Methods, systems, equipment, and media for defect screening of unmanned aerial vehicles (UAVs) in substations

Abstract

This invention relates to a method for screening defects in substation drones, comprising the following steps: acquiring historical operational data and operational defect logs of the substation drones, and establishing a first training set; training a first machine learning model based on the first training set to generate an operational defect screening model for the substation drones; taking original image sets of the substation drones with and without defects, and combining the images to generate a composite image; classifying the appearance defect types of the substation drones based on the composite image, establishing a second training set based on the composite image and the appearance defect types, training a second machine learning model, and generating an appearance defect screening model for the substation drones; acquiring real-time operational data and monitoring images of the substation drones, and inputting them into the operational defect screening model and the appearance defect screening model respectively for operational defect screening and appearance defect screening.

Description

Technical Field

[0001] This invention relates to a method, system, equipment, and medium for screening defects in substation drones, belonging to the field of substation inspection technology. Background Technology

[0002] Substations are a crucial component of the power system, undertaking the vital tasks of power transmission and conversion. To ensure the stable operation of the power system, daily inspections of substations are particularly important. These daily inspections mainly include the following aspects: Equipment appearance inspection: Inspecting the appearance of all equipment within the substation, such as switches, transformers, and cables, to ensure there is no external damage or abnormalities. Equipment operation data monitoring: Monitoring the operating parameters of various equipment, such as voltage, current, and temperature, to ensure they fluctuate within normal ranges. Safety facility inspection: Inspecting fire-fighting equipment, safety warning signs, protective facilities, etc., to ensure the safe and stable operation of the substation, etc.

[0003] Inspection work is complex, and manual inspection requires a lot of manpower and effort, and is likely to miss some items. Therefore, drones, such as intelligent inspection robots, are gradually being deployed in substation inspection work, which greatly improves the efficiency of substation inspection.

[0004] However, defects always appear after the substation drones are put into operation, and how to screen the defects of the equipment in real time has become a new technical problem. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a method, system, equipment, and medium for defect screening of unmanned aerial vehicles (UAVs) in substations.

[0006] The technical solution of the present invention is as follows:

[0007] On the one hand, this invention proposes a method for defect screening of unmanned aerial vehicles (UAVs) in substations, comprising the following steps:

[0008] Acquire historical operation data and operation defect logs of UAVs in substations, construct defect datasets and normal datasets based on historical operation data and operation defect logs, perform data augmentation on defect datasets, and establish the first training set based on the augmented defect datasets and normal datasets;

[0009] The first machine learning model is trained based on the first training set to generate a substation drone operation defect screening model;

[0010] For substation drones, original image sets of substation drones without defects and with defects are captured separately. A first comparison image is generated based on the image without defects, and a second comparison image is generated based on the image with defects. The colors of the first comparison image and the second comparison image are set to have contrast. The first comparison image and the second comparison image after color conversion are combined to generate a combined image.

[0011] For the combined image, the appearance defect type of the substation UAV is classified according to at least one of the defect color and aspect ratio, and a second training set is established based on the generated combined image and appearance defect type;

[0012] A second machine learning model is trained based on the second training set to generate an appearance defect screening model for substation drones.

[0013] The system acquires real-time operational data and monitoring images of the UAV at the substation. The real-time operational data is input into the operational defect screening model for operational defect screening, and the monitoring images are input into the appearance defect screening model for appearance defect screening.

[0014] As a preferred embodiment, the method for training a first machine learning model based on a first training set to generate a substation UAV operational defect screening model specifically includes:

[0015] Construct a first machine learning model that includes a first neural network and a second neural network;

[0016] The amount of data in the first training set is determined, and the first training set is divided into two parts based on the amount of data. One part is input into the first neural network to train the first neural network, and the other part is input into the second neural network to train the second neural network.

[0017] The mean square error is calculated based on the predicted operational defects output by the first and second trained neural networks. The first and second neural networks are then combined based on the calculated mean square error to obtain the operational defect screening model for substation drones.

[0018] In a preferred embodiment, the method of setting the colors of the first comparison image and the second comparison image to have contrast, and then combining the color-converted first comparison image and the second comparison image to generate a combined image, specifically includes:

[0019] Set the first comparison image to red, and set the defective parts in the second comparison image to green, while setting the non-defective parts to red;

[0020] After combining the first and second contrast images after color conversion, a portion of the combined image without defects is displayed in yellow, and a portion of the combined image with defects is displayed in red.

[0021] In a preferred embodiment, in the step of training a second machine learning model based on a second training set to generate an appearance defect screening model for a substation drone, the second machine learning model includes:

[0022] A defect image generation model is used to generate defect images from defects in substation drones based on the color and aspect ratio of defects in combined images.

[0023] A defect type determination model is used to determine the appearance defect type of a substation drone by comparing a generated defect image with a reference defect image.

[0024] On the other hand, the present invention also proposes a substation drone defect screening system, comprising:

[0025] The defect training set construction module is used to acquire historical operation data and operation defect logs of UAVs in substations. Based on the historical operation data and operation defect logs, a defect dataset and a normal dataset are constructed. Data augmentation processing is performed on the defect dataset, and the first training set is established based on the augmented defect dataset and the normal dataset.

[0026] Run the defect screening model training module to train the first machine learning model based on the first training set, and generate the substation drone's operational defect screening model;

[0027] The appearance defect image construction module takes original image sets of the substation drone with and without defects, respectively. It generates a first comparison image based on the image without defects and a second comparison image based on the image with defects. The colors of the first and second comparison images are set to have contrast. The first and second comparison images after color conversion are combined to generate a combined image.

[0028] The appearance defect training set construction module classifies the appearance defect types of substation UAVs based on at least one of the defect's color and aspect ratio for combined images, and establishes a second training set based on the generated combined images and appearance defect types.

[0029] The appearance defect screening model training module trains a second machine learning model based on the second training set to generate an appearance defect screening model for substation drones.

[0030] The defect screening module is used to acquire real-time operating data and monitoring images of the substation drone. The real-time operating data is input into the operation defect screening model for operation defect screening, and the monitoring images are input into the appearance defect screening model for appearance defect screening.

[0031] In a preferred embodiment, the method for generating an operational defect screening model for a substation UAV by training a first machine learning model based on a first training set is as follows:

[0032] Construct a first machine learning model that includes a first neural network and a second neural network;

[0033] The amount of data in the first training set is determined, and the first training set is divided into two parts based on the amount of data. One part is input into the first neural network to train the first neural network, and the other part is input into the second neural network to train the second neural network.

[0034] The mean square error is calculated based on the predicted operational defects output by the first and second trained neural networks. The first and second neural networks are then combined based on the calculated mean square error to obtain the operational defect screening model for substation drones.

[0035] In a preferred embodiment, the method for the appearance defect image construction module to set the colors of the first comparison image and the second comparison image to have contrast, and to combine the first comparison image and the second comparison image after color conversion to generate a combined image, is as follows:

[0036] Set the first comparison image to red, and set the defective parts in the second comparison image to green, while setting the non-defective parts to red;

[0037] After combining the first and second contrast images after color conversion, a portion of the combined image without defects is displayed in yellow, and a portion of the combined image with defects is displayed in red.

[0038] In a preferred embodiment, the step of the appearance defect screening model training module training a second machine learning model based on a second training set to generate an appearance defect screening model for a substation UAV includes:

[0039] A defect image generation model is used to generate defect images from defects in substation drones based on the color and aspect ratio of defects in combined images.

[0040] A defect type determination model is used to determine the appearance defect type of a substation drone by comparing a generated defect image with a reference defect image.

[0041] In another aspect, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method as described in any embodiment of the present invention.

[0042] In another aspect, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0043] The present invention has the following beneficial effects:

[0044] This invention establishes operational defect screening models and appearance defect screening models for substation drones, enabling real-time screening of defects in substation drones. This helps improve inspection efficiency, reduce manual intervention, and ensure the safe and stable operation of the power system. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0048] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0049] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0050] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0051] Example 1:

[0052] See Figure 1 A method for defect screening of unmanned aerial vehicles (UAVs) in substations includes the following steps:

[0053] S100. Acquire historical operational data and operational defect logs of the substation drone. This data includes real-time operational information such as the drone's flight trajectory, altitude, speed, battery level, and sensor data, as well as records of various defects and malfunctions that occurred during operation. When collecting data, it is crucial to ensure its integrity and accuracy to avoid data loss or errors that could lead to biases in subsequent analysis. Based on the historical operational data and operational defect logs, construct a defect dataset and a normal dataset. The defect dataset mainly includes records of various defects and malfunctions that occurred during drone operation, such as sensor failures, insufficient battery power, and flight control anomalies. The normal dataset includes various data from normal drone operation, such as normal flight trajectories and normal sensor data. To increase the diversity and generalization ability of the defect dataset, we need to perform data augmentation. Data augmentation is a method of generating new data samples by transforming and expanding the original data. In the drone defect dataset, we can use techniques such as adding random noise, data rotation, translation, and scaling to augment the original defect data to simulate more real-world operating conditions. This can effectively improve the robustness and generalization ability of the model, enabling the trained model to better adapt to various complex environments. The first training set is established based on the enhanced defective dataset and the normal dataset.

[0054] S200. Train a first machine learning model based on the first training set to generate a substation drone operation defect screening model. In this embodiment, for the task of substation drone operation defect screening, models such as Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN) in deep learning algorithms can be selected. These models have powerful feature extraction and classification capabilities, can handle large amounts of high-dimensional data, and have good generalization performance. By performing multiple rounds of training and adjustment on the first training set, we can obtain a relatively accurate substation drone operation defect screening model. In addition, to further improve the accuracy and generalization performance of the model, this embodiment also adopts the method of transfer learning. Transfer learning can utilize model parameters that have been trained in other fields to initialize our model, thereby accelerating the model training process and improving performance.

[0055] The S300 system, designed for substation drones, captures raw image sets of both defect-free and defective drones. High-definition cameras enable close-up, high-precision imaging of the drones. During capture, factors such as lighting and angle must be carefully considered to ensure clear and accurate images reflecting the actual condition of the equipment. A first comparison image is generated based on the defect-free image. This step establishes a baseline for subsequent comparison with the defective image. Image processing techniques, such as filtering and enhancement, are used to preprocess the raw images, improving image quality and making them more suitable for comparative analysis. A second comparison image is then generated based on the defective image. This step highlights the defective parts for easier subsequent analysis and identification. Image segmentation and edge detection techniques are used to extract the defective parts from the raw images, generating the second comparison image. The colors of both the first and second comparison images are set to have contrast, making the defective parts clearly distinguishable from the normal parts. This can be achieved by adjusting the image's color and contrast parameters. The first and second comparison images after color conversion are combined to generate a composite image, which compares and displays the normal and defective parts to facilitate subsequent analysis and judgment.

[0056] Composite images can be generated using techniques such as image overlay and fusion. When generating composite images, it's crucial to maintain image clarity and contrast, ensuring that normal and defective parts are clearly displayed. Furthermore, annotation and measurement functions can be used to provide detailed descriptions and analyses of defects. The generated composite images can quickly and accurately identify the type and extent of equipment defects.

[0057] S400. For the combined image, image processing techniques are used to extract defect features, which mainly include the defect's color, shape, size, and aspect ratio. Among these, color is one of the most intuitive features; color differences can provide a preliminary indication of the defect type. The aspect ratio reflects the spatial structure information of the defect, helping to distinguish between different types of defects. The appearance defect types of the substation UAV are classified based on features such as color and aspect ratio, categorizing defects into various types such as rust, scratches, and impacts. A second training set is established based on the generated combined image and the appearance defect types.

[0058] S500. Train a second machine learning model based on the second training set to generate a substation drone appearance defect screening model. In this embodiment, a convolutional neural network (CNN) from deep learning algorithms is selected as the second machine learning model. CNN is a neural network specifically designed for processing image data, possessing powerful feature extraction and classification capabilities. By continuously adjusting the model's parameters and structure, we enable the model to achieve a high accuracy rate on the second training set. This model can automatically identify and classify defects in substation appearance images captured by drones, such as common defects like cracks, corrosion, and leakage.

[0059] The S600 acquires real-time operational data and monitoring images of the substation drone, inputs the real-time operational data into the operational defect screening model for operational defect screening, and inputs the monitoring images into the appearance defect screening model for appearance defect screening.

[0060] As a preferred embodiment of this example, the method for training a first machine learning model based on a first training set to generate an operational defect screening model for a substation drone is specifically as follows:

[0061] We construct a first machine learning model comprising a first neural network and a second neural network. These two neural networks can employ different architectures and parameter settings to adapt to different data characteristics and task requirements. For example, the first neural network can use a convolutional neural network (CNN) to process image data, while the second neural network can use a recurrent neural network (RNN) to process sequence data. By combining different types of neural networks, we can fully utilize various data features and improve the model's prediction accuracy.

[0062] The first training set is divided into two parts based on the amount of data. One part is fed into a first neural network for training, and the other part is fed into a second neural network for training. The purpose of this process is to ensure that each neural network receives sufficient training, avoiding overfitting or underfitting. For example, if the training set is large, we can divide it into 80% training and 20% validation sets; if the data is small, we need to divide the dataset more finely to ensure that each neural network receives enough training data. Then, we feed the first part of the data into the first neural network for training, and the second part into the second neural network for training. During training, we need to select appropriate optimization algorithms and loss functions to minimize the error between predicted and actual values. Simultaneously, we also need to perform regularization on the neural networks to prevent overfitting.

[0063] The mean squared error (MSE) is calculated based on the predicted operational defects output from the first and second trained neural networks. MSE is a commonly used error metric that reflects the degree of deviation between predicted and actual values. By calculating the MSE, we can understand the differences and consistency in the prediction results between the two neural networks. The first and second neural networks are then combined based on the calculated MSE. This process can be achieved through weighted averaging, voting, or other ensemble learning methods. By combining the prediction results of the two neural networks, we can obtain a more stable and accurate operational defect screening model for substation drones.

[0064] In a preferred embodiment of this invention, the method of setting the colors of the first comparison image and the second comparison image to have contrast, and then combining the color-converted first comparison image and the second comparison image to generate a combined image, specifically comprises:

[0065] The first comparison image is set to red. This is done to clearly identify this part of the image during the subsequent combination process. Red, as a striking color, helps us visually quickly capture the content of this part of the image. In the second comparison image, defective areas are set to green, and defect-free areas are set to red. Green here represents the presence of differences or defects. By marking defective areas in green, we can more intuitively identify the parts in the second image that differ from the first image. Meanwhile, for defect-free areas in the second comparison image, we set them to red. This is done to maintain consistency with the first comparison image, making the combined image easier to understand. This embodiment uses a color-coding method to make differences more obvious in the combined image.

[0066] When we combine the first and second contrast images after color conversion, two scenarios will occur. One is the defect-free portion of the combined image, which will be displayed in yellow. Yellow is a mixture of red and green, representing the consistency of the two images in that part. The other is the defective portion of the combined image, which will be displayed in red. Red here represents the presence of defects in the second contrast image, creating a sharp contrast with the green. Through this color-coding method, we can more intuitively identify the differences and defects between the two images.

[0067] In a preferred embodiment of this invention, the step of training a second machine learning model based on a second training set to generate a substation drone appearance defect screening model includes:

[0068] A defect image generation model is used to generate defect images from substation drones based on the color and aspect ratio of defects in composite images. This model, based on deep learning and image processing techniques, can generate defect images from substation drones using the color and aspect ratio of defects in composite images. Specifically, the model first preprocesses the substation images captured by the drone, including denoising and enhancement operations to improve image quality. Then, it uses deep learning algorithms such as convolutional neural networks (CNNs) to extract features and identify defects in the preprocessed image. Finally, it generates the corresponding defect image based on features such as the color and aspect ratio of the defects. The defect image generation model can quickly and accurately obtain defect information from substation drones.

[0069] A defect type determination model is used to identify the appearance defect types of substation drones by comparing generated defect images with reference defect images. Specifically, the model first establishes a reference image library containing various defect types. Then, the generated defect image is compared one by one with images in the reference image library, and the defect type is determined by calculating the similarity between the images. The defect type determination model can accurately classify appearance defects of substation drones.

[0070] Example 2:

[0071] This embodiment proposes a substation drone defect screening system, including:

[0072] The defect training set construction module is used to acquire historical operating data and operational defect logs of the UAV in the substation, construct a defect dataset and a normal dataset based on the historical operating data and operational defect logs, perform data augmentation on the defect dataset, and establish a first training set based on the augmented defect dataset and the normal dataset; this module is used to implement the function of step S100 in Example 1, which will not be described in detail here.

[0073] The defect screening model training module is used to train the first machine learning model based on the first training set to generate an operational defect screening model for the substation UAV. This module is used to implement the function of step S200 in Example 1, and will not be described in detail here.

[0074] The appearance defect image construction module, for substation drones, takes original image sets of substation drones with and without defects, respectively. It generates a first comparison image based on the image without defects and a second comparison image based on the image with defects. It sets the colors of the first and second comparison images to have contrast and combines the first and second comparison images after color conversion to generate a combined image. This module is used to implement the function of step S300 in Embodiment 1, and will not be described in detail here.

[0075] The appearance defect training set construction module classifies the appearance defect types of the substation UAV based on at least one of the defect's color and aspect ratio for the combined image, and establishes a second training set based on the generated combined image and appearance defect types; this module is used to implement the function of step S400 in Embodiment 1, and will not be described in detail here.

[0076] The appearance defect screening model training module trains a second machine learning model based on the second training set to generate an appearance defect screening model for the substation UAV. This module is used to implement the function of step S500 in Example 1, and will not be described in detail here.

[0077] The defect screening module is used to acquire real-time operating data and monitoring images of the substation UAV, input the real-time operating data into the operation defect screening model for operation defect screening, and input the monitoring images into the appearance defect screening model for appearance defect screening; this module is used to implement the function of step S600 in Embodiment 1, which will not be described in detail here.

[0078] In a preferred embodiment of this invention, the method by which the operational defect screening model training module trains a first machine learning model based on a first training set to generate an operational defect screening model for a substation UAV is as follows:

[0079] Construct a first machine learning model that includes a first neural network and a second neural network;

[0080] The amount of data in the first training set is determined, and the first training set is divided into two parts based on the amount of data. One part is input into the first neural network to train the first neural network, and the other part is input into the second neural network to train the second neural network.

[0081] The mean square error is calculated based on the predicted operational defects output by the first and second trained neural networks. The first and second neural networks are then combined based on the calculated mean square error to obtain the operational defect screening model for substation drones.

[0082] In a preferred embodiment of this invention, the method for the appearance defect image construction module to set the colors of the first comparison image and the second comparison image to have contrast, and to combine the first comparison image and the second comparison image after color conversion to generate a combined image, is as follows:

[0083] Set the first comparison image to red, and set the defective parts in the second comparison image to green, while setting the non-defective parts to red;

[0084] After combining the first and second contrast images after color conversion, a portion of the combined image without defects is displayed in yellow, and a portion of the combined image with defects is displayed in red.

[0085] In a preferred embodiment of this invention, in the step of the appearance defect screening model training module training a second machine learning model based on a second training set to generate an appearance defect screening model for a substation UAV, the second machine learning model includes:

[0086] A defect image generation model is used to generate defect images from defects in substation drones based on the color and aspect ratio of defects in combined images.

[0087] A defect type determination model is used to determine the appearance defect type of a substation drone by comparing a generated defect image with a reference defect image.

[0088] Example 3:

[0089] This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the substation drone defect screening method as described in any embodiment of the present invention.

[0090] Example 4:

[0091] This embodiment proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the substation drone defect screening method as described in any embodiment of the present invention.

[0092] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0093] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0094] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0095] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for defect screening of unmanned aerial vehicles (UAVs) in substations, characterized in that, Includes the following steps: Acquire historical operation data and operation defect logs of UAVs in substations, construct defect datasets and normal datasets based on historical operation data and operation defect logs, perform data augmentation on defect datasets, and establish the first training set based on the augmented defect datasets and normal datasets; The first machine learning model is trained based on the first training set to generate a substation drone operation defect screening model; For substation drones, original image sets of substation drones without defects and with defects are captured separately. A first comparison image is generated based on the image without defects, and a second comparison image is generated based on the image with defects. The colors of the first comparison image and the second comparison image are set to have contrast. The first comparison image and the second comparison image after color conversion are combined to generate a combined image. For the combined image, the appearance defect type of the substation UAV is classified according to at least one of the defect color and aspect ratio, and a second training set is established based on the generated combined image and appearance defect type; A second machine learning model is trained based on the second training set to generate an appearance defect screening model for substation drones. The system acquires real-time operational data and monitoring images of the UAV at the substation. The real-time operational data is input into the operational defect screening model for operational defect screening, and the monitoring images are input into the appearance defect screening model for appearance defect screening.

2. The method for defect screening of unmanned aerial vehicles (UAVs) in substations according to claim 1, characterized in that, The method for training a first machine learning model based on a first training set to generate a substation drone operation defect screening model is as follows: Construct a first machine learning model that includes a first neural network and a second neural network; The amount of data in the first training set is determined, and the first training set is divided into two parts based on the amount of data. One part is input into the first neural network to train the first neural network, and the other part is input into the second neural network to train the second neural network. The mean square error is calculated based on the predicted operational defects output by the first and second trained neural networks. The first and second neural networks are then combined based on the calculated mean square error to obtain the operational defect screening model for substation drones.

3. The method for defect screening of unmanned aerial vehicles (UAVs) in substations according to claim 1, characterized in that, The method of setting the colors of the first contrast image and the second contrast image to have contrast, and then combining the first contrast image and the second contrast image after color conversion to generate a combined image is as follows: Set the first comparison image to red, and set the defective parts in the second comparison image to green, while setting the non-defective parts to red; After combining the first and second contrast images after color conversion, a portion of the combined image without defects is displayed in yellow, and a portion of the combined image with defects is displayed in red.

4. The method for defect screening of unmanned aerial vehicles (UAVs) in substations according to claim 1, characterized in that, In the step of training a second machine learning model based on a second training set to generate a substation drone appearance defect screening model, the second machine learning model includes: A defect image generation model is used to generate defect images from defects in substation drones based on the color and aspect ratio of defects in combined images. A defect type determination model is used to determine the appearance defect type of a substation drone by comparing a generated defect image with a reference defect image.

5. A substation unmanned aerial vehicle (UAV) defect screening system, characterized in that, include: The defect training set construction module is used to acquire historical operation data and operation defect logs of UAVs in substations. Based on the historical operation data and operation defect logs, a defect dataset and a normal dataset are constructed. Data augmentation processing is performed on the defect dataset, and the first training set is established based on the augmented defect dataset and the normal dataset. Run the defect screening model training module to train the first machine learning model based on the first training set, and generate the substation drone's operational defect screening model; The appearance defect image construction module takes original image sets of the substation drone with and without defects, respectively. It generates a first comparison image based on the image without defects and a second comparison image based on the image with defects. The colors of the first and second comparison images are set to have contrast. The first and second comparison images after color conversion are combined to generate a combined image. The appearance defect training set construction module classifies the appearance defect types of substation UAVs based on at least one of the defect's color and aspect ratio for combined images, and establishes a second training set based on the generated combined images and appearance defect types. The appearance defect screening model training module trains a second machine learning model based on the second training set to generate an appearance defect screening model for substation drones. The defect screening module is used to acquire real-time operating data and monitoring images of the substation drone. The real-time operating data is input into the operation defect screening model for operation defect screening, and the monitoring images are input into the appearance defect screening model for appearance defect screening.

6. A substation unmanned aerial vehicle (UAV) defect screening system according to claim 5, characterized in that, The method for generating the substation UAV operational defect screening model by training a first machine learning model based on a first training set is as follows: Construct a first machine learning model that includes a first neural network and a second neural network; The amount of data in the first training set is determined, and the first training set is divided into two parts based on the amount of data. One part is input into the first neural network to train the first neural network, and the other part is input into the second neural network to train the second neural network. The mean square error is calculated based on the predicted operational defects output by the first and second trained neural networks. The first and second neural networks are then combined based on the calculated mean square error to obtain the operational defect screening model for substation drones.

7. A substation unmanned aerial vehicle (UAV) defect screening system according to claim 5, characterized in that, The method for the appearance defect image construction module to set the colors of the first comparison image and the second comparison image to have contrast, and to combine the first comparison image and the second comparison image after color conversion to generate a combined image is as follows: Set the first comparison image to red, and set the defective parts in the second comparison image to green, while setting the non-defective parts to red; After combining the first and second contrast images after color conversion, a portion of the combined image without defects is displayed in yellow, and a portion of the combined image with defects is displayed in red.

8. A substation unmanned aerial vehicle (UAV) defect screening system according to claim 5, characterized in that, In the step of the appearance defect screening model training module training a second machine learning model based on a second training set to generate an appearance defect screening model for a substation UAV, the second machine learning model includes: A defect image generation model is used to generate defect images from defects in substation drones based on the color and aspect ratio of defects in combined images. A defect type determination model is used to determine the appearance defect type of a substation drone by comparing a generated defect image with a reference defect image.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.

Images