Power transmission line insulator detection method and system based on YOLOv10

Abstract

The invention discloses a YOLOv10-based power transmission line insulator detection method and system, and the method comprises the steps: constructing an insulator data set; then, an insulator detection model based on the lightweight YOLOv10n network is constructed, that is, a ghost convolutional layer GhostConv based on GhostBottleneck module stacking is used for replacing an original common convolutional network of a Backbone network; a feature enhancement module ConcatBiFPN is used for replacing an original connection layer network of the Neck network, and an attention module ContextAggregation is arranged behind a feature extraction module C2FCIB of the Neck network; and finally, the insulator data set is sent to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model for insulator detection, so that the algorithm operation efficiency and accuracy are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing and target detection technology in the field of power transmission line equipment detection technology, and specifically to a transmission line insulator detection method and system based on a lightweight improved YOLOv10 algorithm. Background Art

[0002] High-voltage transmission lines have long been subjected to the test of complex natural environments, such as lightning strikes and acid rain erosion, making them prone to failure. Therefore, it is crucial to conduct regular inspections of the lines. In recent years, drones have become the preferred tool for detecting transmission lines internationally. As a core component of transmission lines, regular inspections of insulators are particularly necessary. Traditional insulator inspections are carried out manually. Although this method is simple, since most transmission lines are distributed in sparsely populated areas, manual inspections are difficult to cover large areas, are inefficient, and are somewhat dangerous. At this stage, the commonly used inspection method is to use drones to take inspection images at specific locations, and then ground staff manually identify and analyze the aerial inspection images for faults. This method not only causes delays in fault detection, but also brings a lot of workload to power inspection personnel.

[0003] With the continuous development of deep learning, computer vision and other technologies, it has become possible to detect insulator defects in aerial images using deep learning and other technologies combined with existing hardware equipment. However, due to the limited computing and storage capabilities of drones, complex detection models, and large amounts of calculations, the current drone inspection method is inefficient and time-consuming, and cannot monitor the insulator status in real time.

[0004] In order to accurately detect the status of insulators in real time, it is necessary to consider reducing the size and complexity of the detection model to meet the requirements of real-time detection on drone equipment, while ensuring the detection accuracy of insulators and improving the efficiency and accuracy of drone inspections of transmission line insulators. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that the edge devices in the task of UAV inspection of transmission line insulators should not be equipped with too heavy models, which leads to the need to process data offline, poor information real-time performance and low efficiency, and thus provide a transmission line insulator detection method and system based on YOLOv10. Among them, the technical solution of the present invention optimizes the structure of the Backbone network layer, that is, the original ordinary convolutional network is replaced with a more lightweight GhostConv network; the connection layer network in the Neck network layer is replaced with the Concat_BiFPN module, and the accuracy and efficiency are better balanced through top-down and bottom-up multi-scale feature fusion; and the ContextAggregation module is additionally added to the Neck network layer to improve attention, achieve better convergence speed, and improve recognition efficiency.

[0006] To this end, the present invention provides the following technical solutions:

[0007] A transmission line insulator detection method based on YOLOv10, comprising the following steps:

[0008] S1: Obtain real-life images of insulators of power transmission and transformation lines and annotate them to obtain an insulator dataset;

[0009] Among them, the insulator position in the real shot image is marked;

[0010] S2: construct an insulator detection model based on a lightweight YOLOv10n network, wherein the lightweight YOLOv10n network includes a Backbone network, a Neck network, and a Head network connected in sequence;

[0011] Among them, the ghost convolution layer GhostConv based on the GhostBottleneck module stack is used to replace the original ordinary convolution network of the Backbone network; and the feature enhancement module Concat_BiFPN is used to replace the original connection layer network of the Neck network, and the attention module ContextAggregation is set after the feature extraction module C2FCIB of the Neck network;

[0012] S3: sending the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model;

[0013] Among them, the trained insulator detection model is used for the detection of insulators in power transmission and transformation lines.

[0014] Optionally, the ghost convolution layer GhostConv is provided with 1 standard convolution module and 9 GhostBottleneck modules in sequence;

[0015] The GhostBottleneck module consists of two stacked Ghost modules, one of which is used as an expansion layer to increase the number of channels, and the second Ghost module is used to reduce the number of channels and match the shortcut path; the output of the previous Ghost module is used as the input of the next Ghost module, and then the shortcut is used to connect the input and output of the two Ghost modules.

[0016] Optionally, the Backbone network is provided with two ghost convolutional layers GhostConv;

[0017] The Backbone network includes a standard convolution layer Conv_1, a ghost convolution layer GhostConv_1, a feature extraction module C2f_1, a ghost convolution layer GhostConv_2, a feature extraction module C2f_2, a downsampling module SCDown_1, a feature extraction module C2f_3, a downsampling module SCDown_2, a feature extraction module C2f_4, an attention module SPPF and an attention module PSA, which are connected in sequence;

[0018] The feature extraction module C2f_2, the downsampling module SCDown_2, and the attention module PSA respectively output feature maps P3, P4, and P5, and input them into the Neck network.

[0019] The Concat_BiFPN module is composed of a BiFPN network and is an existing module, so it will not be described in detail.

[0020] Optionally, after the feature graphs P3, P4, and P5 are input into the Neck network, the feature graphs P3, P4, and P5 are processed in the Neck network as follows:

[0021] The feature map P5 is fused with the feature map P4 in the feature enhancement module Concat_BiFPN_1 of the Neck network after passing through the upsampling layer Upsample_1 of the Neck network;

[0022] The feature map output by the feature enhancement module Concat_BiFPN_1 is extracted by the feature extraction module C2f_5 and passed through the upsampling layer Upsample_2, and then fused with the feature map P3 in the feature enhancement module Concat_BiFPN_2;

[0023] The feature map output by the feature enhancement module Concat_BiFPN_2 is subjected to feature extraction by the feature extraction module C2f_6 and then passed through the standard convolution layer Conv_2, and then fused with the feature map P5 output by the feature extraction module C2f_2 in the feature enhancement module Concat_BiFPN_3;

[0024] The feature map output by the feature enhancement module Concat_BiFPN_3 is subjected to feature extraction by the feature extraction module C2f_7, and then is fused with the feature map P5 in the fully connected layer Concat after passing through the downsampling module SCDown_3;

[0025] The feature map output by the fully connected layer Concat is subjected to feature extraction by the feature extraction module C2FCIB and the attention module ContextAggregation;

[0026] Among them, the feature extraction module C2f_6, the feature extraction module C2f_7, and the feature maps output by the attention module ContextAggregation are all input into the Head network. The Head network correspondingly sets three target detection modules Detect, and the results of the three target detection modules Detect are convolutionally fused to obtain the insulator detection result.

[0027] Optionally, the scale of the standard convolution layer Conv of the Backbone network is P1 / 2, the scale of the ghost convolution layer GhostConv is P2 / 4, P3 / 8, P4 / 16, P5 / 32, and the scale of the feature map is represented by Pn (n=1, 2, 3, 4, 5) respectively.

[0028] Optionally, the operation of the attention module ContextAggregation is expressed as:

[0029] Y=((αA(X)+βA)V)W2+X

[0030] Where Y is the output of the attention module ContextAggregation, X is the input vector of the attention module ContextAggregation, α and β are learnable parameters, A(X) is dynamically generated from X, A is the affinity matrix, V is the transformation of X obtained by linear projection, and W2 is a learnable parameter, i.e., weight.

[0031] In addition, the present invention also provides a detection system based on the above detection method, comprising:

[0032] The insulator dataset construction module is used to obtain real-shot images of insulators of power transmission and transformation lines and annotate the images to obtain the insulator dataset;

[0033] Among them, the insulator position in the real shot image is marked;

[0034] A detection model building module, used to build an insulator detection model based on a lightweight YOLOv10n network, wherein the lightweight YOLOv10n network includes a Backbone network, a Neck network, and a Head network connected in sequence;

[0035] Among them, the ghost convolution layer GhostConv based on the GhostBottleneck module stack is used to replace the original ordinary convolution network of the Backbone network; and the feature enhancement module Concat_BiFPN is used to replace the original connection layer network of the Neck network, and the attention module ContextAggregation is set after the feature extraction module C2FCIB of the Neck network;

[0036] A model training module, used for sending the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model;

[0037] Among them, the trained insulator detection model is used for the detection of insulators in power transmission and transformation lines.

[0038] The present invention also provides a computer device, comprising at least:

[0039] one or more processors;

[0040] a memory storing one or more computer programs;

[0041] The processor calls the computer program to implement:

[0042] A method for detecting insulators of a transmission line based on YOLOv10.

[0043] The present invention also provides a computer storage medium storing a computer program, wherein the computer program is called by a processor to implement:

[0044] A method for detecting insulators of a transmission line based on YOLOv10.

[0045] Beneficial Effects

[0046] The deep learning model adopted by the transmission line insulator detection method provided by the present invention is composed of a Backbone network layer, a Neck network layer, and a Head network layer. Considering the limited computing resources of embedded devices, the structure of the Backbone network layer is optimized, and the original ordinary convolutional network is replaced with a more lightweight GhostConv network, so that it occupies less computing resources, further ensures the real-time nature of information, and improves the efficiency of UAV inspection of transmission line insulators; in addition, in order to ensure lightweight while improving the accuracy of target detection, the connection layer network in the Neck network layer is replaced with the Concat_BiFPN module, and the accuracy and efficiency are better balanced through top-down and bottom-up multi-scale feature fusion; the ContextAggregation module is also added to the Neck network layer to improve attention, achieve better convergence speed, and improve recognition efficiency.

[0047] The essential difference from the existing YOLO application and insulator detection field is that the technical solution of the present invention innovatively balances the accuracy and efficiency selection problem faced by lightweight, and has a high accuracy while taking into account efficiency, reducing the size and complexity of the detection model. In addition, it has good robustness against the interference of complex backgrounds on insulator images, can effectively locate insulator targets in inspection images, and provide a basis for subsequent defect detection. At the same time, it can screen the images obtained during inspections, reduce the burden of manual screening, improve inspection efficiency, and reduce the workload of inspectors, thus safeguarding the normal operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 This is a flow chart of the power transmission line insulator detection method of the present invention;

[0050] Figure 2 This is a Backbone network structure diagram in the detection model of the present invention;

[0051] Figure 3 It is a structural diagram of the Neck network and the Head network in the detection model of the present invention;

[0052] Figure 4 This is a network structure diagram of the Ghost module in the ghost convolution layer GhostConv in the detection model of the present invention;

[0053] Figure 5 It is a network structure diagram of GhostBottleneck in the detection model of the present invention;

[0054] Figure 6 This is the network structure diagram of the ghost convolution layer GhostConv in the detection model of the present invention. DETAILED DESCRIPTION

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] An embodiment of the present invention provides a transmission line insulator detection method based on YOLOv10, comprising the following steps:

[0057] Step 1: Obtain real-life images of insulators of power transmission and transformation lines and annotate them to obtain an insulator dataset. That is, a training set and a test set of insulators are constructed.

[0058] In this embodiment, 1500 images of insulators taken by drones are selected, the image resolution is adjusted to 1152×864, the images are labeled using the LabelImg tool, the corresponding XML files are obtained, and an insulator dataset is established; 900 labeled insulator images are selected as the original training set, and the remaining 600 insulator images are used as the test set.

[0059] Preferably, the training set is enhanced to 2,400 images by using geometric transformation, color transformation, contrast and brightness adjustment, and simulation of complex weather, wherein the geometric transformation methods include but are not limited to rotation, translation, cropping, and mirroring; and the simulation of complex weather includes but is not limited to adding rain, snow, and fog.

[0060] It should be understood that the enhanced operation is a preferred mode of the embodiment of the present invention, and in other feasible embodiments, selective operation can be performed according to actual needs.

[0061] Step 2: Construct an insulator detection model based on a lightweight YOLOv10n network. The present invention uses YOLOv10n as the basic framework to improve the lightweight structure. The lightweight YOLOv10n network includes a Backbone network, a Neck network, and a Head network connected in sequence.

[0062] like Figure 2As shown, the Backbone network includes a standard convolution layer Conv_1, a ghost convolution layer GhostConv_1, a feature extraction module C2f_1, a ghost convolution layer GhostConv_2, a feature extraction module C2f_2, a downsampling module SCDown_1, a feature extraction module C2f_3, a downsampling module SCDown_2, a feature extraction module C2f_4, an attention module SPPF, and an attention module PSA, which are connected in sequence. Based on the above network structure, the feature extraction module C2f_2, the downsampling module SCDown_2, and the attention module PSA correspond to the output feature map P3, the feature map P4, and the feature map P5, respectively, and are input into the Neck network. Among them, in this embodiment, the scale of the standard convolution layer Conv is P1 / 2, and the scale of the ghost convolution layer GhostConv is P2 / 4, P3 / 8, P4 / 16, and P5 / 32.

[0063] From the above network architecture, it can be seen that in order to achieve network lightweighting, the technical solution of the present invention will use the ghost convolution layer GhostConv based on the GhostBottleneck module stack to replace the original ordinary convolution network of the Backbone network. Figure 6 As shown in , the ghost convolution layer GhostConv is equipped with 1 standard convolution module and 9 GhostBottleneck modules in sequence. Figure 5 As shown, the GhostBottleneck module consists of two stacked Ghost modules (such as Figure 4 As shown in the figure, one Ghost module is used as an expansion layer to increase the number of channels, and the second Ghost module is used to reduce the number of channels and match the shortcut path; the output of the previous Ghost module is used as the input of the next Ghost module, and the shortcut is used to connect the input and output of the two Ghost modules. This structure setting greatly reduces the number of network parameters and calculations while ensuring that the size of the output feature map remains unchanged.

[0064] In addition to optimizing the Backbone network, the present invention also optimizes the Neck network as follows: using the feature enhancement module Concat_BiFPN to replace the original connection layer network of the Neck network, and setting the attention module ContextAggregation after the feature extraction module C2FCIB of the Neck network.

[0065] Among them, after the feature maps P3, P4, and P5 are input into the Neck network, the processing process of the feature maps P3, P4, and P5 in the Neck network is as follows:

[0066] like Figure 3As shown, after the feature map P5 passes through the upsampling layer Upsample_1 of the Neck network, it is fused with the feature map P4 in the feature enhancement module Concat_BiFPN_1 of the Neck network; the feature map output by the feature enhancement module Concat_BiFPN_1 is extracted by the feature extraction module C2f_5 and passes through the upsampling layer Upsample_2, and then is fused with the feature map P3 in the feature enhancement module Concat_BiFPN_2; the feature map output by the feature enhancement module Concat_BiFPN_2 is extracted by the feature extraction module C2f_6 and then passes through the standard convolution layer Conv_2, and then is fused with the feature map output by the feature extraction module C2f_2 in the feature enhancement module Concat_BiFPN_3; the feature map output by the feature enhancement module Concat_BiFPN_3 is extracted by the feature extraction module C2f_7, and then passes through the downsampling module SCDown_3, and is fused with the feature map P5 in the fully connected layer Concat; the feature map output by the fully connected layer Concat is subjected to feature extraction by the feature extraction module C2FCIB and the attention module ContextAggregation;

[0067] Among them, the feature maps output by the feature extraction module C2f_6, the feature extraction module C2f_7, and the attention module ContextAggregation are all input into the Head network. The Head network correspondingly sets three target detection modules Detect, and the results of the three target detection modules Detect are convolutionally fused to obtain the insulator detection results.

[0068] The Concat_BiFPN module of this embodiment is composed of a BiFPN network. On the basis of the FPN network, learnable weights are introduced to learn the importance of different input features, and top-down and bottom-up multi-scale feature fusion are repeatedly applied, which has a better accuracy and efficiency trade-off. Since the network architecture of this module is existing, it will not be elaborated in detail.

[0069] The ContextAggregation module in this embodiment is an additional attention mechanism module. It is a general building block for multi-head context aggregation, which can utilize long-range interactions like Transformers while still utilizing the inductive bias of local convolution operations to achieve faster convergence. The principle is as follows:

[0070] Y=((αA(X)+βA)V)W2+X (1)

[0071] Among them, Y is the output, X is the input vector, α and β are learnable parameters, A(X) is dynamically generated from X, A is the affinity matrix, V is the transformation of X obtained by linear projection, and W2 is also a learnable parameter, that is, the weight.

[0072] The Head network of this embodiment includes a target detection module Detect_1, a target detection module Detect_2 and a target detection module Detect_3; the target detection module Detect_1 performs target detection on the features extracted by the feature extraction module C2f_6 in the Neck network, and outputs target 1; the target detection module Detect_2 performs target detection on the features extracted by the feature extraction module C2f_7, and outputs target 2; the target detection module Detect_3 performs target detection on the features extracted by the ContextAggregation module, and outputs target Detect_3; finally, target 1, target 2 and target 3 are convolutionally fused to output the final insulator target.

[0073] In this embodiment, the input feature of the insulator detection model is 640×640×3, the Backbone network layer is used to extract insulator image features, and the sizes of the extracted image features are 160×160×64, 80×80×128, 40×40×256, 20×20×512, and 10×10×1024 respectively; the Neck network layer fuses the features of the three scales of 40×40, 80×80, and 160×160, the three inputs of the feature fusion network layer are 40×40×256, 80×80×128, and 10×10×1024 respectively, and the three outputs of the feature fusion network layer are 40×40×256, 80×80×128, and 10×10×1024 respectively; the input of the target detection layer is connected to the output of the feature fusion network layer, and the target detection layer predicts the insulator images of the three scales of 160×160, 80×80, and 40×40 respectively.

[0074] It should be understood that the above image size is an example of this embodiment, and in other feasible embodiments, adaptive adjustments are made according to actual application requirements.

[0075] Step 3: Send the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model; wherein the trained insulator detection model is used for the detection of insulators of power transmission and transformation lines.

[0076] In summary, the embodiment of the present invention proposes a deep learning model for transmission line insulator detection. The model is constructed based on the lightweight optimized YOLOv10 algorithm, which greatly reduces the size and complexity of the network, occupies less computing resources, and has a certain improvement in accuracy. The YOLOv10 network architecture covers three layers of Backbone, Neck and Head. This model shows strong robustness to the interference of insulator images in complex backgrounds, and can accurately locate insulator targets in inspection images, laying a solid foundation for subsequent defect detection. In addition, the model also has an image screening function, automatically filters non-critical images, significantly reduces the workload of manual screening, and indicates a wide range of application potential. The detection method of the present invention achieves efficient operation under low computing resource consumption by streamlining the model scale and complexity, and can quickly and accurately identify and locate insulators from complex backgrounds, which not only improves the efficiency of inspection operations, but also effectively reduces the labor intensity of inspection personnel, and provides strong support for the stable and efficient operation of the power system.

[0077] Embodiment 2:

[0078] An embodiment of the present invention further provides a detection system based on the above-mentioned insulator detection method, comprising: an insulator data set construction module, a detection model construction module and a model training module.

[0079] Among them, the insulator dataset construction module is used to obtain the real-shot images of insulators of power transmission and transformation lines and perform image annotation to obtain the insulator dataset; that is, to annotate the positions of insulators in the real-shot images.

[0080] The detection model construction module is used to construct an insulator detection model based on a lightweight YOLOv10n network, wherein the lightweight YOLOv10n network includes a Backbone network, a Neck network and a Head network connected in sequence.

[0081] Among them, the ghost convolution layer GhostConv based on the GhostBottleneck module stack is used to replace the original ordinary convolution network of the Backbone network; and the feature enhancement module Concat_BiFPN is used to replace the original connection layer network of the Neck network, and the attention module ContextAggregation is set after the feature extraction module C2FCIB of the Neck network.

[0082] The model training module is used to send the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model.

[0083] In some embodiments, a detection module is also included for performing insulator detection using a trained insulator detection model.

[0084] It should be understood that the specific implementation process of each module please refer to the above method content, the present invention will not be repeated here, and the division of the above functional modules is only for example illustration. In some embodiments, some functional modules can be merged, and some functional modules can be split. Each functional module can be implemented in software or hardware or a combination of software and hardware. Among them, the hardware and software equipment includes but is not limited to general-purpose computer equipment, programmable gate arrays, digital signal processors, microprocessors and their corresponding programming or burning software.

[0085] Embodiment 3:

[0086] An embodiment of the present invention further provides a computer device, comprising at least: one or more processors; and a memory storing one or more computer programs;

[0087] Among them, the processor calls the computer program to implement: the steps of a transmission line insulator detection method based on YOLOv10.

[0088] That is, specific implementation:

[0089] Step 1: Obtain real-life images of insulators of power transmission and transformation lines and annotate them to obtain an insulator dataset.

[0090] Step 2: Build an insulator detection model based on the lightweight YOLOv10n network.

[0091] Step 3: Send the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model; wherein the trained insulator detection model is used for the detection of insulators of power transmission and transformation lines.

[0092] For the specific implementation process of each step, please refer to the description of the above method.

[0093] It should be understood that in the embodiments of the present invention, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the type of device.

[0094] Embodiment 4:

[0095] An embodiment of the present invention provides a computer storage medium storing a computer program, wherein the computer program is called by a processor to implement: steps of a transmission line insulator detection method based on YOLOv10.

[0096] That is, specific implementation:

[0097] Step 1: Obtain real-life images of insulators of power transmission and transformation lines and annotate them to obtain an insulator dataset.

[0098] Step 2: Build an insulator detection model based on the lightweight YOLOv10n network.

[0099] Step 3: Send the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model; wherein the trained insulator detection model is used for the detection of insulators of power transmission and transformation lines.

[0100] For the specific implementation process of each step, please refer to the description of the above method.

[0101] The readable storage medium is a computer-readable storage medium, which may be an internal storage unit of the software and hardware device described in any of the aforementioned embodiments, such as a hard disk or memory of a controller. The readable storage medium may also be an external storage device of the controller, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the controller. Furthermore, the readable storage medium may also include both an internal storage unit of the controller and an external storage device. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0102] Based on this understanding, the technical solution of the present invention, in essence or in other words, the part that contributes to the prior art, or all or 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 enabling 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 invention. The aforementioned readable storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., various media that can store program codes.

[0103] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is a device for realizing the function specified in one flow or multiple flows of the flow chart and / or one or multiple blocks of the block diagram with reference to the instructions executed by the processor according to the method, device (system) and computer program product of the embodiment of the present application. These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which realizes the function specified in one flow or multiple flows of the flow chart and / or one or multiple blocks of the block diagram. These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0104] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A transmission line insulator detection method based on YOLOv10, characterized in that: The following steps are involved: S1: Obtain real-life images of insulators of power transmission and transformation lines and annotate them to obtain an insulator dataset; Among them, the insulator position in the real shot image is marked; S2: construct an insulator detection model based on a lightweight YOLOv10n network, wherein the lightweight YOLOv10n network includes a Backbone network, a Neck network, and a Head network connected in sequence; Among them, the ghost convolution layer GhostConv based on the GhostBottleneck module stack is used to replace the original ordinary convolution network of the Backbone network; and the feature enhancement module Concat_BiFPN is used to replace the original connection layer network of the Neck network, and the attention module ContextAggregation is set after the feature extraction module C2FCIB of the Neck network; S3: sending the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model; Among them, the trained insulator detection model is used for the detection of insulators in power transmission and transformation lines.

2. The method according to claim 1, characterized in that: The ghost convolution layer GhostConv is provided with 1 standard convolution module and 9 GhostBottleneck modules in sequence; The GhostBottleneck module consists of two stacked Ghost modules, one of which is used as an expansion layer to increase the number of channels, and the second Ghost module is used to reduce the number of channels and match the shortcut path; the output of the previous Ghost module is used as the input of the next Ghost module, and then the shortcut is used to connect the input and output of the two Ghost modules.

3. The method according to claim 2, characterized in that: The Backbone network is provided with two ghost convolutional layers GhostConv; The Backbone network includes a standard convolution layer Conv_1, a ghost convolution layer GhostConv_1, a feature extraction module C2f_1, a ghost convolution layer GhostConv_2, a feature extraction module C2f_2, a downsampling module SCDown_1, a feature extraction module C2f_3, a downsampling module SCDown_2, a feature extraction module C2f_4, an attention module SPPF and an attention module PSA, which are connected in sequence; The feature extraction module C2f_2, the downsampling module SCDown_2, and the attention module PSA respectively output feature maps P3, P4, and P5, and input them into the Neck network.

4. The method according to claim 3, characterized in that: After the feature graphs P3, P4 and P5 are input into the Neck network, the processing process of the feature graphs P3, P4 and P5 in the Neck network is as follows: The feature map P5 is fused with the feature map P4 in the feature enhancement module Concat_BiFPN_1 of the Neck network after passing through the upsampling layer Upsample_1 of the Neck network; The feature map output by the feature enhancement module Concat_BiFPN_1 is extracted by the feature extraction module C2f_5 and passed through the upsampling layer Upsample_2, and then fused with the feature map P3 in the feature enhancement module Concat_BiFPN_2; The feature map output by the feature enhancement module Concat_BiFPN_2 is subjected to feature extraction by the feature extraction module C2f_6 and then passed through the standard convolution layer Conv_2, and then fused with the feature map P5 output by the feature extraction module C2f_2 in the feature enhancement module Concat_BiFPN_3; The feature map output by the feature enhancement module Concat_BiFPN_3 is subjected to feature extraction by the feature extraction module C2f_7, and then is fused with the feature map P5 in the fully connected layer Concat after passing through the downsampling module SCDown_3; The feature map output by the fully connected layer Concat is subjected to feature extraction by the feature extraction module C2FCIB and the attention module ContextAggregation; Among them, the feature extraction module C2f_6, the feature extraction module C2f_7, and the feature maps output by the attention module ContextAggregation are all input into the Head network. The Head network correspondingly sets three target detection modules Detect, and the results of the three target detection modules Detect are convolutionally fused to obtain the insulator detection result.

5. The method according to claim 1, characterized in that: The operation of the attention module ContextAggregation is expressed as: Y=((αA(X)+βA)V)W2+X Where Y is the output of the attention module ContextAggregation, X is the input vector of the attention module ContextAggregation, α and β are learnable parameters, A(X) is dynamically generated from X, A is the affinity matrix, V is the transformation of X obtained by linear projection, and W2 is a learnable parameter, i.e., weight.

6. A detection system based on the method according to any one of claims 1 to 5, characterized in that: include: The insulator dataset construction module is used to obtain real-shot images of insulators of power transmission and transformation lines and annotate the images to obtain the insulator dataset; Among them, the insulator position in the real shot image is marked; A detection model building module, used to build an insulator detection model based on a lightweight YOLOv10n network, wherein the lightweight YOLOv10n network includes a Backbone network, a Neck network, and a Head network connected in sequence; Among them, the ghost convolution layer GhostConv based on the GhostBottleneck module stack is used to replace the original ordinary convolution network of the Backbone network; and the feature enhancement module Concat_BiFPN is used to replace the original connection layer network of the Neck network, and the attention module ContextAggregation is set after the feature extraction module C2FCIB of the Neck network; A model training module, used for sending the insulator data set to an insulator detection model based on lightweight YOLOv10n for model training to obtain an insulator detection model; Among them, the trained insulator detection model is used for the detection of insulators in power transmission and transformation lines.

7. A computer device, characterized in that: At least: one or more processors; a memory storing one or more computer programs; The processor calls the computer program to implement: The steps of the method according to any one of claims 1 to 5.

8. A computer storage medium, characterized in that: A computer program is stored, which is called by a processor to implement: The steps of the method according to any one of claims 1 to 5.

Images