Power grid fault point high-definition imaging and intelligent identification system and method based on unmanned aerial vehicle

Abstract

The invention relates to a power grid fault point high-definition imaging and intelligent identification system and method based on an unmanned aerial vehicle, and the system comprises a data collection module which is used for collecting the data of an inspection image shot by the unmanned aerial vehicle in an inspection task; the data processing module is used for receiving and preprocessing the inspection image data; processing the inspection image data through a high-precision image processing unit, and outputting fault category prediction data of the inspection image data; based on the fault category prediction data, fault target detection is carried out on the inspection image data through a target detection unit; and the data display module is used for displaying a fault inspection result which comprises the inspection image data, corresponding fault category prediction data output by the high-precision image processing unit and a fault target detection result output by the target detection unit.

Description

Technical Field

[0001] The present invention relates to a high-definition imaging and intelligent identification system and method for power grid fault points based on an unmanned aerial vehicle, and belongs to the technical field of power grid inspection. Background Art

[0002] At present, there are some solutions for power grid inspection using drones on the market. These solutions usually consist of drone flight platforms, high-definition camera equipment, and basic image processing algorithms. The drone flight platform automatically or semi-automatically inspects the power grid through a preset flight path, and continuously photographs various parts of the power grid during the flight. These high-definition camera equipment can capture high-resolution image data to ensure that the details of the power grid are clearly presented. After the inspection is completed, a large amount of image data is stored and transmitted to the ground control center or remote server. At present, these solutions usually rely on manual analysis or the application of simple image processing algorithms for preliminary fault detection. Although manual analysis has certain flexibility, it is inefficient and easily affected by human factors; while the basic image processing algorithm has a certain effect in dealing with some simple power grid faults, such as obvious disconnection and damage, but for the fault types in complex environments, such as small cracks, hidden structural damage, or faults under environmental clutter interference, the recognition effect is often not ideal. In addition, these systems usually have large delays in data processing and analysis, which makes it difficult to meet the requirements of real-time detection. These factors limit the effectiveness of existing drone inspection solutions in practical applications.

[0003] At present, when using existing technical solutions for power grid inspection, there are mainly the following disadvantages:

[0004] Limited detection accuracy: As existing solutions rely on basic image processing algorithms, they are unable to fully identify and analyze complex fault types, such as tiny cracks, hidden structural damage, etc. This method is easily interfered by environmental factors, such as lighting changes, background clutter, etc., resulting in low accuracy in fault identification and the risk of false positives and false negatives.

[0005] Reliance on manual analysis: Many existing drone inspection solutions still rely on manual analysis of large amounts of image data. This is not only time-consuming and laborious, but the accuracy and consistency of manual analysis may vary depending on the analyst’s experience and status, resulting in inefficient inspections and unstable results. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present invention proposes a high-definition imaging and intelligent identification system and method for power grid fault points based on drones.

[0007] The technical solution of the present invention is as follows:

[0008] On the one hand, the present invention proposes a high-definition imaging and intelligent identification system for power grid fault points based on drones, comprising:

[0009] The data acquisition module is used to collect inspection image data taken by the drone during the inspection mission;

[0010] A data processing module is used to receive and pre-process the inspection image data; process the inspection image data through a high-precision image processing unit, and output fault category prediction data of the inspection image data; and based on the fault category prediction data, perform fault target detection on the inspection image data through a target detection unit;

[0011] The data display module is used to display the fault inspection results, including the inspection image data and the corresponding fault category prediction data output by the high-precision image processing unit and the fault target detection results output by the target detection unit.

[0012] As a preferred implementation, the high-precision image processing unit adopts a CNN convolutional neural network model;

[0013] The input of the CNN convolutional neural network model is the inspection image data, and the output is the category prediction result;

[0014] The convolutional neural network model includes an input layer, a convolution layer, a pooling layer, a deep convolution and residual connection layer, a fully connected layer and an output layer; wherein:

[0015] The input layer is used to input inspection image data;

[0016] The convolution layer is used to extract local features of the input inspection image data and generate a preliminary feature map;

[0017] The pooling layer is used to perform the maximum pooling operation on the preliminary feature map;

[0018] The deep convolution and residual connection layer introduces multiple convolution blocks and residual connection blocks, performs deep convolution and residual connection on the preliminary feature map after the maximum pooling operation, and obtains the final feature map;

[0019] The fully connected layer is used to flatten the final feature map;

[0020] The output layer uses a softmax function to convert the output of the fully connected layer into a category probability distribution.

[0021] As a preferred embodiment, the inspection image data is further preprocessed before being input into the high-precision image processing unit, including:

[0022] Perform image enhancement and denoising processing on the inspection image data.

[0023] As a preferred implementation, the target detection unit adopts the YOLOv5 model, and the YOLOv5 algorithm includes:

[0024] Backbone network, neck network and detection head, where:

[0025] The backbone network is used to extract features from the input inspection image data;

[0026] The neck network is used to perform feature fusion operations on features of different scales output by the backbone network;

[0027] The detection head is used to perform target detection based on the fused features and output the target’s bounding box coordinates, category probability, and confidence.

[0028] On the other hand, the present invention also proposes a method for high-definition imaging and intelligent identification of power grid fault points based on drones, comprising the following steps:

[0029] Collect inspection image data taken by drones during inspection missions;

[0030] Preprocess the inspection image data; process the inspection image data through a high-precision image processing unit to output fault category prediction data of the inspection image data; and based on the fault category prediction data, perform fault target detection on the inspection image data through a target detection unit;

[0031] Display fault inspection results, including inspection image data and the corresponding fault category prediction data output by the high-precision image processing unit and fault target detection results output by the target detection unit.

[0032] As a preferred implementation, the high-precision image processing unit adopts a CNN convolutional neural network model;

[0033] The input of the CNN convolutional neural network model is the inspection image data, and the output is the category prediction result;

[0034] The convolutional neural network model includes an input layer, a convolution layer, a pooling layer, a deep convolution and residual connection layer, a fully connected layer and an output layer; wherein:

[0035] The input layer is used to input inspection image data;

[0036] The convolution layer is used to extract local features of the input inspection image data and generate a preliminary feature map;

[0037] The pooling layer is used to perform the maximum pooling operation on the preliminary feature map;

[0038] The deep convolution and residual connection layer introduces multiple convolution blocks and residual connection blocks, performs deep convolution and residual connection on the preliminary feature map after the maximum pooling operation, and obtains the final feature map;

[0039] The fully connected layer is used to flatten the final feature map;

[0040] The output layer uses a softmax function to convert the output of the fully connected layer into a category probability distribution.

[0041] As a preferred embodiment, the inspection image data is further preprocessed before being input into the high-precision image processing unit, including:

[0042] Perform image enhancement and denoising processing on the inspection image data.

[0043] As a preferred implementation, the target detection unit adopts the YOLOv5 model, and the YOLOv5 algorithm includes:

[0044] Backbone network, neck network and detection head, where:

[0045] The backbone network is used to extract features from the input inspection image data;

[0046] The neck network is used to perform feature fusion operations on features of different scales output by the backbone network;

[0047] The detection head is used to perform target detection based on the fused features and output the target’s bounding box coordinates, category probability, and confidence.

[0048] On the other hand, the present invention further proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present invention when executing the program.

[0049] In yet another aspect, the present invention further provides 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.

[0050] Additional aspects and advantages of the present invention will be set forth in the following description, and some of them will be apparent from the description, or may be understood by practicing the present invention. In addition, the various aspects and advantages of the present invention may be realized and obtained by the method steps and combinations particularly pointed out in the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a schematic diagram of system modules of Embodiment 1 of the present invention;

[0052] Figure 2 Schematic diagram of a UAV power grid inspection management system constructed according to the system in an embodiment of the present invention;

[0053] Figure 3This is an example diagram of the workflow of the system in Example 1 of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0056] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0057] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0058] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0059] Embodiment 1:

[0060] In order to solve the problems of limited detection accuracy, excessive reliance on manual analysis, and challenges to storage and analysis efficiency in existing power grid inspection technologies, this embodiment provides a high-definition imaging and intelligent identification system for power grid fault points based on drones, which improves the accuracy of identifying complex fault types by improving image processing algorithms. The introduction of intelligent and automated fault detection systems reduces reliance on manual analysis and improves detection efficiency and accuracy.

[0061] See also Figure 1 The high-definition imaging and intelligent identification system for power grid fault points includes:

[0062] The data acquisition module is used to collect inspection image data taken by the drone during the inspection mission;

[0063] A data processing module is used to receive and pre-process the inspection image data; process the inspection image data through a high-precision image processing unit, and output fault category prediction data of the inspection image data; and based on the fault category prediction data, perform fault target detection on the inspection image data through a target detection unit;

[0064] The data display module is used to display the fault inspection results, including the inspection image data and the corresponding fault category prediction data output by the high-precision image processing unit and the fault target detection results output by the target detection unit. In the data display module, the Web interface or a special data visualization tool is used to display the data, so that users can intuitively understand the data analysis results. Finally, the inspection data is summarized and the data report is used for post-task evaluation, anomaly detection and data analysis.

[0065] As an implementation method, see Figure 2 In this embodiment, a comprehensive UAV power grid inspection management system is constructed based on the high-definition imaging and intelligent identification system of power grid fault points, which is mainly divided into the following levels:

[0066] 1. Data layer: responsible for processing and storing power grid inspection data from drones.

[0067] 2. System service layer: provides core services such as data processing, fault detection and analysis services.

[0068] 3. Application service layer: processes user requests and business logic, and provides front-end interface interaction and application functions.

[0069] 4. Access layer: responsible for communication with the user end, including browser access, API interface and data transmission.

[0070] This architectural design not only improves the flexibility and scalability of the system, but also ensures the efficient management and processing of power grid inspection data.

[0071] As a preferred implementation of this embodiment, the high-precision image processing unit adopts a CNN convolutional neural network model; the CNN convolutional neural network model performs multi-level feature extraction and analysis on the power grid images taken by the drone to achieve accurate identification of complex fault types. In particular, for fault types that are difficult to detect with traditional methods, such as tiny cracks and hidden structural damage, the deep feature extraction capability of CNN can effectively capture subtle changes and hidden features in the image, significantly improving the accuracy of fault detection.

[0072] The input of the CNN convolutional neural network model is the inspection image data taken by the drone. After preprocessing, these images are converted into a format suitable for neural network processing. Usually, the size of the input image is fixed, and each pixel contains the color information of the RGB three channels, forming a 3D tensor (Height×Width×Channels).

[0073] The output of the CNN convolutional neural network model is a multi-classification or multi-label prediction result. For the identification of a single fault type, the output is a vector containing n categories (n is the number of types of power grid faults), and each element represents the confidence value of the corresponding category. For multi-label scenarios (such as the presence of multiple fault types at the same time), the output is a multi-dimensional vector, and each dimension represents the possibility of the corresponding fault (such as tower collapse, landslide, line break, etc.).

[0074] The convolutional neural network model includes an input layer, a convolution layer, a pooling layer, a deep convolution and residual connection layer, a fully connected layer and an output layer; wherein:

[0075] The input layer is used to input inspection image data; the image size is H×W×3, where H and W are the height and width of the image respectively, and 3 represents the RGB three channels.

[0076] The convolution layer is used to extract the local features of the input inspection image data and generate preliminary feature maps. The first convolution layer applies k1 convolution kernels (Filter), each of which has a size of F1×F1, to generate k1 feature maps (FeatureMaps), which are expressed as:

[0077] F1=σ(W1*I+b1);

[0078] Where: W1 is the convolution kernel matrix, I is the input image, b1 is the bias term, σ is the activation function (ReLU), and * represents the convolution operation.

[0079] The pooling layer is used to perform the maximum pooling operation on the preliminary feature map; it is used to reduce the size of the feature map while retaining the main features. Usually, the maximum pooling (Max Pooling) is used:

[0080] P1 = MaxPool(F1);

[0081] For example, a 2×2 pooling operation reduces the feature map size to half of its original size.

[0082] The deep convolution and residual connection layers introduce multiple convolution blocks and residual connection blocks, and perform deep convolution and residual connection on the preliminary feature map after the maximum pooling operation to capture more complex features. The output of these deep convolution layers is expressed as:

[0083] F n=σ(W n *F n-1 +b n );

[0084] And the feature representation capability is enhanced through residual connection to obtain the final feature map:

[0085] F n ′=F n +F n-2 ;

[0086] The fully connected layer is used to flatten the final feature map for final fault classification:

[0087] z=W fc f+b fc ;

[0088] Where: f is the flattened eigenvector, W fc and b fc are the weight matrix and bias term of the fully connected layer respectively.

[0089] The output layer uses the softmax function to convert the output of the fully connected layer into a category probability distribution:

[0090]

[0091] in: represents the predicted probability of category i, and C is the total number of fault categories.

[0092] As a preferred implementation of this embodiment, the inspection image data is also pre-processed before being input into the high-precision image processing unit, including image enhancement processing and denoising processing, so as to improve the robustness of the model in complex environments. Image enhancement processing includes operations such as brightness adjustment, contrast enhancement, and color balance to ensure the quality of the image under various lighting conditions. In addition, denoising processing reduces the impact of environmental noise on the image through filtering, noise suppression and other methods, improves image clarity, and thus provides cleaner input data for CNN.

[0093] As a preferred implementation of this embodiment, the target detection unit adopts the YOLOv5 model. Based on the high-precision image processing unit, the target detection unit further realizes a fully automated fault detection system. The unit uses the YOLOv5 algorithm to identify power grid faults through automated image analysis, reducing reliance on manual analysis while improving detection efficiency and consistency.

[0094] YOLOv5 (You Only Look Once Version 5) is an advanced object detection algorithm that performs well in various visual tasks with its efficient and accurate performance. It mainly includes the following parts:

[0095] Backbone network, neck network and detection head, where:

[0096] The backbone network is used to extract features from the input inspection image data; the backbone network is based on CSPNet (Cross-Stage Partial Network). CSPNet divides the feature map into two parts and merges them at different stages, which can improve the efficiency of feature extraction and the accuracy of the model.

[0097] The neck network is used to perform feature fusion operations on features of different scales output by the backbone network; the neck network is based on PANet (Path Aggregation Network), which fuses features of different layers through the path aggregation module, thereby improving the model's ability to detect targets.

[0098] There are three detection heads, and three detection heads of different scales are used for object detection. These detection heads are responsible for object detection of different scales, and finally get the detection results. Each detection head will output the coordinates of the bounding box, category probability and confidence.

[0099] As a preferred implementation of this embodiment, the system also integrates big data collection and analysis functions, see Figure 3 ,After the data acquisition module collects the inspection image data through the ,UAV, the data acquisition module sends it to the data processing module. ,This process is divided into two stages: offline and real-time ,online.

[0100] Specifically, if it is in an offline state, the collected inspection image data is first stored offline; then enter the abnormality detection process, in which an algorithm is used to identify abnormal images or outlier images in the inspection image data, and some bad images or meaningless images can be eliminated first. The inspection image data after abnormality detection is sent to the data cleaning process, which can eliminate some images with insufficient pixels to ensure the quality of the image data. The cleaned inspection image data is stored offline, and the storage method can be used in the HDFS distributed file storage system for subsequent batch processing and analysis. The inspection image data stored offline will be processed in the offline computing stage, and the abnormal classification and target detection preprocessing of the inspection image can be performed in the offline computing, such as image denoising, feature calculation, etc. In addition, when the system is online, some of the inspection image data stored offline can be instantly analyzed through real-time online processing to quickly respond to monitored events.

[0101] Specifically, if it is in the real-time processing process, the inspection image data is received and temporarily stored through the message queue, such as the Kafka system, which can be used for asynchronous data transmission between multiple terminals to improve the efficiency and reliability of data processing; then the inspection image data is detected by the YOLOv5 detection algorithm. The target detection results are transferred to the database system through Talend. The database system can use relational databases such as MySQL and Oracle, as well as possible NoSQL solutions, to store processing results and generate data reports.

[0102] Finally, the data display module is uniformly based on the relational database and displayed through a web interface or a specialized data visualization tool, allowing users to intuitively understand the data analysis results.

[0103] Based on the above implementation, this embodiment first uses drones to capture high-definition images of the power grid during the inspection process. These images are pre-processed, including resizing, normalization, and data enhancement, to meet the needs of subsequent analysis. Next, the CNN convolutional neural network model is used to classify the input images into fault categories to determine the fault category of each image; this embodiment pre-trains target detection models corresponding to different fault categories using the YOLOv5 algorithm for different fault categories; then, according to the determined fault category, the target detection outputs the input image to the target detection model corresponding to the fault category;

[0104] The object detection model uses the core component of the YOLOv5 algorithm, namely the backbone network (Backbone-CSPNet) based on CSPDarknet53, to extract features and segment and merge feature maps. Subsequently, the path aggregation network (Neck-PANet) enhances the detection ability of objects of different sizes through multi-scale feature fusion. The detection head (Head-YOLO Head) of YOLOv5 is responsible for object detection at different scales, predicting bounding box coordinates, category probability and confidence. Non-maximum suppression (NMS) is used to remove overlapping predictions, and the results are integrated and optimized to ensure the accuracy of the detection results. Finally, the fault detection and recognition results of each image are output.

[0105] Ultimately, these precisely analyzed fault detection results will be saved and provided to the data display module, providing decision support and maintenance suggestions for operators, enabling them to quickly respond to potential faults in the power grid, thereby significantly improving the efficiency and consistency of power grid inspections. This process not only reduces reliance on manual analysis, but also ensures the stable operation of the power grid and the timeliness of maintenance work through intelligent means.

[0106] Based on the above implementation, this embodiment improves the fault detection accuracy and automation level of power grid inspection: by adopting advanced convolutional neural network (CNN) and YOLOv5 algorithm, the detection accuracy of power grid faults is significantly improved. At the same time, the automated fault detection process reduces the dependence on manual analysis and improves the detection efficiency and consistency.

[0107] Optimize system architecture and user interaction: Provides a flexible and scalable system architecture that supports multi-source data integration and multi-model management. The intuitive display of the user interface further enhances the user experience, making fault detection results easy to understand and operate, and speeding up fault response and maintenance processes.

[0108] Reduce operating costs and improve grid stability: The automated and intelligent inspection process significantly reduces manpower and material costs, while improving the real-time and accuracy of grid fault detection, helping to promptly discover and repair potential problems, thereby ensuring the stable operation and reliability of the grid.

[0109] Embodiment 2:

[0110] This embodiment proposes a method for high-definition imaging and intelligent identification of power grid fault points based on a drone, comprising the following steps:

[0111] Collect inspection image data taken by the drone during the inspection mission; this step is used to implement the function of the data acquisition module in Example 1, and will not be repeated here;

[0112] Preprocess the inspection image data; process the inspection image data through a high-precision image processing unit, and output fault category prediction data of the inspection image data; and based on the fault category prediction data, perform fault target detection on the inspection image data through a target detection unit; this step is used to implement the function of the data processing module in the first embodiment, and will not be repeated here;

[0113] Display the fault inspection results, including the inspection image data and the corresponding fault category prediction data output by the high-precision image processing unit and the fault target detection results output by the target detection unit; this step is used to implement the function of the data display module in Example 1 and will not be repeated here.

[0114] As a preferred implementation of this embodiment, the high-precision image processing unit adopts a CNN convolutional neural network model;

[0115] The input of the CNN convolutional neural network model is the inspection image data, and the output is the category prediction result;

[0116] The convolutional neural network model includes an input layer, a convolution layer, a pooling layer, a deep convolution and residual connection layer, a fully connected layer and an output layer; wherein:

[0117] The input layer is used to input inspection image data;

[0118] The convolution layer is used to extract local features of the input inspection image data and generate a preliminary feature map;

[0119] The pooling layer is used to perform the maximum pooling operation on the preliminary feature map;

[0120] The deep convolution and residual connection layer introduces multiple convolution blocks and residual connection blocks, performs deep convolution and residual connection on the preliminary feature map after the maximum pooling operation, and obtains the final feature map;

[0121] The fully connected layer is used to flatten the final feature map;

[0122] The output layer uses a softmax function to convert the output of the fully connected layer into a category probability distribution.

[0123] As a preferred implementation of this embodiment, the inspection image data is also pre-processed before being input into the high-precision image processing unit, including:

[0124] Perform image enhancement and denoising processing on the inspection image data.

[0125] As a preferred implementation of this embodiment, the target detection unit adopts the YOLOv5 model, and the YOLOv5 algorithm includes:

[0126] Backbone network, neck network and detection head, where:

[0127] The backbone network is used to extract features from the input inspection image data;

[0128] The neck network is used to perform feature fusion operations on features of different scales output by the backbone network;

[0129] The detection head is used to perform target detection based on the fused features and output the target’s bounding box coordinates, category probability, and confidence.

[0130] Embodiment three:

[0131] 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, the high-definition imaging and intelligent identification method for power grid fault points based on a drone as described in any embodiment of the present invention is implemented.

[0132] Embodiment 4:

[0133] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for high-definition imaging and intelligent identification of power grid fault points based on a drone as described in any embodiment of the present invention is implemented.

[0134] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated 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 single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0135] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0136] Those skilled in the art can 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.

[0137] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0138] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A high-definition imaging and intelligent identification system for power grid fault points based on drones, characterized in that: include: The data acquisition module is used to collect inspection image data taken by the drone during the inspection mission; A data processing module, used for receiving inspection image data and performing preprocessing; The inspection image data is processed by a high-precision image processing unit to output fault category prediction data of the inspection image data; Based on the fault category prediction data, the object detection unit performs fault target detection on the inspection image data; The data display module is used to display the fault inspection results, including the inspection image data and the corresponding fault category prediction data output by the high-precision image processing unit and the fault target detection results output by the target detection unit.

2. According to claim 1, a high-definition imaging and intelligent identification system for power grid fault points based on drones is characterized in that: The high-precision image processing unit adopts a CNN convolutional neural network model; The input of the CNN convolutional neural network model is the inspection image data, and the output is the category prediction result; The convolutional neural network model includes an input layer, a convolution layer, a pooling layer, a deep convolution and residual connection layer, a fully connected layer and an output layer; wherein: The input layer is used to input inspection image data; The convolution layer is used to extract local features of the input inspection image data and generate a preliminary feature map; The pooling layer is used to perform the maximum pooling operation on the preliminary feature map; The deep convolution and residual connection layer introduces multiple convolution blocks and residual connection blocks, performs deep convolution and residual connection on the preliminary feature map after the maximum pooling operation, and obtains the final feature map; The fully connected layer is used to flatten the final feature map; The output layer uses a softmax function to convert the output of the fully connected layer into a category probability distribution.

3. The high-definition imaging and intelligent identification system for power grid fault points based on drones according to claim 1 is characterized in that: The inspection image data is also pre-processed before being input into the high-precision image processing unit, including: Perform image enhancement and denoising processing on the inspection image data.

4. The high-definition imaging and intelligent identification system for power grid fault points based on drones according to claim 1 is characterized in that: The target detection unit adopts the YOLOv5 model, and the YOLOv5 algorithm includes: Backbone network, neck network and detection head, where: The backbone network is used to extract features from the input inspection image data; The neck network is used to perform feature fusion operations on features of different scales output by the backbone network; The detection head is used to perform target detection based on the fused features and output the target’s bounding box coordinates, category probability, and confidence.

5. A method for high-definition imaging and intelligent identification of power grid fault points based on drones, characterized in that: The following steps are involved: Collect inspection image data taken by drones during inspection missions; Preprocessing inspection image data; The inspection image data is processed by a high-precision image processing unit to output fault category prediction data of the inspection image data; Based on the fault category prediction data, the object detection unit performs fault target detection on the inspection image data; Display fault inspection results, including inspection image data and the corresponding fault category prediction data output by the high-precision image processing unit and fault target detection results output by the target detection unit.

6. The method for high-definition imaging and intelligent identification of power grid fault points based on drones according to claim 5 is characterized in that: The high-precision image processing unit adopts a CNN convolutional neural network model; The input of the CNN convolutional neural network model is the inspection image data, and the output is the category prediction result; The convolutional neural network model includes an input layer, a convolution layer, a pooling layer, a deep convolution and residual connection layer, a fully connected layer and an output layer; wherein: The input layer is used to input inspection image data; The convolution layer is used to extract local features of the input inspection image data and generate a preliminary feature map; The pooling layer is used to perform the maximum pooling operation on the preliminary feature map; The deep convolution and residual connection layer introduces multiple convolution blocks and residual connection blocks, performs deep convolution and residual connection on the preliminary feature map after the maximum pooling operation, and obtains the final feature map; The fully connected layer is used to flatten the final feature map; The output layer uses a softmax function to convert the output of the fully connected layer into a category probability distribution.

7. The method for high-definition imaging and intelligent identification of power grid fault points based on drones according to claim 5 is characterized in that: The inspection image data is also pre-processed before being input into the high-precision image processing unit, including: Perform image enhancement and denoising processing on the inspection image data.

8. The method for high-definition imaging and intelligent identification of power grid fault points based on drones according to claim 5 is characterized in that: The target detection unit adopts the YOLOv5 model, and the YOLOv5 algorithm includes: Backbone network, neck network and detection head, where: The backbone network is used to extract features from the input inspection image data; The neck network is used to perform feature fusion operations on features of different scales output by the backbone network; The detection head is used to perform target detection based on the fused features and output the target’s bounding box coordinates, category probability, and confidence.

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, the method according to any one of claims 5 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 5 to 8 is implemented.

Images