Multi-mode warning and bird repelling method and system based on solar power supply

Abstract

The embodiment of the invention provides a multi-mode warning and bird repelling method and system based on solar power supply, and the method comprises the steps: controlling an image collection unit to collect a target object image comprising a target object when an object detection unit detects the target object in a monitoring region; controlling a distance measuring unit to detect the distance of the target object; identifying the target object image by using a preset target detection model to obtain the position and the target type of the target object; performing filtering processing on the position and the distance of the target object to obtain optimal distance estimation; determining a driving strategy for the target object according to the target type of the target object and the optimal distance estimation; and controlling a driving unit to drive the target object according to the driving strategy. According to the method and the device, accurate target identification can be realized by utilizing limited resources, and the driving effect is improved by adopting differentiated warning and driving strategies.

Description

Technical Field

[0001] This application relates to the field of bird deterrent technology, and in particular to a multimodal warning and bird deterrent method and system based on solar power. Background Technology

[0002] Bird activity on outdoor power facilities can easily cause faults in lines and facilities, affecting the safe and stable operation of the power grid. Traditional bird deterrents have limited computing resources, resulting in low accuracy in identifying different birds in complex and changing outdoor environments, false triggering, and a single bird deterrent mode, which birds can easily adapt to, leading to poor bird deterrence effects. Summary of the Invention

[0003] In view of this, the purpose of this application is to propose a multimodal warning and bird deterrence method and system based on solar power.

[0004] To achieve the above objectives, this application provides a solar-powered multimodal warning and bird-repelling method, including: When the object detection unit detects a target object within the monitored area, it controls the image acquisition unit to acquire an image of the target object, including the target object; and controls the ranging unit to detect the distance to the target object. The target object image is identified using a preset target detection model to obtain the location and type of the target object; The position and distance of the target object are filtered to obtain the optimal distance estimate; Based on the target type and optimal distance estimate of the target object, a driving strategy for the target object is determined; The control and driving unit drives away the target object according to the driving strategy.

[0005] Optionally, the feature extraction layer of the object detection model introduces a SimAM attention module, which is used to determine the attention weight of each neuron in the extracted feature map. The method is as follows: Calculating neurons using energy functions t energy value e t : (2) in, It is the channel excluding neurons t In addition, the mean of the pixel values ​​output by all other neurons. It is the channel excluding neurons t In addition, the variance of the pixel values ​​output by all other neurons, N For the number of neurons, x i For the firsti The pixel value output by each neuron. x t Neuron t Output pixel values, It's a hyperparameter; The energy values ​​are converted into attention weights using the Sigmoid function to obtain the neuron. t Attention weights for: (3).

[0006] Optionally, the target detection model includes a feature fusion layer, which is used to fuse the input multi-scale feature map and intermediate features extracted from the multi-scale feature map.

[0007] Optionally, the bounding box loss function of the object detection model is: (5) (6) (7) in, L WIoUv3 This represents the loss value of the bounding box loss function; r Gradient gain is used to focus the function on a prediction box of average quality, thus avoiding gradient explosion. IOU The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box. , These are the width and height dimensions of the smallest enclosing box that can simultaneously enclose both the predicted box and the ground truth box; x , y The center pixel coordinates of the predicted bounding box (anchor box). , The coordinates of the center point of the true bounding box (pixel coordinates). The loss value for the current predicted bounding box. These are steady-state baseline values, typically taken as the mean within the sliding window or the loss value from the previous iteration; δ, α, β It is used to adjust gradient gain. r Hyperparameters of scale and distribution.

[0008] Optionally, the position and distance of the target object are filtered to obtain the optimal distance estimate, including: Based on the location and distance of the target object, an observation vector is constructed; The optimal position estimate of the target object is predicted based on the observation vector using the Kalman filter method. Based on the optimal location estimate, the corresponding optimal distance estimate is determined.

[0009] Optionally, the step of using the Kalman filter method to predict the optimal location estimate of the target object based on the observation vector includes: Based on the observed vectors, the Kalman filter method is used to predict the optimal state estimate: (18) in, z k Let be the observation vector. K k For Kalman gain, H For the observation matrix, for k The state vector at any given time, wherein the state vector includes the position and velocity of the target object. for k The optimal state estimate at time -1 for k The optimal state estimate at time t is the optimal state estimate, which includes the optimal position estimate.

[0010] Optionally, the driving strategy includes warning distances, danger distances, and multiple driving modes corresponding to different types of target objects; Based on the target type and optimal distance estimate of the target object, a driving strategy for the target object is determined, including: Based on the target type of the target object, determine the corresponding warning distance and danger distance; Based on the relationship between the optimal distance estimate and the warning distance and danger distance, the required driving mode is determined.

[0011] Optionally, based on the relationship between the optimal distance estimate and the warning distance and danger distance, the required driving mode is determined, including: If the optimal distance estimate is greater than the warning distance, the monitoring mode is activated. If the optimal distance estimate is less than the warning distance or greater than the danger distance, a first-level driving mode is adopted, and the driving intensity of the driving unit in the first-level driving mode is determined according to the optimal distance estimate. If the optimal distance estimate is less than the danger distance, a two-level driving mode is adopted, and the driving intensity of the driving unit in the two-level driving mode is determined according to the optimal distance estimate.

[0012] Optionally, the object detection unit includes a microwave radar and / or an infrared detector; the method further includes: The microwave radar is used to detect whether the target object appears within the monitored area; Alternatively, the infrared detector can be used to detect whether the target object appears within the monitored area; Alternatively, the microwave radar can be used to detect whether the target object appears within the monitoring area. When a moving suspected target object is detected, the infrared detector can be used to detect whether the target object appears within the monitoring area within a preset time. The detection results of the infrared detector can be combined to determine whether the suspected target object is the target object.

[0013] This application also provides a solar-powered multimodal warning and bird-repelling system, including: The object detection unit is used to detect target objects within the monitored area; The control unit is configured to, when the object detection unit detects a target object within the monitoring area, control the image acquisition unit to acquire an image of the target object including the target object; control the ranging unit to detect the distance to the target object; use a preset target detection model to identify the target object image to obtain the position and target type of the target object; filter the position and distance of the target object to obtain an optimal distance estimate; determine a driving strategy for the target object based on the target type and the optimal distance estimate; and control the driving unit to drive the target object according to the driving strategy.

[0014] As can be seen from the above, the multimodal warning and bird deterrence method and system based on solar power provided in this application involves the following steps: When the object detection unit detects a target object within the monitoring area, the image acquisition unit acquires an image of the target object, the ranging unit detects the distance to the target object, the target detection model identifies the target object image to obtain its location and type, the location and distance are filtered to obtain an optimal distance estimate, and a deterrence strategy is determined based on the target object's type and the optimal distance estimate. The deterrence unit then drives the target object according to the strategy. This application can achieve accurate target identification with limited resources and employs a graded and differentiated warning and deterrence strategy to improve the deterrence effect. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the method flow of an embodiment of this application; Figure 2 This is a schematic diagram of the system architecture of an embodiment of this application; Figure 3 This is a schematic diagram of the system composition of some embodiments of this application; Figure 4 This is a schematic diagram of the target detection process in an embodiment of this application; Figure 5 This is a partial structural diagram of the target detection model according to an embodiment of this application; Figure 6 A schematic diagram of the FasterNet module structure with added SimAM attention mechanism in an embodiment of this application; Figure 7 This is a schematic diagram of the weighted bidirectional feature pyramid structure according to an embodiment of this application; Figure 8 These are schematic diagrams of pyramid structures in some embodiments; Figure 9 This is a block diagram of the electronic device structure according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] like Figure 1 As shown in the embodiment of this application, a multimodal warning and bird deterrence method based on solar power is provided, including: S101: When the object detection unit detects a target object within the monitoring area, it controls the image acquisition unit to acquire an image of the target object including the target object; and controls the ranging unit to detect the distance to the target object. In this embodiment, the intelligent bird deterrence method is implemented based on an intelligent bird deterrence system. This system can be deployed in locations requiring bird control, such as substations, power poles, farmland, and airports. Figure 2 , 3 As shown, the intelligent bird deterrence system includes an object detection unit, a control unit, an image acquisition unit, a ranging unit, and a deterrence unit. The object detection unit is used to detect whether a target object appears in the monitored area, such as different kinds of birds. When a target object is detected in the monitored area, the control unit controls the image acquisition unit to acquire an image of the target object within the monitored area, including the target object. At the same time, the control unit controls the ranging unit to detect the distance of the target object relative to the intelligent bird deterrence system.

[0020] In some embodiments, the object detection unit includes a microwave radar and / or an infrared detector; using the object detection unit to detect whether a target object appears within the monitored area includes: Microwave radar is used to detect whether a target object appears within the monitored area; Alternatively, infrared detectors can be used to detect whether a target object appears within the monitored area; Alternatively, microwave radar can be used to detect whether a target object appears in the monitored area. When a moving suspected target object is detected, an infrared detector can be used to detect whether a target object appears in the monitored area within a preset time. The detection results of the infrared detector are then combined to determine whether the suspected target object is the target object.

[0021] In this embodiment, a microwave radar can be used to detect target objects within the monitored area. The presence of a target object is determined when the microwave radar detects valid movement, offering a fast response time suitable for scenarios with extremely high alarm timeliness requirements. Alternatively, an infrared detector can be used to detect target objects within the monitored area. The presence of a target object is determined when typical bird heat source characteristics are detected. To improve detection speed while ensuring accuracy, both microwave radar and an infrared detector can be used simultaneously. When the microwave radar detects valid movement, the infrared detector is activated. If the infrared detector detects bird heat source characteristics within a set short time (e.g., 100-500 milliseconds), the presence of a target object can be definitively confirmed, providing a fast, accurate, and effective solution. If the microwave radar detects valid movement but the infrared detector does not detect bird heat source characteristics within the set time, it is determined to be interference, and no valid target object has been detected.

[0022] In other embodiments, the object detection unit may also employ an ultrasonic detector, which emits detection pulses at a set period and determines whether a moving target exists within the monitored area based on the echo time and intensity. To improve accuracy, the detection results of an infrared detector may also be combined to determine whether a target object has appeared. The above are merely illustrative examples; the type and specific configuration of the object detection unit can be selected and adapted according to the application scenario, and are not limited here.

[0023] In some embodiments, to improve recognition accuracy, the image acquisition unit may employ an industrial-grade high-definition camera, which features a wide dynamic range and automatic infrared filter switching, enabling it to adapt to changes in day and night lighting conditions and acquire high-quality images and videos under various lighting environments. The above is merely an illustrative example; the specific configuration of the image acquisition unit can be selected and adapted according to the application scenario, and is not limited here.

[0024] In some embodiments, the ranging unit can use an ultrasonic probe to achieve the ranging function; that is, when a target object is detected within the monitored area, the ultrasonic probe is used to detect the distance from the target object to the system. Distance is measured based on the time-of-flight principle, and the calculation method is as follows: (1) in, v The speed of sound wave propagation at the current ambient temperature T (degrees Celsius) is calculated as follows: , This is the time difference between the emission and reception of the ultrasonic wave.

[0025] In other methods, distance measurement can also be achieved using distance measurement units such as visual distance measurement and laser distance measurement. This is only an example for illustration. The specific configuration of the distance measurement unit can be selected and adapted according to the application scenario, and there is no specific limitation.

[0026] In some embodiments, the deterrent unit includes an ultrasonic speaker, a sound wave player, a green laser, and a light-emitting warning device. Optionally, two ultrasonic speakers with center frequencies of 25kHz and 36kHz are selected, supporting frequency sweep switching at certain intervals. The sound pressure level is adjustable within the range of 75-100dB, and a certain area is covered by a directional beam angle. The audio player uses a waterproof dual-channel speaker and has built-in memory to store audio files of the calls of predators of different birds, such as sparrowhawks and peregrine falcons, as well as various sudden noises. Dual-channel phase shift technology is used to simulate the effect of sound source movement to increase realism. The duration of each playback can be randomly set (e.g., 3-10 seconds) to prevent birds from adapting. The green laser uses an adjustable-intensity green laser module with a wavelength of 520-525nm, a rated power of 100mW, a rated voltage of 3~5V, and a light scattering range of 60°. The light intensity is continuously adjustable through a pulse-width modulation circuit. The green laser and ultrasonic speaker are coaxially mounted to ensure synchronized sound and light. An aluminum focusing cup is used to increase the illumination distance and also provides heat dissipation. The luminous warning device is based on an LED array and can provide a warning effect through strobe, periodic flashing, etc. The above is only an illustrative example; the type and specific configuration of the deterrent unit can be selected and adapted according to the application scenario, and are not limited here.

[0027] S102: Use a preset target detection model to identify the target object image and obtain the target object's location and target type; In this embodiment, the target object image acquired by the image acquisition unit is input into a pre-constructed target detection model. The model is then used to process and recognize the target object image to obtain the recognition result of the target object. In some methods, the acquired target object image can be pre-processed to obtain an image suitable for input into the model for processing, and then the pre-processed target object is input into the target detection model for processing and recognition.

[0028] In some implementations, the object detection model is based on the YOLOv8 network framework, replacing the C2f module of the YOLOv8 framework with a lightweight FasterNet network that incorporates the parameter-free attention mechanism SimAM (Simple Parameter-Free Attention Module). For example... Figure 4 , 5 As shown, the FasterNet network structure consists of four FasterNet blocks, which are used for four stages of processing. Each FasterNet block is preceded by either a 4×4 convolutional kernel with a stride of 4 embedding layer or a 2×2 convolutional kernel with a stride of 2 merging layer for spatial downsampling and channel expansion. Each FasterNet block consists of a PConv structure and two 1×1 convolutional layers (Conv), presenting an inverted residual block. The PConv structure only needs to perform convolution operations on a portion of the input channels, leaving the remaining channels unchanged, which can make full use of limited computational resources while simultaneously achieving feature extraction.

[0029] like Figure 6 As shown, to improve the feature extraction capability of the FasterNet network, a SimAM attention module is introduced into the FasterNet module. The SimAM attention module is connected after the last convolutional layer of the FasterNet module and before the residual connections. Utilizing the idea that neurons with greater differences from surrounding neurons are more important, the SimAM attention module assigns attention weights to each neuron extracting the feature map. Specifically, for the feature map... Where H, W, and C are the height, width, and number of channels, respectively, and are reshaped into... ,in, For each channel, the energy function is used to calculate the value of each neuron. t The energy value is expressed as: (2) in, It is the average pixel value output by all neurons in this channel except neuron t. In other words, it's the sum of the feature values ​​output by all neurons except neuron t in the current channel, divided by (...). N -1) is the average value obtained; It is the variance of the pixel values ​​output by all neurons in this channel except for neuron t. N For the number of neurons, x i For the first i The pixel value output by each neuron. x t Neuron t Output pixel values, It is a hyperparameter (usually set to 0.5).

[0030] Energy function The smaller the value, the greater the difference between the neuron and its surrounding neurons, and the higher its importance. Formula (2) provides an energy function based on channel statistics to directly quantify the linear separability of a single neuron from its surrounding neurons.

[0031] Then, the energy values ​​are converted into attention weights using the Sigmoid function to obtain the neurons. t The attention weights are: (3) The weights of each neuron are weighted by the corresponding extracted feature values ​​to obtain a weighted feature map. By directly applying weights to the corresponding neurons, fine-grained pixel-level attention allocation is achieved. Without introducing additional parameters, joint spatial and channel attention is realized, improving the accuracy and efficiency of feature extraction. On lightweight networks, performance can be improved without increasing model complexity.

[0032] In some embodiments, the object detection model includes a feature fusion layer for fusing an input multi-scale feature map with intermediate features extracted from the multi-scale feature map.

[0033] Specifically, YOLOv8 employs a Path Aggregation Network (PANet), which adds bottom-up paths to the FPN, leveraging rich, shallow location information to enhance the network's ability to classify and locate targets. For example... Figure 4 , 7 As shown, this embodiment improves PANet by using a weighted bidirectional feature pyramid network (BiFPN), compared to the conventional pyramid structure (such as...). Figure 8 As shown), on the one hand, delete nodes with only one input edge ( Figure 8 (Redundant nodes are not shown). These nodes contribute little to feature fusion; removing them reduces model complexity while maintaining performance, decreasing the number of parameters and computation. On the other hand, adding an extra path between nodes at the same level, such as... Figure 7 The Conv layers on the left side utilize convolutional layers to extract input features. P Extracting intermediate features from the feature extraction layer allows for richer feature fusion without significantly increasing computational cost. This is achieved by extracting intermediate features from the multi-scale feature map from the feature extraction layer. , k This refers to the number of layers in the feature extraction network that output features. Each feature map has a different resolution, and each layer corresponds to a feature representation at a different depth in the network. Lower-level features have higher spatial resolution and richer detail information, while higher-level features contain stronger semantic information. To achieve multi-scale feature fusion, BiFPN is used through learnable weights. Each input feature and intermediate feature is weighted and then fused to achieve semantic and spatial information complementarity across different scales, thereby improving the detection capability of target objects at different scales. The feature fusion method is as follows: (4) in, O i For the first i The fused feature map of each fusion node. It is the first i The weight of each fusion node, It is the first j The weight of each fusion node, M The number of fusion nodes; It is a very small number (such as 0.0001) used to prevent numerical instability and to determine the weight. Learning is achieved through gradient descent. Since BiFPN generates additional fusion nodes O in both the top-down and bottom-up bidirectional paths, therefore... M Greater than k In formula (4) U i Including feature maps from the feature extraction layer And the intermediate features extracted from the feature map by the additional convolutional layers, more specifically, U i Including all feature maps and intermediate features after uniform size transformation, the total number of all feature maps and intermediate features is... M This fusion method improves the robustness and convergence of the fusion results through normalization and stability constraints, and significantly improves the fusion effect of multi-scale features.

[0034] In some embodiments, the YOLOv8n network uses CIoU-Loss as the bounding box regression loss function, which alleviates the optimization problem in the intersection process of the predicted box and the ground truth box to some extent. However, it does not adequately consider the relative position and aspect ratio of the bounding boxes. In outdoor bird detection tasks, considering the small size of the detection target and the complex background, higher accuracy of bounding box localization is required. This embodiment adopts the Wise-IoUv3 bounding box loss function to achieve higher bird recognition accuracy and efficiency. Wise-IoUv3 uses a dynamic non-monotonic focusing mechanism to reduce the influence of harmful gradients, enabling the model to formulate the gradient gain allocation strategy most suitable for the current situation in real time. The specific formula is as follows: (5) (6) (7) in, L WIoUv3 The loss value is the bounding box loss function of Wise-IoUv3. β Characterize the degree of anomaly in the predicted bounding box. β The larger the value, the worse the quality of the predicted box and the lower the matching degree with the ground truth box; r Gradient gain is used to focus the function on a prediction box of average quality, thus avoiding gradient explosion. IOU The intersection-union ratio (IU) of the predicted bounding box and the ground truth bounding box. , These are the width and height dimensions of the smallest enclosing box that can simultaneously enclose both the predicted box and the ground truth box; x , y The center pixel coordinates of the predicted bounding box (anchor box). , The coordinates of the center point of the true bounding box (pixel coordinates). The loss value for the current predicted bounding box. These are steady-state baseline values, typically taken as the mean within the sliding window or the loss value from the previous iteration; δ, α, β It is used to adjust gradient gain. r Hyperparameters of scale and distribution.

[0035] In some methods, the acquired target object image is input into a target detection model based on an improved YOLOv8 network architecture. The model outputs the detection results, including the target object's anchor box position, confidence score, and target object type. For birds, the type is the identified species information, such as sparrow, crow, magpie, etc. In other methods, when the confidence score is higher than a set confidence threshold (e.g., 0.7), subsequent processing continues.

[0036] This embodiment implements a target detection model based on an improved YOLOv8 network framework to identify birds approaching the monitoring area. By introducing a lightweight FasterNet backbone network that integrates SimAM attention, the feature extraction efficiency and model perception capability are significantly improved. Combined with an improved weighted bidirectional feature pyramid network, multi-scale feature fusion is achieved, enhancing the robustness of bird target detection at different scales. Furthermore, the Wise-IoUv3 loss function is used to optimize the predicted bounding box regression process, effectively improving the localization accuracy. While maintaining high-precision detection, the number of parameters is significantly reduced, making it suitable for devices with limited resources.

[0037] S103: Filter the position and distance of the target object to obtain the optimal distance estimate; In this embodiment, the target object's position is detected using a target detection model, and the distance to the target object is detected using a ranging unit. Then, the optimal distance estimate for the target object is determined through filtering. The target object's position output by the target detection model is the pixel coordinates in the acquired target object image. Based on the transformation relationships between the image coordinate system, camera coordinate system, and world coordinate system, and combined with the intrinsic parameters and spatial calibration parameters of the image acquisition unit, the pixel coordinates need to be converted into spatial position coordinates in the world coordinate system. This application does not provide a detailed description of the specific methods and principles of coordinate transformation.

[0038] In some embodiments, the position and distance of the target object are filtered to obtain the optimal distance estimate, including: Construct observation vectors based on the location and distance of the target object; The Kalman filter method is used to predict the optimal position estimate of the target object based on the observation vector. Based on the optimal location estimate, determine the corresponding optimal distance estimate.

[0039] In this embodiment, the Kalman filter method is used to fuse multi-source data of visual and distance information. Through the recursive mechanism of prediction update, the optimal estimation of the position and velocity of the target object is achieved in a noisy environment, providing a basis for intelligent bird deterrence decision-making.

[0040] Specifically, by utilizing the intrinsic and extrinsic calibration parameters of the image acquisition unit, the pixel coordinates (u,v) in the detected target image are converted into spatial position coordinates and direction vectors in the camera coordinate system, which are then combined with the distance detected by the ranging unit. d By combining the spatial direction vector with the depth (distance), the spatial coordinates of the target object in three-dimensional space are obtained, represented as: (8) in, K This is the intrinsic parameter matrix of the image acquisition unit. R It is an extrinsic parameter matrix.t This is a spatial translation vector, representing the translation of the camera coordinate system relative to the world coordinate system. Considering that bird control systems primarily focus on the horizontal distribution and relative distance changes of the monitored area (e.g., around the pole), the spatial position coordinates are... Projected onto the ground plane, defining the position of the two-dimensional plane. Construct fused observation vectors , which serves as the input for the Kalman filter.

[0041] To characterize the planar motion characteristics of birds in a short time, we assume that the target object follows a uniform or gradually changing velocity model and define a state vector. , where subscript k Represents a discrete-time index. The coordinates of the target object on the ground plane. The corresponding velocity components are hidden states automatically learned through continuous position changes. During the time update process, the Kalman filter utilizes the assumption of uniform motion: (9) (10) During state prediction, the prior value of the current position is estimated based on the position and velocity of the previous moment. Velocity is used as a hidden state in the prediction. Utilizing the assumption that birds move at a constant velocity over a short period, the prediction is made at time [time value missing]. k The optimal posterior estimate of -1 Used for predicting time k Given the prior state, the state transition equation is: (11) in, Here is the state transition matrix. The actual time interval between the two observations. For process noise, the covariance matrix Q represents the model uncertainty, and its form is usually as follows: (12) in, , Let be the variance of the process noise in the x and y directions.

[0042] The state covariance prediction is: (13) in, The state covariance matrix estimated at the previous time step. The estimated state covariance matrix at the current time. Q The process noise covariance reflects model incompleteness and target maneuverability. In engineering, it should be selected based on the typical maneuverability of the target and the sampling frequency.Q For example, the variance of the velocity component is the dominant term and follows... enlarge.

[0043] During the update phase, the current observation vector is received. , , This represents the current observed two-dimensional plane position. The distance to the current observation is given; the relationship between the observation vector and the state vector is expressed as: The state vector is mapped to the observation space using the observation matrix H, which is: (14) The first two lines correspond to the two-dimensional planar position information, and the third line is the distance constraint term. The geometric projection coefficient between the plane and the distance measurement is calculated as follows: (15) (16) To measure noise, R The noise covariance matrix can be determined based on the calibration results of visual positioning error and ranging error.

[0044] The specific process of Kalman filtering is as follows: based on the optimal state estimate of the previous time step... Predict the current state according to formula (11) The uncertainty of the prediction is calculated according to formula (13). During the update, calculate the Kalman gain: (17) By fusing the predicted and observed vectors, the optimal state estimate is obtained. (18) in, It is the state vector at the current moment. It is the optimal state estimate at the current moment.

[0045] Updated post-test covariance: (19) During the update phase, the observation vector It provides actual measured two-dimensional planar position and distance information, and compares the observed vector with the predicted value. By comparison, the residuals are obtained. .

[0046] Kalman gain K kThe fusion weights of prediction and observation are determined; when distance noise is low and target recognition is stable, K k If the error is large, it can be corrected based on actual observations; conversely, it can be corrected based on the predictions of the motion model. Through adaptive weight adjustment, the filter can dynamically balance the influence of the two types of information under different environments, and obtain the optimal estimate in the sense of minimum mean square error.

[0047] Through alternating iterations of prediction and update, the optimal state estimate of the target object can be obtained: (20) in, For the optimal position estimation of the fused plane, This provides the optimal velocity component estimation for the plane. Based on the optimal position estimation for the plane, the spatial distance and azimuth of the target object can be calculated: (twenty one) (twenty two) in, This is the optimal distance estimate in the sense of minimum mean square error after fusion, and it is more stable than the distance measurement results of the ranging unit. Let be the direction angle. The magnitude of the optimal velocity estimate is: (twenty three) The prediction phase of Kalman filtering ensures the temporal continuity of state estimation, while the update phase achieves observation correction through the fusion of visual and distance information. The two phases compensate for each other and iterate cyclically, thereby achieving stable tracking of the target object in noisy environments.

[0048] S104: Determine the driving strategy for the target object based on the target type and optimal distance estimate; In this embodiment, the target object and its target type are determined using a target detection model. After obtaining the optimal distance estimate by filtering the position and distance, the optimal driving strategy for effectively driving away the target object is determined based on the target type of the target object and the optimal distance estimate.

[0049] In some embodiments, the driving-away strategy includes warning distances, danger distances, and multiple driving-away modes corresponding to different types of target objects; Based on the target object's target type and optimal distance estimation, determine the driving strategy for the target object, including: Determine the corresponding warning distance and danger distance based on the target type of the target object; Based on the relationship between the optimal distance estimate and the warning distance and danger distance, the required driving mode is determined.

[0050] In this embodiment, different driving modes can be set according to the type of target object and the distance from the target object to the system, forming diverse driving strategies. Considering the different characteristics of different target objects, the set distances also differ. For example, considering the differences between sparrows and crows in size, flight speed, and degree of harm, the warning distance for sparrows is set to 10 meters and the danger distance to 5 meters, while the warning distance for crows is set to 15 meters and the danger distance to 8 meters. Based on the optimal distance estimate from the target object to the system detected in real time, the relationship between the optimal distance estimate and the warning distance and danger distance for the corresponding type of target object is determined, thus determining the best driving strategy for driving away the target object and improving driving effectiveness.

[0051] In some embodiments, the required driving mode is determined based on the relationship between the optimal distance estimate and the warning distance / danger distance, including: If the optimal distance estimate is greater than the warning distance, the surveillance mode is used. If the optimal distance estimate is less than the warning distance but greater than the danger distance, the first-level driving mode is adopted, and the driving intensity of the driving unit in the first-level driving mode is determined according to the optimal distance estimate. If the optimal distance estimate is less than the danger distance, a two-stage driving mode is adopted, and the driving intensity of the driving unit in the two-stage driving mode is determined according to the optimal distance estimate.

[0052] In this embodiment, the driving modes for various target objects can be as follows: if the optimal distance estimate is greater than the warning distance, that is, the distance from the monitoring area is far, the monitoring mode can be used, which only monitors without triggering an alarm to drive away the target object; if the optimal distance estimate is less than the warning distance but greater than the danger distance, that is, the target object has entered the warning range of the monitoring area, the first-level driving mode is used, which employs a relatively mild driving strategy to drive away the target object; if the optimal distance estimate is less than the danger distance, that is, the target object has entered the danger range of the monitoring area, the second-level driving mode needs to be used, which employs a stronger driving strategy to drive away the target object and prevent it from entering the monitoring area and affecting the facilities.

[0053] In some methods, the strength of the driving unit can be estimated based on the current actual optimal distance. It is confirmed that the frequency of an ultrasonic loudspeaker can be: (twenty four) For a green laser, its laser intensity can be: (25) For an audio player, its sound pressure level can be: (26) in, As the reference frequency, As the reference strength,S base As the reference sound pressure level, , , k s is the gain coefficient, which varies under different driving modes. According to equations (24)-(26), as the optimal distance estimate decreases, that is, the closer the target object is to the monitoring area, the greater the intensity of the driving unit.

[0054] For example, when a bird is identified as a sparrow and the optimal distance estimate is greater than 10 meters, monitoring is conducted solely by recording relevant information, without any attempt to scare it away. If the optimal distance estimate is between 5 and 10 meters, a first-level scare mode is adopted, calculating the laser intensity based on the optimal distance estimate, controlling the laser to emit laser light at that intensity, and calculating the frequency of the ultrasonic horn. (Hz), the ultrasonic loudspeaker is controlled to emit ultrasonic waves at this frequency to achieve a deterrent effect; if the optimal distance is estimated to be less than 5 meters, a two-stage deterrent mode is adopted, and the frequency of the ultrasonic loudspeaker is calculated based on the optimal distance estimate. (Hz), control the ultrasonic speaker to emit ultrasonic waves at this frequency, and simultaneously control the audio player to play the call of the sparrow's natural enemy, and calculate the sound pressure level of the playback. (dB), which makes the sound pressure level increase as the optimal distance estimate decreases.

[0055] When a bird is identified as a crow and the optimal distance estimate is greater than 15 meters, monitoring is conducted solely by recording relevant information, without any attempt to scare it away. If the optimal distance estimate is between 8 and 15 meters, a first-level scare mode is employed, controlling the luminous warning device to flash its lights in a deterrent effect. If the optimal distance estimate is less than 8 meters, a second-level scare mode is employed, calculating the sweep frequency cycle of the ultrasonic loudspeaker based on the optimal distance estimate. At 0.05 seconds, the ultrasonic speaker is controlled to scan according to this frequency sweep cycle, and the luminous warning device flashes a bright light to warn the approaching target.

[0056] In some implementations, a driving strategy for the target object can be comprehensively determined based on multiple factors, including the target object's behavior, optimal distance estimation, other monitored objects within the monitoring area, and weather conditions. For example, if the monitored object is critical equipment, a spatial weighting factor can be determined based on the optimal distance estimation from the target object to the monitored object. Prioritize the protection of critical equipment by setting spatial weighting factors. The distance weighting factor is set based on the optimal distance estimate. The closer the distance, the larger the distance weighting factor. Behavioral weighting factors are set based on the dynamic behavioral characteristics of the target object. If the target object is identified to exhibit aggressive or settlement-intentioned behaviors such as loitering, nesting, pecking, or swarming, the value of the behavior weighting factor is increased. An environmental weighting factor is set based on weather and environmental parameters collected by external sensors. For example, increasing the environmental weight factor during severe weather can raise the alert level. Therefore, by combining factors such as weather conditions, the monitored object, and the behavior of the target object, a comprehensive decision weight is determined, expressed as: (27) in, These are the weighting coefficients.

[0057] The overall threat level of the monitored area is determined based on the comprehensive decision weight value, and the most suitable bird deterrence strategy is selected. For example, when the comprehensive decision weight value exceeds the maximum risk threshold, the highest level of bird deterrence strategy is directly activated to maximize bird deterrence, which is applicable to various application scenarios.

[0058] S105: Control the driving unit to drive the target object according to the driving strategy.

[0059] In this embodiment, after determining the driving-off strategy, the control unit controls the corresponding driving-off unit actions according to the strategy, including controlling the frequency of the ultrasonic speaker, controlling the audio file and sound pressure level played by the audio player, and controlling the intensity and frequency of the green laser. Thus, this application, by accurately detecting the type, location, and distance of target objects approaching the monitoring area, and employing a graded and differentiated driving-off strategy, can accurately and effectively drive away different objects, preventing facilities within the monitoring area from being affected and meeting the intelligent operation and maintenance needs of unattended outdoor facilities.

[0060] like Figure 3 As shown, in some embodiments, after driving away the target object, the monitoring area is continuously monitored. If the target object moves away from the monitoring area within a certain period of time (e.g., 30 seconds), the driving away is marked as successful. To improve the long-term effectiveness of the strategy, the control unit uploads the relevant results of the identified target object, the driving strategy adopted, and the driving result to the remote monitoring unit. The remote monitoring unit periodically calculates the success rate of the driving strategy adopted, and calculates the ratio of the number of successful drivings using the driving strategy to the total number of drivings using the driving strategy. This ratio is compared with a set threshold. If the ratio is lower than the threshold, the corresponding driving strategy needs to be optimized and adjusted. For example, the confidence threshold, warning distance, danger distance, and driving mode of the driving unit may be adjusted. The adjusted driving strategy is then sent to the control unit, which drives away the birds according to the optimized driving strategy, thereby improving the bird driving effect and preventing birds from developing adaptation.

[0061] The intelligent bird deterrence method provided in this application, when a target object is detected within a monitored area, acquires an image of the target object, detects the distance to the target object, identifies the target object image using a target detection model to obtain the target object's location and type, filters the target object's location and distance to obtain an optimal distance estimate, determines a deterrence strategy based on the target object's type and the optimal distance estimate, and then deters the target object according to the deterrence strategy. This application, through model improvements, can accurately and reliably identify target objects on resource-limited devices, and adaptively triggers the most effective bird deterrence mode based on conditions such as the target object's type and distance. It also dynamically adjusts the intensity of the deterrence unit based on distance, achieving a graded and differentiated bird deterrence strategy.

[0062] Experiments show that the method of this application achieves a detection accuracy (mAP@0.5) of 92.3% for common birds in the complex environment of substations, with a ranging error of less than 5%. The model weight is reduced by 42% compared with the benchmark YOLOv8 network structure. While effectively improving the accuracy and response speed of bird control, it also significantly improves the lightweight level and deployment flexibility of the system, providing an effective and reliable intelligent bird control solution for bird pest control in outdoor scenarios such as substations.

[0063] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0064] It should be noted that the above description describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims may be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] like Figure 2 As shown in the embodiments of this application, a multimodal warning and bird deterrent system based on solar power is also provided, including: The object detection unit is used to detect target objects within the monitored area; The control unit is used to control the image acquisition unit to acquire an image of the target object, including the target object, when the object detection unit detects a target object within the monitoring area; control the ranging unit to detect the distance to the target object; use a preset target detection model to identify the target object image to obtain the target object's location and target type; filter the target object's location and distance to obtain the optimal distance estimate; determine a driving strategy for the target object based on its target type and the optimal distance estimate; and control the driving unit to drive the target object according to the driving strategy.

[0066] In some cases, bird deterrent systems are powered by solar energy, eliminating the need for an external power source. This makes them suitable for various scenarios and is environmentally friendly.

[0067] For ease of description, the above system is described by dividing it into various modules based on their functions. Of course, when implementing the embodiments of this application, the functions of each module can be implemented in one or more software and / or hardware.

[0068] The system described above is used to implement the corresponding methods in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0069] Figure 9 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0070] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0071] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0072] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0073] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0074] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0075] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0076] The electronic devices described above are used to implement the corresponding methods in the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0077] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0078] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this disclosure (including the claims) is limited to these examples; within the framework of this disclosure, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0079] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this disclosure, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0080] Although this disclosure has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0081] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this disclosure.

Claims

1. A multi-modal warning, bird repelling method based on solar power supply, characterized in that, The method comprises the following steps: When the object detection unit detects a target object in the monitoring area, the image acquisition unit is controlled to acquire a target object image comprising the target object; The distance measurement unit is controlled to detect the distance of the target object; A preset target detection model is used to identify the target object image, to obtain the position and target type of the target object; The position and distance of the target object are filtered to obtain an optimal distance estimate; According to the target type and optimal distance estimate of the target object, a driving strategy for the target object is determined; The driving unit is controlled to drive the target object according to the driving strategy.

2. The method of claim 1, wherein, The feature extraction layer of the target detection model introduces a SimAM attention module, which is used to determine the attention weight of each neuron of the feature map, and the method is as follows: Calculating energy values for neurons using an energy function t e t :​ （2） wherein, is the mean of the pixel values output by all neurons in the channel other than neuron t , is the variance of the pixel values output by all neurons in the channel other than neuron t , N is the number of neurons, x i is the pixel value output by the i th neuron, x t is the pixel value output by neuron t , is a hyperparameter; The energy value is converted into attention weight by a Sigmoid function, to obtain the attention weight of the neuron t is:​ （3）。 3. The method of claim 1, wherein, The target detection model comprises a feature fusion layer, which is used to fuse the input multi-scale feature map and the intermediate feature extracted from the multi-scale feature map.

4. The method of claim 1, wherein, The bounding box loss function of the target detection model is as follows: （5） （6） （7） wherein, L WIoUv3 is the loss value of the bounding box loss function; r is the gradient gain, which is used to focus the function on the common quality of the predicted box, and avoid the occurrence of gradient explosion; IOU is the intersection over union of the predicted box and the real box, 、 are the width and height of the minimum enclosing box that can simultaneously enclose the predicted box and the real box, respectively; x 、 y is the center pixel coordinate of the predicted box (anchor box), 、 is the center pixel coordinate of the real box; is the loss value of the current predicted box, is the steady-state reference value, which is usually the mean value in the sliding window or the loss value of the last iteration; δ, α, β is the gradient gain adjustment factor r scale and distribution hyperparameters.

5. The method of claim 1, wherein, The position and distance of the target object are filtered to obtain an optimal distance estimate, which comprises the following steps: Based on the position and distance of the target object, an observation vector is constructed; A Kalman filtering method is used to predict the optimal position estimate of the target object based on the observation vector; According to the optimal position estimate, a corresponding optimal distance estimate is determined.

6. The method of claim 5, wherein, The Kalman filtering method is used to predict the optimal position estimate of the target object based on the observation vector, which comprises the following steps: Based on the observation vector, a Kalman filtering method is used to predict an optimal state estimate: （18） wherein, z k is the observation vector, K k is the Kalman gain, H is the observation matrix, is k a state vector at time instant t, the state vector comprising a position and a velocity of the target object, is k an optimal state estimate at time instant t-1, is k an optimal state estimate at time instant t, the optimal state estimate comprising an optimal position estimate.

7. The method of claim 1, wherein, The driving strategy comprises the alert distance, dangerous distance and various driving modes corresponding to different types of target objects; According to the target type and optimal distance estimate of the target object, a driving strategy for the target object is determined, which comprises the following steps: According to the target type of the target object, the corresponding alert distance and dangerous distance are determined; According to the size relationship between the optimal distance estimate and the alert distance and dangerous distance, the driving mode to be used is determined.

8. The method of claim 7, wherein, According to the size relationship between the optimal distance estimate and the alert distance and dangerous distance, the driving mode to be used is determined, which comprises the following steps: If the optimal distance estimate is greater than the alert distance, a monitoring mode is used; If the optimal distance estimate is less than the alert distance and greater than the dangerous distance, a first-level driving mode is used, and the driving strength of the driving unit in the first-level driving mode is determined according to the optimal distance estimate; If the optimal distance estimate is less than the dangerous distance, a second-level driving mode is used, and the driving strength of the driving unit in the second-level driving mode is determined according to the optimal distance estimate.

9. The method of claim 1, wherein, The object detection unit comprises a microwave radar and / or an infrared detector; the method further comprises the following steps: The microwave radar is used to detect whether the target object appears in the monitoring area; Or, the infrared detector is used to detect whether the target object appears in the monitoring area; Or, whether the target object appears in the monitoring area is detected by using the microwave radar, when a moving suspected target object is detected, whether the target object appears in the monitoring area is detected by using the infrared detector within a preset time, and whether the suspected target object is the target object is determined in combination with the detection result of the infrared detector.

10. A multi-modal warning, bird repelling system powered by solar energy, characterized in that, Comprise: An object detection unit is used for detecting a target object in a monitoring area; A control unit is used for controlling an image acquisition unit to acquire a target object image including the target object when the object detection unit detects a target object in the monitoring area; and controlling a ranging unit to detect the distance of the target object; A preset target detection model is used for identifying the target object image, and the position and target type of the target object are obtained; The position and distance of the target object are filtered to obtain an optimal distance estimation; a driving strategy for the target object is determined according to the target type of the target object and the optimal distance estimation; and a driving unit is controlled to drive the target object according to the driving strategy.

Images