight trajectories, making the extracted spatiotemporal features more accurate and comprehensive. Simultaneously, the weighted processing of the spatiotemporal attention module further highlights the importance of key time nodes and spatial regions, enabling the model to focus more on key information during classification. Finally, by concatenating the weighted features with the original LSTM input, the optimized trajectory feature parameters not only improve the expressive power of the features but also provide richer and more accurate information for the LSTM classifier, thereby achieving a significant improvement in recognition accuracy.

[0096] In one embodiment, the bird flight trajectory identification result is matched with the bird flight standard trajectory data in the constructed bird flight standard trajectory database using feature value similarity matching to obtain the bird type identification result, including:

[0097] A standard flight trajectory database for birds will be constructed to store the standard flight trajectories and their characteristic values ​​for different birds.

[0098] The bird flight trajectory in the bird flight trajectory recognition results is matched with the standard flight trajectory by feature value similarity. If the similarity matching value is greater than the set matching threshold, the bird flight trajectory is determined to match the standard bird flight trajectory, thereby determining the bird type corresponding to the bird flight trajectory.

[0099] The working principle of the above technical solution is as follows: First, through extensive observation and recording of the flight trajectories of various birds, a detailed database of standard bird flight trajectories is constructed. This database not only contains the unique flight patterns of different birds, but also records their flight characteristic values ​​under different environmental conditions, such as flight speed, altitude changes, and turning radius. These standard trajectories and their characteristic values ​​provide a reliable reference for subsequent feature value similarity matching. In practical applications, when the radar detects a bird's flight trajectory, it compares the identified bird's flight trajectory with the data in the standard trajectory database one by one. This comparison process is based on the similarity of feature values. It calculates the similarity score between the feature values ​​in the identification result and the feature values ​​of the standard trajectory. If the similarity score of a certain standard trajectory is higher than the preset matching threshold, it is determined that the flight trajectory matches the standard trajectory, thereby accurately identifying the bird type corresponding to the flight trajectory.

[0100] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, the accuracy and comprehensiveness of the standard trajectory database are fully utilized. Through precise matching of feature values, the rapid and accurate identification of bird types is achieved. This not only provides more accurate bird information for airport bird control systems, but also helps to formulate more scientific and reasonable bird control strategies, thereby effectively ensuring airport flight safety.

[0101] In one embodiment, the bird flight trajectory in the bird flight trajectory recognition result is matched with a standard flight trajectory by feature value similarity. If the similarity matching value is greater than a set matching threshold, the bird flight trajectory is determined to match the standard bird flight trajectory, thereby determining the bird type corresponding to the bird flight trajectory, including:

[0102] The similarity of the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result is calculated by fast Fourier transform. If the similarity of the trajectory spectrum is greater than the set similarity threshold, an improved dynamic time warping algorithm is used to calculate the morphological similarity between the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result. Specifically, based on the basic dynamic time warping algorithm, a constraint window is introduced to limit the search range of the matching path, and the minimum cumulative cost path between the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result is obtained.

[0103] The working principle of the above technical solution is as follows: First, frequency domain analysis of bird flight trajectories is performed using Fast Fourier Transform to extract the spectral features of the trajectories. These features can reflect key information such as the periodicity and speed changes of bird flight. Next, these spectral features are compared with the spectral features of standard flight trajectories to calculate the similarity. If the similarity is high, it indicates that the two have a large similarity in the frequency domain, which may mean that the bird's flight mode or pattern is similar to the standard trajectory. After confirming that the spectral similarity meets certain conditions, an improved dynamic time warping algorithm is further used to finely match the morphological similarity. Dynamic time warping algorithm is a technique used to measure the similarity between two time series. By finding the optimal matching path between the two, the cumulative cost between them is calculated. On this basis, a constraint window is introduced to limit the search range of the matching path, making the matching process more efficient and accurate. By solving for the minimum cumulative cost path, the morphological similarity between the bird flight trajectory and the standard trajectory can be quantified.

[0104] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, the bird type corresponding to the bird flight trajectory can be accurately identified through dual matching of spectral similarity and morphological similarity, providing reliable data support for the airport intelligent bird control system.

[0105] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A radar detection and identification method for intelligent bird control systems at airports based on bird flight trajectories, characterized in that, include: Based on real-time bird data acquired by bird-detecting radar in the airport area, temporal flight trajectory data of birds is obtained. Based on LSTM classifier and recurrent neural network, the time-series flight trajectory data is classified to obtain bird flight trajectory recognition results; The bird flight trajectory identification results are matched with the bird flight standard trajectory data in the established bird flight standard trajectory database by feature value similarity matching to obtain the bird type identification results.

2. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 1, characterized in that, Based on real-time bird activity data acquired by bird-detecting radar in the airport area, temporal flight trajectory data of birds is obtained, including: Based on multiple configured bird detection radars, real-time bird data of the airport area is acquired. The bird data includes flight trajectory, echo intensity, signal characteristics, flight speed and altitude. A dynamically updated bird information database is built based on bird information data; Based on the bird information database, the temporal flight trajectory data of birds is extracted, and the trajectory feature parameters of the temporal flight trajectory data are calculated and obtained. The trajectory feature parameters include longitude, latitude, altitude, curvature, speed and acceleration.

3. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 2, characterized in that, Based on real-time bird activity data of the airport area acquired by bird detection radar, the temporal flight trajectory data of birds is obtained. The process also includes: signal processing of the bird detection radar signal, which includes: signal sampling, digital down-conversion, pulse compression, amplitude and phase correction, target aggregation and trajectory tracking.

4. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 2, characterized in that, The trajectory feature parameters of the acquired time-series flight trajectory data are calculated, including: Obtain the coordinates of preset points in the time-series flight trajectory data; The preset point coordinates are substituted into the set curve equation to calculate the parameter values ​​of the curve equation. Based on the parameter values, the derivative vector of the curve equation at the preset point is obtained. Specifically, for each flight trajectory, a segmented interpolation connection is performed at every three points to obtain the derivative vector set of the three points on the segmented interpolation curve, which constitutes the characteristic value of the segmented curve. The set of characteristic values ​​obtained by solving segment by segment is used to generate the trajectory characteristic parameters of the time-series flight trajectory data.

5. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 2, characterized in that, Extracting temporal flight trajectory data of birds includes: employing an adaptive segmentation algorithm to dynamically adjust the trajectory segment length based on the rate of change of bird flight speed, and enriching the trajectory points, specifically: The rate of change of velocity is defined as the absolute value of the velocity difference between adjacent time points, and a dynamic threshold for the rate of change of velocity is set. If the rate of change of velocity is greater than the dynamic threshold, adaptive segmentation is initiated, and the current point in the time-series flight trajectory data corresponding to each segment is used as the segmentation starting point. If the rate of change of velocity is less than the dynamic threshold for N consecutive time points, then the adaptive segmentation ends. After each segmentation is completed, the threshold is dynamically updated based on the distribution of the rate of change of the current segment using an exponentially weighted moving average algorithm. For each trajectory point within a segment, sharp turn features, acceleration features, and direction change angles are calculated. These features are then concatenated with the global trajectory features and used as input to the LSTM classifier.

6. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 1, characterized in that, Based on an LSTM classifier and a recurrent neural network, time-series flight trajectory data is classified to obtain bird flight trajectory recognition results, including: Input the trajectory feature parameters into the LSTM classifier and output the classification label of the bird's flight trajectory; Based on the configured recurrent neural network, the flight trajectory is modeled and classified according to the classification labels of the bird flight trajectory to obtain the bird flight trajectory recognition result. The recurrent neural network is configured with a structure with recurrently connected hidden layer units, the hidden layer activation function is the hyperbolic tangent function (tanh), and the output layer activation function is the sigmoid function.

7. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 6, characterized in that, The trajectory feature parameters are input into an LSTM classifier, which outputs classification labels for bird flight trajectories. The process also includes: confidence evaluation and manual verification of the classification labels, specifically: Obtain the probability distribution vector of the classification labels, which includes the probability distribution of bird types; The difference between the highest probability value and the second highest probability value in the probability distribution vector is defined as the confidence level. If the confidence level is greater than or equal to the set confidence threshold, the classification label is considered reliable; if the confidence level is less than the set confidence threshold, the classification is considered low-confidence. The confidence threshold is dynamically updated using a sliding window mechanism based on the statistical results of historical classification data. For classification labels determined to be of low confidence, the original radar signal data, feature parameters, and classification results of the corresponding bird flight trajectories are extracted and manually verified.

8. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 6, characterized in that, Inputting trajectory feature parameters into an LSTM classifier also includes: extracting spatiotemporal features of bird flight patterns based on a convolutional neural network through deep learning of historical data, and optimizing the trajectory feature parameters; specifically: Historical flight trajectory data is converted into a spatiotemporal grid matrix, and the training dataset is expanded by random time slicing, spatial translation, and noise injection. A 3D convolutional neural network is used to extract spatiotemporal features. Through a spatiotemporal attention module, key time nodes and spatial regions are weighted to output weighted features. The weighted features are concatenated with the original LSTM input to optimize the trajectory feature parameters.

9. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 1, characterized in that, The bird flight trajectory identification results are matched with the standard bird flight trajectory data in the constructed standard bird flight trajectory database using feature value similarity analysis to obtain bird type identification results, including: A standard flight trajectory database for birds will be constructed to store the standard flight trajectories and their characteristic values ​​for different birds. The bird flight trajectory in the bird flight trajectory recognition results is matched with the standard flight trajectory by feature value similarity. If the similarity matching value is greater than the set matching threshold, the bird flight trajectory is determined to match the standard bird flight trajectory, thereby determining the bird type corresponding to the bird flight trajectory.

10. The radar detection and identification method for an airport intelligent bird control system based on bird flight trajectories according to claim 9, characterized in that, The bird flight paths identified in the bird flight path recognition results are matched with standard flight paths by feature value similarity. If the similarity match value is greater than a set matching threshold, the bird flight path is determined to match the standard bird flight path, thereby identifying the bird type corresponding to the flight path, including: The similarity of the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result is calculated by fast Fourier transform. If the similarity of the trajectory spectrum is greater than the set similarity threshold, an improved dynamic time warping algorithm is used to calculate the morphological similarity between the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result. Specifically, based on the basic dynamic time warping algorithm, a constraint window is introduced to limit the search range of the matching path, and the minimum cumulative cost path between the bird flight trajectory and the standard flight trajectory in the bird flight trajectory recognition result is obtained.

Images