An AI recognition-based agricultural intelligent bird repelling method and system

Abstract

The application relates to an AI recognition-based agricultural intelligent bird driving method and system, through utilization of deep learning and multi-source heterogeneous information fusion technology, high-precision perception and recognition of farmland bird information are realized, the pertinence and effectiveness of the final early warning strategy are significantly improved, a multi-source time sequence data is constructed according to a first time sequence model and a second time sequence model, a final driving strategy is formulated according to a bird activity rhythm, and the interference on the normal physiological rhythm of birds is minimized. Through integration of multiple driving means such as laser, ultrasonic wave and flash light, the driving duration is greatly prolonged, meanwhile, the driving strategy optimization based on a multi-target optimization model enables the system to have self-learning and adaptation capabilities, the requirement for manual intervention is greatly reduced, and the operation and maintenance cost is reduced by more than 60% compared with a traditional method.

Description

Technical Field

[0001] This invention belongs to the field of smart agriculture technology, and specifically relates to an intelligent bird-repelling method and system for agriculture based on AI recognition. Background Technology

[0002] With the rapid development of modern agriculture, bird damage in farmland has become one of the important factors affecting crop yield and quality. Traditional bird control methods, such as manual scare and sound and light bird deterrence, often suffer from short-lived effects, high labor intensity, and high costs. Although new bird deterrence technologies such as ultrasound and lasers have been applied in recent years, the effectiveness of single bird deterrence methods is often unsatisfactory in the face of complex and ever-changing farmland environments and bird behaviors.

[0003] Meanwhile, with the advancement of ecological civilization construction, agricultural production has placed higher demands on environmental friendliness and ecological sustainability. How to effectively control bird damage while minimizing interference with farmland ecosystems and achieving harmonious coexistence among humans, birds, and crops has become an urgent problem to be solved.

[0004] Furthermore, the complexity of farmland environments, the randomness of bird damage occurrences, and the limitations of farmland power supply all pose significant challenges to the design of intelligent bird control systems. Therefore, there is an urgent need to develop a comprehensive farmland bird control system integrating intelligent sensing, multimodal deterrence, and ecological regulation to address the bird control needs of modern agricultural production. Summary of the Invention

[0005] In view of the above problems, the purpose of this invention is to propose an intelligent agricultural bird control method and system based on AI recognition, which solves the problem that bird damage cannot be effectively solved in existing agricultural production.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides an intelligent agricultural bird-repelling method based on AI recognition, the method comprising:

[0008] Acquire a first image sequence within a farmland scene, wherein the first image sequence consists of multiple first images within a first preset sampling period;

[0009] The first image sequence is input into the first convolutional neural network model for feature extraction to obtain the first image feature information, which includes at least one of multiple first bird activity trajectories and multiple first bird behavior patterns in the current first image sequence.

[0010] And simultaneously acquire thermal infrared information in the farmland scene, the thermal infrared information being multiple thermal infrared image sequences within the first preset sampling period;

[0011] The thermal infrared information is input into the second convolutional neural network model for feature extraction to obtain the second thermal infrared feature information, which includes at least one of the multiple second bird activity trajectories and multiple second bird behavior patterns in the current thermal infrared information.

[0012] Obtain bird attribute information associated with farmland scenes, construct a bird attribute semantic dictionary, map attributes to semantic vectors through attribute embedding matrix, and obtain third attribute semantic vectors. The third attribute semantic vectors include at least one of multiple third bird activity trajectories and multiple third bird behavior patterns in the current bird attribute information.

[0013] The first image feature information, the second thermal infrared feature information, and the third attribute semantic vector are input into the multimodal gated fusion unit for feature interaction and fusion to obtain a semantically enhanced multimodal fusion feature sequence;

[0014] The multimodal fusion feature sequence is input into a long short-term memory network to model the temporal dependency of bird activities, resulting in a first temporal model. The first temporal model includes first bird temporal information, which includes the bird's activity trajectory and behavior pattern in the first time series.

[0015] Acquire a second image sequence of a farmland scene within a second preset sampling period. The second image sequence consists of multiple second images within the second preset sampling period.

[0016] The second image sequence is input into the third convolutional neural network model for feature extraction to obtain the fourth image feature information. Based on the fourth image feature information, the bird category information is obtained, and the fourth bird activity trajectory and fourth bird behavior pattern are obtained based on the second image sequence.

[0017] The second time series model is obtained by encoding the activity trajectory, behavior pattern, and bird category information of the fourth bird. The second time series model includes the time series information of the second bird, which includes the activity trajectory and behavior pattern of the bird in the second time series.

[0018] The final deterrence strategy and / or final early warning strategy are generated based on the first time series model, bird category information, and the second time series model.

[0019] In some embodiments, the first image sequence is input into the first convolutional neural network model for feature extraction, and the first image feature information is represented by formula (1), which is as follows:

[0020] F img =f cnn (I;θ cnn );

[0021] In formula (1), F img For the first image feature information, f cnn For feature extractor, θ cnn Here, represents the learning parameters of the first convolutional neural network model, and I represents the first image sequence. T is the first preset sampling period;

[0022] The first convolutional neural network model is generated based on the convolutional neural network model. The performance evaluation of the first convolutional neural network model is obtained through the cross-entropy loss function, which is expressed by formula (2). Formula (2) is as follows:

[0023]

[0024] In formula (2), L is the loss function value, y i To determine the true categories of the sample bird data during the training of the first convolutional neural network model, The probability distribution for predicting sample bird data during the training of the first convolutional neural network model;

[0025] The thermal infrared information is input into the second convolutional neural network model for feature extraction. The second thermal infrared feature information is represented by formula (3), which is as follows:

[0026] F ir =f cnn (H;θ ir );

[0027] In formula (3), F ir For the second thermal infrared feature information, f cnn For feature extractor, θ ir Here, H represents the learning parameters of the second convolutional neural network model. T is the first preset sampling period.

[0028] In some embodiments, the bird attribute semantic dictionary is obtained through the following steps:

[0029] A semantic dictionary is constructed, which is represented by formula (4), as follows:

[0030] D = d1, d2, ..., d N ;

[0031] In formula (4), D is the semantic dictionary, and d i These are attribute words, and each attribute word represents a bird attribute.

[0032] For each attribute word, a semantic embedding vector is matched to form an attribute embedding matrix, thus forming a semantic representation space;

[0033] The attribute embedding matrix is ​​represented by formula (5), which is as follows:

[0034] E = [e1, e2, ..., e N ];

[0035] In formula (5), E is the attribute embedding matrix, e i For semantic embedding vectors, the semantic embedding vectors represent the positions of attribute words in the semantic representation space.

[0036] In some embodiments, the first image feature information, the second thermal infrared feature information, and the third attribute semantic vector are input into the multimodal gated fusion unit for feature interaction and fusion, and the semantically enhanced multimodal fusion feature sequence is represented by formulas (6) to (8), where formula (6) is as follows:

[0037] F′ img =F img +f att (F img ,F ir ,E)⊙F ir ;

[0038] Formula (7) is as follows:

[0039] F′ ir =F ir +f att (F ir ,F img ,E)⊙F img ;

[0040] Formula (8) is as follows:

[0041] F fused =W1F′img+W2F′ ir +W3(F′ img ⊙F′ ir );

[0042] In formulas (6) to (8), F' img F is the first feature information after fusion. img For the first image feature information, f att Let F' be the first attention mechanism function, E be the attribute embedding matrix, and F' be the function of the first attention mechanism. ir F is the second feature information after fusion. fused W1, W2, and W3 are the first fused feature information, and W1, W2, and W3 are the learning parameters of the fused gating.

[0043] In some embodiments, the multimodal fused feature sequence is input into a long short-term memory network to model the temporal dependency of bird activity, resulting in a first temporal model. The first temporal model includes first bird temporal information, which includes the bird's activity trajectory and behavioral patterns in the first time series.

[0044] The multimodal fusion feature sequence is encoded using a first long short-term memory network, and expressed by formula (9):

[0045]

[0046] In formula (9), h t and c t Let f represent the hidden state and memory cell state of the first long short-term memory network at time t, respectively. lstm For the first long short-term memory network function, θ lstm These are the parameters of the first long short-term memory network;

[0047] The first time series model is trained using a multi-task learning mode to obtain the trained first time series model. The training process is represented by formula (10):

[0048]

[0049] In formula (10), This is the total loss function value. This represents the loss function for the bird detection task. This represents the loss function for the bird tracking task. Let λ1, λ2, and λ3 represent the loss function for the bird behavior recognition task, where λ1, λ2, and λ3 are weighting coefficients that balance different tasks.

[0050] In some embodiments, inputting the second image sequence into a third convolutional neural network model for feature extraction to obtain fourth image feature information, obtaining bird category information based on the fourth image feature information, and obtaining the fourth bird activity trajectory and fourth bird behavior pattern based on the second image sequence include:

[0051] A second attention mechanism function is added to the third convolutional neural network model, and it is expressed by formula (11), which is as follows:

[0052] F = A⊙X;

[0053] In formula (11), F is the attention enhancement feature in the second attention mechanism function, A is the attention weight matrix, X is the convolution feature map in the third convolutional neural network model, and ⊙ is element-wise multiplication;

[0054] The attention weight matrix is ​​represented by formula (12), which is as follows:

[0055]

[0056] In formula (12), W a and W x These are the learning parameters in the second attention mechanism function.

[0057] In some embodiments, a second time-series model is obtained by encoding based on the fourth bird's activity trajectory, the fourth bird's behavior pattern, and bird category information. The second time-series model includes second bird time-series information, which includes the bird's activity trajectory and behavior pattern in the second time series.

[0058] The activity trajectory, behavior pattern, and bird category information of the fourth bird were encoded using a second long short-term memory network and expressed by formulas (13) to (18). Formula (13) is as follows:

[0059] i t =σ(W ii x t +W hi h t-1 +b i );

[0060] Formula (14) is as follows:

[0061] f t =σ(W if xt+W hf h t-1 +b f );

[0062] Formula (15) is as follows:

[0063] o t =σ(W io x t +W ho h t-1 +b o );

[0064] Formula (16) is as follows:

[0065]

[0066] Formula (17) is as follows:

[0067]

[0068] Formula (18) is as follows:

[0069] h t ′=ot ⊙tanh(C t );

[0070] In formulas (13) to (18), i t f is the input gate in the second long short-term memory network. t For the forgetting gate in the second long short-term memory network, o t As the output gate in the second long short-term memory network, C t This refers to the cellular state in the second long short-term memory network. For candidate cell states in the second long short-term memory network, h t 'W' represents a hidden state in the second long short-term memory network. ii W hi Let b be the weight matrix of the input gate. i W is the bias vector of the input gate. if W hf Let b be the weight matrix of the forget gate. f W is the bias vector for the forget gate. io W ho Let b be the weight matrix of the output gate. o This is the bias vector for the output gate;

[0071] The second time series model is trained using a multi-task learning mode to obtain the first time series model after training. The training process is represented by formula (19):

[0072] L'=λ'1L recog +λ'2L' behav In formula (19), L′ is the total loss function value. Let L' be the loss function for the bird recognition task. behav Let λ'1 and λ'2 be the loss function for the bird behavior understanding task, and let λ'1 and λ'2 be the weight coefficients that balance the two tasks.

[0073] In some embodiments, generating the final expulsion strategy based on the first time-series model and the second time-series model further includes:

[0074] An intelligent regulation model is constructed, and the initial driving strategy is obtained by inputting observation information into the intelligent regulation model. The observation information includes environmental parameters, the first time series model, the second time series model, and bird category information.

[0075] The initial driving strategy is evaluated for its effectiveness, and the initial driving strategy is optimized based on the evaluation to obtain the final driving strategy.

[0076] The final driving strategy is expressed by formula (20), which is as follows:

[0077]

[0078] In formula (20), π * (s) represents the final driving strategy, and s represents the observed state. Let Q(s,a;θ) be the set of device parameters for the driving equipment used in the intelligent regulation model, where a is the driving action in the final driving strategy. * ) is the state-action value function, θ * These are the optimal parameters for the state-action value function.

[0079] In some embodiments, generating the final early warning strategy based on the first time series model and the second time series model further includes:

[0080] Multi-source time series data were constructed based on the first and second time series models. The multi-source time series data included agricultural ecological calendar, bird migration calendar and meteorological calendar.

[0081] Based on multi-source time series data, a macroscopic temporal pattern of bird damage is generated, and a multi-scale grid is divided into farmland scenes to generate a spatial distribution heat map of farmland bird damage.

[0082] A multi-objective optimization model is constructed. The input data of the multi-objective optimization model are multi-source time series data, spatial distribution heat map, bird species diversity index and degree of human intervention. The output of the multi-objective optimization model is the final early warning strategy.

[0083] In a second aspect, the present invention also provides an intelligent agricultural bird deterrence system based on AI recognition, applicable to the intelligent agricultural bird deterrence method of the first aspect. The system includes a control unit, a communication module, an energy supply module, a bird deterrence device, an image acquisition module, and an AI recognition module.

[0084] The control unit is used to execute the method described in the first aspect; the communication module includes at least one of a Bluetooth module, a Wi-Fi module, and an antenna, and is used to establish a communication connection with an external device; the energy supply module includes at least one of a solar panel, a micro wind turbine, and a battery pack, and is used to provide power; the bird deterrent device includes at least one of a laser emitter, an ultrasonic generator, and an LED flash array, and is used to execute a final bird deterrent strategy; the image acquisition module includes a high-definition camera and an infrared imager, and is used to acquire a first image sequence, a second image sequence, and thermal radiation information; the AI ​​recognition module is used to perform image processing on the first image sequence, the second image sequence, and the thermal radiation information, and is used to train and correct the first neural network model, the second neural network model, and the third neural network model.

[0085] By employing the aforementioned technical solution, this approach leverages deep learning and multi-source heterogeneous information fusion technology to achieve high-precision perception and identification of bird information in farmland, significantly improving the targeting and effectiveness of the final early warning strategy. Furthermore, by constructing multi-source time-series data based on the first and second time-series models and formulating the final deterrence strategy guided by bird activity rhythms, interference with the birds' normal physiological rhythms is minimized. Integrating multiple deterrence methods such as lasers, ultrasound, and flashlights greatly extends the deterrence duration. Simultaneously, the deterrence strategy optimization based on a multi-objective optimization model enables the system to possess autonomous learning and adaptive capabilities, significantly reducing the need for manual intervention and lowering maintenance costs by more than 60% compared to traditional methods.

[0086] The above description of the invention is merely an overview of the technical solution of the present invention. In order to enable those skilled in the art to better understand the technical solution of the present invention and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of the present invention easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of the present invention. Attached Figure Description

[0087] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on the present invention.

[0088] In the accompanying drawings of the instruction manual:

[0089] Figure 1 The generation principle diagram of the first timing module is shown in the specific implementation diagram;

[0090] Figure 2 This is a schematic diagram illustrating the generation principle of the second timing module in a specific implementation method;

[0091] Figure 3 This is a schematic diagram illustrating the generation principle of the final expulsion strategy described in a specific implementation method.

[0092] Figure 4 This is a schematic diagram illustrating the generation principle of the final early warning strategy described in this specific embodiment;

[0093] Figure 5 This is a schematic diagram of the agricultural intelligent bird-repelling system described in a specific implementation. Detailed Implementation

[0094] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this invention in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this invention and are therefore intended only as examples, not as limiting the scope of protection of this invention.

[0095] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this invention, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0096] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit the invention.

[0097] In the description of this invention, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " generally indicates that the preceding and following objects have an "or" logical relationship.

[0098] In this invention, terms such as “first” and “second” are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy, or order between these entities or operations.

[0099] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this invention is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0100] Similar to the understanding in the Examination Guidelines, in this invention, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments of this invention, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0101] In the description of the embodiments of the present invention, the spatial related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," "circumferential," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of the present invention or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of the present invention.

[0102] Unless otherwise explicitly stated or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this invention, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral arrangement; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this invention according to the specific circumstances.

[0103] Please see Figures 1 to 5 In a first aspect, this embodiment provides an AI-based intelligent bird-repelling method for agriculture, comprising:

[0104] Acquire a first image sequence within a farmland scene, wherein the first image sequence consists of multiple first images within a first preset sampling period;

[0105] The first image sequence is input into the first convolutional neural network model for feature extraction to obtain the first image feature information, which includes at least one of multiple first bird activity trajectories and multiple first bird behavior patterns in the current first image sequence; and simultaneously, thermal infrared information in the farmland scene is acquired, which consists of multiple thermal infrared sequences within a first preset sampling period.

[0106] The thermal infrared information is input into the second convolutional neural network model for feature extraction to obtain the second thermal infrared feature information, which includes at least one of the multiple second bird activity trajectories and multiple second bird behavior patterns in the current thermal infrared information.

[0107] Obtain bird attribute information associated with farmland scenes, construct a bird attribute semantic dictionary, map attributes to semantic vectors through attribute embedding matrix, and obtain third attribute semantic vectors. The third attribute semantic vectors include at least one of multiple third bird activity trajectories and multiple third bird behavior patterns in the current bird attribute information.

[0108] The first image feature information, the second thermal infrared feature information, and the third attribute semantic vector are input into the multimodal gated fusion unit for feature interaction and fusion to obtain a semantically enhanced multimodal fusion feature sequence;

[0109] The multimodal fusion feature sequence is input into a long short-term memory network to model the temporal dependency of bird activities, resulting in a first temporal model. The first temporal model includes first bird temporal information, which includes the bird's activity trajectory and behavior pattern in the first time series.

[0110] Acquire a second image sequence of a farmland scene within a second preset sampling period. The second image sequence consists of multiple second images within the second preset sampling period.

[0111] The second image sequence is input into the third convolutional neural network model for feature extraction to obtain the fourth image feature information. Based on the fourth sequence feature information, the bird category information is obtained, and the fourth bird activity trajectory and fourth bird behavior pattern are obtained based on the second image sequence.

[0112] The second time series model is obtained by encoding the activity trajectory, behavior pattern, and bird category information of the fourth bird. The second time series model includes the time series information of the second bird, which includes the activity trajectory and behavior pattern of the bird in the second time series.

[0113] The final deterrence strategy and / or final early warning strategy are generated based on the first time series model, bird category information, and the second time series model.

[0114] For further details, please refer to Figure 1 In some embodiments, the first image sequence is input into the first convolutional neural network model for feature extraction, and the first image feature information is represented by formula (1), which is as follows:

[0115] F img =f cnn (I;θ cnn );

[0116] In formula (1), F img For the first image feature information, f cnn For feature extractor, θ cnn Here, represents the learning parameters of the first convolutional neural network model, and I represents the first image sequence. T is the first preset sampling period;

[0117] The first convolutional neural network model is generated based on the convolutional neural network model. The performance evaluation of the first convolutional neural network model is obtained through the cross-entropy loss function, which is expressed by formula (2). Formula (2) is as follows:

[0118]

[0119] In formula (2), L is the loss function value, y i To determine the true categories of the sample bird data during the training of the first convolutional neural network model, The probability distribution for predicting sample bird data during the training of the first convolutional neural network model;

[0120] The thermal infrared information is input into the second convolutional neural network model for feature extraction. The second thermal infrared feature information is represented by formula (3), which is as follows:

[0121] F ir =f cnn (H;θ ir );

[0122] In formula (3), F ir For the second thermal infrared feature information, f cnn For feature extractor, θ ir Here, H represents the learning parameters of the second convolutional neural network model. T is the first preset sampling period.

[0123] In this embodiment, a feature extractor is used to obtain the first image sequence. To fully extract spatial structural information from the images, the system employs a convolutional neural network (CNN) to extract features from the first image sequence:

[0124] F img =fcnn(I; θ cnn )

[0125] Where, f cnn θ represents the feature extractor. cnn F represents the learning parameters of the first neural network model. img The first feature information is extracted.

[0126] The first neural network model achieves high-precision identification of birds such as egrets, sparrows, and crows by extracting spatial features from images. When using the first neural network model for image recognition, the performance of the model is evaluated using the cross-entropy loss function, defined as follows:

[0127]

[0128] Where y i It is a real category. It is the probability predicted by the model.

[0129] In this formula, L represents the Cross-Entropy Loss function. It is a commonly used loss function for evaluating the performance of classification models. In this formula, L calculates the probability predicted by the model. Compared with the real category y i The differences between them.

[0130] Specifically, for each sample i, y i It is the true category of the sample, and This represents the probability distribution predicted by the model for that sample. In the formula... This means taking the logarithm of the predicted probability to convert it to a positive value. Then, the log probability of each true class is multiplied by -1; the negative sign is used to convert the loss value to a positive value. Finally, the loss values ​​of all samples are summed to obtain the overall loss function value L.

[0131] The cross-entropy loss function can be used for multi-class classification problems. By minimizing this loss function, the model can better fit the training data and improve the performance of the classification model.

[0132] The system uses classic CNN architectures (such as ResNet and DenseNet) as the backbone network and makes targeted fine-tuning on farmland scene data to improve the scene adaptability of feature representation.

[0133] In terms of thermal infrared modes, the system utilizes a thermal imager to acquire thermal infrared information sequences of farmland scenes.

[0134] Unlike visible light images, thermal infrared information primarily reflects the temperature distribution on an object's surface, offering a unique advantage in capturing biological features (such as those of birds). The system employs a similar CNN architecture to extract feature representations of the second feature information:

[0135] F ir =f cnn (H;θ ir );

[0136] Where θir represents the learning parameters of the second neural network model. Considering the differences in imaging principles between thermal infrared information and the first image, the system has made adaptive adjustments to the design of the feature extractor, such as using smaller convolutional kernel sizes and shallower network layers, to adapt to the low resolution and high noise characteristics of thermal infrared images.

[0137] Please see Figure 1 In some embodiments, the bird attribute semantic dictionary is obtained through the following steps:

[0138] A semantic dictionary is constructed, which is represented by formula (4), as follows:

[0139] D = d1, d2, ..., d N ;

[0140] In formula (4), D is the semantic dictionary, and d i These are attribute words, and each attribute word represents a bird attribute.

[0141] For each attribute word, a semantic embedding vector is matched to form an attribute embedding matrix, thus forming a semantic representation space;

[0142] The attribute embedding matrix is ​​represented by formula (5), which is as follows:

[0143] E = [e1, e2, ..., e N ];

[0144] In formula (5), E is the attribute embedding matrix, e i For semantic embedding vectors, the semantic embedding vectors represent the positions of attribute words in the semantic representation space.

[0145] In this embodiment, to further enhance the semantic representation capability of bird targets, the system also introduces bird attribute information based on prior knowledge. This is achieved by constructing a semantic dictionary of bird attributes D = d1, d2, ..., d... N Each attribute word d i Corresponding to a semantic embedding vector e i The system utilizes the attribute embedding matrix E = [e1, e2, ..., e N This maps discrete attributes to a continuous semantic representation space.

[0146] Specifically, the system enhances the semantic representation capability of bird targets by introducing bird attribute information based on prior knowledge: First, the system constructs a semantic dictionary of bird attributes, which contains N attribute words d. i Each attribute word corresponds to a semantic embedding vector e. i These attribute words can describe the characteristics or properties of a bird, such as color, size, and wing shape.

[0147] Next, the system utilizes the attribute embedding matrix E = [e1, e2, ..., e NThe system maps discrete attributes to a continuous semantic representation space. The attribute embedding matrix is ​​a matrix composed of semantic embedding vectors corresponding to attribute words, with each column representing a semantic embedding vector for one attribute word. By mapping discrete attributes to a continuous semantic representation space, the system can better express the attribute features of flying birds.

[0148] In this way, the system can provide a more accurate and comprehensive description of bird targets in the semantic representation space. By using an attribute embedding matrix, the system can transform discrete attribute information into a continuous semantic representation, thereby better capturing the features and characteristics of bird targets. This enhanced semantic representation capability helps the system better understand and handle tasks and problems related to bird targets.

[0149] Based on this, the logical steps for constructing a semantic dictionary of bird attributes can be referenced in the following example:

[0150] 1. Collect information about the attributes of birds as prior knowledge;

[0151] 2. Based on the collected attribute information, construct a semantic dictionary D, where each attribute word d i Represents a bird attribute;

[0152] 3. For each attribute word d i Assign a corresponding semantic embedding vector e to it. i This vector represents the position of the attribute in the semantic space;

[0153] 4. For each attribute word d i Its corresponding semantic embedding vector e i Pair them up and form them into an attribute embedding matrix E = [e1, e2, ..., e N ];

[0154] 5. By using the attribute embedding matrix E, discrete bird attributes are mapped to a continuous semantic representation space, thereby enhancing the semantic representation capability of bird targets.

[0155] Through the above logical steps, a semantic dictionary of bird attributes can be constructed, in which each attribute word corresponds to its corresponding semantic embedding vector, thereby enhancing the semantic representation capability of bird targets.

[0156] Please see Figure 1 In some embodiments, the first image feature information, the second thermal infrared feature information, and the third attribute semantic vector are input into the multimodal gating fusion unit for feature interaction and fusion, and the semantically enhanced multimodal fusion feature sequence is represented by formulas (6) to (8), where formula (6) is as follows:

[0157] F′ img=F img +f att (F img ,F ir ,E)⊙F ir ;

[0158] Formula (7) is as follows:

[0159] F′ ir =F ir +f att (F ir ,F img ,E)⊙F img ;

[0160] Formula (8) is as follows:

[0161] F fused =W1F′img+W2F′ ir +W3(F′ img ⊙F′ ir );

[0162] In formulas (6) to (8), F' img F is the first feature information after fusion. img For the first image feature information, f att Let F' be the first attention mechanism function, E be the attribute embedding matrix, and F' be the function of the first attention mechanism. ir F is the second feature information after fusion. fused W1, W2, and W3 are the first fused feature information, and W1, W2, and W3 are the learning parameters of the fused gating.

[0163] In this embodiment, after obtaining the first image feature information, the second thermal infrared feature information, and the third semantic attribute vector, the system realizes the interaction and fusion of different modal information through the multimodal feature fusion module.

[0164] For a more detailed understanding, please refer to the example below: Constructing a multimodal fusion algorithm model based on an attention mechanism:

[0165]

[0166] Where, f att The first attention mechanism function is used to dynamically calculate the correlation weights between features of different modalities. W1, W2, and W3 are the learning parameters of the fusion gating, used to control the importance of features of different modalities. Through the attention mechanism and gating unit, MGFU can adaptively adjust the interaction intensity of information from different modalities and generate a semantically enhanced cross-modal fusion feature representation F. fused .

[0167] Please see Figure 1In some embodiments, the multimodal fused feature sequence is input into a long short-term memory network to model the temporal dependencies of bird activity, resulting in a first temporal model. The first temporal model includes first bird temporal information, which includes the bird's activity trajectory and behavioral patterns in the first time series.

[0168] The multimodal fusion feature sequence is encoded using a first long short-term memory network, and expressed by formula (9):

[0169]

[0170] In formula (9), h t and c t Let f represent the hidden state and memory cell state of the first long short-term memory network at time t, respectively. lstm For the first long short-term memory network function, θ lstm These are the parameters of the first long short-term memory network;

[0171] The first time series model is trained using a multi-task learning mode to obtain the trained first time series model. The training process is represented by formula (10):

[0172]

[0173] In formula (10), This is the total loss function value. This represents the loss function for the bird detection task. This represents the loss function for the bird tracking task. Let λ1, λ2, and λ3 represent the loss function for the bird behavior recognition task, where λ1, λ2, and λ3 are weighting coefficients that balance different tasks.

[0174] In this embodiment, based on multimodal feature fusion, the system further introduces a first temporal model to capture the continuity and correlation of bird activity over time. The system uses a Long Short-Term Memory (LSTM) network to encode the fused feature sequence:

[0175]

[0176] Where, h t and c t represents the hidden state and memory cell state of the LSTM at time t, respectively, capturing the temporal dependency of bird activity.

[0177] In multimodal feature fusion, the system integrates features from different sensors or modalities to obtain more comprehensive and richer information. These features may include different types of data such as images, sound, and motion. To achieve temporal modeling, the system chose LSTM as the encoder. LSTM is a recurrent neural network capable of processing and modeling temporal data. By inputting the fused feature sequence into the LSTM network, the system can learn and capture temporal patterns and correlations in the sequence. Through the LSTM encoding process, the system can transform the fused feature sequence into a hidden state representation with temporal modeling capabilities. The hidden state can be used for subsequent tasks, such as predicting bird behavior or identifying bird states.

[0178] In summary, by introducing a time-series modeling module and using LSTM to encode the fused feature sequence, this embodiment enables the system to better capture the continuity and correlation of bird activities in the time dimension, thereby improving the understanding and prediction capabilities of bird behavior.

[0179] Specifically, during the training phase, the system adopts a multi-task learning paradigm, simultaneously optimizing the objective functions for tasks such as bird detection, tracking, and behavior recognition.

[0180]

[0181] in, and Let λ1, λ2, and λ3 represent the loss functions for the bird detection, bird tracking, and bird behavior recognition tasks, respectively, and let λ1, λ2, and λ3 be the weighting coefficients balancing the different tasks. Through end-to-end joint optimization, the system achieves deep fusion of cross-modal information and collaborative enhancement between tasks.

[0182] Furthermore, to further enhance the model's robustness to various environmental changes, the system also incorporates data augmentation strategies, such as random cropping, rotation, scaling, and other image transformations, as well as noise simulation of thermal infrared information. Simultaneously, the system employs transfer learning, initializing the feature extractor using a CNN model pre-trained on large-scale datasets like ImageNet, and then fine-tuning it on farmland monitoring data to alleviate the challenge of learning from small samples.

[0183] This embodiment utilizes multiple sensors to acquire information about farmland scenes, including images, thermal infrared data, and semantic attributes. The system employs deep learning methods for feature extraction and fusion to achieve accurate perception of bird activity. Specifically, the system uses a convolutional neural network to extract features from image and thermal infrared data, fuses multimodal information through attention mechanisms and gating units, and encodes the fused features using a temporal modeling module. During the training phase, the system optimizes bird detection, tracking, and behavior recognition tasks through multi-task learning. Furthermore, the system employs data augmentation and transfer learning methods to enhance the model's robustness and generalization ability.

[0184] Please see Figure 2 In some embodiments, the second image sequence is input into a third convolutional neural network model for feature extraction to obtain fourth image feature information. Based on the fourth image feature information, bird category information is obtained. Furthermore, based on the second image sequence, the activity trajectory and behavioral pattern of the fourth bird are obtained, including:

[0185] A second attention mechanism function is added to the third convolutional neural network model, and it is expressed by formula (11), which is as follows:

[0186] F = A⊙X;

[0187] In formula (11), F is the attention enhancement feature in the second attention mechanism function, A is the attention weight matrix, X is the convolution feature map in the third convolutional neural network model, and ⊙ is element-wise multiplication;

[0188] The attention weight matrix is ​​represented by formula (12), which is as follows:

[0189]

[0190] In formula (12), W a and W x These are the learning parameters in the second attention mechanism function.

[0191] In this embodiment, for bird recognition, the system is based on a Convolutional Neural Network (CNN) and designs a bird recognition model for fine-grained classification. This model draws on classic CNN structures (such as VGG and ResNet), employing multi-layer convolution and pooling operations to extract hierarchical feature representations from the second image sequence. Simultaneously, to further enhance the model's ability to characterize subtle features, an attention mechanism is introduced at the network's end. By learning the saliency weight distribution of different parts of the bird (such as the head, wings, and tail), the recognition model can adaptively allocate the importance of features, focusing on the most discriminative local regions, thereby improving the accuracy of fine-grained classification.

[0192] Specifically, the second attention mechanism function adaptively weights the convolutional feature map X by introducing a weight matrix A, resulting in an attention-enhanced feature representation F:

[0193] F = A⊙X;

[0194] Where ⊙ denotes element-wise multiplication. The attention weight matrix A is obtained by learning the similarity between each position in the feature map and the standard saliency template:

[0195]

[0196] W a and W x These are the learning parameters in the second attention mechanism function. Through end-to-end joint optimization, the attention mechanism and convolutional feature extraction achieve synergistic enhancement, resulting in a significant performance improvement for the recognition model on fine-grained bird classification tasks.

[0197] Please see Figure 2 In some embodiments, a second time-series model is obtained by encoding the activity trajectory, behavioral pattern, and bird category information of the fourth bird. The second time-series model includes second bird time-series information, which includes the bird's activity trajectory and behavioral pattern in the second time series.

[0198] The activity trajectory, behavior pattern, and bird category information of the fourth bird were encoded using a second long short-term memory network and expressed by formulas (13) to (18). Formula (13) is as follows:

[0199] i t =σ(W ii x t +W hi h t-1 +b i );

[0200] Formula (14) is as follows:

[0201] f t =σ(W if xt+W hf h t-1 +b f );

[0202] Formula (15) is as follows:

[0203] o t =σ(W io x t +W ho h t-1 +b o );

[0204] Formula (16) is as follows:

[0205]

[0206] Formula (17) is as follows:

[0207]

[0208] Formula (18) is as follows:

[0209] h t ′=o t ⊙tanh(C t );

[0210] In formulas (13) to (18), i t f is the input gate in the second long short-term memory network. t For the forgetting gate in the second long short-term memory network, o t As the output gate in the second long short-term memory network, C t This refers to the cellular state in the second long short-term memory network. For candidate cell states in the second long short-term memory network, h t 'W' represents a hidden state in the second long short-term memory network. ii W hi Let b be the weight matrix of the input gate. i W is the bias vector of the input gate. if W hf Let b be the weight matrix of the forget gate. f W is the bias vector for the forget gate. io W ho Let b be the weight matrix of the output gate. o This is the bias vector for the output gate;

[0211] The second time series model is trained using a multi-task learning mode to obtain the first time series model after training. The training process is represented by formula (19):

[0212] L'=λ'1L recog +λ'2L' behav ;

[0213] In formula (19), L′ is the total loss function value. Let L' be the loss function for the bird recognition task. behav Let λ'1 and λ'2 be the loss function for the bird behavior understanding task, and let λ'1 and λ'2 be the weight coefficients that balance the two tasks.

[0214] In this embodiment, for understanding bird behavior, the system designs a behavior analysis framework based on spatiotemporal sequence modeling. For continuous bird movement sequences, the system employs a Long Short-Term Memory (LSTM) network to model their temporal dependencies. The second LSTM network, by introducing gating mechanisms (including input gate, forget gate, and output gate), can effectively capture the long-distance dependencies of bird movement in the temporal dimension, thereby characterizing complex dynamic behavioral patterns such as flight trajectories and feeding rhythms. Given a bird movement sequence x1, x2, ..., x... T LSTM updates the hidden state h step by step t To integrate historical information:

[0215] i t =σ(W ii x t +Whih t-1 +b i );

[0216] f t =σ(W if xt+W hf h t-1 +b f );

[0217] o t =σ(W io x t +W ho h t-1 +b o );

[0218]

[0219] h t =o t ⊙tanh(C t );

[0220] Where, i t f t o t C represents the input gate, forget gate, and output gate in the second long short-term memory network, respectively. tThese are memory cells in the second long short-term memory network (LSTM). The LSTM dynamically adjusts the gating states to adaptively model the long-short-term dependencies in bird behavior sequences. In specific deterrence application scenarios, the system also introduces an attention mechanism, learning the importance weights of different behavioral states on deterrence decisions, further enhancing the relevance and interpretability of behavior analysis. For model training, the system employs an end-to-end joint optimization strategy. The bird recognition and behavior understanding models share the underlying convolutional feature extraction network, enabling full information interaction in the feature space. Simultaneously, the system introduces a multi-task learning paradigm, connecting recognition and behavior understanding tasks through a shared feature representation layer and constructing a joint loss function.

[0221] L'=λ'1L recog +λ'2L' behav

[0222] in, and L' behav Let λ'1 and λ'2 represent the loss functions for the bird recognition and behavior understanding tasks, respectively, and λ'1 and λ'2 be the weight coefficients balancing the two tasks. Through end-to-end backpropagation optimization, the recognition model and the behavior analysis model achieve synergistic enhancement, uncovering the intrinsic correlation between appearance features and behavioral patterns at the data level.

[0223] In this embodiment, to further improve the robustness and generalization performance of the model, the system also incorporates data augmentation and transfer learning strategies during training. Data augmentation expands the diversity of training samples by randomly rotating, scaling, and cropping the original bird images, enhancing the model's adaptability to changes in lighting and perspective. Transfer learning utilizes pre-trained CNN models (such as ResNet) on large general datasets like ImageNet for parameter initialization and fine-tunes them in bird-driving scenarios, significantly accelerating the model's convergence speed and improving performance.

[0224] This embodiment extracts feature representations from bird images through multi-layer convolution and pooling operations, and introduces an attention mechanism to enhance the ability to characterize subtle features. In bird recognition, the model focuses on the most discriminative local regions through the attention mechanism, improving the accuracy of fine-grained classification. For bird behavior understanding, the model uses LSTM to model the temporal dependencies of bird movement sequences, capturing dynamic behavioral patterns. Simultaneously, an attention mechanism is introduced to learn the importance weights of behavioral states on deterrence decisions, enhancing the relevance and interpretability of behavioral analysis. Model training employs an end-to-end joint optimization strategy, sharing the convolutional feature extraction network, and introducing a multi-task learning paradigm to balance recognition and behavior understanding tasks. To improve the model's robustness and generalization performance, the system also employs data augmentation and transfer learning strategies. Experimental results show that the system achieves good performance in fine-grained bird recognition and bird deterrence behavior warning tasks.

[0225] In practical applications, through the collaborative design and optimization of bird intelligent recognition and behavior understanding models, this system has constructed a behavioral intelligent analysis engine for bird control in farmland. Experimental results show that the analysis engine achieved a Top-1 accuracy of 92.5% in fine-grained bird recognition tasks, and also demonstrated good performance in early warning of foraging behaviors of typical harmful birds such as egrets and sparrows, capturing high-risk dynamics on average 2-3 observation cycles in advance.

[0226] Please see Figure 3 In some embodiments, generating the final expulsion strategy based on the first time-series model and the second time-series model further includes:

[0227] An intelligent regulation model is constructed, and the initial driving strategy is obtained by inputting observation information into the intelligent regulation model. The observation information includes environmental parameters, the first time series model, the second time series model, and bird category information.

[0228] The initial driving strategy is evaluated for its effectiveness, and the initial driving strategy is optimized based on the evaluation to obtain the final driving strategy.

[0229] The final driving strategy is expressed by formula (20), which is as follows:

[0230]

[0231] In formula (20), π * (s) represents the final driving strategy, and s represents the observed state. Let Q(s,a;θ) be the set of device parameters for the driving equipment used in the intelligent regulation model, where a is the driving action in the final driving strategy. * ) is the state-action value function, θ * These are the optimal parameters for the state-action value function.

[0232] In this embodiment, the system takes environmental parameters (such as temperature, humidity, wind speed, etc.), a first time-series model, a second time-series model, and bird category information (species, quantity, distribution, activity patterns, etc.) and deterrence effect evaluation (such as deterrence rate, warning distance, duration, etc.) as inputs, and gradually optimizes the deterrence strategy through a reinforcement learning agent. The agent's observed state s is provided by a multi-source heterogeneous information fusion module, including multi-dimensional continuous states (environmental parameters) and discrete states (such as bird species). The agent's action space It consists of parameters of each driving subsystem, such as laser power, ultrasonic frequency, and flash brightness, which are used to execute actions. The system adjusts various deportation methods. The reward r given to the agent by the environment is provided by the deportation effectiveness evaluation model, comprehensively considering the effectiveness and energy consumption of deportation. The system uses a Deep Q-Learning algorithm based on value function approximation to train the agent. By continuously interacting with the environment and updating the state-action value function Q(s,a;θ), the agent eventually learns the optimal deportation strategy.

[0233] Please see Figure 3 In some embodiments, generating the final early warning strategy based on the first time series model and the second time series model further includes:

[0234] Multi-source time series data were constructed based on the first and second time series models. The multi-source time series data included agricultural ecological calendar, bird migration calendar and meteorological calendar.

[0235] Based on multi-source time series data, a macroscopic temporal pattern of bird damage is generated, and a multi-scale grid is divided into farmland scenes to generate a spatial distribution heat map of farmland bird damage.

[0236] A multi-objective optimization model is constructed. The input data of the multi-objective optimization model are multi-source time series data, spatial distribution heat map, bird species diversity index and degree of human intervention. The output of the multi-objective optimization model is the final early warning strategy.

[0237] In this embodiment, to accurately grasp the spatiotemporal dynamics of bird damage in farmland, a multi-scale, multi-granularity farmland bird perception model was constructed. In the time dimension, the system integrates multi-source time-series data, including agricultural ecological calendars, bird migration calendars, and meteorological calendars, to characterize the macro-temporal patterns of bird damage. Studies show that over 90% of farmland bird damage in China is concentrated in spring (March-May) and autumn (September-November), with spring being the most prevalent. By performing spatiotemporal clustering on bird monitoring data from 600 observation points nationwide over five consecutive years, the system constructed a digital "bird calendar" for farmland bird damage, dividing my country into 10 bird damage zones, including Northeast, North China, Northwest, and the middle and lower reaches of the Yangtze River. The system also meticulously characterized the key phenological periods of high bird damage incidence in each zone, with an average positioning accuracy better than 3 days.

[0238] For example, in terms of spatial dimension, the system employs a multi-scale gridding method to divide farmland areas into nested grids (e.g., 100 meters, 500 meters, 1 kilometer). Within each grid, the system deploys multimodal bird deterrent devices using sound, light, and electricity, along with bird monitoring units (such as infrared cameras and sound sensors) to achieve real-time monitoring of bird activity within the grid. Through spatial interpolation and rule tree mapping of multi-grid bird data, the system can generate a spatial distribution heat map of bird damage in farmland, intuitively reflecting the spatial differentiation characteristics of bird damage at different scales, including local, regional, and global. Measured data shows that the bird damage heat map generated by this system has a spatial resolution better than 80% at the 100-meter grid scale, effectively guiding zoning measures in farmland and the precise deployment of bird deterrent devices.

[0239] Based on accurate perception of bird activity in farmland, the system further constructs an intelligent optimization-based ecological regulation model for bird control in farmland. Tailored to different agricultural seasons and regional characteristics, this model can adaptively adjust the operating parameters of the multimodal bird control devices (such as frequency, intensity, and duration of bird deterrence) to effectively control bird damage while minimizing bird control costs and ecological disturbance.

[0240] Specifically, this regulation model uses farmland bird monitoring data as input and constraints such as crop bird damage control rate, bird species diversity index, and human intervention rate to construct a multi-objective optimization problem:

[0241]

[0242] g i (x)≤0,i=1,2,…,ph j (x)=0,j=1,2,…,qx min ≤x≤x max This means minimizing a vector function F(x), where F(x) contains m objective functions f. i(x). x is the parameter vector of the bird deterrent device, F(x) is a vector function that includes multiple optimization objectives such as bird deterrent cost and ecological impact, g i (x) and h j (x) represent the inequality and equality constraints, respectively. This model uses a multi-objective evolutionary algorithm (such as NSGA-III) to optimize the bird deterrence parameters, selects the optimal parameter combination that satisfies the constraints on the Pareto front, and distributes it to each bird deterrence node for execution.

[0243] In areas with high rates of bird damage in rice paddies, the system underwent a trial demonstration of the control model over a one-season growing season. Through weekly online parameter optimization, the bird damage rate was effectively controlled to below 5%, a 3 percentage point reduction compared to conventional bird deterrence methods. Meanwhile, the populations of more than 10 common farmland bird species, such as sparrows and egrets, remained largely unchanged compared to before the trial, and the Shannon diversity index remained at a high level above 2.5. Compared to manual bird deterrence, intelligent control reduced manual input by more than 90% and overall costs by more than 60%.

[0244] A Bird Repelling Strategy Guided by Bird Activity Rhythms in Farmland: To further reduce the impact of bird repelling behavior on bird physiological rhythms and population behavior, and to enhance the ecological friendliness of bird repelling, a novel bird repelling strategy guided by bird activity rhythms in farmland is systematically and innovatively proposed. This strategy uses key behavioral rhythms such as bird diurnal rhythms and seasonal migration as constraints. Bird repelling is concentrated during "ecological respite periods" when bird activity is relatively infrequent, while isolation and induction are the main methods during "ecologically sensitive periods" when bird activity is frequent, minimizing disturbance from repelling and maintaining the normal physiological rhythms of birds.

[0245] Studies have shown that the foraging activities of most farmland birds exhibit a distinct diurnal rhythm, concentrated in the two hours after sunrise and the one hour before sunset. Based on this, a "peak-avoidance deterrence" diurnal bird control strategy was designed, which involves increasing the frequency of deterrence during off-peak bird activity periods and appropriately reducing the intensity during peak periods. This strategy uses the bird's biological clock to regulate bird activity and reduce bird droppings pollution. Field data show that "peak-avoidance deterrence" reduces daytime bird droppings in farmland by 20-30% and reduces potential threats from predators by more than 15%.

[0246] On a seasonal scale, the system formulates a "southward retreat and northward avoidance" bird control strategy based on bird migration patterns and local conditions. Specifically, it involves increasing the driving force 1-2 weeks before the southward migration period to induce birds to migrate south earlier; while during the northward return period, habitat isolation and alternative food sources are used to attract birds and minimize their movement, providing them with a buffer adaptation period. Monitoring data shows that the "southward retreat and northward avoidance" strategy results in a staggered distribution of farmland birds and migratory populations in time and space, reducing the feeding rate of migratory birds in farmland by more than 25%.

[0247] Early Warning of Bird Damage in Farmland Based on Niche Simulation: To further expand the application value of this system in regional farmland bird situation early warning, the research group constructed a bird damage risk early warning model based on niche simulation. This model is supported by bird occurrence survey data and remote sensing habitat element data. It constructs a niche model of bird species distribution using the maximum entropy principle and, combined with bird damage sensitivity assessment, generates a regional-scale bird damage risk distribution map.

[0248] Niche models estimate species' environmental preferences using speciation site data and environmental variables. The models assume that environmental conditions at a species' known distribution site represent its niche requirements; therefore, the environmental combination closest to the known distribution site is most likely to provide suitable habitat for the species. This study selected 300 bird survey sampling points and extracted over 40 habitat factors within a 5-kilometer radius, including elevation, vegetation index, distance to water bodies, and farmland area ratio, as model inputs. The distribution probabilities of each species were obtained through iterative optimization.

[0249]

[0250] Where y represents the presence or absence of a species, x is an environmental variable vector, and f(x) is a niche function. Based on the species-habitat preference relationship, the model is spatially extrapolated to the entire study area to obtain a probability map of species distribution suitability. By weighted fusion of the various species distribution maps, the spatial distribution of regional bird richness can be generated. Based on the bird richness distribution map, the system further introduces a bird damage sensitivity assessment model to map the bird damage risk of farmland within the study area. The sensitivity assessment model comprehensively scores farmland from three dimensions: exposure, damage, and vulnerability, obtaining a bird damage risk index.

[0251] R = E × H × V;

[0252] Where E represents the degree of farmland exposure to harmful birds, H" represents the ability of harmful birds to damage crops, and V represents the vulnerability of crops after being damaged. Through AHP hierarchical analysis and expert scoring, a weight matrix of assessment indicators for each dimension was established, and spatial mapping of bird damage risk in farmland in the study area was carried out accordingly.

[0253] IoT-based Farmland Bird Monitoring System: To achieve real-time monitoring of bird activity and bird deterrence processes in farmland, a farmland bird monitoring system based on Low Power Wide Area Network (LPWAN) was designed. This system consists of a terminal sensing layer, a near-field communication layer, and a cloud platform layer. It can monitor bird activity dynamics and the operational status of bird deterrence devices in farmland at any time, and transmit the monitored big data to the cloud management platform in real time.

[0254] At the terminal sensing layer, the system deploys various bird monitoring units, such as infrared cameras, sound sensors, and droppings collectors, on different carriers including fields, agricultural machinery, and bird deterrent devices, to acquire multi-source data on bird activity in farmland, including images, sounds, and droppings. Each monitoring unit is powered by solar energy and can operate continuously for more than two years.

[0255] At the near-field communication layer, the system employs LPWAN communication protocols such as LoRa and NB-IoT to transmit data collected by the monitoring unit to regional aggregation nodes with low power consumption. A single aggregation node can cover a radius of 5-10 kilometers, with a transmission rate of 10-50Kbps. The aggregation node intelligently filters and compresses the received multi-source heterogeneous data and transmits it back to the cloud platform via 4G / 5G networks.

[0256] At the cloud platform layer, the system has built an intelligent monitoring cloud platform for bird activity in farmland. This platform stores, normalizes, and integrates massive amounts of monitoring data uploaded from farmland, and provides intelligent application services such as data analysis and visualization. Users can view the overall bird activity in farmland in real time via a web browser or mobile app and obtain bird damage warnings for the next 1-7 days. The cloud platform can also remotely diagnose faulty devices and push maintenance plans based on the operating status data of bird deterrent devices.

[0257] Please see Figure 5 In a second aspect, this embodiment also provides an agricultural intelligent bird deterrence system based on AI recognition, applicable to the agricultural intelligent bird deterrence method of the first aspect. The system includes a control unit, a communication module, an energy supply module, a bird deterrence device, an image acquisition module, and an AI recognition module.

[0258] The control unit is used to execute the method described in the first aspect; the communication module includes at least one of a Bluetooth module, a Wi-Fi module, and an antenna, and is used to establish a communication connection with an external device; the energy supply module includes at least one of a solar panel, a micro wind turbine, and a battery pack, and is used to provide power; the bird deterrent device includes at least one of a laser emitter, an ultrasonic generator, and an LED flash array, and is used to execute a final bird deterrent strategy; the image acquisition module includes a high-definition camera and an infrared imager, and is used to acquire a first image sequence, a second image sequence, and thermal infrared information; the AI ​​recognition module is used to perform image processing on the first image sequence, the second image sequence, and the thermal infrared information, and is used to train and correct the first neural network model, the second neural network model, and the third neural network model.

[0259] The following are specific application examples based on the aforementioned system:

[0260] The laser-based, acoustic-optical-electric multimodal bird deterrence system intelligently combines various deterrence methods such as lasers, ultrasound, and flashlights to form a comprehensive and three-dimensional bird control system for farmland. Targeting the physiological characteristics and behavioral habits of different bird species, the system designs a multimodal deterrence strategy optimization framework based on reinforcement learning, achieving dynamic scheduling and synergistic efficiency of deterrence methods.

[0261] Laser deterrence subsystem: The laser deterrence system utilizes a directional high-energy laser beam to repel birds through thermal effects and visual disturbance. The system employs a laser parameter optimization model based on avian physiological characteristics. By adjusting parameters such as laser wavelength, power density, and pulse frequency, the system maximizes the deterrence effect without causing permanent harm to the birds. Specifically, the system selects a 532nm green laser, whose spectral distribution highly matches the visually sensitive areas of birds, creating a significant visual stimulus. Simultaneously, the system uses pulse modulation technology to convert continuous laser light into a frequency-variable pulse sequence, generating thermal and startling effects within a safe threshold. Experimental data shows that the 532nm laser at 10-50mW / cm²... 2 At a power density of 100 ppm, it is effective in driving away the vast majority of farmland birds. Continuous irradiation for 5-10 seconds can drive away more than 95% of flying birds.

[0262] Ultrasonic Bird Repelling Subsystem: Ultrasonic bird repelling utilizes birds' sensitivity to specific frequency sound waves. An array of ultrasonic speakers forms a highly directional sound beam, creating auditory interference for birds. The system employs an ultrasonic parameter optimization model based on bird auditory characteristics. By adjusting parameters such as ultrasonic frequency, sound pressure level, and modulation method, it achieves precise repelling of different bird species. Studies have shown that ultrasonic waves with frequencies between 20-30 kHz have a significant repelling effect on most farmland pests (such as sparrows and herons). The system uses phased array-based beamforming technology, with more than 40 ultrasonic transducers working collaboratively to form a directional ultrasonic beam, effectively covering 0.5-5 hectares of farmland. Field test data shows that at sound pressure levels above 120 dB, the system achieves a 92% effective repelling rate for birds such as sparrows, with an average repelling time of 2-3 seconds.

[0263] Flashing Light Bird Repelling Subsystem: This subsystem utilizes the sensitivity of birds' visual systems to changes in light intensity. A flash array generates high-frequency flashes, visually disturbing birds. The system employs a flash parameter optimization model based on bird visual characteristics. By adjusting parameters such as flash frequency, brightness, and wavelength, it achieves precise repelling of different bird species. Research has found that flashes with a frequency of 5-10Hz have a significant repelling effect on the vast majority of birds. The system uses a high-brightness LED array, with PWM control enabling precise modulation of flash frequency and brightness. Field deployment data shows that at flash brightness levels of 500-1000 lux, the system achieves a bird repellency rate of over 90%, with an average repelling distance of 50-100 meters.

[0264] To verify the effectiveness of multimodal coordinated deterrence, large-scale field tests were conducted in different seasons and crop-growing areas. Experimental plots covered typical farmland and orchards such as rice, grape, and corn, with a cumulative observation sample size exceeding 10,000 bird visits. Test data showed that in rice paddies infested with egrets, the combined effectiveness of the laser-ultrasound-flash ternary coordinated deterrence reached 98.5%, an improvement of 15.2% compared to single-modal deterrence. In wheat fields frequently infested with sparrows, the average deterrence duration of the laser-ultrasound-flash synergistic deterrence exceeded 2 hours, extending it by more than 50 minutes compared to single-modal deterrence. Compared to traditional manual deterrence, the system saves over 95% of manual time and reduces crop loss by more than 5 percentage points. Through intelligent multimodal coordinated deterrence, the system can autonomously adjust the parameter combinations of various deterrence methods based on the real-time status of bird damage, minimizing energy consumption while ensuring long-term deterrence. Real-world data shows that, compared to fixed parameter combinations, the adaptive multimodal deterrence strategy improves deterrence efficiency by 10% while reducing system energy consumption by 20-30%. From an eco-friendly perspective, strategic deterrence also significantly reduces the potential risk of harm to birds; optimized control of laser and ultrasound reduces the incidence of stress behaviors in birds to below 5%.

[0265] Modular design is a method of dividing a complex system into several independent, reusable functional units, aiming to reduce system coupling and improve maintainability and scalability. This system adopts a layered modular architecture, dividing the underlying hardware, middle-layer communication, and upper-layer applications into relatively independent subsystems, which are further subdivided into modules such as sensing, dispersal, edge computing, data transmission, and energy management.

[0266] Each functional module is connected via standardized interfaces, adopting a plug-and-play design philosophy. This loosely coupled architecture allows each module to be designed, developed, tested, and upgraded independently, significantly reducing the complexity of system integration. For example, the bird deterrent device can be flexibly equipped with laser, ultrasonic, and flashlight modules depending on the protected target; the edge computing module can be equipped with high-performance GPUs, FPGAs, and other accelerators to meet the computing power requirements of AI algorithms; and the data transmission module can flexibly select from multiple communication protocols such as Zigbee, LoRa, and NB-IoT according to communication distance and bandwidth requirements.

[0267] Furthermore, this system adopts a microservices architecture, breaking down business logic into fine-grained, loosely coupled service units. For example, functions such as bird perception, bird damage warning, bird deterrence strategies, and equipment management are encapsulated as independent microservices, and communication and collaboration between services are achieved through lightweight RESTful APIs. The microservices architecture enables the system to flexibly respond to changing requirements, and the development team can choose the most suitable technology stack for individual services, improving development efficiency and system performance.

[0268] To ensure interoperability and replaceability between modules, this system strictly adheres to common industrial interface standards. At the hardware level, an industrial-grade M12 interface is used for the physical connection between modules. This interface has reliable waterproof and dustproof performance (IP67), enabling it to withstand harsh agricultural environments. For data transmission, the system uses the RS-485 bus protocol, realizing serial communication and multi-point interconnection between modules. RS-485 has good anti-interference capabilities and long-distance transmission capabilities, with a maximum transmission distance of up to 1200 meters.

[0269] At the software level, the system is designed with a unified API specification based on a RESTful architecture. Each microservice provides a standardized HTTP interface, using JSON format for data serialization and deserialization. This web standards-based interface design enables seamless integration of services developed in different languages ​​and on different platforms, greatly improving the ease of system integration. Simultaneously, the system also introduces an API Gateway pattern, providing a unified entry point and access control for external calls, simplifying the complexity of interaction between clients and microservices.

[0270] Bird deterrent devices in farmland are typically deployed in areas far from the power grid, and a stable energy supply is a prerequisite for ensuring the long-term operation of the system. For farmland applications, this system has developed a diversified energy supply solution, comprehensively utilizing new energy technologies such as solar energy, wind energy, and environmental energy harvesting to minimize dependence on external power sources.

[0271] The system uses high-efficiency monocrystalline silicon solar panels as the main power generation unit, with a maximum output power of 50W per panel and a photoelectric conversion efficiency of over 22%. Considering the wind conditions in farmland environments, the system also integrates a micro wind turbine, which can output up to 20W per generator under an average wind speed of 3m / s. The system utilizes maximum power point tracking (MPPT) technology to adjust the operating status of the photovoltaic and wind turbines in real time, ensuring they always operate at their optimal power output points, thus improving energy conversion efficiency.

[0272] In low light and light wind conditions, the output of solar and wind energy is low, making it difficult to support the continuous operation of the equipment. Therefore, the system also integrates environmental energy harvesting technology, collecting minute amounts of energy from bird activity, environmental vibrations, and electromagnetic infrared radiation through piezoelectric, thermoelectric, and RF methods to provide supplementary power for low-power sensors and controllers. For example, the system deploys piezoelectric tiles around the bird deterrent device, utilizing the mechanical vibrations of birds landing to generate electricity; a single tile can produce 1.5mW of output power under a 100g load.

[0273] To balance the volatility of energy output and load demand, the system is equipped with a smart microgrid control unit. This unit employs a Model Predictive Control (MPC) algorithm, combining historical power generation data, load forecasts, and SOC estimation to optimize the power flow scheduling of solar, wind, and energy storage, minimizing the probability of power shortages. Simultaneously, the control unit also supports multiple safety protection measures, including overcharge and over-discharge protection and short-circuit protection, extending the lifespan of the energy storage batteries.

[0274] Through modular design and diversified power supply, this system achieves high flexibility, scalability, and continuous operation capability. Field test data shows that the system can operate stably for more than 18 months with an equipment failure rate of less than 1% and an energy deficit probability of less than 0.1%.

[0275] The above example has the following beneficial effects:

[0276] The system utilizes deep learning and multi-source heterogeneous information fusion technology to achieve high-precision perception and identification of bird activity in farmland. Experimental data shows that the system achieves a bird identification accuracy of 92.5% in complex farmland environments, significantly surpassing traditional methods. This accurate identification capability provides a reliable basis for subsequent bird control strategies, significantly improving the targeting and effectiveness of prevention and control.

[0277] By integrating multiple repelling methods such as lasers, ultrasound, and flashlights, the system overcomes the drawback of single-mode repelling that can easily lead to avoidance behavior. Field test data shows that in a paddy field environment, the overall effectiveness of multimodal synergistic repelling reaches 98.5%, an improvement of 15.2% compared to single-mode repelling. This synergistic effect greatly extends the duration of repelling, averaging over 2 hours.

[0278] The system innovatively proposes a bird-based activity rhythm-guided repelling strategy. Through measures such as "peak-avoidance repelling" and "southward and northward retreat," it minimizes interference with the birds' normal physiological rhythms. Monitoring data shows that this strategy reduces daytime bird droppings in farmland by 20-30%, while maintaining a relatively stable bird population size and diversity index, reflecting a good ecological balance.

[0279] The reinforcement learning-based expulsion strategy optimization enables the system to learn and adapt autonomously. Long-term operational data shows that, compared to fixed parameter combinations, the adaptive multimodal expulsion strategy reduces system energy consumption by 20-30% while maintaining expulsion efficiency. Simultaneously, IoT-based remote monitoring significantly reduces the need for manual intervention, lowering maintenance costs by over 60% compared to traditional methods.

[0280] The integrated ecological niche model provides a scientific basis for early warning of bird damage risks in regional farmland. Within a 600-square-kilometer experimental area, the model achieved a spatial positioning accuracy of over 80% for high-risk areas, enabling the issuance of early warning information one month in advance, thus providing strong support for macro-level decision-making by agricultural management departments.

[0281] The layered modular architecture and standardized interface design give the system strong scalability and adaptability. Field tests show that the system can be quickly deployed and optimized in different farmland environments, with a failure rate of less than 1% and can operate stably for more than 18 months.

[0282] Comprehensive evaluation shows that the system effectively controls bird damage while saving more than 95% of manpower time compared to traditional manual methods, and reducing crop loss by an average of 5 percentage points. Considering equipment investment and maintenance costs, the payback period for the system in large-scale application scenarios is approximately 2-3 years.

[0283] As a typical application of smart agriculture, this system not only solves specific bird-related problems but also provides a model for the refined management of farmland ecosystems. The large amount of farmland ecological data generated by the system provides valuable data support for agricultural research and policy making, and powerfully promotes the digital and intelligent transformation of agriculture.

[0284] In summary, this AI-based intelligent agricultural bird control system demonstrates significant advantages in terms of technological innovation, eco-friendliness, economic benefits, and industry-driving effect, representing the future development direction of agricultural pest control.

[0285] In summary, this technical solution utilizes deep learning and multi-source heterogeneous information fusion technology to achieve high-precision perception and identification of bird information in farmland, significantly improving the pertinence and effectiveness of the final early warning strategy. Furthermore, by constructing multi-source time series data based on the first and second time series models and formulating the final deterrence strategy guided by bird activity rhythms, interference with the birds' normal physiological rhythms is minimized. By integrating multiple deterrence methods such as lasers, ultrasound, and flashlights, the duration of deterrence is greatly extended. Simultaneously, the deterrence strategy optimization based on a multi-objective optimization model enables the system to possess autonomous learning and adaptive capabilities, significantly reducing the need for manual intervention and lowering maintenance costs by more than 60% compared to traditional methods.

[0286] Finally, it should be noted that although the above embodiments have been described in the description and drawings of this invention, this should not limit the scope of patent protection of this invention. Any technical solutions that are based on the essential concept of this invention, utilize the content described in the description and drawings of this invention to make equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this invention.

Claims

1. An intelligent bird-repelling method for agriculture based on AI recognition, characterized in that, include: Acquire a first image sequence within a farmland scene, wherein the first image sequence consists of multiple first images within a first preset sampling period; The first image sequence is input into the first convolutional neural network model for feature extraction to obtain first image feature information. The first image feature information includes at least one of multiple first bird activity trajectories and multiple first bird behavior patterns in the current first image sequence. And simultaneously acquire thermal infrared information within the farmland scene, wherein the thermal infrared information is a sequence of multiple thermal infrared images within a first preset sampling period; The thermal infrared information is input into a second convolutional neural network model for feature extraction to obtain second thermal infrared feature information. The second thermal infrared feature information includes at least one of multiple second bird activity trajectories and multiple second bird behavior patterns in the current thermal infrared information. Obtain the bird attribute information associated with the farmland scene, construct a bird attribute semantic dictionary, and map the attributes to semantic vectors through the attribute embedding matrix to obtain a third attribute semantic vector. The third attribute semantic vector includes at least one of multiple third bird activity trajectories and multiple third bird behavior patterns in the current bird attribute information. The first image feature information, the second thermal infrared feature information, and the third attribute semantic vector are input into the multimodal gating fusion unit for feature interaction and fusion to obtain a semantically enhanced multimodal fusion feature sequence. The multimodal fusion feature sequence is input into a long short-term memory network to model the temporal dependency of bird activities, resulting in a first temporal model. The first temporal model includes first bird temporal information, which includes the bird's activity trajectory and behavior pattern in the first time series. Acquire a second image sequence of the farmland scene within a second preset sampling period, wherein the second image sequence consists of multiple second images within the second preset sampling period; The second image sequence is input into the third convolutional neural network model for feature extraction to obtain the fourth image feature information. Based on the fourth image feature information, the bird category information is obtained, and the fourth bird activity trajectory and fourth bird behavior pattern are obtained according to the second image sequence. The second time series model is obtained by encoding the fourth bird activity trajectory, the fourth bird behavior pattern and the bird category information. The second time series model includes the second bird time series information, which includes the bird activity trajectory and behavior pattern in the second time series. The final deterrence strategy and / or final early warning strategy are generated based on the first time series model, the bird category information, and the second time series model.

2. The AI-based intelligent bird-repelling method for agriculture according to claim 1, characterized in that, The first image sequence is input into the first convolutional neural network model for feature extraction, and the feature information of the first image is represented by formula (1), which is as follows: F img =f cnn (I;θ cnn ); In formula (1), F img For the first image feature information, f cnn For feature extractor, θ cnn Let I be the learning parameters of the first neural network model, and let I be the first image sequence. T is the first preset sampling period; The first convolutional neural network model is generated based on the convolutional neural network model. The performance evaluation of the first convolutional neural network model is obtained through the cross-entropy loss function, which is expressed by formula (2), as follows: In formula (2), L is the loss function value, y i To determine the true categories of the sample bird data during the training of the first convolutional neural network model, The probability distribution for predicting sample bird data during the training of the first convolutional neural network model; The thermal infrared information is input into the second convolutional neural network model for feature extraction, and the second thermal infrared feature information is represented by formula (3), which is as follows: F ir =f cnn (H;θ ir ); In formula (3), F ir For the second thermal infrared feature information, f cnn For feature extractor, θ ir Here, H represents the learning parameters of the second convolutional neural network model. T is the first preset sampling period.

3. The AI-based intelligent bird-repelling method for agriculture according to claim 1, characterized in that, The bird attribute semantic dictionary is obtained through the following steps: A semantic dictionary is constructed, which is represented by formula (4), as follows: D=d1,d2,...,d N ; In formula (4), D is the semantic dictionary, and d i The attribute term represents a bird attribute; For each attribute word, a semantic embedding vector is matched to form an attribute embedding matrix, thus forming a semantic representation space; The attribute embedding matrix is ​​represented by formula (5), which is as follows: E=[e1,e2,…,e N ]: In formula (5), E is the attribute embedding matrix, e i The semantic embedding vector represents the position of the attribute word in the semantic representation space.

4. The AI-based intelligent bird-repelling method for agriculture according to claim 1, characterized in that, The first image feature information, the second thermal infrared feature information, and the third attribute semantic vector are input into the multimodal gating fusion unit to obtain the multimodal fusion feature sequence, which is represented by formulas (6) to (8). Formula (6) is as follows: F img =F img +f att (F img ,F ir ,E)⊙F ir ; The formula (7) is as follows: F ir =F ir +f att (F ir ,F img ,E)⊙F img ; The formula (8) is as follows: F fused =W1F′img+W2F′ ir +W3(F′ img ⊙F′ ir ); In formulas (6) to (8), F' img F is the first feature information after fusion. img For the first image feature information, f att Let F' be the first attention mechanism function, E be the attribute embedding matrix, and F' be the function of the first attention mechanism. ir F is the second feature information after fusion. fused W1, W2, and W3 are the first fused feature information, and W1, W2, and W3 are the learning parameters of the fused gating.

5. The AI-based intelligent bird-repelling method for agriculture according to claim 4, characterized in that, The multimodal fused feature sequence is input into a long short-term memory network to model the temporal dependencies of bird activity, resulting in a first temporal model. This first temporal model includes first bird temporal information, which includes the bird's activity trajectory and behavioral patterns in the first time series. The multimodal fusion feature sequence is encoded using a first long short-term memory network, and expressed by formula (9): In formula (9), h t and c t Let f represent the hidden state and memory cell state of the first long short-term memory network at time t, respectively. lstm For the first long short-term memory network function, θ lstm These are the parameters of the first long short-term memory network; The first time series model is trained using a multi-task learning mode to obtain the trained first time series model. The training process is represented by formula (10): In formula (10), This is the total loss function value. This represents the loss function for the bird detection task. This represents the loss function for the bird tracking task. Let λ1, λ2, and λ3 represent the loss function for the bird behavior recognition task, where λ1, λ2, and λ3 are weighting coefficients that balance different tasks.

6. The AI-based intelligent bird-repelling method for agriculture according to claim 5, characterized in that, The second image sequence is input into a third convolutional neural network model for feature extraction to obtain fourth image feature information. Based on the fourth image feature information, bird category information is obtained, and the fourth bird activity trajectory and fourth bird behavior pattern are obtained based on the second image sequence, including: A second attention mechanism function is added to the third convolutional neural network model, and is expressed by formula (11), which is as follows: F = A⊙X; In formula (11), F is the attention enhancement feature in the second attention mechanism function, A is the attention weight matrix, X is the convolution feature map in the third convolutional neural network model, and ⊙ is element-wise multiplication; The attention weight matrix is ​​represented by formula (12), which is as follows: In formula (12), W a and W x These are the learning parameters in the second attention mechanism function.

7. The AI-based intelligent bird-repelling method for agriculture according to claim 6, characterized in that, Encoding is performed based on the fourth bird's activity trajectory, the fourth bird's behavior pattern, and the bird category information to obtain a second time-series model. The second time-series model includes second bird time-series information, which includes the bird's activity trajectory and behavior pattern in the second time series. The activity trajectory, behavior pattern, and category information of the fourth bird are encoded using a second long short-term memory network, and are expressed by formulas (13) to (18), wherein formula (13) is as follows: i t =σ(W ii x t +W hi h t-1 +b i ); The formula (14) is as follows: f t =σ(W if xt+W hf h t-1 +b f ); The formula (15) is as follows: o t =σ(W io x t +W ho h t-1 +b o ); The formula (16) is as follows: The formula (17) is as follows: The formula (18) is as follows: h t ’=o t ⊙tanh(C t ); In formulas (13) to (18), i t f is the input gate in the second long short-term memory network. t For the forgetting gate in the second long short-term memory network, o t As the output gate in the second long short-term memory network, C t This refers to the cellular state in the second long short-term memory network. For candidate cell states in the second long short-term memory network, h t 'W' represents a hidden state in the second long short-term memory network. ii W hi Let b be the weight matrix of the input gate. i W is the bias vector of the input gate. if W hf Let b be the weight matrix of the forget gate. f W is the bias vector for the forget gate. io W ho Let b be the weight matrix of the output gate. o This is the bias vector for the output gate; The second time series model is trained using a multi-task learning mode to obtain the first time series model after training. The training process is represented by formula (19): L'=λ'1L recog +λ'2L' behav ; In formula (19), L' is the total loss function value. Let L' be the loss function for the bird recognition task. behav Let λ'1 and λ'2 be the loss function for the bird behavior understanding task, and let λ'1 and λ'2 be the weight coefficients that balance the two tasks.

8. The AI-based intelligent bird-repelling method for agriculture according to claim 1, characterized in that, The final expulsion strategy generated based on the first time series model and the second time series model also includes: An intelligent regulation model is constructed, and observation information is input into the intelligent regulation model to obtain an initial driving strategy. The observation information includes environmental parameters, a first time series model, a second time series model, and the bird category information. The initial driving-away strategy is evaluated for its effectiveness, and the initial driving-away strategy is optimized based on the evaluation to obtain the final driving-away strategy. The final expulsion strategy is expressed by formula (20), which is as follows: In formula (20), π * (s) represents the final driving strategy, and s represents the observed state. Let Q(s,a;θ) be the set of device parameters for the driving equipment used by the intelligent adjustment model, where a is the driving action in the final driving strategy. * ) is the state-action value function, θ * These are the optimal parameters for the state-action value function.

9. The AI-based intelligent bird-repelling method for agriculture according to claim 1, characterized in that, The final early warning strategy generated based on the first time series model and the second time series model also includes: Multi-source time series data are constructed based on the first time series model and the second time series model. The multi-source time series data includes agricultural ecological calendar, bird migration calendar and meteorological calendar. Based on the multi-source time series data, a macroscopic temporal pattern of bird damage is generated, and the farmland scene is divided into multi-scale grids to generate a spatial distribution heat map of farmland bird damage. A multi-objective optimization model is constructed. The input data of the multi-objective optimization model are the multi-source time series data, spatial distribution heat map, bird species diversity index and degree of human intervention. The output of the multi-objective optimization model is the final early warning strategy.

10. An intelligent bird-repelling system for agriculture based on AI recognition, characterized in that, The agricultural intelligent bird-repelling method according to any one of claims 1 to 9, the system includes a control unit, a communication module, an energy supply module, a bird-repelling device, an image acquisition module, and an AI recognition module; The control unit is used to perform the method according to any one of claims 1 to 9; The communication module includes at least one of a Bluetooth module, a Wi-Fi module, and an antenna, and is used to establish a communication connection with an external device. The energy supply module includes at least one of a solar panel, a micro wind turbine, and a battery pack, and the energy supply system is used to provide power. The bird deterrent device includes at least one of a laser emitter, an ultrasonic generator, and an LED flash array, and is used to execute a final bird deterrent strategy. The image acquisition module includes a high-definition camera and an infrared imager. The image acquisition module is used to acquire a first image sequence, a second image sequence, and thermal radiation information. The AI ​​recognition module is used to perform image processing on the first image sequence, the second image sequence, and the thermal radiation information, and the AI ​​recognition module is used to train and correct the first neural network model, the second neural network model, and the third neural network model.

Images