Intelligent bird identifying, tracking and repelling device

Abstract

The invention relates to the technical field of bird repelling, in particular to an intelligent bird recognizing, tracking and repelling device, which comprises a video monitoring module, a tracking module, a bird repelling module, a control module and a power supply module, and is characterized in that the video monitoring module is used for monitoring bird conditions in a no-fly zone in real time by using a camera; the laser bird repelling module is used for repelling birds by using a laser; the control holder module is used for installing the video monitoring module and the laser bird repelling module; and the core control module is used for collecting the video information transmitted by the camera in the video monitoring module through identification and analysis, and detecting and tracking the birds in the no-fly zone. According to the invention, multi-algorithm real-time processing of the STC target tracking algorithm, the ECO algorithm and the YOLO-V3 detection algorithm is realized, the operation efficiency of the system is improved, the YOLO-V3 detection algorithm can adjust the size of a tracking frame according to the size of an identified target, targets of different sizes can be better adapted, the tracking accuracy is improved, laser is used for repelling birds, and the tracking accuracy is improved. The bird repelling effect is improved, and meanwhile safety is considered.

Description

Technical Field

[0001] The invention relates to the technical field of bird driving away, in particular to a bird intelligent identification, tracking and driving away device. Background Art

[0002] With the accelerated process of urban construction and transformation and the sustained rapid development of my country's social economy, the demand for electricity continues to grow at a high speed. As a basic industry, the construction scale of power transmission and transformation projects continues to increase. Due to the particularity of outdoor structural equipment of transmission lines and substations, many birds use them as nesting sites. Bird excrement and nesting accelerate the pollution and aging of power facilities, which can easily cause line flashover and short circuit faults in transmission lines. The number of bird activities around power facilities in cities and suburbs has increased significantly, and the number of bird damage defects has increased sharply, posing a threat to the reliable and stable operation of the power system, while increasing the workload of power operators in dealing with bird damage.

[0003] At present, the commonly used methods of repelling birds without causing harm are: changing the ecological environment and stimulating signal repelling. However, these two types of bird repelling methods have their advantages and disadvantages. Changing the ecological environment is suitable for long-term bird repelling, and its effect is obvious, but it still cannot be completely eliminated. Stimulating signal bird repelling has a very obvious short-term effect, but it faces the problem of losing its bird repelling effect due to long-term use causing birds to adapt.

[0004] Therefore, there is a need for a bird intelligent identification, tracking and driving away device, which has a significant and long-term bird-repelling effect. It is also necessary to take into account the benefits and try to drive away birds only in places where birds appear, reduce resource waste and ensure speed. Summary of the invention

[0005] The purpose of the present invention is to provide a bird intelligent identification, tracking and driving away device to solve the problems raised in the above background technology.

[0006] In order to solve the above technical problems, the present invention provides a device for intelligently identifying, tracking and driving away birds, comprising:

[0007] Video monitoring module: used to use cameras to monitor the situation of birds in the no-fly zone in real time;

[0008] Laser bird repellent module: used to repel birds using lasers;

[0009] Control PTZ module: used to install the video monitoring module and the laser bird-repelling module;

[0010] Core control module: used to collect video information transmitted by the camera in the video monitoring module, and detect and track flying birds in the no-fly zone through identification and analysis.

[0011] As a further improvement of the present technical solution, the laser in the laser bird repellent module uses basic bird repellent strategies and advanced bird repellent strategies to achieve the bird repellent effect; the basic bird repellent strategy is to drive away flying birds and keep them away from the monitored protection area; the advanced bird repellent strategy is to drive away flying birds in a given direction, specifically including: first, identifying the orientation and movement direction characteristics of the bird flock, and then controlling the laser movement trajectory according to the positional relationship between the orientation of the bird flock and the monitored protection area.

[0012] As a further improvement of the technical solution, the control gimbal module includes various axis systems of the turntable for changing the position, and each axis system of the turntable is a three-loop control structure, including a current loop, a speed loop and a position loop.

[0013] As a further improvement of the present technical solution, the current loop in the gimbal control module is composed of current feedback, a current controller and a PWM power amplifier. The current loop is implemented inside the power amplifier. The current controller is used for current correction. The voltage of the control motor is proportional to the torque of the motor, and the control voltage directly controls the acceleration. The input variable of the speed loop in the gimbal control module is the difference between the speed command and the actual speed calculated by the position sensor. The angle sensor and the angle measurement circuit are used to obtain the actual deflection angle of the turntable, and then the actual speed is obtained based on the rate estimate. The input variable of the position loop in the gimbal control module is the difference between the position planning command and the actual position. The deviation is adjusted by the position correction algorithm, and its output is used as the original value of the speed command.

[0014] As a further improvement of the present technical solution, the control gimbal module realizes target tracking based on the perception of the target position, and calculates the offset of the target position relative to the visual axis through the core control module. The offset is obtained by solving the miss amount of the target in the detector image coordinate system. The image coordinate system is defined on the target surface of the detection device. The origin of the coordinate system is the intersection of the optical axis center and the detector target surface. The coordinate axes of the image coordinate system are established in the horizontal and vertical directions on the target surface with the u axis and the v axis as the coordinate axes of the image coordinate system. The coordinate value of the target feature point in the image coordinate system is (x, y), which is the miss amount of the target in the image coordinate system. The angle of the tracked target offset relative to the optical visual axis is obtained as follows:

[0015]

[0016] Where f is the focal length of the optical detection device, θ u The target is equivalent to the misalignment angle of the optical axis in the u-axis direction, θ v is the misalignment angle of the target relative to the optical axis in the v-axis direction.

[0017] As a further improvement of the present technical solution, when the core control module detects and tracks flying birds in a no-fly zone, the ECO algorithm is used to track the target, the STC target tracking algorithm is responsible for short-term tracking when the ECO algorithm is initialized, and the YOLO-V3 target detection algorithm is responsible for detecting the target to assist the ECO algorithm in target tracking.

[0018] As a further improvement of the present technical solution, the specific steps for detecting and tracking flying birds in the no-fly zone in the core control module are as follows: the YOLO-V3 target detection algorithm and the ECO target tracking algorithm are processed in parallel; the STC target tracking algorithm performs a short target tracking when the ECO algorithm is initialized; after the ECO algorithm is initialized, the tracking result coordinate information of the previous frame STC is passed to the ECO module, and the ECO module is used to replace the STC target tracking; the YOLO-V3 algorithm extracts features of different scales through the backbone network and then post-processes the multi-scale high-latitude features; the ECO algorithm extracts multi-scale Hog features and CN features from the initial frame, and uses the YOLO-V3 detection algorithm to assist in correcting the tracking frame during the target tracking process; the STC target tracking algorithm extracts the template frame and establishes a spatiotemporal context model; the spatiotemporal context model is used to track and locate the target by calculating the maximum posterior probability for subsequent frames.

[0019] As a further improvement of the present technical solution, when detecting and tracking birds in a no-fly zone in the core control module, in the YOLO-V3 target detection algorithm, a pyramid network is used for multi-size detection, Softmax is removed and logistic regression is selected as the classifier, the Darknet-53 network architecture is used, all pooling layers and fully connected layers are discarded in the network structure, and a convolution kernel with a step size of 2 is used to replace the downsampling operation in the previous model. At the same time, a residual structure is added and a shortcut connections structure is added.

[0020] As a further improvement of the present technical solution, when the YOLO-V3 target detection algorithm is used in the core control module, the convolution layer is composed of multiple convolution kernels. In the forward propagation process, different convolution kernels traverse the input image, find the position with the largest filter response, screen out effective features and output corresponding feature maps; extract features through continuous training of the network; at the output, the model outputs three feature maps of different sizes, and then selects corresponding AnchorBoxes at each size for prediction. The Anchor Boxes size is preset in advance through clustering, and Anchor Boxes of 9 sizes, namely large, medium and small, are selected. Three corresponding Anchor Boxes of appropriate sizes are allocated at each size, corresponding to the three feature maps output by the network to detect targets of each size respectively, and the classifier uses multiple logistic classifiers.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. The bird intelligent identification, tracking and driving away device uses laser bird driving. Compared with traditional bird driving methods, laser bird driving is more efficient and can cover a large area in a very short time without contacting birds, thus avoiding contact between people and birds. Laser bird driving uses non-lethal laser beams, which will not cause harm to birds or humans. Laser bird driving can also be adjusted according to different environments and different conditions to achieve the desired bird driving effect.

[0023] 2. The bird intelligent identification, tracking and repelling device realizes real-time processing of multiple algorithms including STC target tracking algorithm, ECO algorithm and YOLO-V3 detection algorithm, which improves the operation efficiency of the system. The YOLO-V3 detection algorithm can adjust the size of the tracking frame according to the size of the identified target, which can better adapt to targets of different sizes and improve the tracking accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is the overall flow chart of the present invention;

[0025] Figure 2 A diagram showing a method for calculating an offset in the present invention;

[0026] Figure 3 It is a flow chart of controlling the turntable in the present invention;

[0027] Figure 4 A flowchart of tracking targets in the present invention;

[0028] Figure 5 It is a structural diagram of the YOLO-V3 detection algorithm in the present invention.

[0029] The meanings of the marks in the figure are as follows: 100, video monitoring module; 200, laser bird-repelling module; 300, pan / tilt control module; 400, core control module. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] like Figure 1-Figure 5 As shown, the purpose of this embodiment is to provide a bird intelligent identification, tracking and driving away device, including:

[0032] Video monitoring module 100: used to use a camera to monitor the situation of birds in the no-fly zone in real time, and use multiple cameras to monitor to ensure no blind spots;

[0033] Laser bird-repellent module 200: used to drive away birds using lasers. It uses the characteristic that birds are highly sensitive to the green spectrum with a wavelength of 532nm, and uses a green laser beam to sweep the pole tower and its surrounding area along a programmable path. When working, it emits a beam or line of green laser to simulate the conditions for biological visual reflection. After being stimulated by the bright green laser beam, the birds fly away from the pole tower and the protected area of ​​the transmission line out of a survival reaction;

[0034] Control PTZ module 300: used to install video monitoring module 100 and laser bird repellent module 200, motion elements include position, speed and torque;

[0035] Core control module 400: used to collect video information transmitted by the camera in the video monitoring module 100, and detect and track flying birds in the no-fly zone through identification and analysis;

[0036] The use of bird intelligent identification, tracking and removal devices can effectively drive away birds flying around power equipment and facilities such as substations and transmission lines, as well as birds resting on facilities, effectively reducing the pollution and harm to power facilities caused by bird excrement and nesting, and improving the reliability and stability of power system operation.

[0037] In this embodiment, the working principle of the laser bird repellent is to achieve the repelling effect by interfering with the bird's visual system through the laser beam. The light and heat energy of the laser beam are used to stimulate the birds, making them feel uncomfortable and leaving the target area. Pulsed laser technology is usually used to emit a series of high-intensity, short-pulse laser beams. These pulsed lasers form bright and visible light spots or dynamic line patterns in the air, which can illuminate the birds at a long distance. When the laser beam is irradiated to the bird's body or surrounding environment, part of the laser light energy will be absorbed or scattered to generate light and heat energy. This light and heat energy stimulates the visual, auditory and sensory systems of birds, attracting their attention and tension, making them feel uneasy and uncomfortable, and tending to leave the area irradiated by the laser. The effect of the bird-repelling laser depends on multiple factors, including the power, frequency, wavelength, pulse mode, etc. of the laser. Generally speaking, high power, frequency and wavelength suitable for bird visual perception, and pulse mode with a high frequency of change will increase the effect of the laser. It has the characteristics of environmental protection, non-toxicity, and high efficiency, and is widely used in farmland, orchards, power facilities, airports, docks, etc. The laser in the laser bird-repelling module 200 uses basic bird-repelling strategies and advanced bird-repelling strategies to achieve the bird-repelling effect. The role of the basic bird-repelling strategy is to drive away flying birds, keep flying birds away from the monitoring and protection area, and control the intensity and stroboscopic frequency of the laser to drive away flying birds. The advanced bird-repelling strategy is to drive away flying birds in a given direction, specifically including: first, identifying the characteristics of the bird flock's orientation, movement direction, etc., and then controlling the laser movement trajectory according to the positional relationship between the bird flock's orientation and the monitoring and protection area, and evaluating the driving-away efficiency function in real time during the driving-away process, and adaptively adjusting the laser working and movement parameters.

[0038] When using laser bird repellent, attention should be paid to the safety of birds and humans to avoid harm to birds. At the same time, when a person is identified on the laser track, the person's range should be avoided. After the bird is driven away, the laser automatically turns off to reduce power consumption and prevent it from affecting the normal state of other equipment or the environment.

[0039] refer to Figure 3 In this embodiment, the gimbal control module 300 includes various axis systems of the turntable for changing the position. Each axis system of the turntable is a three-loop control structure, including a current loop, a rate loop and a position loop. The rate PI and the position PID are both implemented by the control software.

[0040] Furthermore, the current loop in the control pan-tilt module 300 is composed of current feedback, current controller and PWM power amplifier. The current loop is implemented inside the power amplifier. The current controller is used for current correction. Current negative feedback is introduced to make the voltage of the control motor proportional to the torque of the motor. The control voltage can directly control the acceleration, thereby improving the dynamic anti-interference ability of the system and preventing the power conversion circuit and the motor from overcurrent. The input variable of the speed loop in the control pan-tilt module 300 is the difference between the speed command and the actual speed calculated by the position sensor. The angle sensor and the angle measurement circuit are used to obtain the actual deflection angle of the turntable, and then the actual speed is obtained according to the rate estimation. The deviation is adjusted by the speed PI controller, and its output is used as the original value of the torque current command. The introduction of the rate loop can increase the damping of the system and improve the open-loop gain of the entire system, so that the position loop has higher servo stiffness and better dynamic performance. The input variable of the position loop in the control pan-tilt module 300 is the difference between the position planning command and the actual position. The deviation is adjusted by the position correction algorithm, and its output is used as the original value of the speed command. The position loop is the main feedback loop of the system, which is used to ensure the accuracy of the system.

[0041] In addition, based on the classic PID, a feedforward control method is introduced to improve the dynamic quality of the system, increase the response speed and tracking command capability, and expand the system bandwidth; an active resistance torque friction compensation strategy is introduced to compensate for the influence of mechanical dead zone and friction on positioning accuracy;

[0042] In addition, to avoid mechanical resonance, a second-order low-pass filter or notch filter is introduced in the forward channel of the system to eliminate the influence of mechanical resonance; multiple sets of controllers are designed to achieve the best control effect in response to changes in system load; during the debugging process, under the premise of suppressing interference, the bandwidth of the system should be increased as much as possible to improve the response speed of the system and suppress the dynamic coupling torque.

[0043] refer to Figure 4 , further, the control gimbal module 300 realizes target tracking based on the perception of the target position, and calculates the offset of the target position relative to the visual axis through the core control module 400. The offset is obtained by solving the miss amount of the target in the detector image coordinate system. The image coordinate system is defined on the target surface of the detection device. The origin of the coordinate system is the intersection of the optical axis center and the detector target surface. The coordinate axes of the image coordinate system are established in the horizontal and vertical directions on the target surface with the u axis and the v axis as the coordinate axes of the image coordinate system. The coordinate value of the target feature point in the image coordinate system is (x, y), which is the miss amount of the target in the image coordinate system. The offset angle of the tracked target relative to the optical visual axis is obtained as:

[0044]

[0045] Where f is the focal length of the optical detection device, θ uThe target is equivalent to the misalignment angle of the optical axis in the u-axis direction, θ v It is the misalignment angle of the target relative to the optical axis in the v-axis direction. It can calculate the misalignment angle position that each frame axis needs to rotate during target tracking, and complete the compensation of the target offset relative to the optical axis.

[0046] refer to Figure 4 In this embodiment, when the core control module 400 detects and tracks flying birds in a no-fly zone, the ECO algorithm is used to track the target, the STC target tracking algorithm is responsible for short-term tracking when the ECO algorithm is initialized, and the YOLO-V3 target detection algorithm is responsible for detecting the target to assist the ECO algorithm in target tracking.

[0047] Furthermore, the specific steps of detecting and tracking flying birds in the no-fly zone in the core control module 400 are as follows: the YOLO-V3 target detection algorithm and the ECO target tracking algorithm are processed in parallel. When the ECO algorithm is initialized, the STC target tracking algorithm performs a short target tracking. After the ECO algorithm is initialized, the tracking result coordinate information of the previous frame STC is passed to the ECO module, and the ECO module is used to replace the STC target tracking. The YOLO-V3 target detection algorithm is a single-stage one-shot algorithm and does not include the proposal selection in the two-stage algorithms such as Faster-RCNN. After the YOLO-V3 target detection algorithm extracts features of different scales through the backbone network, it directly post-processes the multi-scale high-latitude features to achieve target detection; the ECO algorithm extracts rich multi-scale Hog features and CN features from the initial frame, and uses the YOLO-V3 detection algorithm to assist in correcting the tracking frame during the target tracking process. The STC target tracking algorithm extracts the template frame and establishes a spatiotemporal context model. The spatiotemporal context model is used to track and locate the target by calculating the maximum posterior probability for subsequent frames.

[0048] refer to Figure 5 Furthermore, the convolutional neural network extracts the features of the input image through the convolution operation of the network, and can retain the spatial relationship between pixels; a typical convolutional neural network mainly includes four operations: convolution, nonlinear processing, pooling and classification recognition; when detecting and tracking birds in the no-fly zone in the core control module 400, the YOLO-V3 target detection algorithm strengthens the feature extraction and classification capabilities of the image, uses a pyramid network for multi-size detection, removes Softmax and uses logistic regression as a classifier, uses the Darknet-53 network architecture, discards all pooling layers and fully connected layers in the network structure, and uses a convolution kernel with a step size of 2 to replace the downsampling operation in the previous model. At the same time, a residual structure is added, and a shortcut connections structure is added.

[0049] Furthermore, when the YOLO-V3 target detection algorithm is used in the core control module 400, the convolution layer is composed of multiple convolution kernels. In the forward propagation process, different convolution kernels traverse the input image to find the position with the largest filter response, thereby screening out effective features and outputting corresponding feature maps; through continuous training of the network, the parameters in the convolution kernel will also become more targeted, so that features such as texture and edges can be effectively extracted; in terms of output, the model directly outputs three feature maps of different sizes, and then selects the corresponding Anchor at each size. Boxes is used for prediction. The specific implementation details are as follows: the feature map Y1 is obtained by downsampling the input image 32 times in total at the 79th layer of the entire network, which is used to detect larger objects. Before obtaining Y1, its input is upsampled by 2 times, and then fused with the feature map of the previous 61st layer to obtain a feature map with a cumulative downsampling of 16 times, which is used to detect medium-sized objects. Before obtaining Y2, its input is also upsampled by 2 times, and then fused with the feature map of the 36th layer to obtain a feature map with a cumulative downsampling of 8 times, which is used to detect small objects. In multi-scale detection, the size of Anchor Boxes is preset in advance through clustering, and finally 9 sizes of Anchor Boxes, large, medium and small, are selected. Three corresponding Anchor Boxes of appropriate sizes are allocated to each size, corresponding to the three feature maps output by the network, to detect targets of each size respectively, and the classifier uses multiple logistic classifiers.

[0050] During training, the training data set is VAL2017, the test set is VOC2007, the number of categories is 80, the training set and the test set are both 3,000 images, and the network input is a 416×416 three-channel color image; the training set and the test set are both open source test sets and can be provided to users for later use. For the special scene recognition of bird repelling, the network is additionally trained with videos and images provided by users, and the new module is injected into the image module to enhance the image tracker's recognition ability for specific scenes.

[0051] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bird intelligent identification, tracking and driving away device, It is characterized in that include: Video monitoring module (100): used to monitor the situation of flying birds in the no-fly zone in real time using a camera; Laser bird-repelling module (200): used to repel birds using laser; A pan / tilt control module (300): used for installing the video monitoring module (100) and the laser bird-repelling module (200); The core control module (400) is used to collect video information transmitted by the camera in the video monitoring module (100), and detect and track flying birds in the no-fly zone through identification and analysis.

2. The bird intelligent identification, tracking and driving away device according to claim 1, Features: The laser in the laser bird-repelling module (200) uses a basic bird-repelling strategy and an advanced bird-repelling strategy to achieve a bird-repelling effect; the basic bird-repelling strategy is used to drive away flying birds and keep them away from the monitored protection area; The advanced bird-repelling strategy is to drive away flying birds in a given direction. Specifically, it includes: first, identifying the orientation and movement direction characteristics of the bird flock, and then controlling the laser movement trajectory based on the positional relationship between the bird flock orientation and the monitored protection area.

3. The bird intelligent identification, tracking and driving away device according to claim 1, Features: The pan / tilt control module (300) comprises various axis systems of a turntable for changing position, and each axis system of the turntable is a three-loop control structure, including a current loop, a speed loop and a position loop.

4. The bird intelligent identification, tracking and driving away device according to claim 3, Features: The current loop in the pan-tilt control module (300) is composed of current feedback, a current controller and a PWM power amplifier. The current loop is implemented inside the power amplifier. The current controller is used for current correction. The voltage of the control motor is proportional to the torque of the motor. The control voltage directly controls the acceleration. The speed loop input variable in the pan-tilt control module (300) is the difference between the speed command and the actual rotation speed calculated by the position sensor. The angle measurement sensor and the angle measurement circuit are used to obtain the actual deflection angle of the turntable, and then the actual rotation speed is obtained according to the speed estimation. The position loop input variable in the pan-tilt control module (300) is the difference between the position planning command and the actual position. The deviation is adjusted by a position correction algorithm, and its output is used as the original value of the speed command.

5. The bird intelligent identification, tracking and driving away device according to claim 4, Features: The control pan / tilt module (300) realizes target tracking based on the perception of the target position, and calculates the offset of the target position relative to the visual axis through the core control module (400), the offset is obtained by solving the miss amount of the target in the detector image coordinate system, and an image coordinate system is defined on the target surface of the detection device, the origin of the coordinate system is the intersection of the optical axis center and the detector target surface, and the coordinate axes of the image coordinate system are established in the horizontal and vertical directions of the target surface with the u axis and the v axis as the coordinate axes of the image coordinate system, the coordinate value of the target feature point in the image coordinate system is (x, y), and the coordinate value is the miss amount of the target in the image coordinate system, and the offset angle of the tracked target relative to the optical visual axis is obtained as follows: Where f is the focal length of the optical detection device, θ u The target is equivalent to the misalignment angle of the optical axis in the u-axis direction, θ v is the misalignment angle of the target relative to the optical axis in the v-axis direction.

6. The bird intelligent identification, tracking and driving away device according to claim 1, Features: When the core control module (400) detects and tracks flying birds in a no-fly zone, the ECO algorithm is used to track the target, the STC target tracking algorithm is responsible for short-term tracking when the ECO algorithm is initialized, and the YOLO-V3 target detection algorithm is responsible for detecting the target to assist the ECO algorithm in tracking the target.

7. The bird intelligent identification, tracking and driving away device according to claim 6, It is characterized in that The specific steps of detecting and tracking flying birds in the no-fly zone in the core control module (400) are as follows: the YOLO-V3 target detection algorithm and the ECO target tracking algorithm are processed in parallel; the STC target tracking algorithm performs short-term target tracking when the ECO algorithm is initialized; after the ECO algorithm is initialized, the tracking result coordinate information of the previous frame of STC is transferred to the ECO module, and the ECO module is used to replace the STC target tracking; the YOLO-V3 algorithm extracts different scale features through the backbone network, and then performs post-processing on the multi-scale high-latitude features; The ECO algorithm extracts multi-scale Hog features and CN features from the initial frame, and uses the YOLO-V3 detection algorithm to assist in correcting the tracking frame during target tracking. The STC target tracking algorithm extracts the template frame and establishes a spatiotemporal context model. The spatiotemporal context model is used to track and locate the target by calculating the maximum posterior probability for subsequent frames.

8. The bird intelligent identification, tracking and driving away device according to claim 7, Features: When detecting and tracking flying birds in a no-fly zone in the core control module (400), in the YOLO-V3 target detection algorithm, a pyramid network is used for multi-scale detection, Softmax is removed and logistic regression is selected as a classifier, a Darknet-53 network architecture is used, all pooling layers and fully connected layers are discarded in the network structure, a convolution kernel with a step size of 2 is used to replace the downsampling operation in the previous model, and a residual structure is added, and a shortcut connections structure is added.

9. The bird intelligent identification, tracking and driving away device according to claim 8, Features: When the YOLO-V3 target detection algorithm is used in the core control module (400), the convolution layer is composed of multiple convolution kernels. In the forward propagation process, different convolution kernels traverse the input image to find the position with the largest filter response, screen out effective features and output corresponding feature maps; extract features through continuous network training; in output, the model outputs three feature maps of different sizes, and then selects corresponding Anchor Boxes at each size for prediction, and presets the size of Anchor Boxes in advance through clustering, selects Anchor Boxes of 9 sizes, namely large, medium and small, and allocates three corresponding Anchor Boxes of appropriate sizes at each size to correspond to the three feature maps output by the network to detect targets of different sizes respectively, and the classifier uses multiple logistic classifiers.

Images