Unmanned aerial vehicle line simulation method and system based on image recognition algorithm

Abstract

The invention discloses an unmanned aerial vehicle line simulation method and system based on an image recognition algorithm, and the method comprises the steps: constructing a lightweight power transmission line conductor detection model, and collecting a real-time image through a camera carried by an unmanned aerial vehicle; the lightweight transmission line conductor detection model identifies the conductor in the real-time image, and when the identification result is that the conductor exists in the real-time image, the focal length of the lens of the camera is adjusted; performing image acquisition on the lead by using the camera with the adjusted focal length to obtain a lead image, and processing the lead image based on an improved Kalman filtering algorithm to track the lead; the distance between the unmanned aerial vehicle and the wire is measured through a millimeter wave radar carried on the unmanned aerial vehicle, and the flight condition of the unmanned aerial vehicle is controlled by adopting an active control strategy so that the distance between the unmanned aerial vehicle and the wire can be kept as a fixed preset distance. According to the invention, the cost of the line-imitating inspection equipment of the unmanned aerial vehicle is reduced, and line-imitating inspection of the small multi-rotor unmanned aerial vehicle can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles (UAVs), and in particular relates to an UAV line imitation method and system based on an image recognition algorithm. Background Art

[0002] Transmission lines play a crucial role in power transmission within the power system. Manually operated drones are difficult to inspect for defects in conductors and ground wires. Drone-guided inspections, where drones autonomously follow the lines while maintaining a fixed distance, can reduce the workload of manual drone inspections. Drone-guided inspections require precise identification of conductors and ground wires, enabling the captured images to clearly reveal defects.

[0003] Current drone-based wire tracking inspection methods have the following shortcomings: image recognition algorithms struggle to accurately detect small targets like wires, as they easily blend in with the sky, making it difficult to accurately track the target. Commonly used LiDAR ranging solutions for wire tracking are costly and cannot be mounted on small multi-rotor drones. While market-leading technologies exist that use drones for image recognition to identify and track wires, current drone control algorithms for wire tracking inspections still lack the ability to accurately control flight attitude based on target detection to ensure safe flight. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention proposes a UAV line imitation method based on image recognition algorithm, including: Build a lightweight transmission line conductor detection model and collect real-time images using a camera mounted on a drone; The lightweight transmission line conductor detection model identifies conductors in real-time images and adjusts the camera lens focal length when the recognition result indicates that there are conductors in the real-time image. The wire image is acquired by using a camera with adjusted focal length, and the wire image is processed based on an improved Kalman filter algorithm to track the wire. The distance between the drone and the wire is measured by the millimeter-wave radar carried by the drone, and the flight of the drone is controlled by adopting an active control strategy to keep the distance between the drone and the wire at a fixed preset distance.

[0005] Furthermore, the process of building a lightweight transmission line conductor detection model is as follows: A transmission line conductor detection model is constructed. The backbone network in the transmission line conductor detection model is pruned by weight partitioning to obtain the pruned transmission line conductor detection model. The activation values ​​in the pruned transmission line conductor detection model are quantized using a two-step optimization algorithm and a clustering sparse algorithm. The weights in the pruned transmission line conductor detection model are sparsely quantized based on the quantized results of the activation values ​​to obtain a lightweight transmission line conductor detection model. The lightweight transmission line conductor detection model is deployed in the edge computing terminal of the drone.

[0006] Furthermore, the process of constructing a transmission line conductor detection model is as follows: Acquire an image dataset of the conductor, preprocess sample images in the image dataset, and classify and label the preprocessed sample images; The migrated and improved YOLOX model is trained through the classified and labeled image dataset to build a transmission line conductor detection model.

[0007] Furthermore, the transferred and improved YOLOX model was trained using the classified and annotated image dataset to obtain a transmission line conductor detection model. During the training process, the Harris Eagle optimization algorithm was used to optimize the model parameters of the transferred and improved YOLOX model. The optimization process is as follows: Initialize the model parameters of the migrated and improved YOLOX model as the initial vector within the preset global range; The probability value corresponding to the target position of the initial vector is calculated through the fitness function. If the probability value is greater than or equal to the preset value, the target position is updated arbitrarily within the entire domain, and the probability value corresponding to the updated target position is calculated through the fitness function; if the probability value meets the condition of being less than the preset value, the preset behavior and random walk measurement are used to determine the optimal position of the target position according to the size of the probability value; the fitness function adopts a loss function based on the global vector.

[0008] Furthermore, the preprocessing includes cropping, rotation, contrast stretching, image enhancement and filtering; The contrast stretching process is expressed as: ; in, and are the grayscale values ​​of the sample image before and after stretching; and They are the grayscale values ​​at the 2% and 97% percentiles in the grayscale distribution of the sample image before stretching; The image enhancement process is expressed as: ; in, is a constant and is set according to the brightness of the sample image; is the pixel point of the sample image before image enhancement, and the pixel value is The value indicates that Adjust the parameters for the pixels. is the pixel point of the sample image after image enhancement, represents the pixel adjustment function; The filtering process is expressed as: ; in, is the pixel value of the pixel point in the i-th sample image after filtering, is the pixel value of the input i-th sample image, is the pixel value at any position in the sample image before filtering, Represents the index of any position in the sample image; is the pixel set of the sample image before filtering, is the standard deviation of the Gaussian function, is the weight vector of the i-th sample image, represents the exponential function, Indicates taking the absolute value.

[0009] Furthermore, the specific construction method of the migrated and improved YOLOX model is as follows: Get the original YOLOX model; Build some convolution modules on the backbone network of the original YOLOX model to replace the original 4-layer ordinary convolution modules; Among them, the structure of some convolution modules is 1 The partial convolutional layer and 2 The ordinary convolutional layers are connected in series; In the original YOLOX backbone network, the Nix module is introduced after the first two C2f modules. The Nix module includes a self-attention layer, a convolutional combination layer, and a feature aggregation layer. The convolution feature of the input feature is obtained through the convolution combination layer in the Nix module, the self-attention feature of the input feature is obtained through the self-attention layer in the Nix module, and the convolution feature and the self-attention feature are aggregated through the feature aggregation layer in the Nix module to obtain the self-attention feature aggregation feature. The process of obtaining the self-attention feature aggregation feature is expressed as: ; in, is the self-attention feature aggregation feature, and are all preset weight matrices. is the convolution feature, is the self-attention feature.

[0010] Furthermore, the loss function based on the global vector Expressed as: ; ; ; ; in, is the attention coefficient, is the penalty amount, and are all hyperparameters, The outliers of the anchor boxes of the YOLOX model improved for migration, is the predicted box of the wire in the sample image, are the horizontal and vertical coordinates of the prediction box of the wire, is the ground-truth box of the wire in the sample image, are the horizontal and vertical coordinates of the ground truth box of the wire in the sample image; and Respectively represent the width and height of the minimum outer frame of the prediction box of the wire in the sample image, is the compensation penalty function, Penalty function for the improved YOLOX model for migration, Represents the exponential function.

[0011] Furthermore, the process of pruning the backbone network in the transmission line conductor detection model by weight partitioning is as follows: ; in, is the model output function, are the model learnable parameters, is the pixel value of the input sample image, represents the weight score; According to the weight score, the connection weights below the threshold in the backbone network are deleted.

[0012] Furthermore, the process of adjusting the focal length of the camera lens is as follows: Constructing the adjustment vector , expressed as: ; in, They are focus adjustment vector, angle adjustment vector and position adjustment vector respectively; Adjusting the lens based on the adjustment vector, photographing the wire with the adjusted lens to obtain a target image, evaluating the target image to obtain an image evaluation vector, and determining whether the image evaluation vector reaches a set score; When the image evaluation vector does not reach the set score, the adjustment vector is adjusted, the focal length of the lens is adjusted using the adjusted adjustment vector, and the wire is photographed again using the adjusted lens to obtain a target image, until the image evaluation vector of the target image reaches the set score, and the target image whose image evaluation vector reaches the set score is used as the wire image; The improved Kalman filter method for tracking wires is expressed as: ; in, represents the state of wire m at the tth moment, Indicates that the conductor m is The motion function at each moment, Indicates the The noise sequence at each moment, represents the position of wire m at the tth moment, Indicates that the conductor m is The position change function at each moment is Indicates the The measurement noise at each moment; The active control strategy is expressed as: ; ; in, For the drone at the next discrete moment status, For the drone at the current discrete moment status, For the drone at the current discrete moment The control input, For the drone at the current discrete moment The output status, is the state transition matrix, is the control input matrix, is the output matrix, is the direct transmission matrix.

[0013] A UAV line-mimicking system based on an image recognition algorithm, used in the above-mentioned UAV line-mimicking method based on an image recognition algorithm, comprising: The data acquisition and preprocessing module is used to collect image datasets of overhead ground wires and conductors, preprocess and classify the image datasets, and construct sample sets, verification sets, and test sets; The model training and optimization module is used to carry the YOLOX model and perform training based on the YOLOX model, optimize the YOLOX model parameters, and obtain the transmission line conductor detection model; Model lightweighting and deployment module, used to lightweight the transmission line conductor detection model and deploy it on the UAV computing terminal; The drone control module is used to adjust the lens focal length to the appropriate position and angle to take pictures of the wires, and control the drone to maintain a fixed distance from the wires and fly autonomously.

[0014] The beneficial effects of the present invention are: 1. This paper introduces filtering and the Nix module based on the attention mechanism into the image processing process, optimizes model parameters through the Harris Eagle algorithm, and further lightweights the model through weight partitioning. Ultimately, the transmission line conductor detection model's target recognition capability for conductors is improved with less computing power, avoiding confusion between conductors and the sky in the image, and enabling drones to more accurately follow the conductors.

[0015] 2. The present invention improves the Kalman filter algorithm and combines it with an active control strategy to improve the image quality of the drone's line-simulation inspection. Compared with the existing line-simulation technology, it is better in focus quality, picture size and angle, and the inspection effect is better. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a flowchart of the steps of a UAV line imitation method based on an image recognition algorithm of the present invention. DETAILED DESCRIPTION

[0017] Reference Figure 1 , a UAV line imitation method based on image recognition algorithm, including: Build a lightweight transmission line conductor detection model and collect real-time images using a camera mounted on a drone; The lightweight transmission line conductor detection model identifies conductors in real-time images and adjusts the camera lens focal length when the recognition result indicates that there are conductors in the real-time image. The wire image is acquired by using a camera with adjusted focal length, and the wire image is processed based on an improved Kalman filter algorithm to track the wire. The distance between the drone and the wire is measured by the millimeter-wave radar carried by the drone, and the flight of the drone is controlled by adopting an active control strategy to keep the distance between the drone and the wire at a fixed preset distance.

[0018] Furthermore, the process of building a lightweight transmission line conductor detection model is as follows: A transmission line conductor detection model is constructed. The backbone network in the transmission line conductor detection model is pruned by weight partitioning to obtain the pruned transmission line conductor detection model. The activation values ​​in the pruned transmission line conductor detection model are quantized using a two-step optimization algorithm and a clustering sparse algorithm. The weights in the pruned transmission line conductor detection model are sparsely quantized based on the quantized results of the activation values ​​to obtain a lightweight transmission line conductor detection model. The lightweight transmission line conductor detection model is deployed in the edge computing terminal of the drone.

[0019] Furthermore, the process of constructing a transmission line conductor detection model is as follows: Acquire an image dataset of the conductor, preprocess sample images in the image dataset, and classify and label the preprocessed sample images; The migrated and improved YOLOX model is trained through the classified and labeled image dataset to build a transmission line conductor detection model.

[0020] Furthermore, the transferred and improved YOLOX model was trained using the classified and annotated image dataset to obtain a transmission line conductor detection model. During the training process, the Harris Eagle optimization algorithm was used to optimize the model parameters of the transferred and improved YOLOX model. The optimization process is as follows: Initialize the model parameters of the migrated and improved YOLOX model as the initial vector within the preset global range; The probability value corresponding to the target position of the initial vector is calculated through the fitness function. If the probability value is greater than or equal to the preset value, the target position is updated arbitrarily within the entire domain, and the probability value corresponding to the updated target position is calculated through the fitness function; if the probability value meets the condition of being less than the preset value, the preset behavior and random walk measurement are used to determine the optimal position of the target position according to the size of the probability value; the fitness function adopts a loss function based on the global vector.

[0021] Furthermore, the preprocessing includes cropping, rotation, contrast stretching, image enhancement and filtering; The contrast stretching process is expressed as: ; in, and are the grayscale values ​​of the sample image before and after stretching; and They are the grayscale values ​​at the 2% and 97% percentiles in the grayscale distribution of the sample image before stretching; The image enhancement process is expressed as: ; in, is a constant and is set according to the brightness of the sample image; is the pixel point of the sample image before image enhancement, and the pixel value is The value indicates that Adjust the parameters for the pixels. is the pixel point of the sample image after image enhancement, represents the pixel adjustment function; The filtering process is expressed as: ; in, is the pixel value of the pixel point in the i-th sample image after filtering, is the pixel value of the input i-th sample image, is the pixel value at any position in the sample image before filtering, Represents the index of any position in the sample image; is the pixel set of the sample image before filtering, is the standard deviation of the Gaussian function, is the weight vector of the i-th sample image, represents the exponential function, Indicates taking the absolute value.

[0022] Furthermore, the specific construction method of the migrated and improved YOLOX model is as follows: Get the original YOLOX model; Build some convolution modules on the backbone network of the original YOLOX model to replace the original 4-layer ordinary convolution modules; Among them, the structure of some convolution modules is 1 The partial convolutional layer and 2 The ordinary convolutional layers are connected in series; In the original YOLOX backbone network, the Nix module is introduced after the first two C2f modules. The Nix module includes a self-attention layer, a convolutional combination layer, and a feature aggregation layer. The convolution feature of the input feature is obtained through the convolution combination layer in the Nix module, the self-attention feature of the input feature is obtained through the self-attention layer in the Nix module, and the convolution feature and the self-attention feature are aggregated through the feature aggregation layer in the Nix module to obtain the self-attention feature aggregation feature. The process of obtaining the self-attention feature aggregation feature is expressed as: ; in, is the self-attention feature aggregation feature, and are all preset weight matrices. is the convolution feature, is the self-attention feature.

[0023] Furthermore, the loss function based on the global vector Expressed as: ; ; ; ; in, is the attention coefficient, is the penalty amount, and are all hyperparameters, The outliers of the anchor boxes of the YOLOX model improved for migration, is the predicted box of the wire in the sample image, are the horizontal and vertical coordinates of the prediction box of the wire, is the ground-truth box of the wire in the sample image, are the horizontal and vertical coordinates of the ground truth box of the wire in the sample image; and Respectively represent the width and height of the minimum outer frame of the prediction box of the wire in the sample image, is the compensation penalty function, Penalty function for the improved YOLOX model for migration, Represents the exponential function.

[0024] Furthermore, the process of pruning the backbone network in the transmission line conductor detection model by weight partitioning is as follows: ; in, is the model output function, are the model learnable parameters, is the pixel value of the input sample image, represents the weight score; According to the weight score, the connection weights below the threshold in the backbone network are deleted.

[0025] Furthermore, the process of adjusting the focal length of the camera lens is as follows: Constructing the adjustment vector , expressed as: ; in, They are focus adjustment vector, angle adjustment vector and position adjustment vector respectively; Adjusting the lens based on the adjustment vector, photographing the wire with the adjusted lens to obtain a target image, evaluating the target image to obtain an image evaluation vector, and determining whether the image evaluation vector reaches a set score; When the image evaluation vector does not reach the set score, the adjustment vector is adjusted, the focal length of the lens is adjusted using the adjusted adjustment vector, and the wire is photographed again using the adjusted lens to obtain a target image, until the image evaluation vector of the target image reaches the set score, and the target image whose image evaluation vector reaches the set score is used as the wire image; The improved Kalman filter method for tracking wires is expressed as: ; in, represents the state of wire m at the tth moment, Indicates that the conductor m is The motion function at each moment, Indicates the The noise sequence at each moment, represents the position of wire m at the tth moment, Indicates that the conductor m is The position change function at each moment is Indicates the The measurement noise at each moment; The active control strategy is expressed as: ; ; in, For the drone at the next discrete moment status, For the drone at the current discrete moment status, For the drone at the current discrete moment The control input, For the drone at the current discrete moment The output status, is the state transition matrix, is the control input matrix, is the output matrix, is the direct transmission matrix.

[0026] A UAV line-mimicking system based on an image recognition algorithm, used in the above-mentioned UAV line-mimicking method based on an image recognition algorithm, comprising: The data acquisition and preprocessing module is used to collect image datasets of overhead ground wires and conductors, preprocess and classify the image datasets, and construct sample sets, verification sets, and test sets; The model training and optimization module is used to carry the YOLOX model and perform training based on the YOLOX model, optimize the YOLOX model parameters, and obtain the transmission line conductor detection model; Model lightweighting and deployment module, used to lightweight the transmission line conductor detection model and deploy it on the UAV computing terminal; The drone control module is used to adjust the lens focal length to the appropriate position and angle to take pictures of the wires, and control the drone to maintain a fixed distance from the wires and fly autonomously.

[0027] The above-described specific embodiments further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0028] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A drone line-simulation method based on image recognition algorithm, characterized in that: include: Build a lightweight transmission line conductor detection model and collect real-time images using a camera mounted on a drone; The lightweight transmission line conductor detection model identifies conductors in real-time images and adjusts the camera lens focal length when the recognition result indicates that there are conductors in the real-time image. The wire image is acquired by using a camera with adjusted focal length, and the wire image is processed based on an improved Kalman filter algorithm to track the wire. The distance between the drone and the wire is measured by the millimeter-wave radar carried by the drone, and the flight of the drone is controlled by adopting an active control strategy to keep the distance between the drone and the wire at a fixed preset distance.

2. The UAV line imitation method based on image recognition algorithm according to claim 1 is characterized in that: The specific process of building a lightweight transmission line conductor detection model is as follows: A transmission line conductor detection model is constructed. The backbone network in the transmission line conductor detection model is pruned by weight partitioning to obtain the pruned transmission line conductor detection model. The activation values ​​in the pruned transmission line conductor detection model are quantized using a two-step optimization algorithm and a clustering sparse algorithm. The weights in the pruned transmission line conductor detection model are sparsely quantized based on the quantized results of the activation values ​​to obtain a lightweight transmission line conductor detection model. The lightweight transmission line conductor detection model is deployed in the edge computing terminal of the drone.

3. The UAV line imitation method based on image recognition algorithm according to claim 2 is characterized in that: The specific process of building a transmission line conductor detection model is as follows: Acquire an image dataset of the conductor, preprocess sample images in the image dataset, and classify and label the preprocessed sample images; The migrated and improved YOLOX model is trained through the classified and labeled image dataset to build a transmission line conductor detection model.

4. The UAV line imitation method based on image recognition algorithm according to claim 3 is characterized in that: The transferred and improved YOLOX model is trained using the classified and labeled image dataset to obtain a transmission line conductor detection model. During the training process, the Harris Eagle optimization algorithm is used to optimize the model parameters of the transferred and improved YOLOX model. The optimization process is as follows: Initialize the model parameters of the migrated and improved YOLOX model as the initial vector within the preset global range; The probability value corresponding to the target position of the initial vector is calculated through the fitness function. If the probability value is greater than or equal to the preset value, the target position is updated arbitrarily within the entire domain, and the probability value corresponding to the updated target position is calculated through the fitness function; if the probability value meets the condition of being less than the preset value, the preset behavior and random walk measurement are used to determine the optimal position of the target position according to the size of the probability value; the fitness function adopts a loss function based on the global vector.

5. The method for imitating a line of an unmanned aerial vehicle based on an image recognition algorithm according to claim 3, characterized in that: The preprocessing includes cropping, rotation, contrast stretching, image enhancement and filtering; The contrast stretching process is expressed as: ； in, and are the grayscale values ​​of the sample image before and after stretching; and They are the grayscale values ​​at the 2% and 97% percentiles in the grayscale distribution of the sample image before stretching; The image enhancement process is expressed as: ； in, is a constant and is set according to the brightness of the sample image; is the pixel point of the sample image before image enhancement, and the pixel value is The value indicates that Adjust the parameters for the pixels. is the pixel point of the sample image after image enhancement, represents the pixel adjustment function; The filtering process is expressed as: ； in, is the pixel value of the pixel point in the i-th sample image after filtering, is the pixel value of the input i-th sample image, is the pixel value at any position in the sample image before filtering, Represents the index of any position in the sample image; is the pixel set of the sample image before filtering, is the standard deviation of the Gaussian function, is the weight vector of the i-th sample image, represents the exponential function, Indicates taking the absolute value.

6. The method for imitating a line of an unmanned aerial vehicle based on an image recognition algorithm according to claim 4, characterized in that: The specific construction method of the migrated and improved YOLOX model is as follows: Get the original YOLOX model; Build some convolution modules on the backbone network of the original YOLOX model to replace the original 4-layer ordinary convolution modules; Among them, the structure of some convolution modules is 1 The partial convolutional layer and 2 The ordinary convolutional layers are connected in series; In the original YOLOX backbone network, the Nix module is introduced after the first two C2f modules. The Nix module includes a self-attention layer, a convolutional combination layer, and a feature aggregation layer. The convolution feature of the input feature is obtained through the convolution combination layer in the Nix module, the self-attention feature of the input feature is obtained through the self-attention layer in the Nix module, and the convolution feature and the self-attention feature are aggregated through the feature aggregation layer in the Nix module to obtain the self-attention feature aggregation feature. The process of obtaining the self-attention feature aggregation feature is expressed as: ； in, is the self-attention feature aggregation feature, and are all preset weight matrices. is the convolution feature, is the self-attention feature.

7. The method for imitating a line of an unmanned aerial vehicle based on an image recognition algorithm according to claim 4, characterized in that: Loss function based on global vector Expressed as: ； ； ； ； in, is the attention coefficient, is the penalty amount, and are all hyperparameters, The outliers of the anchor boxes of the YOLOX model improved for migration, is the predicted box of the wire in the sample image, are the horizontal and vertical coordinates of the prediction box of the wire, is the ground-truth box of the wire in the sample image, are the horizontal and vertical coordinates of the ground truth box of the wire in the sample image; and Respectively represent the width and height of the minimum outer frame of the prediction box of the wire in the sample image, is the compensation penalty function, Penalty function for the improved YOLOX model for migration, Represents the exponential function.

8. The method for imitating a line of an unmanned aerial vehicle based on an image recognition algorithm according to claim 1, characterized in that: The specific process of pruning the backbone network in the transmission line conductor detection model through weight partitioning is as follows: ； in, is the model output function, are the model learnable parameters, is the pixel value of the input sample image, represents the weight score; According to the weight score, the connection weights below the threshold in the backbone network are deleted.

9. The method for imitating a line of an unmanned aerial vehicle based on an image recognition algorithm according to claim 1, characterized in that: The process of adjusting the camera lens focal length is as follows: Constructing the adjustment vector , expressed as: ； in, They are focus adjustment vector, angle adjustment vector and position adjustment vector respectively; Adjusting the lens based on the adjustment vector, photographing the wire with the adjusted lens to obtain a target image, evaluating the target image to obtain an image evaluation vector, and determining whether the image evaluation vector reaches a set score; When the image evaluation vector does not reach the set score, the adjustment vector is adjusted, the focal length of the lens is adjusted using the adjusted adjustment vector, and the wire is photographed again using the adjusted lens to obtain a target image, until the image evaluation vector of the target image reaches the set score, and the target image whose image evaluation vector reaches the set score is used as the wire image; The improved Kalman filter method for tracking wires is expressed as: ； in, represents the state of wire m at the tth moment, Indicates that the conductor m is The motion function at each moment, Indicates the The noise sequence at each moment, represents the position of wire m at the tth moment, Indicates that the conductor m is The position change function at each moment is Indicates the The measurement noise at each moment; The active control strategy is expressed as: ； ； in, For the drone at the next discrete moment status, For the drone at the current discrete moment status, For the drone at the current discrete moment The control input, For the drone at the current discrete moment The output status, is the state transition matrix, is the control input matrix, is the output matrix, is the direct transmission matrix.

10. A UAV line-mimicking system based on an image recognition algorithm, used to implement the UAV line-mimicking method based on an image recognition algorithm as claimed in any one of claims 1 to 9, characterized in that: include: The data acquisition and preprocessing module is used to collect image datasets of overhead ground wires and conductors, preprocess and classify the image datasets, and construct sample sets, verification sets, and test sets; The model training and optimization module is used to carry the YOLOX model and perform training based on the YOLOX model, optimize the YOLOX model parameters, and obtain the transmission line conductor detection model; Model lightweighting and deployment module, used to lightweight the transmission line conductor detection model and deploy it on the UAV computing terminal; The drone control module is used to adjust the lens focal length to the appropriate position and angle to take pictures of the wires, and control the drone to maintain a fixed distance from the wires and fly autonomously.

Images