An airport bird detection and driving integrated method based on artificial intelligence

Abstract

The application relates to the technical field of aviation safety, and discloses an airport bird detection and driving integrated method based on artificial intelligence. The method comprises the following steps: constructing a backbone network based on a convolution layer and a C2f block, constructing a multi-branch feature fusion structure based on a convolution layer, a C2f block and a bidirectional weighted fusion, constructing a fine-grained perception detection head in a cascaded form in multiple dimensions based on residual connection and multi-head attention, and constructing a bird identification model based on the backbone network, the multi-branch feature fusion structure and multiple fine-grained perception detection heads; obtaining a flying bird picture, extracting features of the flying bird picture in different scales by the backbone network, performing feature fusion on different scales by the multi-branch feature fusion structure to obtain multiple fusion features, performing flying bird type prediction on the multiple fusion features by the multiple fine-grained perception detection heads, obtaining a flying bird type prediction result, and dispersing the flying birds in combination with a driving strategy. The method solves the problem that the existing airport bird detection and driving method cannot effectively drive the birds to ensure aviation safety.

Description

Technical Field

[0001] This invention relates to the field of aviation safety technology, and in particular to an integrated method for detecting, identifying, and driving away birds at airports based on artificial intelligence. Background Technology

[0002] Bird strikes not only severely damage aircraft components but can also lead to major flight accidents. Although major airports have established bird prevention systems, in actual operation, existing bird detection and deterrence methods still face severe technical bottlenecks and practical difficulties, mainly in the following aspects.

[0003] The limited detection methods make it difficult to balance false alarm and missed detection rates. Due to the lack of a deep fusion mechanism for multimodal data, a single sensor cannot achieve robust target detection in complex weather and electromagnetic environments, resulting in frequent missed and false alarms.

[0004] The target identification is coarse-grained and lacks accurate bird species classification capabilities. Due to the lack of deep learning algorithms that support bird micro-Doppler features and fine-grained image features, airports are often forced to adopt a "one-size-fits-all" warning approach for all targets, which not only wastes resources on deportation but also, in severe cases, delays critical evacuation opportunities due to the inability to identify high-risk bird species.

[0005] The current methods of deterring birds are "static," making them highly adaptable to birds. Birds quickly adapt to fixed-frequency sounds and a lack of varied light stimuli, causing equipment to fail shortly after deployment. Ground-based equipment has a limited range and cannot target flocks flying at mid-to-high altitudes or far from the runway. Existing deterrence strategies are mostly based on simple rules (such as "activate upon detection"), lacking intelligent decision-making based on avian behavior. These methods cannot predict bird movements based on species, numbers, and flight paths, and cannot achieve the precise operation of "deterring only when birds pose a threat to the aircraft."

[0006] The separation of detection and deterrence, coupled with a lack of real-time, interconnected closed-loop control, results in isolated information systems at many airports. Detection information is displayed on control tower screens, while deterrence operations often rely on ground personnel receiving instructions and driving to the scene. This "human-centric" operational model suffers from significant time lag (typically on the order of minutes), failing to address the high-speed movement (millisecond-level changes) of birds. Furthermore, the lack of coordination between aerial mobile platforms such as drones and ground methods leads to a lack of three-dimensionality and proactivity in deterrence efforts. Often, this not only fails to deter flocks of birds but also startles them into the runway's takeoff and landing paths, causing counterproductive results.

[0007] In summary, existing airport bird control work urgently needs an integrated approach that combines all-weather multimodal detection, AI-based refined identification, adaptive intelligent strategy decision-making, and air-ground coordinated three-dimensional bird removal to address current industry pain points such as high false alarm rates, inaccurate identification, rapid failure of bird removal, and delayed response. Summary of the Invention

[0008] This invention provides an integrated method for bird detection and deterrence at airports based on artificial intelligence, in order to solve the problem that existing methods for bird detection and deterrence at airports cannot effectively drive away birds to ensure aviation safety.

[0009] To achieve the above objectives, the present invention employs the following technical solution:

[0010] This invention provides an integrated method for bird detection and control at airports based on artificial intelligence, comprising the following steps: Step 1: Construct a bird recognition model, which includes a backbone network that acquires features at different scales based on images of flying birds, a multi-branch feature fusion structure that fuses features at different scales at different spatial scales, and multiple fine-grained perception detection heads that predict based on the fused features output by the multi-branch feature fusion structure. The backbone network is constructed based on convolutional layers and C2f blocks, the multi-branch feature fusion structure is constructed based on the bidirectional weighted fusion of convolutional layers and C2f blocks, and the fine-grained perceptual detection head is constructed based on multi-head attention in a cascaded manner. Step 2: Based on radar and photoelectric tracking, bird images are acquired. The bird recognition model uses a backbone network to extract features at different scales based on the bird images. Based on the features at different scales, a multi-branch feature fusion structure is used to fuse features at different scales to obtain multiple fused features. Multiple fine-grained perception detection heads predict the bird species based on the multiple fused features. The one with the highest confidence is selected as the bird species prediction result. Based on the bird species prediction result, a dispersal strategy is combined to disperse the birds.

[0011] Furthermore, the backbone network is constructed based on convolutional layers and C2f blocks, including: constructing backbone units based on convolutional layers and C2f blocks, and constructing the backbone network in the order of convolutional layers, stacked backbone units, and fast pyramid pooling network structure.

[0012] The above design effectively solves the problem of drastic scale changes in bird targets due to flight, and can effectively extract high-level and low-level features to achieve more dimensional feature representation.

[0013] Furthermore, the multi-branch feature fusion structure is constructed by building multiple feature fusion branches at different scales based on convolutional layers and C2f blocks, and the feature fusion branches between adjacent scales are combined with upsampling and downsampling for bidirectional weighted fusion.

[0014] Furthermore, the multi-branch feature fusion structure includes a first feature fusion branch, a second feature fusion branch, and a third feature fusion branch, and is connected to the backbone unit or fast pyramid pooling in the backbone network in descending order of scale. The first feature fusion branch and the second feature fusion branch each include a convolutional layer, a first C2f block and a second C2f block, and the third feature fusion branch includes a convolutional layer and a single C2f block; The first feature fusion branch performs bidirectional weighted fusion by upsampling the features downsampled from the corresponding backbone unit and the features processed by the first C2f block in the second feature fusion branch; and performs bidirectional weighted fusion by downsampling the features processed by the first C2f block and the features of the backbone unit at the previous scale. The second feature fusion branch performs bidirectional weighted fusion by upsampling the features downsampled from the corresponding backbone unit and the features in the third feature fusion branch that have not been processed by the C2f block; and performs bidirectional weighted fusion by downsampling the features processed by the first C2f block and the features in the first feature fusion branch that have been processed by the second C2f block. The third feature fusion branch performs bidirectional weighted fusion of the features downsampled by fast pyramid pooling and the features processed by the second C2f block in the second feature fusion branch through downsampling.

[0015] Furthermore, the bidirectional weighted fusion in the multi-branch feature fusion structure adopts fast normalization fusion.

[0016] Furthermore, the construction of the fine-grained sensing detection head includes setting multi-head attention in the residual branches of multiple residual connections in a cascaded manner, respectively for the scale dimension, spatial dimension and channel dimension.

[0017] Furthermore, the multi-head attention network structure designed for the scale dimension includes, in sequence, average pooling, convolutional layers, activation functions, and hard sigmoid layers; The multi-head attention network for spatial dimension settings includes a residual main branch, a first residual sub-branch, and a second residual sub-branch. The residual main branch is constructed based on a data selection layer and a convolutional layer, the first residual sub-branch is constructed based on a bias layer, and the second residual sub-branch is constructed based on a sigmoid function. The multi-head attention network designed for the channel dimension includes a first residual branch and a second residual branch. The first residual branch includes average pooling, a fully connected layer, an activation function, a fully connected layer, and normalization in sequence. The second residual branch performs an identity mapping on the input features, fuses them with the features output by the first residual branch, and then fuses them with the main network line of the detection head.

[0018] Furthermore, the radar and photoelectric tracking for acquiring bird images includes: determining the flying object based on radar, determining whether the flying object is a bird based on photoelectric tracking, and taking a bird image if it is a bird.

[0019] Furthermore, the expulsion strategies include bird species expulsion strategies, regional linkage strategies, timed expulsion, and ADS-B linkage expulsion. The bird species repelling strategy includes: if the bird species matches the predefined high-risk bird species, then during the linkage period, bird repelling equipment is used to repel the bird corresponding to the bird image; The regional linkage strategy includes: counting the number of bird species in a predetermined area, and using bird-repelling devices to drive away the birds corresponding to the bird images during the linkage period based on the number of birds; The timed bird removal includes: setting a timed task, and using bird-repelling equipment to drive away the bird corresponding to the bird image within the predetermined time of the timed task; The ADS-B linked bird deterrence includes: obtaining flight information; if the flight information indicates that an aircraft has reached a predetermined distance, then using bird deterrence equipment to deter the bird corresponding to the image of that bird.

[0020] Beneficial effects: This invention provides an integrated method for detecting birds at airports based on artificial intelligence. The backbone network, through the stacking of convolutional layers and C2f blocks, can efficiently extract multi-scale basic features of bird images. The multi-branch feature fusion structure introduces a bidirectional weighting mechanism, which fully interweaves shallow details and deep semantic information, thereby enhancing the model's ability to distinguish different types of birds in the air.

[0021] The fine-grained perception detection head, constructed with residual connections and multi-head attention cascades, can focus on subtle discriminative features such as beak, plumage, and body shape from multiple dimensions including scale, space, and channels, significantly improving the accuracy of distinguishing similar bird species. At the same time, the residual structure effectively prevents gradient degradation of deep networks.

[0022] The system employs a combination of radar and optoelectronic equipment to track and acquire images, ensuring high-quality image acquisition in complex backgrounds and dynamic flight conditions. After the model automatically identifies the bird species, it combines bird species with deterrence strategies to achieve targeted and efficient deterrence. Attached Figure Description

[0023] Figure 1This is a schematic diagram of the overall network structure of the prediction model in an embodiment of the present invention (in... Figure 1 In the diagram, U represents upsampling, B represents bidirectional weighted fusion, and SPPF represents fast spatial pooling. Figure 2 This is a schematic diagram showing the detailed network structure of the detection head in an embodiment of the present invention; Figure 3 This is a schematic diagram of the overall network structure of the detection head in an embodiment of the present invention; Figure 4 This is a flowchart of the expulsion strategy according to an embodiment of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a," and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked," and similar terms, are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0026] This application provides an integrated method for bird detection and patrol at airports based on artificial intelligence, comprising the following steps: Step 1: Construct a bird recognition model. The bird recognition model includes a backbone network that acquires features at different scales based on images of flying birds, a multi-branch feature fusion structure that fuses features at different scales in different spatial scales, and multiple fine-grained perception detection heads that make predictions based on the fused features output by the multi-branch feature fusion structure. The backbone network is built based on convolutional layers and C2f blocks, the multi-branch feature fusion structure is built based on the combination of convolutional layers and C2f blocks with bidirectional weighted fusion, and the fine-grained perception detection head is built based on a cascaded form with multi-head attention set in multiple dimensions. Please see Figure 1Regarding the specific network structure of the backbone network, the backbone unit is constructed based on convolutional layers and C2f blocks, and the backbone network is constructed in the order of convolutional layers, stacked 4 backbone units, and fast pyramid pooling network structure. In the backbone network, initial RGB processing is first performed through convolutional layers, then shape and texture features are extracted layer by layer through stacked backbone units to obtain features at different scales, and finally context aggregation is performed through fast pyramid pooling to obtain local features at multiple scales.

[0027] For the multi-branch feature fusion structure, multiple feature fusion branches are constructed at different scales based on convolutional layers and C2f blocks. The feature fusion branches at adjacent scales are combined with upsampling and downsampling for bidirectional weighted fusion to form the multi-branch feature fusion structure. Specifically, the multi-branch feature fusion structure includes a first feature fusion branch, a second feature fusion branch, and a third feature fusion branch, which are connected to the backbone units or fast pyramid pooling in the backbone network in descending order of scale. The first feature fusion branch and the second feature fusion branch both include a convolutional layer, a first C2f block and a second C2f block, and the third feature fusion branch includes a convolutional layer and a single C2f block. The first feature fusion branch performs bidirectional weighted fusion by upsampling the features after downsampling the corresponding backbone unit and the features after processing by the first C2f block in the second feature fusion branch. The features processed by the first C2f block and the features of the backbone unit at the previous scale are then combined by downsampling for bidirectional weighted fusion. The second feature fusion branch performs bidirectional weighted fusion by upsampling the features after downsampling the corresponding backbone unit and the features in the third feature fusion branch that have not been processed by the C2f block, and performs bidirectional weighted fusion by downsampling the features after processing by the first C2f block and the features in the first feature fusion branch that have been processed by the second C2f block. The third feature fusion branch performs bidirectional weighted fusion of the features downsampled by fast pyramid pooling and the features processed by the second C2f block in the second feature fusion branch through downsampling. The first feature fusion branch is connected to the second backbone unit in the backbone network, the second feature fusion branch is connected to the fourth backbone unit in the backbone network, and the third feature fusion branch is connected to the fast pyramid pooling in the backbone network. This bidirectional feature fusion realizes the feature fusion expression of multiple dimensions, which effectively improves the recognition model's ability to identify bird species features such as bird shape, feathers, and neck in bird images, thereby achieving more accurate bird species prediction.

[0028] The bidirectional weighted fusion all employs fast normalization fusion, which is expressed by the following formula: ; in, Indicates the characteristics after fusion; Represents input features; and All represent learnable weights; This represents the minimum value of the constrained numerical oscillation, which is set to 0.0001 in this embodiment; Fast normalization fusion sets weights for each input feature in conjunction with network learning to reflect the importance of different features. This effectively enables the final expressed features to better integrate information and eliminates invalid or noisy features, avoiding the problem of feature information loss caused by simple multi-scale feature addition.

[0029] Please see Figure 2-3 For the construction of a single fine-grained sensing detection head, it is to set multi-head attention in the residual branches of multiple residual connections in a cascaded manner, targeting the scale dimension, spatial dimension and channel dimension respectively; Specifically, the multi-head attention network structure set for the scale dimension includes average pooling, convolutional layers, activation functions, and hard sigmoid layers. The use of the hard sigmoid function here effectively improves the efficiency of the prediction model in actual engineering deployment compared to the conventional sigmoid function, and also maintains effective feature representation during the calculation process without causing a decrease in the prediction accuracy of the prediction model.

[0030] The multi-head attention network for spatial dimension settings includes a residual main branch, a first residual sub-branch, and a second residual sub-branch. The residual main branch is constructed based on a data selection layer and a convolutional layer, the first residual sub-branch is constructed based on a bias layer, and the second residual sub-branch is constructed based on a sigmoid function. The multi-head attention network designed for the channel dimension includes a first residual branch and a second residual branch. The first residual branch includes average pooling, a fully connected layer, an activation function, another fully connected layer, and normalization. The second residual branch performs an identity mapping on the input features, fuses them with the features output by the first residual branch, and then fuses them with the main network line of the detection head.

[0031] By improving the detection head by combining the scale dimension, spatial dimension, and channel dimension with an attention mechanism, we can achieve effective attention to birds at a fine-grained level and improve the prediction accuracy of the prediction model.

[0032] In the prediction model, fine-grained perception detection heads are set after the first feature fusion branch, the second feature fusion branch, and the third feature fusion branch. These three fine-grained perception detection heads adopt a multi-scale parallel prediction mechanism. During the training phase, the model adopts a full-scale training method, and the three detection heads simultaneously receive the gradients of backpropagation and learn the features of birds at different scales (large, medium, and small). During the deployment / inference phase, three fine-grained perceptual detection heads simultaneously output candidate detection results (including bounding box coordinates, category, and confidence score). The system then uses non-maximum suppression (NMS) to merge and deduplicate all candidate boxes from the three heads, selecting the final result with the highest confidence score. This design ensures the model can effectively detect birds of different distances and sizes.

[0033] In the prediction model, the loss function includes classification loss. With positioning loss The loss function is expressed by the following formula: ; in, The weights represent the classification loss settings; Classification loss A binary cross-entropy loss function is used to measure the accuracy of the detection head in predicting bird categories, expressed by the following formula: ; in, Indicates the number of samples; This represents the actual label, and its value is either 0 or 1. This represents the predicted probability output by the prediction model; Location loss The CIoU loss function is then used, which comprehensively considers the overlap area between the predicted and ground truth bounding boxes, the distance between their center points, and the consistency of their aspect ratios. This effectively improves the localization accuracy of bird targets, as expressed by the following formula: ; ; ; in, express Penalties; This represents the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box. This represents the diagonal length of the smallest bounding rectangle that covers both bounding boxes; Indicates the coordinates of the center point of the prediction box; Represents the coordinates of the center point of the true bounding box; This represents the square of the Euclidean distance between the center point of the predicted bounding box and the center point of the ground truth bounding box; Represents the weighting function; Indicates the aspect ratio consistency parameter; These represent the width and height of the actual bounding box, respectively. These represent the width and height of the virtual box, respectively. Step 2: Based on radar and photoelectric tracking, bird images are acquired. The bird recognition model uses a backbone network to extract features at different scales based on the bird images. Based on the features at different scales, a multi-branch feature fusion structure is used to fuse features at different scales to obtain multiple fused features. Multiple fine-grained perception detection heads use multiple fused features to predict bird species. The one with the highest confidence is selected as the bird species prediction result. Based on the bird species prediction result, a dispersal strategy is combined to disperse the birds.

[0034] Specifically, the system uses radar to identify flying objects, and uses photoelectric tracking to determine whether the flying object is a bird. If it is a bird, images of the bird are taken. Please see Figure 4 The expulsion strategies include bird species expulsion strategies, regional joint expulsion strategies, timed expulsion, and ADS-B joint expulsion. The bird species repelling strategy includes: if the bird species matches the predefined high-risk bird species, bird repelling equipment will be used to repel the bird corresponding to the bird image during the linkage period; The regional linkage strategy includes: counting the number of bird species within a predetermined area, and using bird-repelling equipment to drive away the birds corresponding to the bird images during the linkage period based on the number; Timed bird removal includes: setting a timed task, and using bird-repelling equipment to drive away the bird corresponding to the picture of the bird within the predetermined time of the timed task; ADS-B linked bird deterrence includes: obtaining flight information; if the flight information indicates that an aircraft has reached a predetermined distance, then using bird deterrence equipment to deter the bird corresponding to the image of that bird.

[0035] This embodiment presents an AI-based integrated method and predictive model for airport bird detection and control, which, compared to the conventional method of using radar to detect birds and then manually driving vehicles to drive them away, achieves automated monitoring and bird control around the clock. Furthermore, based on multi-scale feature extraction and bidirectional weighted fusion in the pyramid structure of the predictive model, combined with a fine-grained sensing detection head, it can more accurately identify bird species and more effectively drive away identified bird species.

[0036] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. An integrated method for bird detection and control at airports based on artificial intelligence, characterized in that, Includes the following steps: Step 1: Construct a bird recognition model, which includes a backbone network that acquires features at different scales based on images of flying birds, a multi-branch feature fusion structure that fuses features at different scales at different spatial scales, and multiple fine-grained perception detection heads that predict based on the fused features output by the multi-branch feature fusion structure. The backbone network is constructed based on convolutional layers and C2f blocks, the multi-branch feature fusion structure is constructed based on the bidirectional weighted fusion of convolutional layers and C2f blocks, and the fine-grained perceptual detection head is constructed based on multi-head attention in a cascaded manner. The backbone network is constructed based on convolutional layers and C2f blocks, including: constructing backbone units based on convolutional layers and C2f blocks, and constructing the backbone network in the order of convolutional layers, stacked backbone units, and fast pyramid pooling network structure; The multi-branch feature fusion structure is constructed by building multiple feature fusion branches at different scales based on convolutional layers and C2f blocks, and the feature fusion branches between adjacent scales are combined with upsampling and downsampling for bidirectional weighted fusion. The multi-branch feature fusion structure includes a first feature fusion branch, a second feature fusion branch, and a third feature fusion branch, and is connected to the backbone unit or fast pyramid pooling in the backbone network in descending order of scale. The first feature fusion branch and the second feature fusion branch each include a convolutional layer, a first C2f block and a second C2f block, and the third feature fusion branch includes a convolutional layer and a single C2f block; The first feature fusion branch performs bidirectional weighted fusion by upsampling the features after downsampling the corresponding backbone unit and the features after processing by the first C2f block in the second feature fusion branch; and performs bidirectional weighted fusion by downsampling the features after processing by the first C2f block and the features of the backbone unit at the previous scale. The second feature fusion branch performs bidirectional weighted fusion by upsampling the features after downsampling the corresponding backbone unit and the features in the third feature fusion branch that have not been processed by the C2f block, and performs bidirectional weighted fusion by downsampling the features after processing by the first C2f block and the features in the first feature fusion branch that have been processed by the second C2f block. The third feature fusion branch performs bidirectional weighted fusion of the features downsampled by fast pyramid pooling and the features processed by the second C2f block in the second feature fusion branch through downsampling. Step 2: Based on radar and photoelectric tracking, bird images are acquired. The bird recognition model uses a backbone network to extract features at different scales based on the bird images. Based on the features at different scales, a multi-branch feature fusion structure is used to fuse features at different scales to obtain multiple fused features. Multiple fine-grained perception detection heads predict the bird species based on the multiple fused features. The one with the highest confidence is selected as the bird species prediction result. Based on the bird species prediction result, a dispersal strategy is combined to disperse the birds.

2. The integrated method for airport bird detection based on artificial intelligence according to claim 1, characterized in that, In the multi-branch feature fusion structure, the bidirectional weighted fusion adopts fast normalization fusion.

3. The integrated method for airport bird detection based on artificial intelligence according to claim 1, characterized in that, The construction of the fine-grained sensing detection head includes setting multi-head attention in the residual branches of multiple residual connections in a cascaded manner, targeting the scale dimension, spatial dimension, and channel dimension respectively.

4. The integrated method for airport bird detection based on artificial intelligence according to claim 3, characterized in that, The multi-head attention network structure designed for the scale dimension includes, in sequence, average pooling, convolutional layers, activation functions, and hardsigmoid layers; The multi-head attention network for spatial dimension settings includes a residual main branch, a first residual sub-branch, and a second residual sub-branch. The residual main branch is constructed based on a data selection layer and a convolutional layer, the first residual sub-branch is constructed based on a bias layer, and the second residual sub-branch is constructed based on a sigmoid function. The multi-head attention network designed for the channel dimension includes a first residual branch and a second residual branch. The first residual branch includes average pooling, a fully connected layer, an activation function, a fully connected layer, and normalization in sequence. The second residual branch performs an identity mapping on the input features, fuses them with the features output by the first residual branch, and then fuses them with the main network line of the detection head.

5. The integrated method for airport bird detection and control based on artificial intelligence according to claim 1, characterized in that, The process of acquiring bird images using radar and photoelectric tracking includes: identifying the flying object based on radar, determining whether the flying object is a bird based on photoelectric tracking, and taking a picture of the bird if it is a bird.

6. The integrated method for airport bird detection based on artificial intelligence according to claim 1, characterized in that, The expulsion strategies include bird species expulsion strategies, regional joint strategies, timed expulsion, and ADS-B joint expulsion. The bird species repelling strategy includes: if the bird species matches the predefined high-risk bird species, then during the linkage period, bird repelling equipment is used to repel the bird corresponding to the bird image; The regional linkage strategy includes: counting the number of bird species in a predetermined area, and using bird-repelling devices to drive away the birds corresponding to the bird images during the linkage period based on the number of birds; The timed bird removal includes: setting a timed task, and using bird-repelling equipment to drive away the bird corresponding to the bird image within the predetermined time of the timed task; The ADS-B linked bird deterrence includes: obtaining flight information; if the flight information indicates that an aircraft has reached a predetermined distance, then using bird deterrence equipment to deter the bird corresponding to the image of that bird.

Images