A method for identifying the blade tips of a wind turbine by an airborne edge computing platform of an unmanned aerial vehicle

Abstract

The present invention discloses a method for identifying the tip of a wind turbine blade by an airborne edge computing platform of an unmanned aerial vehicle, which relates to the technical field of maintenance of wind power equipment. The method includes the following steps: Step 1, install a flexible copper mesh composed of four brackets on the top of the unmanned aerial vehicle. Conductors are led out from the connection between the brackets and the unmanned aerial vehicle. During the flight of the unmanned aerial vehicle, it is suspended to the ground by a tethered connection and connected to one end of a direct resistance meter. The present invention realizes high-precision identification of the tip of the wind turbine blade by integrating high-precision sensors and a deep learning model. The multi-layer structure of the deep learning model ensures that the model can process complex and variable blade images, improving the accuracy of identification. At the same time, the video tracking algorithm uses gradient information and a pixel point weight matrix to achieve high-dynamic and low-latency tracking of the tip, ensuring real-time performance. It not only improves the identification accuracy but also shortens the response time, providing reliable technical support for the monitoring and maintenance of wind turbine blades.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power equipment maintenance, and specifically provides a method for identifying the tip of a wind turbine blade by an unmanned aerial vehicle (UAV)-borne edge computing platform. Background Art

[0002] In the field of wind energy generation, as a key component, the health status of wind turbine blades directly affects the operation efficiency and safety of wind turbines. However, traditional monitoring methods for wind turbine blades often have many deficiencies.

[0003] On the one hand, most traditional methods rely on manual inspections, which are inefficient and costly, and it is difficult to detect subtle damages or abnormalities of the blades in a timely manner. On the other hand, although existing automated monitoring technologies can reduce manual intervention to a certain extent, there are still obvious shortcomings in terms of recognition accuracy and real-time performance. Especially under complex and variable environmental conditions, such as insufficient light, high wind speed, or when the blades are rotating, traditional monitoring technologies often have difficulty accurately capturing the position information of the blade tips, resulting in deviations in the monitoring results. In addition, existing monitoring systems are also easily limited by computing resources and processing speed when dealing with a large amount of data, and it is difficult to meet the requirements of real-time monitoring and rapid response.

[0004] Therefore, the development of an innovative technology that can accurately and real-time identify the tip of a wind turbine blade has become an urgent need to improve the monitoring efficiency and accuracy of wind turbines. Summary of the Invention

[0005] The purpose of the present invention is to make up for the deficiencies of the existing technology, and provides a method for identifying the tip of a wind turbine blade by an unmanned aerial vehicle (UAV)-borne edge computing platform. It can achieve high-precision identification of the tip of a wind turbine blade by integrating high-precision sensors and deep learning models. The multi-layer structure of the deep learning model ensures that the model can process complex and variable blade images, improving the accuracy of identification. At the same time, the video tracking algorithm uses gradient information and pixel point weight matrix to achieve high-dynamic and low-latency tracking of the tip, ensuring real-time performance. It not only improves the recognition accuracy but also shortens the response time, providing reliable technical support for the monitoring and maintenance of wind turbine blades.

[0006] To solve the above technical problems, the present invention provides the following technical solution: A method for identifying the tip of a wind turbine blade by an unmanned aerial vehicle (UAV)-borne edge computing platform, the method comprising the following steps:

[0007] Step 1: Install a flexible copper mesh composed of four brackets on the top of the drone. Lead wires are drawn out from the connection between the brackets and the drone. During the flight of the drone, it is suspended to the ground by a tether and connected to one end of a direct resistance meter, while the other end of the direct resistance meter is connected to the tower base grounding wire. At the same time, a measurement system is carried on the drone, and corresponding automatic control systems and algorithms are developed. The measurement system includes a sensor module and a data processing module. The sensor module is responsible for collecting data, and the data processing module processes the collected data. The automatic control system is used to control the flight attitude and actions of the drone so that it can approach and identify the blade tip according to a predetermined strategy;

[0008] Step 2: Lock the blade at a Y-shaped angle. The drone flies below the blade tip to ensure that the image of the blade tip is stored in the field of view of the guiding camera. Train a deep learning model through the pre-collected data. The construction of this deep learning model is based on a neural network structure. Let the number of neurons in the input layer be , the value of which is determined by the dimension of the collected wind turbine blade image data set. Let the image data set be , where is the number of images. For each image , its pixel coordinates are , , is the number of pixels. , the hidden layer adopts a multi-layer structure. Let the number of hidden layers be . The number of neurons in the -th hidden layer is , and its determination formula is , where is an adjustment coefficient, and its value range is [0.5, 1.5]. is the number of neurons in the -th hidden layer. The number of neurons in the output layer is , corresponding to the recognition result of the blade tip. Define the loss function , where is the predicted value, is the true value, is the regularization parameter, and its value range is [0.001, 0.1], which is used to prevent overfitting. is the weight of the -th neuron in the -th layer. Adjust the model parameters through the backpropagation algorithm to minimize the loss function ;

[0009] Step 3: Based on the algorithm of video tracking, in this algorithm, the position coordinates of the blade tip in the image are defined as , the center coordinates of the image are . Let the gray value matrix of the image be , and its size is , and are the number of rows and columns of the image grayscale value matrix respectively. By calculating the gradient matrix of the image , whose elements are and , the preliminary position of the leaf tip is determined using the gradient information. Then, according to the grayscale value distribution characteristics of the pixel points around the leaf tip, a weight matrix is defined, whose elements are determined by the distance between the pixel points around the leaf tip and the leaf tip and the grayscale value difference. Let the distance be , and the grayscale value difference be , then , where is the weight adjustment coefficient, and its value range is [0.1, 1]. Calculate the weighted position coordinates to obtain the position of the leaf tip in the image on-site;

[0010] Step 4: Combine the leaf tip height with the high-precision RTK information of the UAV. Let the leaf tip height be , and the horizontal coordinate in the UAV RTK information be . Define a spatial transformation matrix , which is constructed based on the comprehensive consideration of the earth curvature, atmospheric refraction, and the attitude factors of the UAV itself. Let the earth curvature parameter be , the atmospheric refraction coefficient be , and the UAV attitude angle be , then The elements are derived from the formula through complex geometric and physical model derivations. Through the spatial transformation formula , where is the coordinate of the leaf tip in the initial reference frame. Calculate the three-dimensional spatial deviation of the leaf tip relative to the UAV, obtain the horizontal control amount of the UAV, and control the horizontal position of the UAV through the interface of the UAV SDK to make the leaf tip at the center of the image, and at the same time control the UAV to rise in height.

[0011] Furthermore, the installation angle and position of the flexible copper mesh installation structure are precisely designed. The flexible copper mesh composed of four brackets can adapt to different flight postures and environmental interferences while ensuring the connection stability with the UAV. From the perspective of structural mechanics, the length and diameter of the brackets satisfy specific proportional relationships. Let the proportionality coefficients be , then there is , , , This proportional relationship takes into account the influence of different wind speeds , wind directions , the flight speed of the drone and attitude angles on the stability of the copper mesh. At the same time, the material of the copper mesh, its conductivity and flexibility meet specific requirements. Let the thresholds be and respectively. Then there is and .

[0012] Furthermore, the automated control system adopts a hierarchical control architecture, including a hardware drive layer at the bottom, a control algorithm layer in the middle, and a task scheduling layer at the top. The hardware drive layer is responsible for communicating with and controlling the hardware devices of the drone, including motor drive and sensor data acquisition. The control algorithm layer adopts a hybrid control strategy, combining proportional-integral-derivative (PID) control and fuzzy logic control. For the horizontal position control of the drone, let the desired position be , the current position be , the error be , and the parameters of the PID control be the proportional coefficient , the integral coefficient , and the derivative coefficient respectively. Their determination methods are as follows: Let the dynamic characteristic parameters of the drone flight environment be , including the wind speed , the wind direction and the air density . , where is a coefficient determined through experiments, and its value range is [0.1, 10]. At the same time, fuzzy logic control is used to handle uncertainties and nonlinear problems. By defining fuzzy sets and fuzzy rules, the parameters of the PID control are dynamically adjusted. The upper task scheduling layer allocates system resources according to the priorities and time requirements of the tasks.

[0013] Furthermore, the data collected for training the deep learning model includes image data of different types of fan blades under different lighting conditions, different angles, and different operating states. For the lighting conditions, they are divided into three cases: strong light, weak light, and natural light. Relevant image data are collected respectively, and the images are normalized. Let the normalization function be , where is the original image data, and They are the minimum and maximum values of the data respectively. For the blade angle, it covers various angle ranges from horizontal to vertical. The actual angle of the blade is obtained through a precise angle measurement device and marked in the image data. For the operating state, it includes two cases: stationary and rotating. In the rotating state, blade image data at different rotational speeds are collected. At the same time, the collected data is augmented, and data enhancement techniques are adopted, including rotation, flipping, and scaling operations. The probability of each operation is determined through experiments, and the value range is [0.1, 1].

[0014] Furthermore, during the implementation of the video tracking algorithm, a multi-scale feature extraction method is adopted. Let the original scale of the image be , and through downsampling and upsampling operations, image features of different scales are obtained, which are respectively . For the image features of each scale, different feature extractors and different layer structures of the convolutional neural network (CNN) are used. At the same time, during the tracking process, an adaptive window adjustment strategy is adopted. Let the initial tracking window size be . During the tracking process, according to the movement speed and direction of the leaf tip in the image, the size and position of the tracking window are dynamically adjusted. Let the adjustment coefficient of the tracking window be , then the window adjustment formula is . When calculating the weights of the pixel points within the window, the color features and texture features of the pixel points are considered. Let the color value of the pixel point be , and the texture value be . The weight function is defined as , where is the weight adjustment coefficient, and

[0015] are the reference color value and texture value, which are determined through experiments. Furthermore, in step four, by obtaining real-time data related to the atmospheric refraction coefficient and the earth's curvature, the spatial transformation matrix is dynamically adjusted. The adjustment method is based on an adaptive algorithm. Let the adaptive coefficient be . When it is detected that the change in the atmospheric refraction coefficient or the earth's curvature parameter exceeds the threshold , the elements of are updated. The update formula is , where

[0016] Furthermore, when the UAV SDK interface controls the horizontal position and altitude of the UAV, the parameters of the interface are finely set and adjusted. Let the horizontal control parameter of the interface be , and the vertical control parameter be . For horizontal position control, when the deviation between the desired position and the current position is within a certain range, let the deviation range be , then the value of satisfies , where and are the lower and upper limit values determined by experiments. For altitude control, when the deviation between the desired altitude and the current altitude is within a certain range, let the deviation range be , then the value of satisfies , where and are the lower and upper limit values determined by experiments. At the same time, during the control process, factors such as the flight characteristics of the UAV and the sensor accuracy are considered, and the interface parameters are dynamically adjusted. The adjustment method is based on an adaptive algorithm. Let the adaptive coefficient be . When it is detected that the change in the flight characteristics of the UAV or the sensor accuracy exceeds the threshold , and are updated, and the update formula is , where and are the parameter change amounts calculated according to the new situation.

[0017] Furthermore, the measurement system includes a sensor module and a data processing module. The sensor module uses a high-precision camera and a resistance measurement sensor. The resolution of the camera is , where and are the number of pixels in the horizontal and vertical directions respectively. By increasing the resolution of the camera, a clearer tip image is obtained. The measurement accuracy of the resistance measurement sensor is . By increasing the measurement accuracy of the sensor, the data processing module uses a digital signal processing algorithm to filter, amplify, and digitize the data collected by the sensor. Let the filtering algorithm be , the amplification factor be , and the digitization processing function be , then the processed data is . At the same time, in order to improve the accuracy of the measurement system, the sensor module and the data processing module are jointly optimized, and the optimal parameter matching relationship between the sensor module and the data processing module is determined through experiments, including the relationship between the camera resolution and the filtering algorithm parameters, and the relationship between the accuracy of the resistance measurement sensor and the amplification factor.

[0018] Furthermore, the tip recognition and control process is collaborative. Throughout the process, the recognition and tracking of the blade tip and the control of the UAV are interrelated and collaborative. When the blade tip recognition algorithm detects a change in the position of the blade tip, it promptly transmits the information to the tracking algorithm and the UAV control algorithm. The tracking algorithm adjusts the tracking strategy based on the change in the position of the blade tip, and the UAV control algorithm adjusts the attitude and altitude of the UAV according to the tracking result and the spatial solution result. At the same time, the control state of the UAV is also fed back to the blade tip recognition and tracking algorithms. This collaboration is achieved through a communication protocol, which is set as , and it includes regulations on message format, transmission rate, and error handling. The message format stipulates the encoding methods of the blade tip position information and the UAV control instructions. The transmission rate is determined according to the UAV flight speed and mission requirements. The error handling stipulates the retransmission and error correction operations when information transmission errors occur.

[0019] Compared with the prior art, the blade tip recognition algorithm of the UAV-borne edge computing platform has the following beneficial effects:

[0020] First, through the combination of the deep learning model and the video tracking algorithm, the present invention realizes the high-precision and real-time recognition and tracking of the blade tips of wind turbine blades. The deep learning model is based on a neural network structure and is trained with a large amount of image data, enabling accurate recognition of blade tips under different lighting conditions, angles, and operating states. The video tracking algorithm uses multi-scale feature extraction and adaptive window adjustment strategies to real-time track the position of the blade tip in the image. Even when the blade tip moves at high speed or the lighting conditions change, it can maintain a small tracking error. This high-precision and real-time recognition and tracking ability provides reliable data support for subsequent UAV control, ensuring that the UAV can accurately and stably approach and contact the blade tip.

[0021] Second, the present invention has strong adaptability and can cope with different environmental conditions (such as wind speed, wind direction, lighting, and temperature) and blade types (wind turbine blades of different models and designs). By introducing an adaptive mechanism, the algorithm can automatically adjust parameters and strategies according to environmental changes to ensure stable performance and accuracy under different conditions. At the same time, the algorithm also takes into account the performance differences of UAVs. By calibrating and adjusting the UAV parameters, it ensures the effective operation of the algorithm on UAVs of different brands and models. This strong adaptability makes the algorithm have broad application prospects and practical value in practical applications.

[0022] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0024] Figure 1 It is a flowchart of a method for identifying the tip of a wind turbine blade by an unmanned aerial vehicle (UAV) - borne edge - computing platform. Specific embodiments

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0026] Embodiment 1: Identification of the tip of a specific - model wind turbine blade in a strong - light environment

[0027] In an open area with sufficient sunlight, the tip of a wind turbine blade of model XX is identified. The on - site wind speed is , the wind direction is , the flight speed of the UAV is , and the attitude angle is .

[0028] Algorithm parameter settings

[0029] Deep - learning model:

[0030] The collected image data set contains 1000 images , and the number of pixel points in each image is , so the number of neurons in the input layer ;

[0031] Number of hidden layers , adjustment coefficient , for the first hidden layer , the number of neurons , the second hidden layer , , the third hidden layer , , regularization parameter .

[0032] Video tracking algorithm:

[0033] Image grayscale value matrix with a size of , weight adjustment coefficient , and the initial size of the tracking window is .

[0034] Spatial resolution formula:

[0035] Earth curvature parameter , atmospheric refraction coefficient , the attitude angle of the UAV is , then the elements of the spatial transformation matrix

[0036] .

[0037] Process and monitoring: First, the UAV flies according to the predetermined attitude, the flexible copper mesh remains stable, data is collected through the sensor module, the camera resolution is , the accuracy of the resistance measurement sensor is , the deep learning model processes the collected images, continuously adjusts the weights through the backpropagation algorithm, and after 1000 iterations, the loss function decreases from the initial 100 to 1.5, achieving a preliminary identification of the leaf tip. The video tracking algorithm accurately tracks the leaf tip according to the image gradient and pixel point weights. During the movement of the leaf tip, the window size is dynamically adjusted according to the leaf tip speed and direction, and the tracking error always remains within 5 pixels. The spatial resolution formula combines the leaf tip height and the UAV position information to accurately calculate the three-dimensional spatial deviation of the leaf tip relative to the UAV, controls the horizontal position and height of the UAV, and controls the deviation of the leaf tip from the image center within pixels and the height deviation within . Finally, the UAV successfully approaches the leaf tip, the metal mesh contacts the leaf tip, and the direct resistance meter accurately measures the leaf tip - tower base resistance value as .

[0038] Example 2: Identification of the leaf tips of different models of wind turbine blades in low-light environments

[0039] On a dim evening, the leaf tips of two types of wind turbine blades, models YY and ZZ, are identified. The on-site wind speed is , and the wind direction is °, , and the attitude angle is .

[0040] Algorithm parameter settings

[0041] Deep learning model:

[0042] The collected image data set contains 1500 images ​​, the number of pixels in each image is , then the number of neurons in the input layer ;

[0043] Number of hidden layers , adjustment coefficient , for the first hidden layer ( ), the number of neurons , the second hidden layer , , the third hidden layer , , the fourth hidden layer , , regularization parameter .

[0044] Video tracking algorithm:

[0045] Image grayscale value matrix The size is , weight adjustment coefficient , the initial size of the tracking window is .

[0046] Earth curvature parameter , atmospheric refraction coefficient , the attitude angle of the UAV is , then the elements of the spatial transformation matrix .

[0047] Process and monitoring: The UAV flies in low-light environment, the flexible copper mesh has good stability, the sensor module collects data, the camera resolution is , the accuracy of the resistance measurement sensor is , after 1500 iterations of the deep learning model, the loss function decreases from the initial 120 to 2.5, realizing the preliminary identification of the leaf tip. The video tracking algorithm accurately tracks the leaf tip by adjusting the weight in low light, the tracking error is within 8 pixels, the spatial calculation formula accurately calculates the deviation, and the UAV is controlled to make the deviation between the leaf tip and the image center within pixels, and the height deviation is controlled within . For two different types of blades, the leaf tip - tower base resistance values can be accurately measured, which are .

[0048] ​It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. A method for identifying the tip of a wind turbine blade by an edge computing platform on a drone, characterized in that: The method comprises the following steps: Step 1: Install a flexible copper mesh consisting of four brackets on the top of the drone. Lead wires are drawn out from the connection between the brackets and the drone. During the flight of the drone, the wires are tethered to the ground and connected to one end of the direct resistance meter. The other end of the direct resistance meter is connected to the tower base ground wire. At the same time, a measurement system is installed on the drone, and a corresponding automatic control system and algorithm are developed. The measurement system includes a sensor module and a data processing module. The sensor module is responsible for collecting data, and the data processing module processes the collected data. The automatic control system is used to control the flight attitude and movement of the drone, so that it can approach and identify the blade tip according to a predetermined strategy. Step 2: Lock the blade at a Y-shaped angle, and fly the drone below the blade tip to ensure that the blade tip is visible in the camera’s field of view. The deep learning model is trained based on the neural network structure. The number of neurons in the input layer is set to , whose value is determined by the dimension of the collected wind turbine blade image data set. Suppose the image data set is ,in is the number of images, for each image , whose pixel coordinates are , , is the number of pixels, , the hidden layer adopts a multi-layer structure, and the number of hidden layers is , No. The number of neurons in the hidden layer is , the formula for determining it is ,in is the adjustment coefficient, the value range is [0.5, 1.5], For the The number of neurons in the hidden layer is , corresponding to the recognition result of the leaf tip, define the loss function ,in is the predicted value, is the true value, is a regularization parameter with a value range of [0.001, 0.1], which is used to prevent overfitting. For the Tier The weights of neurons are adjusted through the back propagation algorithm to make the loss function minimize; Step 3: Based on the video tracking algorithm, the position coordinates of the leaf tip in the image are defined as , the image center coordinates are , let the gray value matrix of the image be , whose size is , and are the number of rows and columns of the image gray value matrix, respectively. By calculating the gradient matrix of the image , whose elements are and , use the gradient information to determine the initial position of the leaf tip, and then define a weight matrix based on the gray value distribution characteristics of the pixels around the leaf tip , whose elements It is determined by the distance between the pixel points around the leaf tip and the leaf tip and the gray value difference. Let the distance be , the gray value difference is ,but ,in is the weight adjustment coefficient, the value range is [0.1, 1], and the weighted position coordinates are calculated , the position of the leaf tip in the image is obtained in real time; Step 4: Combine the tip height with the high-precision RTK information of the UAV and set the tip height to , the horizontal coordinates in the drone RTK information are , define a space transformation matrix , which is constructed based on the comprehensive consideration of the earth curvature, atmospheric refraction and the UAV's own attitude factors. The earth curvature parameter is , the atmospheric refraction coefficient is , the drone attitude angle is ,but Elements Formulas derived from complex geometric and physical models , through the space transformation formula ,in The coordinates of the blade tip in the initial reference system are calculated, and the spatial three-dimensional deviation of the blade tip relative to the UAV is calculated to obtain the horizontal control amount of the UAV. The horizontal position of the UAV is controlled through the interface of the UAV SDK so that the blade tip is in the center of the image, and the UAV is controlled to rise in height.

2. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: The installation angle of the flexible copper mesh installation structure and location After precise design, the flexible copper mesh composed of four brackets can adapt to different flight postures and environmental interference while ensuring the stability of the connection with the drone. From the perspective of structural mechanics, the length of the bracket and diameter To satisfy a specific proportional relationship, the proportional coefficients are , then , , , This proportional relationship takes into account different wind speeds ,wind direction And the flight speed of the drone and posture angle The influence of factors on the stability of the copper mesh, at the same time, the conductivity of the copper mesh material and flexibility To meet specific requirements, the thresholds are set as and , then and .

3. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: The automation control system adopts a hierarchical control architecture, including the bottom hardware driver layer, the middle control algorithm layer and the upper task scheduling layer. The hardware driver layer is responsible for communicating and controlling the hardware devices of the drone, including motor drive and sensor data acquisition. The control algorithm layer adopts a hybrid control strategy, combining proportional integral differential PID control and fuzzy logic control. For the horizontal position control of the drone, the desired position is , the current location is , the error is , the parameters of PID control are proportional coefficient , integral coefficient and the differential coefficient , and the determination method is as follows: Assume that the dynamic characteristic parameters of the UAV flight environment are , including wind speed ,wind direction and air density , ,in is a coefficient determined by experiments, with a value range of [0.1, 10]. At the same time, fuzzy logic control is used to deal with uncertainty and nonlinear problems. By defining fuzzy sets and fuzzy rules, the parameters of PID control are dynamically adjusted. The upper task scheduling layer allocates system resources according to the priority and time requirements of the task.

4. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: The data collected for the deep learning model training data include image data of different models of fan blades under different lighting conditions, different angles and different operating states. The lighting conditions are divided into three conditions: strong light, weak light and natural light. The relevant image data are collected respectively and the images are normalized. The normalization function is set as ,in is the original image data, and They are the minimum and maximum values ​​of the data respectively. For the blade angle, it covers all angles from horizontal to vertical. The actual angle of the blade is obtained through precise angle measurement equipment and marked in the image data. For the running state, it includes two situations: stationary and rotating. In the rotating state, different speeds are collected At the same time, the collected data was expanded and processed by using data enhancement technology, including rotation, flipping and scaling operations. The probability of each operation Determined through experiments, the value range is [0.1, 1].

5. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: In the implementation process of the video tracking algorithm, a multi-scale feature extraction method is adopted. Assume that the original scale of the image is , through downsampling and upsampling operations, we get image features of different scales, which are For each scale of image features, different feature extractors and different layer structures of convolutional neural network (CNN) are used. At the same time, in the tracking process, an adaptive window adjustment strategy is adopted, and the initial tracking window size is set to During the tracking process, according to the movement speed of the leaf tip in the image and direction , dynamically adjust the size and position of the tracking window, and set the adjustment coefficient of the tracking window to , then the window adjustment formula is When calculating the weight of the pixel in the window, the color and texture features of the pixel are taken into account. Let the color value of the pixel be , the texture value is , define the weight function as ,in is the weight adjustment coefficient, are reference color and texture values, determined experimentally.

6. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: In step 4, the spatial transformation matrix is ​​obtained by real-time acquisition of atmospheric refraction coefficient and earth curvature related data. Dynamic adjustment is performed based on an adaptive algorithm. The adaptive coefficient is set as When the atmospheric refractive index or earth curvature parameter changes exceeding the threshold When, yes Elements Update, the update formula is ,in is the change in element calculated based on the new data.

7. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: When the drone SDK interface controls the horizontal position and height of the drone, the interface parameters are finely set and adjusted. Suppose the interface's horizontal control parameters are , the vertical control parameter is For horizontal position control, when the deviation between the expected position and the current position is within a certain range, the deviation range is set to ,but The value of satisfies ,in and are the lower and upper limits determined by the experiment. For altitude control, when the deviation between the desired altitude and the current altitude is within a certain range, the deviation range is ,but The value of satisfies ,in and are the lower and upper limits determined by the experiment. At the same time, in the control process, the flight characteristics of the UAV and the accuracy of the sensor are taken into account, and the interface parameters are dynamically adjusted. The adjustment method is based on an adaptive algorithm. The adaptive coefficient is set to When the flight characteristics of the drone or the sensor accuracy changes beyond the threshold, When, yes and Update, the update formula is ,in and is the parameter change calculated according to the new situation.

8. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: The measurement system includes a sensor module and a data processing module. The sensor module uses a high-precision camera and a resistance measurement sensor. The resolution of the camera is ,in and are the number of pixels in the horizontal and vertical directions respectively. By improving the resolution of the camera, a clearer image of the leaf tip can be obtained. The measurement accuracy of the resistance measurement sensor is By improving the measurement accuracy of the sensor, the data processing module uses a digital signal processing algorithm to filter, amplify and digitize the data collected by the sensor. Suppose the filtering algorithm is , the magnification is , the digital processing function is , then the processed data ,At the same time, in order to improve the accuracy of the ,measurement system, the sensor module and the data processing module were ,coordinately optimized, and the optimal parameter matching relationship between ,the sensor module and the data processing module was determined through ,experiments, including the relationship between the camera resolution and ,filtering algorithm parameters, and the 9. The method for identifying the tip of a wind turbine blade by an edge computing platform on a drone according to claim 1, characterized in that: The blade tip recognition and control process is synergistic. In the whole process, the blade tip recognition, tracking and UAV control are interrelated and synergistic. When the blade tip recognition algorithm detects the position change of the blade tip, it promptly transmits the information to the tracking algorithm and the UAV control algorithm. The tracking algorithm adjusts the tracking strategy according to the position change of the blade tip. The UAV control algorithm adjusts the speed and height of the UAV according to the tracking results and the spatial solution results. At the same time, the control state of the UAV will also be fed back to the blade tip recognition and tracking algorithm. This synergy is achieved through a communication protocol. Suppose the communication protocol is It includes provisions on message format, transmission rate and error handling. The message format specifies the encoding method of blade tip position information and UAV control instructions. The transmission rate is determined according to the UAV flight speed and mission requirements. Error handling specifies retransmission and error correction operations when information transmission errors occur.

Images