Bird nest removing system based on unmanned aerial vehicle detection and deep learning

Abstract

The invention discloses a bird nest removing system based on unmanned aerial vehicle detection and deep learning, and the system employs an image collection module of an unmanned aerial vehicle to carry out the aerial photographing of a target area, the captured image data is transmitted to a ground terminal, and the ground terminal carries out the deep analysis of the transmitted image through a detection module, and generates a detailed detection report. A detection report is input to a judgment module, if it is judged that the bird nest exists in the detection information, an alarm module is triggered immediately, a clearing module of the unmanned aerial vehicle is started and controlled, a clearing or grabbing device is started, fixed-point clearing is conducted on the target bird nest, and meanwhile the unmanned aerial vehicle continuously uses a high-resolution camera to monitor the whole clearing process. And after the clearing task is finished, the unmanned aerial vehicle performs inspection shooting and detection on the target area again to verify whether the bird nest is thoroughly cleared. According to the method, image acquisition, deep learning detection, intelligent discrimination and diversified clearing technologies are combined, and the bird nest on the high-voltage power transmission facility can be quickly and accurately identified and efficiently cleared.

Description

Technical Field

[0001] The present invention relates to intelligent operation and maintenance technology for high-voltage power transmission facilities, and specifically to a bird's nest removal system based on drone detection and deep learning. Background Art

[0002] In the field of power system maintenance, bird nests pose a serious threat to the safety of high-voltage lines and towers. Traditional bird nest removal methods rely primarily on manual climbing of towers, a time-consuming and labor-intensive method that also carries high safety risks. With the rapid development of drone technology, their use in power inspections is becoming increasingly widespread. UAVs can carry high-definition cameras and a variety of sensors, enabling efficient inspections of high-voltage lines and towers. However, existing drone bird nest detection technology suffers from low recognition accuracy and high rates of false detections and missed detections in complex environments. Furthermore, drone methods for removing bird nests are relatively simple, typically employing methods such as mechanical cutting, which makes it difficult to handle bird nests in diverse locations and structures.

[0003] Therefore, developing an efficient and accurate bird nest detection and removal method and device based on drones is of great significance to ensuring the safe and stable operation of the power system. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to address the deficiencies in the prior art and provide a bird nest removal system based on drone detection and deep learning. Through the deep integration of drones and deep learning technology, efficient and accurate identification and removal of bird nests on high-voltage transmission facilities can be achieved. The present invention greatly reduces the cost of manual inspections and the time required for removal, improves the accuracy and reliability of bird nest detection and removal, provides strong technical support for bird pest control in power systems, and effectively ensures the safe and stable operation of high-voltage transmission facilities.

[0005] Technical solution: The present invention is a bird nest removal system based on drone detection and deep learning. The drone is equipped with an image acquisition module, a communication module, a positioning module, and a removal device. The specific removal method is as follows:

[0006] Step 1: After the UAV flies to the target area (such as high-voltage transmission facilities), the image acquisition module uses the camera to collect the corresponding image and transmits it to the ground control terminal using the communication module;

[0007] Step 2: The ground control terminal uses the judgment module to analyze the received image and generate a detection report. The detection report includes the specific location, bounding box, category information and confidence level of the bird's nest in the target area. The detection judgment module uses the improved YOLO11 model. After the image is input into the detection module, it is analyzed and outputs the location and confidence level of the automatic bird's nest. The specific method is as follows:

[0008] Step 2.1: The input image is first scaled to a preset fixed size. The downsampling Conv module and C3K2 module of the backbone network in the detection module perform feature extraction on the scaled image to obtain original feature maps of three different scales.

[0009] Step 2.2: The obtained feature maps of the three sizes are fed into the Neck layer of the detection module for feature fusion. Before the Neck layer feature fusion, a convolutional block attention module (CBAM) is added to dynamically adjust the feature weights through the channel-spatial dual attention mechanism to enhance the regional features of the bird's nest (such as texture and shape) and suppress background interference such as high-voltage lines.

[0010] Step 2.3: The three feature maps output by the Neck layer are fed into the detection head of the Head layer, and candidate detection boxes and category scores are generated through the convolutional network. Subsequently, an adaptive non-maximum suppression (NMS) mechanism is introduced to dynamically adjust the suppression threshold according to the distribution density of bird nests in the image. The fused features enhanced by the CBAM of the Neck layer are obtained.

[0011] Step 2.4: Dynamically balance missed detections and false detections, and finally output the optimized detection results (including the location of the bird's nest, bounding box, confidence score, and timestamp).

[0012] Step 3: Input the detection report into the judgment module. If the judgment module recognizes that there is a bird's nest in the detection information, the alarm module is triggered and the detection information is stored in the data storage module;

[0013] Step 4: Based on the inspection report obtained in step 2, the ground operator of the UAV starts and controls the removal device on the UAV to remove the bird's nest; the removal device includes a high-pressure gas purge module and a manipulator removal module.

[0014] By using a high-pressure gas device or equipping a drone with a manipulator, for example, during flight, a bird's nest can be scanned using a bird-related foreign object distance sensor to obtain the distance and offset angle between the nest and the drone in real time. Based on this information, the direction and length of the telescopic rod and manipulator are adjusted. The manipulator then grabs and removes the nest, while the drone uses a high-resolution camera to monitor the removal process in real time.

[0015] During the cleaning process, the positioning module obtains real-time position information to dynamically adjust the cleaning device;

[0016] Step 5: After the removal task is completed, the drone will collect images of the target area again and repeat steps 1 to 4 to test the removal effect. If the judgment module still detects a bird's nest, the ground operator will decide whether to re-execute steps 4 and 5 until it is confirmed that the bird's nest has been completely removed.

[0017] Step 6. After the cleaning is completed, a detailed report will be generated based on the status records before and after the cleaning. The detailed report includes the three-dimensional spatial coordinates of the bird's nest in the target area, the bird's nest feature matrix, the cleaning operation log, and the comparison image, and is archived in the archive system.

[0018] Furthermore, the image acquisition module utilizes a high-resolution camera, which is mounted on the drone to capture images of the target area, including high-voltage power lines and towers. This camera possesses excellent optical performance and autofocus capabilities, enabling clear capture of details of the bird's nest under varying lighting conditions. Furthermore, the camera's zoom capability allows for zoomed-in photography of distant nests, facilitating subsequent image processing and analysis.

[0019] To improve detection efficiency and accuracy, the image acquisition module works together to analyze the collected images using computer vision and image processing technology, effectively eliminating background interference and accurately identifying bird nests. The working process of the convolutional block attention module (CBAM) in the detection module is as follows:

[0020] First, for the input feature map F∈R C×H×W , processed using the Channel Attention mechanism; C is the number of channels, H×W is the spatial dimension;

[0021] Global pooling operation: Average pooling:

[0022] Max Pooling:

[0023] MLP with shared weights:

[0024] Compression ratio r (usually 16):

[0025] MLP(x)=W1·δ(W0·x),W0∈R C / r×C ,W1∈R C×C / r ;

[0026] Activation function δ: ReLU

[0027] Sigmoid activation: M c (F) = σ(MLP(F avg )+MLP(F max ));

[0028] Output feature map:

[0029] is channel-by-channel multiplication;

[0030] Then, the input feature map F′∈R C×H×WUse Spatial Attention to process:

[0031] Channel dimension pooling operation:

[0032] Average Pooling:

[0033] Max Pooling:

[0034] Convolutional fusion:

[0035] Convolution kernel size 7×7:

[0036] Sigmoid activation:

[0037] Finally, the feature map is output:

[0038] is the multiplication of spatial positions;

[0039] (3) CBAM complete formula

[0040]

[0041] The detection module introduces an adaptive non-maximum suppression (NMS) mechanism to dynamically adjust the suppression threshold according to the density of bird nests in the image. The adjustment method is:

[0042] For densely populated areas (such as the top of a high-voltage tower): Lower the NMS threshold to 0.4, relax the inter-frame overlap tolerance, prioritize high recall, and avoid missing dense targets.

[0043] For sparse areas (such as a single transmission line): increase the NMS threshold to 0.6, strengthen the overlap screening between frames, and suppress false detection interference.

[0044] Furthermore, in order to facilitate the accurate operation of the cleaning device and reduce misoperation, the specific work content of the judgment module is as follows:

[0045] Based on the location and size of the bird's nest output by the detection module, a multispectral imaging system is used to analyze the material's spectral characteristics. A lidar scan is used to construct a three-dimensional spatial topology model. An electric field sensor array measures local electromagnetic environment parameters. A millimeter-wave radar monitors ambient airflow disturbances. A stereo vision system is used to analyze the contact surface geometry, thereby constructing a bird's nest feature matrix [N, D, E, S, C]; where N is the material conductivity coefficient (0-1), D is the distance to the charged object (meters), E is the electric field strength (kV / m), S is the wind speed (m / s), and C is the conductor contact area ratio (%).

[0046] The judgment module further analyzes the location and characteristics of the bird's nest, such as material and structural stability, to determine whether it needs to be removed and the best removal strategy;

[0047] If N<0.5 (dead branches / grass leaves), D≥1.2, and E<8 are met simultaneously, the "simple scenario" prompt is triggered and a high-pressure gas removal path is generated;

[0048] If N ≥ 0.5 (metal / plastic characteristic spectrum matching) or C > 30% (calculated by point cloud contact surface fitting) or S > 10 (combined with radar echo turbulence detection), the "complex scene" prompt is activated and the clearing device is started.

[0049] The calculation methods of the above parameters such as N, D, E, S, and C are directly obtained based on the detection of professional equipment.

[0050] To achieve precise positioning and navigation of drones in complex environments such as high-voltage lines and high-voltage towers, the positioning module includes a GPS positioning module auxiliary positioning sensor. The GPS positioning module receives satellite signals through the drone's anti-electromagnetic interference enhanced GPS receiver. It has a built-in multi-layer magnetic shield and adaptive bandpass filter. In a 50kV / m high-voltage electric field environment, the positioning error is ≤0.1m. It obtains the drone's precise location information in real time, including coordinate data such as longitude, latitude, and altitude, enabling the drone to accurately locate the specific location of the bird's nest in the high-voltage line corridor and high-voltage tower area (with an accuracy of ±10cm). The module also needs to eliminate interference from reflected signals from the tower's metal structure through a multipath suppression algorithm to provide accurate navigation guidance for detection and removal operations. The auxiliary positioning sensor module integrates sensors such as visual sensors, lidar, and IMU, and fuses them with GPS data. Specific methods include collaborating with GPS spatial data through a loosely coupled fusion architecture based on coordinate transformation matrices, and using the PTP protocol to achieve μs-level time synchronization to improve positioning and timing accuracy; visual sensors can be considered to be equipped with anti-reflective filters to provide continuous positioning by tracking the sag characteristics of the wires when the GPS signal is lost; lidar constructs a three-dimensional point cloud to assist in obstacle avoidance in complex environments; IMU data and GPS positioning information are fused through quaternion interpolation to ensure that the positioning drift rate is <0.1m / s under strong electromagnetic interference, ensuring continuous and stable flight positioning.

[0051] Furthermore, the drone is also equipped with an alarm module, which monitors the drone's flight status and surrounding environment in real time, such as wind speed, wind direction, obstacles, etc.; after detecting a bird's nest, the alarm module triggers an alarm signal based on the analysis results of the judgment module. If the detected bird's nest is located on a high-voltage wire or is made of materials such as branches and dead grass, and there are potential safety hazards, the alarm module will issue an audible and visual alarm, and push real-time alarm information to the operator through the communication module, prompting the specific location, type and potential risks of the bird's nest. If the bird's nest is made of non-conductive materials (dead branches / grass leaves), is ≥1.2 meters away from the charged body, and the electric field strength is <8kV / m, it can be classified as "simple scene" and sent to the operator as a prompt. If the nest contains non-degradable materials such as metal / plastic or the contact area with the wire is >30%, and there is dynamic interference (wind speed >10m / s, bird activity, etc.), it can be classified as "complex scene" and sent to the operator as a prompt.

[0052] Furthermore, the removal device includes a high-pressure gas purge module and a robotic arm removal module. The high-pressure gas purge module uses a high-pressure gas injection device composed of a carbon fiber gas tank, a vector-adjustable nozzle, and a pulsed solenoid valve equipped on the drone to carry out targeted blowing and removal of bird nests made of loose materials such as branches and dead grass on high-voltage power lines. The operator can remotely adjust the injection pressure of 0.5-0.8MPa through the ground terminal and control the nozzle to deflect by ±15°. Combined with the air pressure-distance linkage algorithm, it realizes non-contact and safe removal. The robotic arm removal module: Through a robotic arm equipped with a six-degree-of-freedom robotic arm, an adaptive pressure feedback gripper, and an insulated wrist joint, it removes bird nests on high-voltage towers that are entangled with metal wires, adhered to conductors, and subject to dynamic interference (wind speed > 10m / s, bird activity, etc.). The joints of the robot are wrapped with composite insulating material, and the laser visual servo system achieves a positioning accuracy of ±0.5mm. The operator can remotely switch the clamping mode (5-100N adjustable pressure) and plan the three-dimensional disassembly path to ensure millimeter-level precision operation in an electrical environment.

[0053] Furthermore, the safety warning module integrates millimeter-wave radar, dual-spectrum cameras, electric field sensors, and other elements to monitor the drone's flight status and surrounding environmental parameters in real time, including wind speed, wind direction, obstacle distribution, and high-voltage electric field strength. When a medium- or long-range obstacle is detected, the device overheats, or a strong electromagnetic field is detected, a multi-level warning mechanism is triggered, sending a warning signal to the ground station via an encrypted channel. This simultaneously initiates an emergency avoidance or automatic return sequence to ensure the safety of the drone and the operator. This safety warning module ensures the drone's normal operation and timely response and return in the event of an emergency. The aforementioned alarm module, upon detecting a bird's nest, issues an alarm to alert personnel for the next step of removal.

[0054] In order to record and organize the relevant data of UAV nest removal and generate detailed reports, the ground control terminal is also equipped with an archiving and reporting module, which includes a data storage module and a report generation module.

[0055] The data storage module stores various data collected by the drone during the bird nest removal process, including nest detection data, drone flight data, and removal operation data, providing a reference for subsequent removal work. The report generation module automatically generates a detailed bird nest removal report based on this data. The report includes basic information about the removal task, such as the time and location of the mission, drone model, and type of removal device. It also details the number, location, and type of nests detected, along with accompanying imagery to visually demonstrate the distribution of the nests. The report also records the specific removal method, time, and effectiveness of each nest, as well as the drone's flight status and operator control commands. Finally, the removal results are evaluated, analyzing the success rate, efficiency, and any remaining issues, and providing recommendations for improvement.

[0056] Furthermore, the drone is also equipped with an energy management module, which monitors the battery power and energy consumption of the drone, reasonably allocates and manages the energy of the drone, ensures that the drone has sufficient power support when performing tasks, and provides early warning when the battery is low, reminding ground operators to replace the battery in time or return the drone for charging, to avoid the drone falling or mission interruption due to depletion of power.

[0057] Beneficial Effects: This invention enables rapid and accurate identification and efficient removal of bird nests on high-voltage transmission facilities. By cleverly combining image acquisition, deep learning detection, intelligent discrimination, and diverse removal technologies, these modules work closely together to significantly reduce the risks and costs of manual inspections while significantly improving the efficiency and accuracy of bird nest removal, providing a strong guarantee for the safe and stable operation of high-voltage transmission facilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0059] Figure 2 This is a diagram of the improved YOLO11 framework and the judgment process of the detection module in the present invention;

[0060] Figure 3 It is a relationship diagram of the modules of the present invention when performing tasks. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is described in detail below, but the protection scope of the present invention is not limited to the embodiments.

[0062] First, drones equipped with high-resolution cameras capture aerial images of target areas (such as power towers or power lines). The captured images are then transmitted to a ground control center. The ground terminal uses a modified YOLO11 model to perform in-depth analysis of the incoming images and generate a detailed inspection report.

[0063] Subsequently, this detection report is input into the identification system. If it is determined that there is a bird's nest in the detection information, the alarm device will be triggered immediately.

[0064] Based on this detection intelligence, the ground drone operator can quickly respond by activating the drone's removal module, activating the cleaning or grabbing mechanism, and performing targeted removal of the nest. Meanwhile, the drone continuously monitors the removal process with its high-resolution camera. After the removal mission is complete, the drone will conduct another round of inspections, photographing and testing the target area to verify complete removal. If an alarm is triggered, personnel will assess the need for additional measures based on the inspection report, restarting the removal process if necessary. Finally, the pre- and post-removal status is recorded in a report and stored securely in the archive system for future reference and evaluation.

[0065] like Figures 1 to 3 As shown in the figure, the bird nest removal system based on drone detection and deep learning of the present invention is designed to efficiently solve the problem of bird nest detection and removal on high-voltage transmission facilities. The drone is equipped with an image acquisition module, a communication module, a positioning module, and a removal device. The specific removal method is as follows:

[0066] Step 1: After the drone flies to the target area, it uses the image acquisition module to collect the corresponding image through the camera and transmits it to the ground control terminal using the communication module;

[0067] Step 2: The ground control terminal uses the judgment module to analyze the received image and generate a detection report. The detection report includes the specific location, bounding box, category information and confidence level of the bird's nest in the target area. The detection module uses the improved YOLO11 model. After the image is input into the detection module, it is analyzed and outputs the location and confidence level of the bird's nest in the image. The specific method is as follows:

[0068] Step 2.1: The input image is first scaled to a preset fixed size. The downsampling Conv module and C3K2 module of the backbone network in the detection module perform feature extraction on the scaled image to obtain feature maps of three different scales.

[0069] Step 2.2: The obtained feature maps of the three sizes are fed into the Neck layer of the detection module for feature fusion. Before the Neck layer feature fusion, a convolutional block attention module (CBAM) is added to dynamically adjust the feature weights through the channel-spatial dual attention mechanism.

[0070] Step 2.3: The three feature maps output by the Neck layer are fed into the detection head of the Head layer, where candidate detection boxes and category scores are generated through a convolutional network. Subsequently, an adaptive non-maximum suppression (NMS) mechanism is introduced to dynamically adjust the suppression threshold based on the density of bird nests in the image.

[0071] Step 2.4: Dynamically balance missed detections and false detections, and finally output optimized detection results;

[0072] Step 3: Input the detection report into the judgment module. If the judgment module recognizes that there is a bird's nest in the detection information, the alarm module is triggered and the detection information is stored in the data storage module;

[0073] Step 4: Based on the inspection report obtained in step 2, the ground operator of the UAV activates and controls the removal device on the UAV to remove the bird's nest; the removal device includes a high-pressure gas purge module and a robotic removal module; during the removal process, the positioning module obtains real-time position information to dynamically adjust the removal device;

[0074] Step 5: After the removal task is completed, the drone will collect images of the target area again and repeat steps 1 to 4 to test the removal effect. If the judgment module still detects a bird's nest, the ground operator will decide whether to re-execute steps 4 and 5 until it is confirmed that the bird's nest has been completely removed.

[0075] Step 6. After the cleaning is completed, a detailed report will be generated based on the status records before and after the cleaning. The detailed report includes the three-dimensional spatial coordinates of the bird's nest in the target area, the bird's nest feature matrix, the cleaning operation log, and the comparison image, and is archived in the archive system.

[0076] The image acquisition module of this embodiment uses a high-resolution camera, and the drone is equipped with the camera to capture images containing high-voltage lines and high-voltage towers in the target area.

[0077] like Figure 2 As shown in Figure 2, the working process of the convolutional block attention module CBAM in the detection module is as follows:

[0078] First, for the input feature map F∈R C×H×W , processed using the channel attention mechanism; C is the number of channels, H×W is the spatial dimension;

[0079] Global pooling operation: Average pooling:

[0080] Max Pooling:

[0081] MLP with shared weights:

[0082] Compression ratio r (usually 16):

[0083] MLP(x)=W1·δ(W0·x),W0∈R C / r×C ,W1∈R C×C / r ;

[0084] Activation function δ: ReLU

[0085] Sigmoid activation: M c (F) = σ(MLP(F avg )+MLP(F max ));

[0086] Output feature map:

[0087] is channel-by-channel multiplication;

[0088] Then, the input feature map F′∈R C×H×W Processing using spatial attention mechanism:

[0089] Channel dimension pooling operation:

[0090] Average Pooling:

[0091] Max Pooling:

[0092] Convolutional fusion:

[0093] Convolution kernel size 7×7:

[0094] Sigmoid activation:

[0095] Finally, the output feature map CBAM(F):

[0096] is the multiplication of spatial positions;

[0097] The detection module introduces an adaptive non-maximum suppression (NMS) mechanism to dynamically adjust the suppression threshold according to the density of bird nests in the image. The adjustment method is:

[0098] For dense areas, the NMS threshold is lowered to 0.4, the inter-frame overlap tolerance is relaxed, and high recall is prioritized to avoid missing dense targets.

[0099] For sparse areas, the NMS threshold is increased to 0.6 to strengthen the overlap screening between frames and suppress false detection interference.

[0100] The specific work content of the judgment module in this embodiment is:

[0101] Based on the location and size of the bird's nest output by the detection module, a multispectral imaging system is used to analyze the material's spectral characteristics. A lidar scan is used to construct a three-dimensional spatial topology model. An electric field sensor array measures local electromagnetic environment parameters. A millimeter-wave radar monitors ambient airflow disturbances. A stereo vision system is used to analyze the contact surface geometry, thereby constructing a bird's nest feature matrix [N, D, E, S, C], where N is the material conductivity coefficient, D is the distance to the charged object, E is the electric field intensity, S is the wind speed, and C is the conductor contact area ratio.

[0102] Next, analyze the location and characteristics of the bird's nest, and determine whether it needs to be removed and the best removal strategy:

[0103] If N<0.5, D≥1.2, and E<8 are met at the same time, the "Simple Scenario" prompt is triggered and a high-pressure gas purge path is generated;

[0104] If N≥0.5 or C>30% or S>10, the "complex scene" prompt is activated and the clearing device is started.

[0105] The positioning module of this embodiment includes a GPS positioning module and an auxiliary positioning sensor. The GPS positioning module receives satellite signals through the anti-electromagnetic interference enhanced GPS receiver on the drone, and has a built-in multi-layer magnetic shielding cover and adaptive bandpass filter to obtain the precise location information of the drone in real time, including coordinate data such as longitude, latitude and altitude, so that the drone can accurately locate the specific location of the bird's nest in the high-voltage line corridor and high-voltage tower area; the auxiliary positioning sensor module integrates visual sensors, lidar and IMU sensors and integrates them with GPS data.

[0106] The drone of this embodiment is also equipped with an alarm module, which monitors the flight status and surrounding environment of the drone in real time. After confirming that a bird's nest is detected, the alarm module triggers an alarm signal based on the analysis results of the judgment module.

[0107] The cleaning device of this embodiment includes a high-pressure gas purge module and a robot cleaning module;

[0108] The high-pressure gas purge module uses a high-pressure gas jet device composed of a carbon fiber gas tank, a vector-adjustable nozzle, and a pulsed solenoid valve equipped on the drone to carry out targeted purge and remove bird nests made of loose materials such as branches and dead grass on high-voltage power lines.

[0109] Robotic arm removal module: A robot equipped with a six-degree-of-freedom robotic arm, an adaptive pressure feedback gripper, and an insulated wrist joint is used to remove bird nests on high-voltage towers that are entangled with metal wires, adhered to conductors, or have dynamic interference.

[0110] The safety warning module of this embodiment integrates millimeter-wave radar, dual-spectrum camera, and electric field sensor to monitor the flight status and surrounding environmental parameters of the drone in real time. When medium- and long-distance obstacles are detected, equipment overheating is found, or a strong electromagnetic field is detected, a multi-level warning mechanism is triggered, and a warning signal is sent to the ground station through an encrypted channel, and an emergency avoidance or automatic return program is simultaneously initiated to ensure the safety of the drone and the operator.

[0111] The ground control terminal of this embodiment is also provided with an archiving and reporting module, which records and organizes the relevant data of the drone clearing the bird's nest and generates a detailed report. The specific method includes: the data storage module stores various data collected by the drone during the bird's nest clearing process, including bird's nest detection data, drone flight data, and clearing operation data; the report generation module automatically generates a detailed bird's nest clearing report based on the organized data.

[0112] The drone is also equipped with an energy management module, which monitors the drone's battery level and energy consumption, and distributes and manages the drone's energy.

Claims

1. A bird nest removal system based on drone detection and deep learning, characterized by: The drone is equipped with an image acquisition module, a communication module, a positioning module, and a cleaning device. The specific cleaning method is as follows: Step 1: After the drone flies to the target area, it uses the image acquisition module to collect the corresponding image through the camera and transmits it to the ground control terminal using the communication module; Step 2: The ground control terminal uses the judgment module to analyze the received image and generate a detection report. The detection report includes the specific location, bounding box, category information and confidence level of the bird's nest in the target area; The detection module uses the improved YOLO11 model. After the image is input into the detection module, the location and confidence of the bird's nest in the image are output after analysis. The specific method is as follows: Step 2.1: The input image is first scaled to a preset fixed size. The downsampling Conv module and C3K2 module of the backbone network in the detection module perform feature extraction on the scaled image to obtain feature maps of three different scales. Step 2.2: The obtained feature maps of the three sizes are fed into the Neck layer of the detection module for feature fusion. Before the Neck layer feature fusion, a convolutional block attention module (CBAM) is added to dynamically adjust the feature weights through the channel-spatial dual attention mechanism. Step 2.3: The three feature maps output by the Neck layer are fed into the detection head of the Head layer, where candidate detection boxes and category scores are generated through a convolutional network. Subsequently, an adaptive non-maximum suppression (NMS) mechanism is introduced to dynamically adjust the suppression threshold based on the density of bird nests in the image. Step 2.4: Dynamically balance missed detections and false detections, and finally output optimized detection results; Step 3: Input the detection report into the judgment module. If the judgment module recognizes that there is a bird's nest in the detection information, the alarm module is triggered and the detection information is stored in the data storage module; Step 4: Based on the inspection report obtained in step 2, the ground operator of the UAV activates and controls the removal device on the UAV to remove the bird's nest; the removal device includes a high-pressure gas purge module and a robotic removal module; during the removal process, the positioning module obtains real-time position information to dynamically adjust the removal device; Step 5: After the removal task is completed, the drone will collect images of the target area again and repeat steps 1 to 4 to test the removal effect. If the judgment module still detects a bird's nest, the ground operator will decide whether to re-execute steps 4 and 5 until it is confirmed that the bird's nest has been completely removed. Step 6. After the cleaning is completed, a detailed report will be generated based on the status records before and after the cleaning. The detailed report includes the three-dimensional spatial coordinates of the bird's nest in the target area, the bird's nest feature matrix, the cleaning operation log, and the comparison image, and is archived in the archive system.

2. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The image acquisition module uses a high-resolution camera, and the drone is equipped with the camera to capture images containing high-voltage lines and high-voltage towers in the target area.

3. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The working process of the convolutional block attention module CBAM in the detection module is as follows: First, for the input feature map F∈R C×H×W , processed using the channel attention mechanism; C is the number of channels, H×W is the spatial dimension; Global pooling operation: Average pooling: Max Pooling: MLP with shared weights: Compression ratio r (usually 16): MLP(x)=W1·δ(W0·x),W0∈R c / r×C ,W1∈R C×C / r ; Activation function δ: ReLU Sigmoid activation: M c (F) = σ(MLP(F avg )+MLP(F max )); Output feature map: is channel-by-channel multiplication; Then, the input feature map F′∈R C×H×W Processing using spatial attention mechanism: Channel dimension pooling operation: Average Pooling: Max Pooling: Convolutional fusion: Convolution kernel size 7×7: Sigmoid activation: Finally, the output feature map CBAM(F): is the multiplication of spatial positions; The detection module introduces an adaptive non-maximum suppression (NMS) mechanism to dynamically adjust the suppression threshold according to the density of bird nests in the image. The adjustment method is: For dense areas, the NMS threshold is lowered to 0.4, the inter-frame overlap tolerance is relaxed, and high recall is prioritized to avoid missing dense targets. For sparse areas, the NMS threshold is increased to 0.6 to strengthen the overlap screening between frames and suppress false detection interference.

4. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The specific working content of the judgment module is: Based on the location and size of the bird's nest output by the detection module, a multispectral imaging system is used to analyze the material's spectral characteristics. A lidar scan is used to construct a three-dimensional spatial topology model. An electric field sensor array measures local electromagnetic environment parameters. A millimeter-wave radar monitors ambient airflow disturbances. A stereo vision system is used to analyze the contact surface geometry, thereby constructing a bird's nest feature matrix [N, D, E, S, C], where N is the material conductivity coefficient, D is the distance to the charged object, E is the electric field intensity, S is the wind speed, and C is the conductor contact area ratio. Next, analyze the location and characteristics of the bird's nest, and determine whether it needs to be removed and the best removal strategy: If N<0.5, D≥1.2, and E<8 are met at the same time, the "Simple Scenario" prompt is triggered and a high-pressure gas purge path is generated; If N≥0.5 or C>30% or S>10, the "complex scene" prompt is activated and the clearing device is started.

5. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The positioning module includes a GPS positioning module and an auxiliary positioning sensor. The GPS positioning module receives satellite signals through the anti-electromagnetic interference enhanced GPS receiver on the drone, and has a built-in multi-layer magnetic shield and adaptive bandpass filter to obtain the drone's precise location information in real time, including coordinate data such as longitude, latitude and altitude, so that the drone can accurately locate the specific location of the bird's nest in the high-voltage line corridor and high-voltage tower area; the auxiliary positioning sensor module integrates visual sensors, lidar and IMU sensors and fuses them with GPS data.

6. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The drone is also equipped with an alarm module, which monitors the drone's flight status and surrounding environment in real time. After confirming that a bird's nest has been detected, the alarm module triggers an alarm signal based on the analysis results of the judgment module.

7. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The cleaning device includes a high-pressure gas purge module and a manipulator cleaning module; The high-pressure gas purge module uses a high-pressure gas jet device composed of a carbon fiber gas tank, a vector-adjustable nozzle, and a pulsed solenoid valve equipped on the drone to carry out targeted purge and remove bird nests made of loose materials such as branches and dead grass on high-voltage power lines. Robotic arm removal module: A robot equipped with a six-degree-of-freedom robotic arm, an adaptive pressure feedback gripper, and an insulated wrist joint is used to remove bird nests on high-voltage towers that are entangled with metal wires, adhered to conductors, or have dynamic interference.

8. The bird nest removal system based on drone detection and deep learning according to claim 1 is characterized in that: The safety warning module integrates millimeter-wave radar, dual-spectrum camera, and electric field sensor to monitor the drone's flight status and surrounding environmental parameters in real time. When medium- and long-distance obstacles are detected, equipment overheating is found, or a strong electromagnetic field is detected, a multi-level warning mechanism is triggered, and a warning signal is sent to the ground station through an encrypted channel. The emergency avoidance or automatic return program is simultaneously initiated to ensure the safety of the drone and the operator.

9. The bird nest removal system based on drone detection and deep learning according to claim 1, characterized in that: The ground control terminal also has an archiving and reporting module, which records and organizes the relevant data of the drone's removal of bird nests and generates detailed reports. The specific methods include: The data storage module stores various data collected by the drone during the bird nest removal process, including bird nest detection data, drone flight data, and removal operation data; the report generation module automatically generates a detailed bird nest removal report based on the sorted data.

10. The bird nest removal system based on drone detection and deep learning according to claim 1, characterized in that: The drone is also provided with an energy management module, which monitors the battery level and energy consumption of the drone, and distributes and manages the energy of the drone.

Images