CN113252701A - Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method - Google Patents
Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method Download PDFInfo
- Publication number
- CN113252701A CN113252701A CN202110746924.6A CN202110746924A CN113252701A CN 113252701 A CN113252701 A CN 113252701A CN 202110746924 A CN202110746924 A CN 202110746924A CN 113252701 A CN113252701 A CN 113252701A
- Authority
- CN
- China
- Prior art keywords
- insulator
- picture
- cloud
- data set
- edge
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
Abstract
The invention discloses a power transmission line insulator self-explosion defect detection system and method based on cloud-edge cooperation, and the power transmission line insulator self-explosion defect detection system comprises an unmanned aerial vehicle, an edge end and a cloud end, wherein the edge end is carried on the unmanned aerial vehicle and is in communication connection with the cloud end, a lightweight small neural network without an anchor frame is arranged at the edge end, an unsupervised generation type large neural network is arranged at the cloud end, the unmanned aerial vehicle is used for acquiring an insulator picture in real time to generate an insulator monitoring data set, the edge end detects and positions the insulator monitoring data set to obtain an insulator detection result preliminarily, meanwhile, the insulator monitoring data set generates a normal insulator picture without background information after being preprocessed, then the normal insulator picture without background information is sent to the cloud end, the cloud end accurately detects the normal insulator picture without background information to obtain the defect position of insulator self-explosion, and the detection precision is realized, The network resource utilization rate is high, and the data processing capacity is large.
Description
Technical Field
The invention relates to the technical field of image detection, in particular to a power transmission line insulator self-explosion defect detection system and method based on cloud edge cooperation.
Background
Due to the rapid development of the current social power system, the erection area of power equipment is continuously enlarged, so that the power transmission line is continuously extended, and the method has very important significance for regularly monitoring the state of the power transmission line. The insulator is an important part of the power transmission line and plays a role in mechanical fixation and electrical insulation. However, the insulator is very vulnerable to damage due to long-term exposure to the open air and working in a high-voltage environment, which seriously affects the safe and stable operation of the power system. Therefore, insulator defect detection is one of the important tasks of power transmission line inspection. At present, insulator defect detection generally relies on the human eye to observe, carries out the insulator state observation along the transmission line through dispatching staff, and this kind of artifical mode of patrolling and examining needs a large amount of human input to efficiency is very low. On the other hand, because the transmission lines are widely distributed in complex environments such as mountainous areas, the difficulty in developing the manual inspection mode is greatly improved. The introduction of helicopter inspection alleviates the inefficiency of manual inspection to some extent, but the cost is too high and the flexibility is low to make it impossible to apply on a large scale.
In recent years, the rise of unmanned aerial vehicle platform and deep learning technology has brought new opportunity for power patrol. The unmanned aerial vehicle inspection mode has the characteristics of safety and high efficiency. And the deep learning technology is utilized to identify and detect important electric power components, so that the consumption of human resources is greatly reduced, and the construction of the smart grid is promoted. However, the limited energy of the unmanned aerial vehicle equipment is a key factor for restricting the application of the unmanned aerial vehicle inspection technology. Insulator defect detection is generally divided into two stages, wherein the first stage carries out defect detection positioning on an insulator string, and the second stage carries out defect identification on the basis of insulator detection results. However, the insulator has a large aspect ratio, and a large amount of background information can be detected by using a general target detection network in the first stage, so that subsequent defect identification is interfered. Meanwhile, the deep learning network model with strong robustness excessively depends on a large training data set, and due to the particularity of the insulators, a large amount of data of the insulators, particularly defective insulators, are difficult to obtain, so that certain obstacles are brought to research work.
Disclosure of Invention
In view of this, the technical problem to be solved by the present invention is how to reduce the energy consumption of the unmanned aerial vehicle by deploying the unmanned aerial vehicle on the device in a lightweight manner while ensuring the accuracy, and to solve the problem of overfitting the model due to insufficient defective insulator samples.
In order to achieve the above object, in one aspect, the invention provides a power transmission line insulator based on cloud edge coordination
The edge self-explosion defect detection system comprises an unmanned aerial vehicle, an edge end and a cloud end, wherein the edge end is carried on the unmanned aerial vehicle and is in communication connection with the cloud end, a lightweight small neural network without an anchor frame is arranged at the edge end, an unsupervised generation type large neural network is arranged at the cloud end, the unmanned aerial vehicle is used for collecting insulator pictures in real time and generating an insulator monitoring data set, the edge end detects and positions the insulator monitoring data set based on the lightweight small neural network without the anchor frame to obtain an insulator detection result preliminarily, meanwhile, the insulator monitoring data set generates a normal insulator picture without background information after being preprocessed, and then the normal insulator picture without the background information is sent to the cloud end; and the cloud end accurately detects the normal insulator picture without background information based on the unsupervised generation type large neural network, so that the defect position of the insulator spontaneous explosion is obtained.
Further, the step of generating a normal insulator picture without background information after preprocessing the insulator monitoring data set specifically includes: and the insulator monitoring data set is rotated and cut to obtain a normal insulator picture without background information.
Further, the light-weight small neural network without the anchor frame is a target detection network CenterNet, the target detection network CenterNet selects DLA34 as a backbone network, the unsupervised large-scale neural network is a variation automatic encoder, and the variation automatic encoder only needs to train a normal insulator picture and does not need to label the normal insulator picture.
Further, a rotating boundary frame is obtained by improving regression angle information of the target detection network CenterNet, preprocessing of the insulator monitoring data set is completed according to the rotating boundary frame, and therefore redundancy of background information of the insulator pictures in the insulator monitoring data set is reduced, and normal insulator pictures without background information are obtained.
On the other hand, the invention also provides a power transmission line insulator self-explosion defect detection method based on cloud-edge coordination, the defect detection is carried out by the power transmission line insulator self-explosion defect detection system based on cloud-edge coordination, and the power transmission line insulator self-explosion defect detection method based on cloud-edge coordination comprises the following steps:
s1, training and deploying the lightweight small neural network without the anchor frame and the unsupervised generating large neural network respectively, wherein the trained lightweight small neural network without the anchor frame is deployed at an edge end, and the unsupervised generating large neural network is deployed at a cloud end;
s2, acquiring an insulator picture in real time through an unmanned aerial vehicle, and generating an insulator monitoring data set;
s3, detecting and positioning the insulator monitoring data set by the edge end through a lightweight small neural network without an anchor frame to preliminarily obtain an insulator detection result, and preprocessing the insulator monitoring data set to generate a normal insulator picture without background information;
and S4, accurately detecting the normal insulator picture without background information by the cloud through an unsupervised generation type large neural network, and further obtaining the defect position of the insulator spontaneous explosion.
Further, the step S1 specifically includes the following steps:
s11, collecting and calibrating an insulator monitoring data set for training;
s12, selecting a target detection network CenterNet without an anchor frame as the light-weight small-sized neural network with an edge end, selecting DLA34 with few parameters as a main network by the target detection network CenterNet, improving regression angle information of the target detection network CenterNet to obtain a rotating boundary frame, and finishing preprocessing of an insulator monitoring data set according to the rotating boundary frame to obtain a normal insulator picture without background information; selecting a variational automatic encoder as the unsupervised generation type large neural network at the cloud end, wherein the variational automatic encoder only needs to train a normal insulator picture and does not need to label the normal insulator picture;
s13, training the target detection network CenterNet and the variation automatic encoder, and continuously adjusting respective training weights until the detection accuracy and the recall rate are highest;
and S14, the target detection network CenterNet is deployed on the edge end, and the variational automatic encoder is deployed on the cloud end.
Further, when the insulator monitoring data set for training is calibrated in step S11, a tool capable of performing calibration of the rotating bounding box is selected, an xml file is exported, the insulator monitoring data set is rotated and cut according to the xml file, and a normal insulator picture in the range of the rotating bounding box is obtained and used as the training data of the cloud variational automatic encoder, where the xml file includes 5 pieces of position information of the insulator, which are respectively a horizontal coordinate and a vertical coordinate of a central point, a width and a height of the rotating bounding box, and an angle of the rotating bounding box.
Further, the step S3 specifically includes the following steps:
s31, identifying and positioning insulators by using an anchor frame-free target detection network CenterNet by the edge end to the insulator monitoring data set transmitted by the unmanned aerial vehicle;
s32, if insulators exist in insulator pictures in the insulator monitoring data set, predicting a rotating boundary frame according to the training weight of the target detection network CenterNet in the step S13, cutting the insulator pictures with the insulators according to the rotating boundary frame and rotating the insulator pictures to be horizontal to obtain normal insulator pictures without background information;
and S33, transmitting the normal insulator picture without the background information obtained in the step S32 to a cloud for spontaneous explosion defect detection.
Further, the step S4 specifically includes the following steps:
s41, reconstructing the normal insulator picture without background information transmitted in the step S33 by using the training weight of the variational automatic encoder obtained by training in the step S13 to obtain a reconstruction result picture;
s42, subtracting the reconstruction result picture of the step S41 from the normal insulator picture without background information transmitted to the step S33 to obtain a reconstruction error;
and S43, comparing the preset threshold value with the reconstruction error, judging whether the normal insulator picture without the background information transmitted to the step S33 has the spontaneous explosion defect, judging that a defective insulator exists when the reconstruction error is larger than the preset threshold value, and judging that the normal insulator exists when the reconstruction error is smaller than the preset threshold value.
Further, the step S3 includes the following steps: if no insulator exists in the insulator picture of the insulator monitoring data set, the insulator picture is directly discarded and cannot be transmitted to the cloud.
According to the power transmission line insulator self-explosion defect detection system and method based on cloud-edge cooperation, the edge end is carried on the unmanned aerial vehicle, insulator picture data acquired by the unmanned aerial vehicle can be synchronously acquired, and the real-time performance of data processing is guaranteed; however, the unmanned aerial vehicle cannot provide infinite energy for the edge end, so that the edge end can only perform simple processing tasks with low power consumption. The cloud has stronger computing power, but the real-time requirement cannot be met far away from the inspection site, and meanwhile, if the data obtained by the unmanned aerial vehicle are completely and directly transmitted to the cloud, a large amount of energy loss is caused, so that the detection system and the detection method need to be designed by combining the characteristics and advantages of the cloud and the edge. In the invention, an anchor frame-free target detection frame is combined with a main network with less parameters as a processing algorithm of an edge end to realize the detection and the positioning of the insulator; and a generative variational automatic encoder is deployed at the cloud end, and only normal insulation subdata needs to be trained, and whether the insulator has the self-explosion defect or not is judged by calculating the reconstruction error of an input and output picture. Compared with the prior art, the invention has the beneficial effects that: the advantages of the cloud end and the edge end are fully exerted, the utilization rate of network resources is improved, and the intellectualization of power routing inspection is realized; in combination with the structural characteristics of the insulator, an efficient lightweight small neural network without an anchor frame is designed and deployed at the edge end; an unsupervised generation type large neural network is adopted at the cloud, and the problem that data of defective insulators are limited is solved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate an embodiment of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
fig. 1 is a schematic structural diagram of a power transmission line insulator spontaneous explosion defect detection system based on cloud-edge coordination according to an embodiment of the present invention;
fig. 2 is a flowchart of a power transmission line insulator spontaneous explosion defect detection method based on cloud-edge coordination according to an embodiment of the present invention;
fig. 3 is a flowchart of a method for detecting a self-explosion defect of an insulator of a power transmission line based on cloud-edge coordination according to another embodiment of the present invention;
FIG. 4 is a flow chart of insulator detection and positioning at the edge end according to the present invention;
fig. 5 is a flowchart of the cloud-based accurate detection according to the present invention.
Detailed Description
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict. The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings.
For a better understanding of the present invention, the following words are specifically explained:
cloud edge cooperation refers to a process that a cloud end and an edge end cooperate to process picture data obtained by an unmanned aerial vehicle;
centret, a central point network;
DLA34, Deep Layer Aggregation, 34 is the number of network convolution layers
Fig. 1 is a schematic structural diagram of a power transmission line insulator spontaneous explosion defect detection system based on cloud edge coordination according to an embodiment of the present invention. The detection system comprises an unmanned aerial vehicle, an edge end and a cloud end, wherein the edge end is carried on the unmanned aerial vehicle and is in communication connection with the cloud end, preferably, the edge end and the cloud end carry out data transmission through a 4g network, a lightweight small neural network without an anchor frame is arranged at the edge end, an unsupervised generation type large neural network is arranged at the cloud end, the unmanned aerial vehicle is used for acquiring an insulator picture in real time and generating an insulator monitoring data set, the edge end detects and positions the insulator monitoring data set based on the lightweight small neural network without the anchor frame to obtain an insulator detection result preliminarily, and meanwhile, the insulator monitoring data set generates a normal insulator picture without background information after being preprocessed and then sends the normal insulator picture without background information to the cloud end; the cloud side accurately detects a normal insulator picture without background information based on an unsupervised generation type large neural network, and then the defect position of the insulator spontaneous explosion is obtained. It should be noted that the normal insulator picture refers in particular to a picture containing insulators in the insulator picture acquired by the unmanned aerial vehicle, and the insulator monitoring data set completes the preprocessing process through rotation and cutting to obtain the normal insulator picture without background information.
Through the arrangement, the advantages of the cloud end and the edge end are fully exerted, the utilization rate of network resources is improved, and the intellectualization of power routing inspection is realized; and the insulator is combined with the structural characteristics of the insulator, a lightweight small neural network without an anchor frame is designed and deployed at the side end, and an unsupervised generation type large neural network is deployed at the cloud end, so that the problem that the data of the defective insulator is limited is solved, and the insulator has the advantages of accurate defect detection, high network resource utilization rate and large data processing capacity.
As a preferred embodiment of the present invention, in order to reduce the calculation burden of the edge end and reduce the power consumption of the unmanned aerial vehicle, the light-weight small neural network without an anchor frame is a target detection network centrnet, a rotation boundary frame is obtained by improving regression angle information of the target detection network centrnet, and preprocessing of an insulator monitoring data set is completed according to the rotation boundary frame, so that redundancy of background information of an insulator picture in the insulator monitoring data set is reduced, a normal insulator picture without background information is obtained, and the target detection network centrnet selects DLA34 as a backbone network, thereby reducing the calculation amount; the unsupervised generation type large neural network is a variational automatic encoder, and the variational automatic encoder only needs to train a normal insulator picture and does not need to label the insulator picture. The target detection network CenterNet has no anchor frame, does not need post-treatment such as non-maximum value inhibition and the like, has smaller frame and less network parameters compared with a target detection network frame based on an anchor frame class, and is suitable for being deployed at an edge end; the traditional automatic encoder comprises an encoder and a decoder, and a potential space between the encoder and the decoder is described in a single-value form, while the variational automatic encoder describes the characteristics of the potential space in a probability distribution form, so that the generalization performance of the neural network model is stronger.
As an embodiment of the present invention, as shown in fig. 2, the present invention further provides a cloud-edge-based cooperative detection method for detecting a self-explosion defect of an insulator of a power transmission line, where the detection method is performed by the cloud-edge-based cooperative detection system for detecting a self-explosion defect of an insulator of a power transmission line, and includes the following steps:
s1, training and deploying the lightweight small neural network without the anchor frame and the unsupervised generating large neural network respectively, wherein the trained lightweight small neural network without the anchor frame is deployed at an edge end, and the unsupervised generating large neural network is deployed at a cloud end;
s2, acquiring an insulator picture in real time through an unmanned aerial vehicle, and generating an insulator monitoring data set; the method comprises the following steps: the method comprises the steps that an unmanned aerial vehicle obtains real-time data, the unmanned aerial vehicle carrying a camera is used for shooting insulators on a power transmission line in real time, the shot insulator pictures are transmitted to an edge end, and an insulator monitoring data set is generated;
s3, detecting and positioning the insulator monitoring data set by the edge end through a lightweight small neural network without an anchor frame to preliminarily obtain an insulator detection result, and preprocessing the insulator monitoring data set to generate a normal insulator picture without background information;
and S4, accurately detecting the normal insulator picture without background information by the cloud through an unsupervised generation type large neural network, and further obtaining the defect position of the insulator spontaneous explosion.
After the neural networks trained by the unmanned aerial vehicle are respectively deployed at the cloud end and the edge end, firstly, identifying and detecting pictures acquired by the unmanned aerial vehicle in real time, preliminarily obtaining a positioning detection result of the insulator, and transmitting the detected insulator pictures to the cloud end, but the insulator pictures are preprocessed (rotated and cut) before being sent to the neural network of the cloud end; and finally, the defect position is accurately detected by utilizing the neural network of the cloud, the respective advantages of the cloud and the edge are fully exerted while the structural characteristics of the insulator are combined, and the method has the advantages of accurate defect detection, high network resource utilization rate and large data processing capacity.
Referring to fig. 3, fig. 3 is a flowchart of a flow chart of a power transmission line insulator self-explosion defect detection method based on cloud-edge coordination according to another embodiment of the present invention.
A power transmission line insulator self-explosion defect detection method based on cloud edge coordination performs defect detection through the power transmission line insulator self-explosion defect detection system based on cloud edge coordination, and the detection method comprises the following steps:
s11, collecting and calibrating an insulator monitoring data set for training;
it should be noted that, in the step, the insulator monitoring data set is collected and calibrated, only the position of the insulator in the insulator picture of the insulator monitoring data set needs to be calibrated, the defect position does not need to be calibrated, only a normal insulator picture (i.e., an insulator picture including the insulator) needs to be collected, and no abnormal insulator picture (i.e., an insulator picture not including the insulator) needs to be collected. The position of the insulator is calibrated by software, and in order to enable the background information contained in the rotating boundary frame surrounding the insulator to be less, the RoLabelImg marking tool is selected, and the tool can rotate the boundary frame to perform calibration, so that the insulator can be better surrounded. The specific labeling process is as follows: firstly, opening an insulator picture folder to be marked, and starting marking; secondly, pressing an E key of the keyboard to create a rotating rectangular frame; thirdly, rotating and adjusting the direction of the rectangular frame through Z, X, C and V keys of the keyboard; fourthly, the created rotating rectangular frame is marked as an insulator label; and fifthly, storing the standard picture and exporting an xml file. The xml file generated by calibration contains the position and type information of the insulators, the position information of the insulators is 5, the position information is respectively represented by horizontal and vertical coordinates (x, y), width and height (w, h) of the center point of a rectangular frame and a rotation angle theta, the type information is represented by a preset number or character, and only one type is the insulator. The xml file obtained in the step is used as a training sample of the edge terminal neural network. Meanwhile, the obtained xml file can obtain position information (namely a central point horizontal coordinate and a central point vertical coordinate, the width and the height of the rotating boundary frame and the angle of the rotating boundary frame) of the insulator in the aerial image of the unmanned aerial vehicle, the insulator picture is cut and rotated to be horizontal according to the position information, a normal insulator picture containing a small amount of background information can be obtained, and the cut normal insulator picture is directly used as a training sample of the cloud neural network.
S12, selecting a target detection network CenterNet without an anchor frame as the light-weight small-sized neural network with an edge end, selecting DLA34 with few parameters as a main network by the target detection network CenterNet, improving regression angle information of the target detection network CenterNet to obtain a rotating boundary frame, and finishing preprocessing of an insulator monitoring data set according to the rotating boundary frame to obtain a normal insulator picture without background information; selecting a variational automatic encoder as the unsupervised generation type large neural network at the cloud end, wherein the variational automatic encoder only needs to train a normal insulator picture and does not need to label the normal insulator picture;
s13, training the target detection network CenterNet and the variation automatic encoder, and continuously adjusting respective training weights until the detection accuracy and the recall rate are highest;
and S14, the target detection network CenterNet is deployed on the edge end, and the variational automatic encoder is deployed on the cloud end.
S2, acquiring an insulator picture in real time through an unmanned aerial vehicle, and generating an insulator monitoring data set;
s3, detecting and positioning the insulator monitoring data set by the edge end through a lightweight small neural network without an anchor frame to preliminarily obtain an insulator detection result, and preprocessing the insulator monitoring data set to generate a normal insulator picture without background information;
meanwhile, as shown in fig. 4, the steps are specifically divided into the following steps:
s31, the edge end carries out insulator identification and positioning on an insulator monitoring data set transmitted by the unmanned aerial vehicle by using a target detection network CenterNet without an anchor frame, whether an insulator exists in a picture is predicted, if the insulator is not detected, the picture is discarded and cannot be transmitted to the cloud end; otherwise, executing step S32;
s32, if insulators exist in insulator images in the insulator monitoring data set, predicting a rotating boundary frame according to the training weight of the target detection network CenterNet in the step S13, surrounding the insulator images with the insulators according to the outline of the insulators on the basis of the rotating boundary frame, cutting the insulator images with the insulators according to the predicted position information of the rotating boundary frame, and rotating the insulator images to the horizontal direction to obtain normal insulator images without background information;
the above process can greatly reduce the influence of background information redundancy on subsequent defect detection.
And S33, transmitting the normal insulator picture without the background information obtained in the step S32 to a cloud for spontaneous explosion defect detection.
And S4, accurately detecting the normal insulator picture without background information by the cloud through an unsupervised generation type large neural network, and further obtaining the defect position of the insulator spontaneous explosion.
In a further technical solution, as shown in fig. 5, step S4 specifically includes the following steps:
s41, reconstructing the normal insulator picture without background information transmitted in the step S33 by using the training weight of the variational automatic encoder obtained by training in the step S13 to obtain a reconstruction result picture;
it should be noted that the above-mentioned variational automatic encoder only needs to train a normal insulator picture, and the normal insulator picture is reconstructed into a still normal insulator picture after passing through the variational automatic encoder, and will have the characteristics of a normal insulator. And the training weight obtained after training only learns the characteristics and data distribution of the normal insulator. Further, in the prediction, if a defective insulator picture is input, since the previously obtained training weight is obtained only by training a normal insulator, only the output reconstructed insulator is a normal insulator. Specifically, if an insulator with a self-explosion defect is input, the insulator sheet of the output insulator is filled at the position of the self-explosion defect, and the reconstruction results of other positions are nearly consistent with the input picture, so that a normal insulator picture is obtained. Therefore, the input defective insulator has a larger error than the output thereof, so that the insulator can be judged to have defects, and the defect position can be obtained according to the picture difference.
S42, subtracting the reconstruction result picture of the step S41 from the normal insulator picture without background information transmitted to the step S33 to obtain a reconstruction error; specifically, the reconstruction error of a normal insulator is small, and the reconstruction error of an insulator with a self-explosion defect is large;
and S43, comparing the preset threshold value with the reconstruction error, judging whether the normal insulator picture without the background information transmitted to the step S33 has the spontaneous explosion defect, judging as the insulator with the spontaneous explosion defect when the reconstruction error is larger than the preset threshold value, and judging as the normal insulator when the reconstruction error is smaller than the preset threshold value. Note that the insulator picture judged as having the self-explosion defect is discarded.
Through the process, the insulator detection accuracy is guaranteed, meanwhile, the light-weight small neural network without the anchor frame is arranged on the edge end carried by the unmanned aerial vehicle, so that the energy loss of the unmanned aerial vehicle is reduced, meanwhile, the unsupervised generation type large neural network is arranged at the cloud end, and the problem of model overfitting caused by the fact that a defective insulator sample is insufficient is solved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (10)
1. The electric transmission line insulator self-explosion defect detection system based on cloud-edge cooperation is characterized by comprising an unmanned aerial vehicle, an edge end and a cloud end, wherein the edge end is carried on the unmanned aerial vehicle and is in communication connection with the cloud end, a lightweight small neural network without an anchor frame is arranged at the edge end, an unsupervised generation type large neural network is arranged at the cloud end, the unmanned aerial vehicle is used for acquiring an insulator picture in real time and generating an insulator monitoring data set, the edge end detects and positions the insulator monitoring data set based on the lightweight small neural network without the anchor frame to obtain an insulator detection result preliminarily, and meanwhile, the insulator monitoring data set is preprocessed to generate a normal insulator picture without background information and then sends the normal insulator picture without the background information to the cloud end; and the cloud end accurately detects the normal insulator picture without background information based on the unsupervised generation type large neural network, so that the defect position of the insulator spontaneous explosion is obtained.
2. The cloud-edge-collaboration-based power transmission line insulator self-explosion defect detection system as claimed in claim 1, wherein the generation of the normal insulator picture without background information after preprocessing the insulator monitoring data set specifically comprises: and the insulator monitoring data set is rotated and cut to obtain a normal insulator picture without background information.
3. The cloud-edge-collaboration-based transmission line insulator self-explosion defect detection system as claimed in claim 2, wherein the light-weight small neural network without the anchor frame is a target detection network CenterNet, the target detection network CenterNet selects DLA34 as a backbone network, the unsupervised large generation type neural network is a variation automatic encoder, and the variation automatic encoder only needs to train a normal insulator picture and does not need to label the normal insulator picture.
4. The cloud-edge-collaboration-based power transmission line insulator self-explosion defect detection system as claimed in claim 3, wherein a rotating boundary frame is obtained by improving regression angle information of a target detection network CenterNet, and preprocessing of an insulator monitoring data set is completed according to the rotating boundary frame, so that redundancy of background information of an insulator picture in the insulator monitoring data set is reduced, and a normal insulator picture without background information is obtained.
5. The cloud-edge-collaboration-based self-explosion defect detection method for the insulator of the power transmission line is characterized in that the defect detection is carried out by the cloud-edge-collaboration-based self-explosion defect detection system for the insulator of the power transmission line, which is disclosed by claim 1, and the cloud-edge-collaboration-based self-explosion defect detection method for the insulator of the power transmission line comprises the following steps:
s1, training and deploying the lightweight small neural network without the anchor frame and the unsupervised generating large neural network respectively, wherein the trained lightweight small neural network without the anchor frame is deployed at an edge end, and the unsupervised generating large neural network is deployed at a cloud end;
s2, acquiring an insulator picture in real time through an unmanned aerial vehicle, and generating an insulator monitoring data set;
s3, detecting and positioning the insulator monitoring data set by the edge end through a lightweight small neural network without an anchor frame to preliminarily obtain an insulator detection result, and preprocessing the insulator monitoring data set to generate a normal insulator picture without background information;
and S4, accurately detecting the normal insulator picture without background information by the cloud through an unsupervised generation type large neural network, and further obtaining the defect position of the insulator spontaneous explosion.
6. The cloud-edge-collaboration-based power transmission line insulator self-explosion defect detection method according to claim 5, wherein the step S1 specifically comprises the following steps:
s11, collecting and calibrating an insulator monitoring data set for training;
s12, selecting a target detection network CenterNet without an anchor frame as the light-weight small-sized neural network with an edge end, selecting DLA34 with few parameters as a main network by the target detection network CenterNet, improving regression angle information of the target detection network CenterNet to obtain a rotating boundary frame, and finishing preprocessing of an insulator monitoring data set according to the rotating boundary frame to obtain a normal insulator picture without background information; selecting a variational automatic encoder as the unsupervised generation type large neural network at the cloud end, wherein the variational automatic encoder only needs to train a normal insulator picture and does not need to label the normal insulator picture;
s13, training the target detection network CenterNet and the variation automatic encoder, and continuously adjusting respective training weights until the detection accuracy and the recall rate are highest;
and S14, the target detection network CenterNet is deployed on the edge end, and the variational automatic encoder is deployed on the cloud end.
7. The cloud-edge-collaboration-based detection method for the self-explosion defects of the insulators of the power transmission line, according to the claim 6, when the insulator monitoring data set for training is calibrated in the step S11, a tool capable of calibrating a rotating boundary frame is selected, an xml file is exported, the insulator monitoring data set is rotated and cut according to the xml file, a normal insulator picture in the range of the rotating boundary frame is obtained and serves as training data of the cloud-side variation automatic encoder, and the xml file contains 5 pieces of position information of the insulators, namely horizontal and vertical coordinates of a central point, width and height of the rotating boundary frame and an angle of the rotating boundary frame.
8. The cloud-edge-collaboration-based power transmission line insulator self-explosion defect detection method according to claim 7, wherein the step S3 specifically comprises the following steps:
s31, identifying and positioning insulators by using an anchor frame-free target detection network CenterNet by the edge end to the insulator monitoring data set transmitted by the unmanned aerial vehicle;
s32, if insulators exist in insulator images in the insulator monitoring data set, predicting a rotating boundary frame according to the training weight of the target detection network CenterNet in the step S13, surrounding the insulator images with the insulators according to the outline of the insulators on the basis of the rotating boundary frame, cutting the insulator images with the insulators according to the predicted position information of the rotating boundary frame, and rotating the insulator images to the horizontal direction to obtain normal insulator images without background information;
and S33, transmitting the normal insulator picture without the background information obtained in the step S32 to a cloud for spontaneous explosion defect detection.
9. The cloud-edge-collaboration-based power transmission line insulator self-explosion defect detection method according to claim 8, wherein the step S4 specifically comprises the following steps:
s41, reconstructing the normal insulator picture without background information transmitted in the step S33 by using the training weight of the variational automatic encoder obtained by training in the step S13 to obtain a reconstruction result picture;
s42, subtracting the reconstruction result picture of the step S41 from the normal insulator picture without background information transmitted to the step S33 to obtain a reconstruction error;
and S43, comparing the preset threshold value with the reconstruction error, judging whether the normal insulator picture without the background information transmitted to the step S33 has the spontaneous explosion defect, judging that a defective insulator exists when the reconstruction error is larger than the preset threshold value, and judging that the normal insulator exists when the reconstruction error is smaller than the preset threshold value.
10. The cloud-edge-collaboration-based power transmission line insulator self-explosion defect detection method according to claim 8, wherein the step S3 further comprises the following steps: if no insulator exists in the insulator picture of the insulator monitoring data set, the insulator picture is directly discarded and cannot be transmitted to the cloud.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110746924.6A CN113252701B (en) | 2021-07-02 | 2021-07-02 | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110746924.6A CN113252701B (en) | 2021-07-02 | 2021-07-02 | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113252701A true CN113252701A (en) | 2021-08-13 |
| CN113252701B CN113252701B (en) | 2021-10-26 |
Family
ID=77190436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110746924.6A Active CN113252701B (en) | 2021-07-02 | 2021-07-02 | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113252701B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113627386A (en) * | 2021-08-30 | 2021-11-09 | 山东新一代信息产业技术研究院有限公司 | Visual video abnormity detection method |
| CN114445411A (en) * | 2022-04-11 | 2022-05-06 | 广东电网有限责任公司佛山供电局 | Unmanned aerial vehicle line patrol defect identification system and control method |
| CN115035108A (en) * | 2022-08-10 | 2022-09-09 | 四川中电启明星信息技术有限公司 | Insulator defect detection method based on deep learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108648169A (en) * | 2018-03-20 | 2018-10-12 | 中国科学院自动化研究所 | The method and device of high voltage power transmission tower defects of insulator automatic identification |
| CN109934222A (en) * | 2019-03-01 | 2019-06-25 | 长沙理工大学 | A kind of insulator chain self-destruction recognition methods based on transfer learning |
| CN110322437A (en) * | 2019-06-20 | 2019-10-11 | 浙江工业大学 | A kind of fabric defect detection method based on autocoder and BP neural network |
| CN110705374A (en) * | 2019-09-11 | 2020-01-17 | 武汉敏捷数云信息技术有限公司 | Transmission line defect identification method based on improved RetinaNet network |
| CN111626987A (en) * | 2020-04-29 | 2020-09-04 | 南京邮电大学 | Vehicle detection method based on unmanned aerial vehicle and anchor-frame-free network |
| CN112419401A (en) * | 2020-11-23 | 2021-02-26 | 上海交通大学 | Aircraft surface defect detection system based on cloud edge cooperation and deep learning |
| CN112491982A (en) * | 2020-11-13 | 2021-03-12 | 国网天津市电力公司 | Refined sensing method based on cloud edge cooperative power transmission line |
-
2021
- 2021-07-02 CN CN202110746924.6A patent/CN113252701B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108648169A (en) * | 2018-03-20 | 2018-10-12 | 中国科学院自动化研究所 | The method and device of high voltage power transmission tower defects of insulator automatic identification |
| CN109934222A (en) * | 2019-03-01 | 2019-06-25 | 长沙理工大学 | A kind of insulator chain self-destruction recognition methods based on transfer learning |
| CN110322437A (en) * | 2019-06-20 | 2019-10-11 | 浙江工业大学 | A kind of fabric defect detection method based on autocoder and BP neural network |
| CN110705374A (en) * | 2019-09-11 | 2020-01-17 | 武汉敏捷数云信息技术有限公司 | Transmission line defect identification method based on improved RetinaNet network |
| CN111626987A (en) * | 2020-04-29 | 2020-09-04 | 南京邮电大学 | Vehicle detection method based on unmanned aerial vehicle and anchor-frame-free network |
| CN112491982A (en) * | 2020-11-13 | 2021-03-12 | 国网天津市电力公司 | Refined sensing method based on cloud edge cooperative power transmission line |
| CN112419401A (en) * | 2020-11-23 | 2021-02-26 | 上海交通大学 | Aircraft surface defect detection system based on cloud edge cooperation and deep learning |
Non-Patent Citations (2)
| Title |
|---|
| XINYU LIU ET AL: "Box-Point Detector: A Diagnosis Method for Insulator Faults in Power Lines Using Aerial Images and Convolutional Neural Networks", 《IEEE》 * |
| 陈庆等: "航拍绝缘子卷积神经网络检测及自爆识别研究", 《电子测量与仪器学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113627386A (en) * | 2021-08-30 | 2021-11-09 | 山东新一代信息产业技术研究院有限公司 | Visual video abnormity detection method |
| CN114445411A (en) * | 2022-04-11 | 2022-05-06 | 广东电网有限责任公司佛山供电局 | Unmanned aerial vehicle line patrol defect identification system and control method |
| CN115035108A (en) * | 2022-08-10 | 2022-09-09 | 四川中电启明星信息技术有限公司 | Insulator defect detection method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113252701B (en) | 2021-10-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113252701B (en) | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method | |
| CN110033453A (en) | Based on the power transmission and transformation line insulator Aerial Images fault detection method for improving YOLOv3 | |
| CN108648169A (en) | The method and device of high voltage power transmission tower defects of insulator automatic identification | |
| CN112581443A (en) | Light-weight identification method for surface damage of wind driven generator blade | |
| CN115546630A (en) | Construction site extraction method and system based on remote sensing image characteristic target detection | |
| CN113205039B (en) | Power equipment fault image recognition disaster investigation system and method based on multiple DCNN networks | |
| CN110910440B (en) | Power transmission line length determination method and system based on power image data | |
| CN111209832A (en) | Auxiliary obstacle avoidance training method, equipment and medium for transformer substation inspection robot | |
| CN115909093A (en) | Power equipment fault detection method based on unmanned aerial vehicle inspection and infrared image semantic segmentation | |
| CN113255590A (en) | Defect detection model training method, defect detection method, device and system | |
| CN116188402A (en) | Insulator defect identification method based on improved SSD algorithm | |
| CN115546664A (en) | Cascaded network-based insulator self-explosion detection method and system | |
| CN112614130A (en) | Unmanned aerial vehicle power transmission line insulator fault detection method based on 5G transmission and YOLOv3 | |
| CN114037645A (en) | Coating defect detection method and device for pole piece, electronic equipment and readable medium | |
| CN114648708A (en) | State detection method and device of photovoltaic module, storage medium and electronic equipment | |
| CN114119528A (en) | Defect detection method and device for distribution line insulator | |
| CN113705763B (en) | Real-time detection method and system for distribution network transformer based on nerve computation stick | |
| CN112861670B (en) | Transmission line hardware detection method and system | |
| CN115100592A (en) | Method and device for identifying hidden danger of external damage of power transmission channel and storage medium | |
| Zan et al. | Defect Identification of Power Line Insulator Based on an Improved yolov4-tiny Algorithm | |
| CN112183191A (en) | Method for detecting bird nest in power transmission line based on countermeasure training | |
| Panigrahy et al. | Research on Transmission Line Insulator Defects Detection using YOLOv7 | |
| CN116909320B (en) | Electric power collaborative inspection strategy analysis method based on ant colony algorithm | |
| Liu et al. | Intelligent Detection Method for Avian Nests on Transmission Lines Based on Edge Computing and YOLOv5 | |
| Sun et al. | Threat Target Detection System for UAV Equipment on the End Side of Transmission Lines |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |