CN114332659A - Power transmission line defect inspection method and device based on lightweight model issuing - Google Patents
Power transmission line defect inspection method and device based on lightweight model issuing Download PDFInfo
- Publication number
- CN114332659A CN114332659A CN202210227821.3A CN202210227821A CN114332659A CN 114332659 A CN114332659 A CN 114332659A CN 202210227821 A CN202210227821 A CN 202210227821A CN 114332659 A CN114332659 A CN 114332659A
- Authority
- CN
- China
- Prior art keywords
- target device
- detection model
- image
- transmission line
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The application relates to a method and a device for inspecting defects of a power transmission line issued based on a lightweight model. The method comprises the following steps: compressing the target device detection model to obtain and send the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating the edge equipment to identify a current unmanned aerial vehicle inspection image, and indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain and upload a target device image; and if the target device image is received, processing the target device image by adopting a defect detection model to obtain a defect inspection result of the power transmission line. By adopting the method, the defect detection of the target device can be realized in the inspection process of the power transmission line, and the identification accuracy is high.
Description
Technical Field
The application relates to the technical field of power inspection, in particular to a method and a device for inspecting defects of a power transmission line based on lightweight model issuing.
Background
Power transmission is a crucial link in a power system in China, and numerous power devices and related defects of a power transmission line are main contents of power transmission inspection. However, most key equipment of the power transmission line is deployed outdoors, and has the characteristics of wide distribution, remote areas, harsh geographic environment and the like, which brings difficulty to routing inspection. With the great development of the manufacturing industry and the internet of things, the unmanned aerial vehicle is widely applied to inspection of power transmission lines by power grid enterprises. Compare in the manual work and patrol and examine, unmanned aerial vehicle patrols and examines and has reduced the demand to the manpower by a wide margin, patrols and examines the demand that the in-process exists in time handling a large amount of data of taking photo by plane effectively.
However, the existing transmission line defect inspection method has the problem of low defect identification accuracy of a transmission line target device.
Disclosure of Invention
Therefore, in order to solve the technical problems, a power transmission line defect inspection method and a power transmission line defect inspection device issued based on a lightweight model are needed, wherein the power transmission line defect inspection method and the power transmission line defect inspection device can improve the defect identification accuracy of a power transmission line target device.
In a first aspect, the application provides a transmission line defect inspection method issued based on a lightweight model. The method comprises the following steps:
compressing the target device detection model to obtain and send the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating the edge equipment to identify the current unmanned aerial vehicle inspection image so as to obtain the type and the position of the target device; the type and the position of the target device are used for indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain and upload an image of the target device;
if the target device image is received, processing the target device image by adopting a defect detection model to obtain a defect inspection result of the power transmission line; the defect detection model is obtained by training a second single-stage target detection model based on a target device training set.
In one embodiment, the target device detection model and the defect detection model are both models including a backbone network, an encoder and a decoder which are connected in sequence; the trunk network comprises a residual error neural network and a feature map output by a fifth convolution layer of the feature pyramid, and the residual error neural network is connected with the feature map; the encoder comprises a projection layer and a residual block which are connected; the projection layer is connected with the feature map; the decoder comprises a classification branch and a regression branch; the regression branch comprises an implicit prediction object used for outputting the position of the target device; the classification branch outputs the type of the target device by merging with the implicit prediction object.
In one embodiment, the target device inspection model comprises a plurality of common convolutional layers; the defect detection model includes a number of packed convolutional layers.
In one embodiment, the target device detection model adopts an equilibrium matching strategy to adjust the sample proportion of a training set of the target device for training; the defect detection model employs a cross entropy loss function to adjust the sample proportions of the training set of target devices for training.
In one embodiment, the step of compressing and issuing the target device detection model comprises: dynamically pruning the target device detection model based on the average channel significance of each layer of channels of the target device detection model to obtain a compressed target device detection model; and issuing the compressed target device detection model through a 5G network by adopting an HTTP (hyper text transport protocol).
In one embodiment, the target device comprises a glass insulator, a vibration damper and a grading ring; the target device training set comprises an xml label file in a VOC format; the method further comprises the following steps: and if the target device image is received, optimizing a defect detection model based on the target device image.
In a second aspect, the application further provides a transmission line defect inspection device based on the lightweight model issuing. The device comprises:
the compression issuing module is used for compressing the target device detection model to obtain and issue the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating the edge equipment to identify the current unmanned aerial vehicle inspection image so as to obtain the type and the position of the target device; the type and the position of the target device are used for indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain and upload an image of the target device;
the result output module is used for processing the target device image by adopting a defect detection model if the target device image is received, and obtaining a defect inspection result of the power transmission line; the defect detection model is obtained by training a second single-stage target detection model based on a target device training set.
In a third aspect, the application further provides a cloud device. The cloud device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of the method.
In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method.
In a fifth aspect, the present application further provides a computer program product. The computer program product comprises a computer program which, when being executed by a processor, carries out the steps of the above-mentioned method.
The transmission line defect inspection method, the transmission line defect inspection device, the computer equipment, the storage medium and the computer program product are issued based on the lightweight model, and the compressed target device detection model is obtained and issued by compressing the target device detection model; if the target device image is received, the defect detection model is adopted to process the target device image to obtain a power transmission line defect inspection result, defect detection on the target device can be achieved in the power transmission line inspection process, and the identification accuracy is high.
Drawings
Fig. 1 is a schematic flow chart of a transmission line defect inspection method issued based on a lightweight model in one embodiment;
FIG. 2 is an application environment diagram of the transmission line defect inspection method issued based on the lightweight model in one embodiment;
FIG. 3 is a schematic diagram of the structure of a target device inspection model and a defect inspection model in one embodiment;
FIG. 4 is a schematic flow chart of the transmission line defect inspection step in one embodiment;
FIG. 5 is a block diagram of a power transmission line defect inspection device issued based on a lightweight model in one embodiment;
fig. 6 is an internal structure diagram of the cloud device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
Automatic analysis is carried out around unmanned aerial vehicle image data patrolling, and a machine vision analysis method is utilized to improve the target device and defect identification effect, so that the method becomes one of research and application hotspots of power grid operation and maintenance. With the high-speed development of information communication technologies such as edge computing, 5G communication and cloud-edge cooperation, the unmanned aerial vehicle inspection image identification cooperative optimization of the power grid activates new development power. The automatic analysis method of the machine patrol image mainly comprises a cloud side and a side (edge side or terminal side) by means of calculation carrier division, the image identification method based on the cloud side relies on strong computing power of a GPU (Graphics Processing Unit) server cluster, good identification effect can be achieved on defects of target devices, the identification effect depends on image quality acquired by an unmanned aerial vehicle, and if the unmanned aerial vehicle does not focus on the target devices of a power grid (such as a power transmission line) and shoots too many backgrounds, the defects that the target devices cannot be identified by the cloud side can be caused. The edge side-based image identification method has the main advantages that the real-time performance is good, the defect of a target device can be identified in a short time, but the identification effect of computing resources is limited to be poor, and a plurality of defects are often missed to be identified.
In an embodiment, as shown in fig. 1, a power transmission line defect inspection method issued based on a lightweight model is provided, and by taking an example that the method is applied to a cloud side in fig. 2 (for example, a server, which may be implemented by an independent server or a server cluster composed of a plurality of servers), it can be understood that the method is also applied to a system including a cloud side and an edge side (including an edge side and a terminal side), and is implemented by interaction between the cloud side and the edge side. The edge side can include edge device and unmanned aerial vehicle, and edge device can carry on unmanned aerial vehicle, also can regard as two independent equipment respectively with unmanned aerial vehicle. The method comprises the following steps:
step 110, compressing the target device detection model to obtain and send down the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating the edge equipment to identify the current unmanned aerial vehicle inspection image so as to obtain the type and the position of the target device; the type and the position of the target device are used for indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain and upload an image of the target device;
specifically, the cloud side can label target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line to obtain a target device training set; training a first single-stage target detection model (the single-stage target detection model is a model which can realize target detection only by extracting features once) based on a target device training set to obtain a target device detection model; compressing the target device detection model to obtain and send the compressed target device detection model; by compressing the target device detection model and issuing the compressed target device detection model, the lightweight model can be issued so as to meet the requirements of a receiving end on small model size, low calculation complexity, low battery power consumption and flexible issuing, updating and deploying.
Further, the edge device refers to a device that is deployed on the edge side and can be used for performing edge computing, and the edge device may receive a compressed target device detection model sent by a cloud in an HTTP Protocol (Hyper Text Transfer Protocol) through a 5G (5 th Generation Mobile Communication Technology, fifth Generation Mobile Communication Technology) network; the edge equipment can also identify the current unmanned aerial vehicle inspection image based on the compressed target device detection model to obtain the type and the position of the target device; the edge device can also instruct the unmanned aerial vehicle to adjust the posture based on the type and the position of the target device so as to acquire the image of the target device and upload the image through a 5G network.
In some examples, the edge device receives a compressed target device detection model software package sent by the cloud to a specified position in the system through a 5G network by using an HTTP (hyper text transport protocol), and updates a target device detection model in the edge device; the edge equipment starts a target device detection module, receives a video acquired by a camera of the unmanned aerial vehicle through a local area network, identifies a target device in the video, sends information (including the name and the position of the target device in a picture) of the target device to the unmanned aerial vehicle, and the unmanned aerial vehicle adjusts the posture according to the position of the target device and shoots the target device again by focusing through the camera; and the edge equipment sends the re-shot target device image to the cloud side for defect identification through a 5G network in an HTTP (hyper text transport protocol).
Step 120, if the target device image is received, processing the target device image by adopting a defect detection model to obtain a defect inspection result of the power transmission line; the defect detection model is obtained by training a second single-stage target detection model based on a target device training set.
Specifically, a second single-stage target detection model is trained based on a target device training set to obtain a defect detection model; and if the target device image sent by the side end side is received, processing the target device image by adopting a defect detection model to obtain a defect inspection result of the power transmission line.
In some examples, if the target device image sent by the edge side is received, the target device detection model and the defect detection model may be further iteratively optimized based on the target device image, and are respectively used for next target device detection and defect detection.
In the power transmission line defect inspection method issued based on the lightweight model, the compressed target device detection model is obtained and issued by compressing the target device detection model; if the target device image is received, the defect detection model is adopted to process the target device image to obtain a power transmission line defect inspection result, defect detection on the target device can be achieved in the power transmission line inspection process, and the identification accuracy is high.
In one embodiment, as shown in fig. 3, the target device detection model and the defect detection model are both models including a backbone network, an encoder and a decoder connected in sequence; the trunk network comprises a residual error neural network and a feature map output by a fifth convolution layer of the feature pyramid, and the residual error neural network is connected with the feature map; the encoder comprises a projection layer and a residual block which are connected; the projection layer is connected with the feature map; the decoder comprises a classification branch and a regression branch; the regression branch comprises an implicit prediction object used for outputting the position of the target device; the classification branch outputs the type of the target device by merging with the implicit prediction object.
Specifically, a residual neural network (ResNet) of the backbone network may employ ResNet 50; the Feature Pyramid (FPN) can extract features and output Feature maps of different scales, and extract a Feature map output by the fifth convolutional layer of the Feature Pyramid (i.e., C5/DC5 Feature map); the encoder may comprise a run-up encoder; the expansion encoder comprises a projection layer (Projector) and four Residual Blocks (Residual Blocks) with different convolution kernel expansion rates which are connected in sequence; the decoder comprises two parallel head branches, namely a classification branch and a regression branch, wherein the number of convolutional layers of the classification branch is different from that of the regression branch; and the value of the implicit prediction object of the regression branch is used for multiplying the output of the classification branch to obtain the type of the target device.
In some examples, ResNet50 accesses the C5 feature map of the feature pyramid, with a number of channels of 2048 and a downsampling rate of 32; the projection layer firstly adopts a 1 × 1 convolution layer to reduce the channel dimension, and then adopts a 3 × 3 convolution layer to refine the context semantic information; the output of the projection layer is accessed into four continuously superposed residual blocks with different convolution kernel expansion rates to generate output characteristics with a plurality of receptive fields and cover the scales of all objects; the regression branch includes four convolution layers plus a BN (Batch Normalization) layer and a ReLU (Rectified Linear Unit) layer; the classification branch comprises two convolution layers, and an implicit prediction object (object) is added to each anchor of the regression branch; if the IOU (interaction over unit) of a bbox (bounding box) and a group route (reference standard or labeled data) is larger than that of other bboxes, its object score (which is a binary value for predicting foreground and background) is 1, that is, the best box; if a bbox is not a best box, but its IOU with a ground truth is greater than a threshold (e.g., 0.5), then its object score is 0; the final classification confidence is obtained by multiplying the output of the classification branch with the value of the implicitly predicted object (object score).
In one embodiment, the target device inspection model comprises a plurality of common convolutional layers; the defect detection model includes a number of packed convolutional layers.
Specifically, a target device detection model adopts a common convolution mode; the defect detection model adopts a group convolution (group convolution) mode, so that the calculation amount is further reduced.
In one embodiment, the target device detection model adopts an equilibrium matching strategy to adjust the sample proportion of a training set of the target device for training; the defect detection model employs a cross entropy loss function to adjust the sample proportions of the training set of target devices for training.
Specifically, in order to solve the problem of imbalance of positive and negative samples, a Uniform Matching (Uniform Matching) strategy is introduced into a target device detection model, namely for each group channel frame, only the k nearest anchors are used as positive samples; the defect detection model adopts a cross entropy (cross entropy) loss function to realize the adjustment of the positive and negative proportion of the input classifier sample, and simultaneously realizes the reasonable distinction of the difficult sample and the easy sample in the sample.
In one embodiment, the step of compressing and issuing the target device detection model comprises: dynamically pruning the target device detection model based on the average channel significance of each layer of channels of the target device detection model to obtain a compressed target device detection model; and issuing the compressed target device detection model through a 5G network by adopting an HTTP (hyper text transport protocol).
Specifically, the weight of the target device detection model is dynamically pruned, different thresholds are set for different layers in the target device detection model, that is, the average channel significance of each layer of channel is set as the threshold of the corresponding layer, only the channel larger than the threshold is calculated during reasoning, and the rest channels are skipped, so that the calculated amount is reduced, and the pruned target device detection model is sent to the edge device through a 5G network by an HTTP protocol.
In some examples, taking an L-th layer of the target device detection model as an example, taking N samples, calculating an average channel significance of M channels of the L-th layer, where the channel significance is a value of the channel when the N samples are inferred, and taking the average channel significance as a threshold.
In one embodiment, the target device comprises a glass insulator, a vibration damper and a grading ring; the target device training set comprises an xml label file in a VOC format; the method further comprises the following steps: and if the target device image is received, optimizing a defect detection model based on the target device image.
Specifically, target devices (such as glass insulators, vibration dampers, grading rings and the like) in the image are marked by using a minimum external rectangular frame, and the marked file is an xml file in a VOC format; and if the target device image is received, training and optimizing a current defect detection model based on the target device image so as to improve the accuracy of the next defect detection.
In some examples, if the target device image is received, the target device detection model may be trained and optimized based on the target device image and reissued again, so as to improve the accuracy of the next target device detection of the edge device.
In some examples, as shown in fig. 4, the cloud side trains a target device detection model, compresses the target device detection model to obtain a compressed target device detection model, and issues a model software package; the side end receives and updates a target device detection model software package so as to identify a target device for the video uploaded by the unmanned aerial vehicle and send an identification result to the unmanned aerial vehicle; the unmanned aerial vehicle adjusts the posture according to the recognition result to shoot again and transmits back the video; the unmanned aerial vehicle sends the target device image shot again to the cloud side; and identifying the power grid defects by adopting a defect detection model on the basis of the target device image on the cloud side, and further iteratively optimizing the defect detection model.
It should be understood that, although the steps in the flowcharts related to the embodiments as described above are sequentially displayed as indicated by arrows, the steps are not necessarily performed sequentially as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a part of the steps in the flowcharts related to the embodiments described above may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the execution order of the steps or stages is not necessarily sequential, but may be rotated or alternated with other steps or at least a part of the steps or stages in other steps.
Based on the same inventive concept, the embodiment of the application also provides a transmission line defect inspection device issued based on the lightweight model, which is used for realizing the transmission line defect inspection method issued based on the lightweight model. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the method, so that specific limitations in one or more embodiments of the power transmission line defect inspection device issued based on the lightweight model can be referred to the limitations on the power transmission line defect inspection method issued based on the lightweight model, and details are not repeated herein.
In an embodiment, as shown in fig. 5, the present application further provides a transmission line defect inspection device issued based on a lightweight model. The device comprises:
a compression issuing module 510, configured to compress the target device detection model, obtain a compressed target device detection model, and issue the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating the edge equipment to identify the current unmanned aerial vehicle inspection image so as to obtain the type and the position of the target device; the type and the position of the target device are used for indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain and upload an image of the target device;
a result output module 520, configured to, if the target device image is received, process the target device image by using a defect detection model to obtain a defect inspection result of the power transmission line; the defect detection model is obtained by training a second single-stage target detection model based on a target device training set.
In one embodiment, the compression issuing module 510 is further configured to perform dynamic pruning on the target device detection model based on the average channel significance of each layer of channel of the target device detection model, and obtain a compressed target device detection model; and issuing the compressed target device detection model through a 5G network by adopting an HTTP (hyper text transport protocol).
In one embodiment, the result output module 520 is further configured to optimize the defect detection model based on the target device image if the target device image is received.
All or part of each module in the transmission line defect inspection device issued based on the lightweight model can be realized through software, hardware and combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, the application further provides the cloud device. The cloud device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of the method.
In one embodiment, a cloud device is provided, and the cloud device may be a server, and an internal structure diagram of the cloud device may be as shown in fig. 6. The cloud device comprises a processor, a memory and a network interface which are connected through a system bus. The processor of the cloud device is used for providing computing and control capability. The memory of the cloud device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the cloud equipment is used for storing the transmission line defect inspection data. The network interface of the cloud device is used for being connected and communicated with an external terminal through a network. The computer program is executed by the processor to realize the transmission line defect inspection method issued based on the lightweight model.
Those skilled in the art will appreciate that the structure shown in fig. 6 is only a block diagram of a part of the structure related to the present application, and does not constitute a limitation to the cloud device to which the present application is applied, and a specific cloud device may include more or less components than those shown in the figure, or combine some components, or have a different arrangement of components.
In one embodiment, the present application also provides a computer-readable storage medium. The computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned method.
In one embodiment, the present application also provides a computer program product. The computer program product comprises a computer program which, when being executed by a processor, carries out the steps of the above-mentioned method.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high-density embedded nonvolatile Memory, resistive Random Access Memory (ReRAM), Magnetic Random Access Memory (MRAM), Ferroelectric Random Access Memory (FRAM), Phase Change Memory (PCM), graphene Memory, and the like. Volatile Memory can include Random Access Memory (RAM), external cache Memory, and the like. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others. The databases referred to in various embodiments provided herein may include at least one of relational and non-relational databases. The non-relational database may include, but is not limited to, a block chain based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic devices, quantum computing based data processing logic devices, etc., without limitation.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.
Claims (10)
1. A transmission line defect inspection method issued based on a lightweight model is characterized by comprising the following steps:
compressing the target device detection model to obtain and send the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating edge equipment to identify a current unmanned aerial vehicle inspection image so as to obtain the type and the position of the target device; the type and the position of the target device are used for indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain an image of the target device and upload the image;
if the target device image is received, processing the target device image by adopting a defect detection model to obtain a defect inspection result of the power transmission line; the defect detection model is obtained by training a second single-stage target detection model based on the target device training set.
2. The method of claim 1, wherein the target device detection model and the defect detection model are both models comprising a backbone network, an encoder, and a decoder connected in sequence;
the trunk network comprises a residual error neural network and a feature map output by a fifth convolution layer of the feature pyramid, and the residual error neural network is connected with the feature map; the encoder comprises a projection layer and a residual block which are connected; the projection layer is connected with the feature map; the decoder comprises a classification branch and a regression branch; the regression branch comprises an implicit prediction object for outputting the position of the target device; the classification branch outputs the type of the target device by merging with the implicit prediction object.
3. The method of claim 2, wherein the target device inspection model comprises a number of common convolutional layers; the defect detection model includes a number of packed convolutional layers.
4. The method of claim 3, wherein the target device detection model employs an equilibrium matching strategy to adjust a sample proportion of the target device training set used for training;
the defect detection model employs a cross entropy loss function to adjust the sample proportions of the training set of target devices for training.
5. The method of claim 4, wherein the step of compressing and issuing the target device detection model comprises:
based on the average channel significance of each layer of channel of the target device detection model, dynamically pruning the target device detection model to obtain the compressed target device detection model;
and issuing the compressed target device detection model through a 5G network by adopting an HTTP protocol.
6. The method of any one of claims 1 to 5, wherein the target device comprises a glass insulator, a vibration damper, and a grading ring; the target device training set comprises an xml label file in a VOC format;
the method further comprises the following steps:
and if the target device image is received, optimizing the defect detection model based on the target device image.
7. The utility model provides a transmission line defect inspection device based on lightweight model is issued which characterized in that, the device includes:
the compression issuing module is used for compressing the target device detection model to obtain and issue the compressed target device detection model; the target device detection model is obtained by training a first single-stage target detection model based on a target device training set; the target device training set is obtained by marking target devices in historical unmanned aerial vehicle routing inspection images of the power transmission line by adopting a minimum external rectangular frame; the compressed target device detection model is used for indicating edge equipment to identify a current unmanned aerial vehicle inspection image so as to obtain the type and the position of the target device; the type and the position of the target device are used for indicating the edge equipment to indicate the unmanned aerial vehicle to adjust the posture so as to obtain an image of the target device and upload the image;
the result output module is used for processing the target device image by adopting a defect detection model if the target device image is received, so as to obtain a defect inspection result of the power transmission line; the defect detection model is obtained by training a second single-stage target detection model based on the target device training set.
8. Cloud device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 6 when executing the computer program.
9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of any one of claims 1 to 6 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210227821.3A CN114332659A (en) | 2022-03-09 | 2022-03-09 | Power transmission line defect inspection method and device based on lightweight model issuing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210227821.3A CN114332659A (en) | 2022-03-09 | 2022-03-09 | Power transmission line defect inspection method and device based on lightweight model issuing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114332659A true CN114332659A (en) | 2022-04-12 |
Family
ID=81033869
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210227821.3A Pending CN114332659A (en) | 2022-03-09 | 2022-03-09 | Power transmission line defect inspection method and device based on lightweight model issuing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114332659A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116343132A (en) * | 2023-05-26 | 2023-06-27 | 南方电网数字电网研究院有限公司 | Complex scene power equipment defect identification method and device and computer equipment |
| CN116883391A (en) * | 2023-09-05 | 2023-10-13 | 中国科学技术大学 | Two-stage distribution line defect detection method based on multi-scale sliding window |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015131462A1 (en) * | 2014-03-07 | 2015-09-11 | 国家电网公司 | Centralized monitoring system and monitoring method for unmanned aerial vehicle to patrol power transmission line |
| CN113888514A (en) * | 2021-10-09 | 2022-01-04 | 中国南方电网有限责任公司超高压输电公司广州局 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
| CN113962973A (en) * | 2021-10-27 | 2022-01-21 | 云南电网有限责任公司昆明供电局 | Power transmission line unmanned aerial vehicle intelligent inspection system and method based on satellite technology |
| CN113963333A (en) * | 2021-11-15 | 2022-01-21 | 北京联合大学 | Traffic sign board detection method based on improved YOLOF model |
-
2022
- 2022-03-09 CN CN202210227821.3A patent/CN114332659A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015131462A1 (en) * | 2014-03-07 | 2015-09-11 | 国家电网公司 | Centralized monitoring system and monitoring method for unmanned aerial vehicle to patrol power transmission line |
| CN113888514A (en) * | 2021-10-09 | 2022-01-04 | 中国南方电网有限责任公司超高压输电公司广州局 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
| CN113962973A (en) * | 2021-10-27 | 2022-01-21 | 云南电网有限责任公司昆明供电局 | Power transmission line unmanned aerial vehicle intelligent inspection system and method based on satellite technology |
| CN113963333A (en) * | 2021-11-15 | 2022-01-21 | 北京联合大学 | Traffic sign board detection method based on improved YOLOF model |
Non-Patent Citations (1)
| Title |
|---|
| 危双丰 等: "基于航拍影像的输电线路绝缘子识别及缺陷检测", 《工程勘察》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116343132A (en) * | 2023-05-26 | 2023-06-27 | 南方电网数字电网研究院有限公司 | Complex scene power equipment defect identification method and device and computer equipment |
| CN116343132B (en) * | 2023-05-26 | 2023-09-15 | 南方电网数字电网研究院有限公司 | Complex scene power equipment defect identification method and device and computer equipment |
| CN116883391A (en) * | 2023-09-05 | 2023-10-13 | 中国科学技术大学 | Two-stage distribution line defect detection method based on multi-scale sliding window |
| CN116883391B (en) * | 2023-09-05 | 2023-12-19 | 中国科学技术大学 | Two-stage distribution line defect detection method based on multi-scale sliding window |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019109771A1 (en) | Power artificial-intelligence visual-analysis system on basis of multi-core heterogeneous parallel computing | |
| EP3683572B1 (en) | Battery detection method and device | |
| CN114362367B (en) | Cloud-edge-cooperation-oriented power transmission line monitoring system and method, and cloud-edge-cooperation-oriented power transmission line identification system and method | |
| CN114332659A (en) | Power transmission line defect inspection method and device based on lightweight model issuing | |
| CN109828845B (en) | Insulator thermal imaging real-time diagnosis system based on edge calculation | |
| CN110689539B (en) | Workpiece surface defect detection method based on deep learning | |
| CN110503135B (en) | Deep learning model compression method and system for power equipment edge side recognition | |
| CN114943689A (en) | Method for detecting components of steel cold-rolling annealing furnace based on semi-supervised learning | |
| CN113312957A (en) | off-Shift identification method, device, equipment and storage medium based on video image | |
| CN111369430A (en) | Mobile terminal portrait intelligent background replacement method based on mobile deep learning engine | |
| CN115147383A (en) | Insulator state rapid detection method based on lightweight YOLOv5 model | |
| CN114445336A (en) | Distribution equipment defect detection method and device, computer equipment and storage medium | |
| CN115223009A (en) | Small target detection method and device based on improved YOLOv5 | |
| CN113408630A (en) | Transformer substation indicator lamp state identification method | |
| CN112399177B (en) | Video coding method, device, computer equipment and storage medium | |
| CN113901928A (en) | Target detection method based on dynamic super-resolution, and power transmission line component detection method and system | |
| CN114640669A (en) | Edge calculation method and device | |
| CN116823793A (en) | Device defect detection method, device, electronic device and readable storage medium | |
| CN117218102A (en) | Insulator defect detection method and system based on improved YOLOv5 | |
| CN116580324A (en) | Yolov 5-based unmanned aerial vehicle ground target detection method | |
| CN116524309A (en) | Distributed scrap steel detection method and system based on hierarchical fine granularity fusion federal learning | |
| CN115439738A (en) | Underwater target detection method based on self-supervision cooperative reconstruction | |
| CN112804446B (en) | Big data processing method and device based on cloud platform big data | |
| CN111726592B (en) | Method and apparatus for obtaining architecture of image signal processor | |
| CN113469254A (en) | Target detection method and system based on target detection model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220412 |
|
| RJ01 | Rejection of invention patent application after publication |