CN115439716A - Method, device and equipment for training and detecting aging detection model of distribution network insulator - Google Patents
Method, device and equipment for training and detecting aging detection model of distribution network insulator Download PDFInfo
- Publication number
- CN115439716A CN115439716A CN202211143600.4A CN202211143600A CN115439716A CN 115439716 A CN115439716 A CN 115439716A CN 202211143600 A CN202211143600 A CN 202211143600A CN 115439716 A CN115439716 A CN 115439716A
- Authority
- CN
- China
- Prior art keywords
- distribution network
- insulator
- loss function
- function value
- style
- 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
- 239000012212 insulator Substances 0.000 title claims abstract description 296
- 238000001514 detection method Methods 0.000 title claims abstract description 276
- 238000012549 training Methods 0.000 title claims abstract description 258
- 238000000034 method Methods 0.000 title claims abstract description 68
- 230000032683 aging Effects 0.000 title claims abstract description 58
- 230000005012 migration Effects 0.000 claims abstract description 209
- 238000013508 migration Methods 0.000 claims abstract description 209
- 230000005540 biological transmission Effects 0.000 claims abstract description 94
- 230000007704 transition Effects 0.000 claims abstract description 18
- 230000006870 function Effects 0.000 claims description 152
- 238000000605 extraction Methods 0.000 claims description 29
- 238000010586 diagram Methods 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 17
- 230000004927 fusion Effects 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 3
- 230000002431 foraging effect Effects 0.000 claims 3
- 238000002372 labelling Methods 0.000 abstract description 13
- 238000004891 communication Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 230000007547 defect Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000003287 optical effect Effects 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000009413 insulation Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 239000002253 acid Substances 0.000 description 1
- 238000003916 acid precipitation Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010030 laminating Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000003595 mist Substances 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 229920002379 silicone rubber Polymers 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G06T3/04—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Abstract
The invention discloses a method, a device and equipment for training and detecting an aging detection model of a distribution network insulator, wherein the training method comprises the following steps: obtaining an initial training sample set; inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples; determining a target domain training sample set according to the style transition image sample, and determining a sample transition loss function value according to the initial training sample set and the target domain training sample set; inputting the target domain training sample set into a source domain detection model to obtain a model training loss function value; determining a target loss function value according to the sample migration loss function value and the model training loss function value; network parameters in the source domain detection model are adjusted based on the target loss function value to obtain the target domain detection model, so that the workload of labeling distribution network insulator images is reduced, the model training cost is reduced, and the robustness and the generalization capability of the target domain detection model are improved.
Description
Technical Field
The invention relates to the technical field of power detection, in particular to a training and detecting method, device and equipment for a distribution network insulator aging detection model.
Background
The insulator is characterized by much lighter weight than ceramic, high strength, difficult damage and good pollution resistance, and is widely used as an important device for supporting a lead and preventing current from flowing back to the ground in a power system. However, extreme weather such as corona discharge on the power line, ultraviolet irradiation of sunlight, environmental humidity and acid rain and acid mist can cause the organic silicon rubber material of the umbrella skirt of the insulator to age, which causes the reduction of electrical performance and mechanical performance, thereby causing potential threat to the power supply reliability of the power grid. In severe cases, network failure, large-scale power failure and the like can even occur. Therefore, the aging defect of the distribution network insulator is accurately detected and solved at an early stage, and further damage of power transmission equipment can be effectively prevented, so that the safety and stability of power transmission of a power line are ensured.
At present, the insulator detection party mainly controls a navigation route of unmanned aerial vehicle inspection and collects distribution network insulator images, and analyzes and processes the images based on a target detection algorithm to judge the damage condition of an insulator umbrella skirt. The target detection algorithm aiming at the damage of the umbrella skirt of the distribution network insulator mainly comprises the following steps: detection is performed by conventional image processing techniques and by introducing a deep-learning object detection model.
The traditional detection method mainly comprises the steps of distinguishing insulators to be detected from background information of an image to be detected through different characteristics of contour edge characteristics, color characteristics, gray level characteristics, texture characteristics and the like of a target, and analyzing according to the obtained information of the insulators to obtain a detection result of breakage of an umbrella skirt of the insulator. Therefore, when insulator information and background region indexes are not large, or the insulator target is shielded due to poor shooting angles, or the insulator sub-target with unobvious texture features is processed, the detection accuracy of the traditional detection method is greatly reduced. That is, the conventional detection method is poor in robustness.
Moreover, the reason why the above method can achieve satisfactory results is that: the training set and the testing set come from the same field, the training set and the testing set are independently and simultaneously distributed, and the training data of the model and the real data needing to be detected are from the same type of insulator image. At present, most of insulator training sample images which can be obtained publicly are power transmission insulator images. Although the distribution network insulator detection task and the transmission insulator detection task have extremely high similarity, the distribution network insulator and the transmission insulator are both of multilayer shed structures, but the image background of the transmission insulator of the distribution network insulator has larger difference. The power transmission insulator is arranged in a power transmission network, and the image background of the power transmission insulator contains vegetation, rivers, silt and Gobi; the distribution network insulator is mostly arranged in the distribution network, and the image background of the distribution network insulator contains vehicles, buildings, pedestrians and the like. The detection model obtained by training the data set of the power transmission insulator is used for detecting the damage condition of the umbrella skirt of the distribution network insulator, and the effect is poor. And it takes long time and high cost to re-collect and label a large number of distribution network insulator data sets.
Disclosure of Invention
The invention provides a method, a device and equipment for training and detecting an aging detection model of a distribution network insulator, wherein a target domain training sample set is obtained by utilizing rich source domain marking data of a target transmission insulator image sample subjected to style migration through style migration of a transmission insulator image sample and a distribution network insulator image, so that the workload of marking the target domain training sample set is greatly reduced, and the cost of re-marking the distribution network insulator image is reduced; compared with the traditional detection model, the detection model obtained by training based on the target domain training sample set improves the robustness and generalization capability of the model.
According to one aspect of the invention, a training method for a distribution network insulator aging detection model is provided, which comprises the following steps:
obtaining an initial training sample set; the initial training sample set comprises: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data; inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples;
determining a target domain training sample set according to the style migration image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set;
inputting the target domain training samples in the target domain training sample set into a source domain detection model to obtain a prediction detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample;
determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
According to an aspect of the present invention, a method for detecting aging of a distribution network insulator is provided, including:
acquiring an insulator image of a distribution network to be detected;
and inputting the distribution network insulator image to be detected into a target domain detection model obtained by training by adopting the training method of the distribution network insulator aging detection model in any embodiment, and determining the detection result of the distribution network insulator image to be detected.
According to another aspect of the present invention, there is provided a training apparatus for a distribution network insulator aging detection model, comprising:
the sample set input module is used for acquiring an initial training sample set; the initial training sample set comprises: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data;
the style migration module is used for inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples;
the sample loss determining module is used for determining a target domain training sample set according to the style migration image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set;
a training loss determining module, configured to input the target domain training samples in the target domain training sample set into a source domain detection model to obtain a predicted detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample;
a parameter adjustment module for determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
According to another aspect of the present invention, there is provided an electronic apparatus including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute the training method for the aging detection model of the distribution network insulator according to any embodiment of the present invention, and/or execute the aging detection method for the distribution network insulator according to any embodiment.
According to another aspect of the present invention, a computer-readable storage medium is provided, where the computer-readable storage medium stores computer instructions, and the computer instructions are configured to, when executed by a processor, implement a training method for a distribution network insulator aging detection model according to any embodiment of the present invention, and/or implement a distribution network insulator aging detection method according to any embodiment of the present invention.
The invention discloses a method, a device and equipment for training and detecting an aging detection model of a distribution network insulator, wherein the method for training the aging detection model of the distribution network insulator comprises the following steps: acquiring an initial training sample set; the initial training sample set includes: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data; inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples; determining a target domain training sample set according to the style migration image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set; inputting target domain training samples in the target domain training sample set into a source domain detection model to obtain a prediction detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample; determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model. The method comprises the steps that a source domain detection model is trained by using a style migration image sample obtained by carrying out style migration on a power transmission insulator image sample and a distribution network insulator image to obtain a target domain detection model, so that the problems that the obtained labeled distribution network insulator data set has less data, and the labeling is long and high in cost are solved, the workload of labeling the target domain training sample set is greatly reduced, and the cost and time consumed by manually labeling the distribution network insulator image are reduced; and moreover, parameter adjustment is carried out on the source domain detection model according to the sample migration loss function value and the model training loss function value to obtain a target domain detection model, and the robustness and the generalization capability of the target domain detection model are improved.
It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present invention, nor do they necessarily limit the scope of the invention. Other features of the present invention will become apparent from the following description.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a training method for a distribution network insulator aging detection model according to an embodiment of the present invention;
fig. 2 is a flowchart of a training method for a distribution network insulator aging detection model according to a second embodiment of the present invention;
FIG. 3 is a schematic diagram of the principle of joint maximum mean difference;
fig. 4 is a flowchart of another training method for a distribution network insulator aging detection model according to a second embodiment of the present invention;
fig. 5 is a flowchart of an aging detection method for distribution network insulators according to a third embodiment of the present invention;
fig. 6 is a schematic structural diagram of a training device of a distribution network insulator aging detection model according to a fourth embodiment of the present invention;
fig. 7 is a schematic structural diagram of an aging detection apparatus for distribution network insulators according to a fifth embodiment of the present invention;
fig. 8 is a schematic structural diagram of an electronic device implementing a training method for a distribution network insulator aging detection model and/or an aging detection method for a distribution network insulator according to an embodiment of the present invention.
Detailed Description
In order to make those skilled in the art better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, shall fall within the protection scope of the present invention.
It is noted that the terms "initial", "target", and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Example one
Fig. 1 is a flowchart of a training method for a distribution network insulator aging detection model according to an embodiment of the present invention, which is applicable to a situation where a distribution network insulator aging detection model capable of detecting damage to a shed of a distribution network insulator is obtained by training, where the method may be performed by a training device for the distribution network insulator aging detection model, the training device for the distribution network insulator aging detection model may be implemented in hardware and/or software, and the training device for the distribution network insulator aging detection model may be configured in an electronic device. As shown in fig. 1, the method includes:
s110, obtaining an initial training sample set; the initial training sample set comprises: the method comprises the steps of obtaining a power transmission insulator image sample containing detection result label data and a distribution network insulator image sample not containing the detection result label data.
The initial training sample set is a sample set used for training the source domain detection model after style migration, and the initial training sample set comprises: and the power transmission insulator image sample set and the distribution network insulator image sample set. The power transmission insulator image sample does not contain detection result label data, namely the power transmission insulator image sample is not manually marked; the distribution network insulator image sample comprises detection result label data, namely the distribution network insulator image sample is manually marked with a detection result label. The detection result tag data may include: the shed is normal or the shed is aged.
Specifically, the method for acquiring the initial training sample set can be used for acquiring a transmission insulator image and a distribution network insulator image, and manually labeling detection result label data on the transmission insulator image to obtain a transmission insulator image sample; and (4) carrying out no labeling processing on distribution network insulator images to obtain a power transmission insulator image sample.
And S120, inputting the power transmission insulator image samples and the distribution network insulator images in the initial training sample set into a style migration network for style migration to obtain style migration image samples.
The style migration network is a network model for performing style migration on an input image. The inputs to the style migration network include: an image for providing content and an image for providing a genre. The style migration image sample is an image sample obtained after the style migration of the power transmission insulator image sample and the distribution network insulator image.
In the actual task of the insulator to be detected, the distribution environments of the distribution network insulator and the transmission insulator are greatly different. Because the power transmission insulator is arranged in the power transmission network, the image background of the power transmission insulator contains vegetation, rivers, silt and gobi; the distribution network insulator is mostly arranged in the distribution network, and the image background of the distribution network insulator contains vehicles, buildings, pedestrians and the like. Therefore, in this step, the power transmission insulator image sample containing the detection result label data (i.e., manually labeled) is used to provide the content characteristics, and the distribution network insulator image containing no detection result label data (i.e., not manually labeled) is used to provide the style characteristics. The finally obtained style migration image sample not only contains the content characteristics (namely umbrella skirt characteristics) and the detection result label data of the power transmission insulator image sample, but also contains the style characteristics (namely image background) of the distribution network insulator image.
Specifically, the power transmission insulator image samples which contain the detection result label data in the initial training sample set are used as images for providing content characteristics, the distribution network insulator image samples which do not contain the detection result label data in the initial training sample set are used as images for providing style characteristics, and the images are respectively input into a style migration network; and fusing style characteristics of the distribution network insulator image sample into the power transmission insulator image sample through a style migration network to obtain a style migration image sample.
This has the advantage that: the method comprises the steps that the style of a distribution network insulator image sample is transferred into a power transmission insulator image to obtain a style transfer image which has the style characteristics of the distribution network insulator image and the content characteristics of the power transmission insulator image, the marking of the power transmission insulator image is reserved, the detection model is trained by adopting the style transfer image, the distribution network insulator image does not need to be marked again, and the cost and the time consumed by manually marking the distribution network insulator image are reduced; compared with a target domain detection model obtained by training based on the power transmission insulator image, the target domain detection model obtained by training the style migration image sample can greatly improve the detection accuracy.
S130, determining a target domain training sample set according to the style transition image samples; and determining a sample migration loss function value according to the initial training sample set and the target domain training sample set.
Wherein, the target domain training sample set is a training sample set used for training the source domain detection model. The sample migration loss function value is an error value between the style migration image sample output by the style migration network and the image samples (the power transmission insulator image sample and the distribution network insulator image sample) in the input initial training sample set, and can be understood as a loss value generated by style migration.
Specifically, a target domain training sample set is determined according to the style migration image samples, and a sample migration loss function value generated by style migration is determined according to an initial training sample set input by the style migration network and the style migration image samples output by the style migration network.
For example, the manner of determining the target domain training sample set by using the style migration image samples may be to determine a set determined by the style migration image samples as the target domain training sample set; or determining a set of the style transition image samples and distribution network insulator image samples containing detection result label data obtained through manual labeling as a target domain training sample set; or determining the style migration image sample and the power transmission insulator image sample containing the detection result label data as a target domain training sample set.
The target domain training sample set obtained in the step can be used for training a source domain detection model to obtain a target domain detection model, and compared with a detection model obtained based on power transmission insulator image training, the detection accuracy can be effectively improved.
S140, inputting the target domain training samples in the target domain training sample set into a source domain detection model to obtain a prediction detection result; and determining a model training loss function value according to the predicted detection result and the detection result label data contained in the target domain training sample.
The source domain detection model may be understood as an untrained initial detection model, for example, a yoloV5 model or other target detection models, which is not limited in this embodiment of the present invention. The predicted detection result is a detection result obtained by processing the target domain training sample through the source domain detection model. The model training loss function value is an error value between a prediction detection result output by the source domain detection model and an input target domain training sample, and can be understood as a loss value generated by model training. Specifically, target domain training samples in a target domain training sample set are input into a source domain detection model, and a target of the target domain training samples is detected through the source domain detection model to obtain a prediction detection result; and determining a model training loss function value generated by model training according to the predicted detection result and the detection result label data contained in the target domain training sample.
S150, determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
The target domain detection model is a completely trained detection model obtained by training the source domain detection model.
Specifically, a target loss function value is determined according to a sample migration loss function value and a model training loss function value obtained by each iterative training; and adjusting network parameters in the source domain detection model according to the target loss function value until a preset number of times is obtained or the target loss function value is minimum, so as to obtain the target domain detection model.
In the step, the network parameters in the source domain detection model are adjusted by using the sample migration loss function value and the model training loss function value, so that the accuracy, robustness and generalization capability of the target domain detection model are improved.
According to the technical scheme of the embodiment of the invention, an initial training sample set is obtained; the initial training sample set includes: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data; inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples; determining a target domain training sample set according to the style migration image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set; inputting target domain training samples in the target domain training sample set into a source domain detection model to obtain a prediction detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample; determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model. The method comprises the steps that a source domain detection model is trained by using a style migration image sample obtained by carrying out style migration on a power transmission insulator image sample and a distribution network insulator image to obtain a target domain detection model, so that the problems that the obtained labeled distribution network insulator data set has less data, and the labeling is long and high in cost are solved, the workload of labeling the target domain training sample set is greatly reduced, and the cost and time consumed by manually labeling the distribution network insulator image are reduced; and moreover, parameter adjustment is carried out on the source domain detection model according to the sample migration loss function value and the model training loss function value to obtain a target domain detection model, and the robustness and the generalization capability of the target domain detection model are improved.
Optionally, determining a target domain training sample set according to the style migration image samples includes:
and determining a set of the style migration image samples and the power transmission insulation image samples containing the detection result label data as a target domain training sample set.
In practical engineering application, although a transmission insulator image is richer and easier to collect than a distribution network insulator image, and a large number of marked transmission insulator image samples exist, the distribution network insulator image is difficult to collect; limited by the number of distribution network insulator images, the number of style migration image samples obtained by style migration according to the distribution network insulator images and the transmission insulator images is limited, and the number of training sample sets required by the training model cannot be obtained.
Because the detection task of the transmission insulator and the detection task of the distribution network insulator have high similarity, the embodiment of the invention enriches and expands the style migration image sample set to obtain the target domain training sample set by adopting the collection of the transmission insulator image samples containing the label data of the detection result, thereby ensuring that the data volume of the target domain training sample set meets the training requirement, and effectively improving the detection accuracy compared with the detection model obtained by adopting the transmission insulator image training.
Optionally, inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples, where the method includes:
inputting the power transmission insulator image samples in the initial training sample set into a content feature extraction module of the style migration network to obtain a content feature graph;
inputting distribution network insulator image samples in the initial training sample set into a style feature extraction module of a style migration network to obtain a style feature diagram;
and inputting the content characteristic diagram and the style characteristic diagram into a characteristic fusion module of the style migration network to obtain a style migration image sample.
The style migration network may adopt a VGG19 network, and the style migration network may include: the system comprises a content feature extraction module, a style feature extraction module and a feature fusion module. The content feature extraction module is used for extracting a content feature map of the input image, the style feature extraction module is used for extracting a style feature map of the input image, and the feature fusion module is used for performing feature fusion on the content feature map input by the content feature extraction module and the style feature map output by the style feature extraction module.
Specifically, a power transmission insulator image sample in an initial training sample set is used as an image for providing content characteristics, the image is input into a content characteristic extraction module of the style migration network, and the content characteristics of the power transmission insulator image sample are extracted through the content characteristic extraction module to obtain a content characteristic diagram; and taking the distribution network insulator image samples in the style migration training sample set as images for providing style characteristics, inputting the images into a style characteristic extraction module of the style migration network, and extracting the style characteristics of the distribution network insulator image samples through the style characteristic extraction module to obtain a style characteristic diagram. And performing feature fusion on the content feature graph output by the content feature extraction module and the style feature graph output by the style feature extraction module through a feature fusion module to obtain a style migration image sample.
Example two
Fig. 2 is a flowchart of a training method for a distribution network insulator aging detection model according to a second embodiment of the present invention, and this embodiment further defines "determining a sample migration loss function value according to an initial training sample set and a target domain training sample set" in step S120 between the above embodiments. As shown in fig. 2, the method includes:
s210, obtaining an initial training sample set; the initial training sample set includes: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data.
S220, inputting the power transmission insulator image samples and the distribution network insulator images in the initial training sample set into a style migration network for style migration to obtain style migration image samples.
And S230, determining a target domain training sample set according to the style transition image samples.
And S240, determining a style migration loss function value according to the style migration image sample, the power transmission insulator image sample and the distribution network insulator image sample.
And the style migration loss function value is a function value calculated based on the style migration loss function.
Specifically, the style migration loss function value can be determined according to the content loss function value between the style migration image sample and the power transmission insulator image sample and the style loss function value between the style migration image sample and the distribution network insulator image sample.
And S250, determining the combined maximum average deviation according to the style transition image sample and the distribution network insulator image sample.
Wherein Joint Maximum Mean Difference (JMMD) is a distance metric for determining the distribution between source domain data (distribution network insulator image samples) and target domain data (style migration image samples).
Specifically, the distribution difference between the power transmission insulator image sample and the style migration image sample can be reduced by combining the iterative optimization with the maximum mean difference. The principle schematic of the combined maximum mean difference is shown in fig. 3.
And S260, determining the sum of the style migration loss function value and the joint maximum average deviation as a sample migration loss function value.
Wherein the sample migration loss function value is a function value calculated based on the sample migration loss function.
Specifically, the sample migration loss function is obtained according to the sum of the style migration loss function and the joint maximum average deviation, that is:
L sample ＝L migrate +JMMD；
wherein L is sample For a sample migration loss function, JMMD is the combined maximum mean difference of the power transmission insulator image sample and the target domain training sample, L migrate Is a style migration loss function.
And inputting the style migration loss function value and the combined maximum average deviation into a sample migration loss function to obtain a sample migration loss function value.
S270, inputting the target domain training samples in the target domain training sample set into a source domain detection model to obtain a prediction detection result; and (4) determining a model training loss function value according to the predicted detection result and the detection result label data contained in the target domain training sample.
S280, determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
Specifically, the target loss function is:
L total ＝L sample +L model ＝L migrate +JMMD+L model ；
wherein L is total As a function of the target loss, L model Training loss function for the model, L sample The migration loss function for the sample includes: joint maximum mean difference JMMD of power transmission insulator image sample and target domain training sample, and style migration loss function L migrate 。
For example, an adaptive moment estimation (ADAM) algorithm may be used to optimize the network, and the objective loss function is iteratively solved to obtain an optimal solution (i.e., an objective loss function value). The specific method is to firstly give a parameter w to be optimized, an objective loss function f and an initial learning rate alpha. Iterative optimization can then begin. The following operations are performed at each iteration: calculating the gradient of the objective function with respect to the current parameter:
calculating one from historical gradientsOrder moment m t ＝β 1 *m t-1 +(1-β 1 )*g t And second moment
β 1 Control of first moment, beta 2 Controlling the second moment; calculating the falling gradient at the current moment:
updating omega of the value of the objective function as a function of the decreasing gradient t+1 ＝ω t -η t . And finally, realizing detection and classification of targets through a Support Vector Machine (SVM) classifier, and finishing a distribution network insulator aging defect detection task. And the model performance is strictly evaluated by using two indexes of accuracy and recall rate, so that the effectiveness of the model is ensured.
By the aid of a domain self-adaptive algorithm in transfer learning, the network can more easily obtain characteristic information of aging defects of distribution network insulator images in a target domain training sample in a training process, training time and storage cost are reduced, training precision is remarkably improved, and the problem of poor model generalization capability caused by insufficient model training data volume is solved.
According to the technical scheme of the embodiment of the invention, an initial training sample set is obtained; the initial training sample set comprises: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data; inputting the power transmission insulator image samples and the distribution network insulator images in the initial training sample set into a style migration network for style migration to obtain style migration image samples; determining a target domain training sample set according to the style migration image samples; determining a style migration loss function value according to the style migration image sample, the transmission insulator image sample and the distribution network insulator image sample; determining a joint maximum average deviation according to the style migration image sample and the distribution network insulator image sample; determining the sum of the style migration loss function and the joint maximum average deviation as a sample migration loss function value; inputting target domain training samples in the target domain training sample set into a source domain detection model to obtain a prediction detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample; determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model. The method comprises the steps that a source domain detection model is trained by using a style migration image sample obtained by carrying out style migration on a power transmission insulator image sample and a distribution network insulator image to obtain a target domain detection model, so that the problems that the obtained labeled distribution network insulator data set has less data, and the labeling is long in time and high in cost are solved, the workload of labeling the target domain training sample set is greatly reduced, and the cost and time consumed by manually labeling the distribution network insulator image are reduced; and the source domain detection model is subjected to parameter adjustment according to the style migration loss function value, the combined maximum average deviation and the model training loss function value to obtain the target domain detection model, so that the distribution difference between source domain data and target domain data is reduced, data alignment is realized, and the robustness and generalization capability of the target domain detection model are improved.
Optionally, determining a style migration loss function value according to the style migration image sample, the transmission insulator image sample, and the distribution network insulator image sample, includes:
determining a content loss function value according to the style migration image sample and the power transmission insulator image sample;
determining a style loss function value according to the style migration image sample and the distribution network insulator image sample;
a weighted sum of the content loss function value and the style loss function value is determined as a style migration loss function value.
Specifically, the feature extraction module of the style migration network includes multiple convolution layers, each convolution layer includes multiple filters, and the number of feature maps output by each convolution layer can be determined according to the number of the filters. Vectorizing the feature map to obtain a feature map with a size ofM l (i.e., the size of the filter). Laminating N of each layer l The eigenvectors (i.e., the number of filters included in each convolutional layer) are stored in the matrix
Of (1), its elements
The output of the ith filter at position j of the ith layer is shown.
Setting up
A sample of a power transmission insulator image is represented,
representing a style-shifted image sample, P l Representing the response of a Transmission insulator image sample to the layer I of the convolutional layer of the content feature extraction Module, F l Representing the desired image response to the L convolutional layers of the content feature extraction module, the content loss function for L layers is:
the content loss function is:
wherein r is L And the L layers of convolution layer weights of the content feature extraction module.
Setting up
A sample of the distribution network insulator picture is shown,
representing a style migration image sample, A l Insulation diagram for representing distribution networkResponse of image sample to the 1-layer convolution layer of the style feature extraction Module, G l Representing the response of the desired image to the L convolutional layers of the style feature extraction module, the style loss function for the L layers is:
the style loss function is:
wherein r is L The L-layer convolution layer weight of the style feature extraction module.
The style migration loss function is:
optionally, determining a joint maximum average deviation according to the style transition image sample and the distribution network insulator image sample includes:
the distribution network insulator image sample and the style transition image sample with the same detection result are determined according to the detection result label data contained in the prediction detection result and the style transition image sample corresponding to the distribution network insulator image sample;
calculating the maximum average deviation of samples between distribution network insulator image samples and style migration image samples with the same detection result;
and determining the sum of the maximum average deviations of the samples corresponding to each detection result as the joint maximum average deviation.
Specifically, the distribution network insulator image sample and the style transition image sample with the same detection result are determined according to the detection result label data contained in the prediction detection result and the style transition image sample corresponding to the distribution network insulator image sample, so that data alignment is realized. The theoretical accurate value of the maximum average deviation of the samples between the distribution network insulator image samples and the style migration image samples with the same detection result is as follows:
wherein, P is the distribution formed by the distribution network insulator image samples, and Q is the distribution formed by the style transition image samples; c is the center of the distribution, z s,1 Extracting features of a sample through a first layer of feature extraction layer L1;
the sample is passed through 1,2, \8230 \ 8230; | L | feature extraction layer to extract a combination of features (i.e., z) s,1 ,z s,2 ,…,z s,L )；
Is a tensor product;
namely, it is
The distribution network insulator image sample is distributed in the distribution network insulator image sample by the distribution network insulator image sample distribution method, wherein the distribution P passes through the distribution center formed by the combination of the characteristics extracted by the L | characteristic extraction layer 1,2, \8230;,
the distribution Q constructed for the style migration image sample passes through the center of the distribution composed of the features extracted by the L-feature extraction layer 1,2, \8230 |.
According to the specific technical scheme of the embodiment of the invention, as shown in fig. 4, a power transmission insulator image sample containing detection result label data is used as an image for providing content characteristics, a distribution network insulator image sample not containing detection result label data is used as an image for providing style characteristics, and the images are respectively input into a style migration network for style migration to obtain a style migration image sample; determining a set of style migration image samples and power transmission insulator image samples containing detection result label data as a target domain training sample set, and inputting the target domain training sample set into a source domain detection model; determining a model training loss function value according to a prediction detection result output by a source domain detection model and detection result label data contained in the target domain training sample; determining a sample migration loss function value according to the power transmission insulator image sample, the distribution network insulator image sample and the target domain training sample set; determining a joint maximum average deviation according to distribution network insulator image samples and style migration image samples containing prediction detection results, and obtaining a target loss function value according to a model training loss function value, a sample migration loss function value and the joint maximum average deviation; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
EXAMPLE III
Fig. 5 is a flowchart of a method for detecting aging of a distribution network insulator according to a third embodiment of the present invention, where this embodiment is applicable to detecting a damaged umbrella skirt of the distribution network insulator, and the method may be implemented by an aging detection apparatus for the distribution network insulator, where the aging detection apparatus for the distribution network insulator may be implemented in a hardware and/or software manner, and the aging detection apparatus for the distribution network insulator may be configured in an electronic device. As shown in fig. 5, the method includes:
and S310, acquiring an insulating image of the distribution network to be detected.
The distribution network insulator image to be detected can be understood as a distribution network insulator image waiting for detection.
S320, inputting the distribution network insulator image to be detected into a target domain detection model obtained by training through a distribution network insulator aging detection model, and determining a target detection result of the distribution network insulator image to be detected.
Specifically, a power transmission insulator image sample containing detection result label data in an initial training sample set and a distribution network insulator image sample not containing detection result label data are input into a style migration network for style migration, and a style migration image sample is obtained; determining a target domain training sample set according to the style transition image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set; inputting target domain training samples in the target domain training sample set into a source domain detection model to obtain a predicted detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample; determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model. And inputting the distribution network insulator image to be detected into a target domain detection model, and determining a target detection result of the distribution network insulator image to be detected.
According to the technical scheme of the embodiment of the invention, the distribution network insulator image to be detected is obtained; the distribution network insulator image to be detected is input into the target domain detection model obtained by training through the distribution network insulator aging detection model, the target detection result of the distribution network insulator image to be detected is determined, the detection accuracy of the distribution network insulator image can be improved, and the high-precision and high-recall distribution network insulator damage defect detection is realized.
Example four
Fig. 6 is a schematic structural diagram of a training device for a distribution network insulator aging detection model according to a fourth embodiment of the present invention. As shown in fig. 6, the apparatus includes: a sample set input module 410, a style migration module 420, a sample loss determination module 430, a training loss determination module 440, and a parameter adjustment module 450;
a sample set input module 410, configured to obtain an initial training sample set; the initial training sample set comprises: the method comprises the steps that a power transmission insulator image sample containing detection result label data and a distribution network insulator image sample not containing the detection result label data are obtained;
the style migration module 420 is configured to input the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration, so as to obtain style migration image samples;
a sample loss determining module 430, configured to determine a target domain training sample set according to the style migration image sample; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set;
a training loss determining module 440, configured to input the target domain training samples in the target domain training sample set into a source domain detection model to obtain a predicted detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample;
a parameter adjustment module 450 configured to determine a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
Optionally, the sample loss determining module 430 includes:
and the training sample set determining unit is used for determining the style transition image sample and the set of the power transmission insulator image samples containing the detection result label data as a target domain training sample set.
Optionally, the style migration module 420 is specifically configured to:
inputting the power transmission insulator image samples in the initial training sample set into a content feature extraction module of the style migration network to obtain a content feature diagram;
inputting distribution network insulator image samples in the initial training sample set into a style feature extraction module of the style migration network to obtain a style feature map;
and inputting the content feature map and the style feature map into a feature fusion module of the style migration network to obtain a style migration image sample.
Optionally, the sample loss determining module 430 includes:
the style migration loss determining unit is used for determining a style migration loss function value according to the style migration image sample, the power transmission insulator image sample and the distribution network insulator image sample;
a maximum average deviation determining unit, configured to determine a joint maximum average deviation according to the style transition image sample and the distribution network insulator image sample;
a sample migration loss determining unit, configured to determine a sum of the style migration loss function value and the joint maximum average deviation as the sample migration loss function value.
Optionally, the style migration loss determining unit is specifically configured to:
determining a content loss function value according to the style migration image sample and the transmission insulator image sample;
determining a style loss function value according to the style migration image sample and the distribution network insulator image sample;
determining a weighted sum of the content loss function value and the style loss function value as a style migration loss function value.
Optionally, the maximum average deviation determining unit is specifically configured to:
determining distribution network insulator image samples and style migration image samples with the same detection result according to the prediction detection result corresponding to the distribution network insulator image samples and the detection result label data contained in the style migration image samples;
calculating the maximum average deviation of samples between distribution network insulator image samples and style migration image samples with the same detection result;
and determining the sum of the maximum average deviations of the samples corresponding to each detection result as the combined maximum average deviation.
The training device for the distribution network insulator aging detection model provided by the embodiment of the invention can execute the training method for the distribution network insulator aging detection model provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.
EXAMPLE five
Fig. 7 is a schematic structural diagram of an aging detection device for distribution network insulators according to a fifth embodiment of the present invention. As shown in fig. 7, the apparatus includes: an image acquisition module 510 and an image detection module 520;
the image acquisition module 510 is configured to acquire an insulator image of the distribution network to be detected;
the image detection module 520 is configured to input the to-be-detected distribution network insulator image into a target domain detection model obtained by training through a distribution network insulator aging detection model, and determine a target detection result of the to-be-detected distribution network insulator image.
The aging detection device for the distribution network insulator provided by the embodiment of the invention can execute the aging detection method for the distribution network insulator provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.
Example six
FIG. 8 illustrates a schematic diagram of an electronic device 10 that may be used to implement an embodiment of the invention. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.
As shown in fig. 8, the electronic device 10 includes at least one processor 11, and a memory communicatively connected to the at least one processor 11, such as a Read Only Memory (ROM) 12, a Random Access Memory (RAM) 13, and the like, wherein the memory stores a computer program executable by the at least one processor, and the processor 11 can perform various suitable actions and processes according to the computer program stored in the Read Only Memory (ROM) 12 or the computer program loaded from a storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data necessary for the operation of the electronic apparatus 10 may also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input/output (I/O) interface 15 is also connected to bus 14.
A number of components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, or the like; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.
The processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, or the like. The processor 11 performs the above described methods and processes, such as a training method of a distribution network insulator aging detection model or an aging detection method of a distribution network insulator.
In some embodiments, the training method of the aging detection model of the distribution network insulator or the aging detection method of the distribution network insulator may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 10 via the ROM 12 and/or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the training method of the aging detection model of the distribution network insulator or the aging detection method of the distribution network insulator described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured by any other suitable means (e.g., by means of firmware) to perform the training method of the aging detection model of the distribution network insulator or the aging detection method of the distribution network insulator.
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
A computer program for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be performed. A computer program can execute entirely on a machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the electronic device. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), blockchain networks, and the internet.
The computing system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical host and VPS service are overcome.
It should be understood that various forms of the flows shown above, reordering, adding or deleting steps, may be used. For example, the steps described in the present invention may be executed in parallel, sequentially, or in different orders, and are not limited herein as long as the desired results of the technical solution of the present invention can be achieved.
The above-described embodiments should not be construed as limiting the scope of the invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A training method for a distribution network insulator aging detection model is characterized by comprising the following steps:
obtaining an initial training sample set; the initial training sample set comprises: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data; inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples;
determining a target domain training sample set according to the style migration image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set;
inputting the target domain training samples in the target domain training sample set into a source domain detection model to obtain a predicted detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample;
determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
2. The method of claim 1, wherein determining a set of target domain training samples from the style transition image samples comprises:
and determining the style transition image sample and the collection of the power transmission insulator image samples containing the detection result label data as a target domain training sample set.
3. The method of claim 1, wherein the step of inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples comprises:
inputting the power transmission insulator image samples in the initial training sample set into a content feature extraction module of the style migration network to obtain a content feature diagram;
inputting the distribution network insulator image samples in the initial training sample set into a style feature extraction module of the style migration network to obtain a style feature diagram;
and inputting the content characteristic diagram and the style characteristic diagram into a characteristic fusion module of the style migration network to obtain a style migration image sample.
4. The method of claim 1, wherein determining a sample migration loss function value from the initial training sample set and the target domain training sample set comprises:
determining a style migration loss function value according to the style migration image sample, the power transmission insulator image sample and the distribution network insulator image sample;
determining a joint maximum average deviation according to the style migration image sample and the distribution network insulator image sample;
determining a sum of the style migration loss function value and the joint maximum mean deviation as the sample migration loss function value.
5. The method of claim 4, wherein determining a style migration loss function value from the style migration image sample, the transmission insulator image sample, and the distribution network insulator image sample comprises:
determining a content loss function value according to the style migration image sample and the power transmission insulator image sample;
determining a style loss function value according to the style migration image sample and the distribution network insulator image sample;
determining a weighted sum of the content loss function value and the style loss function value as a style migration loss function value.
6. The method of claim 4, wherein determining a joint maximum mean deviation from the style transition image samples and the distribution network insulator image samples comprises:
determining distribution network insulator image samples and style migration image samples with the same detection result according to the prediction detection result corresponding to the distribution network insulator image samples and the detection result label data contained in the style migration image samples;
calculating the maximum average deviation of samples between distribution network insulator image samples and style migration image samples with the same detection result;
and determining the sum of the maximum average deviations of the samples corresponding to each detection result as the combined maximum average deviation.
7. The aging detection method for the distribution network insulator is characterized by comprising the following steps:
acquiring an insulator image of a distribution network to be detected;
inputting the distribution network insulator image to be detected into a target domain detection model obtained by training through the distribution network insulator aging detection model training method according to any one of claims 1 to 6, and determining a target detection result of the distribution network insulator image to be detected.
8. The utility model provides a join in marriage trainer of net insulator aging testing model which characterized in that includes:
the sample set input module is used for acquiring an initial training sample set; the initial training sample set comprises: the power transmission insulator image samples containing the detection result label data and the distribution network insulator image samples not containing the detection result label data;
the style migration module is used for inputting the power transmission insulator image samples and the distribution network insulator image samples in the initial training sample set into a style migration network for style migration to obtain style migration image samples;
the sample loss determining module is used for determining a target domain training sample set according to the style migration image samples; determining a sample migration loss function value according to the initial training sample set and the target domain training sample set;
a training loss determining module, configured to input the target domain training samples in the target domain training sample set into a source domain detection model to obtain a predicted detection result; determining a model training loss function value according to the predicted detection result and detection result label data contained in the target domain training sample;
a parameter adjustment module for determining a target loss function value according to the sample migration loss function value and the model training loss function value; and adjusting network parameters in the source domain detection model based on the target loss function value to obtain a target domain detection model.
9. An electronic device, characterized in that the electronic device comprises:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein, the first and the second end of the pipe are connected with each other,
the memory stores a computer program executable by the at least one processor, the computer program being executable by the at least one processor to enable the at least one processor to perform the method for training a model for aging detection of a distribution network insulator of any of claims 1-6, and/or to perform the method for aging detection of a distribution network insulator of claim 7.
10. A computer readable storage medium, wherein the computer readable storage medium stores computer instructions for causing a processor to execute the method for training the aging detection model of the distribution network insulator of any one of claims 1 to 6, and/or the method for aging detection of the distribution network insulator of claim 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211143600.4A CN115439716A (en) | 2022-09-20 | 2022-09-20 | Method, device and equipment for training and detecting aging detection model of distribution network insulator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211143600.4A CN115439716A (en) | 2022-09-20 | 2022-09-20 | Method, device and equipment for training and detecting aging detection model of distribution network insulator |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115439716A true CN115439716A (en) | 2022-12-06 |
Family
ID=84249116
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211143600.4A Pending CN115439716A (en) | 2022-09-20 | 2022-09-20 | Method, device and equipment for training and detecting aging detection model of distribution network insulator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115439716A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116151034A (en) * | 2023-04-17 | 2023-05-23 | 广东电网有限责任公司揭阳供电局 | Insulator core rod crisping prediction method, device, equipment and medium |
-
2022
- 2022-09-20 CN CN202211143600.4A patent/CN115439716A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116151034A (en) * | 2023-04-17 | 2023-05-23 | 广东电网有限责任公司揭阳供电局 | Insulator core rod crisping prediction method, device, equipment and medium |
| CN116151034B (en) * | 2023-04-17 | 2023-06-27 | 广东电网有限责任公司揭阳供电局 | Insulator core rod crisping prediction method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112967248B (en) | Method, apparatus, medium and program product for generating defect image samples | |
| CN110992307A (en) | Insulator positioning and identifying method and device based on YOLO | |
| CN115439716A (en) | Method, device and equipment for training and detecting aging detection model of distribution network insulator | |
| CN113361572A (en) | Training method and device of image processing model, electronic equipment and storage medium | |
| CN114937025A (en) | Image segmentation method, model training method, device, equipment and medium | |
| CN113204614B (en) | Model training method, method for optimizing training data set and device thereof | |
| CN115293366B (en) | Model training method, information prediction method, device, equipment and medium | |
| CN116468112A (en) | Training method and device of target detection model, electronic equipment and storage medium | |
| CN116468967A (en) | Sample image screening method and device, electronic equipment and storage medium | |
| CN115359322A (en) | Target detection model training method, device, equipment and storage medium | |
| CN117197517A (en) | Insulator pollution classification method and device, electronic equipment and medium | |
| CN113706705B (en) | Image processing method, device, equipment and storage medium for high-precision map | |
| CN114037052A (en) | Training method and device for detection model, electronic equipment and storage medium | |
| CN114494782A (en) | Image processing method, model training method, related device and electronic equipment | |
| CN113361524B (en) | Image processing method and device | |
| CN114841993A (en) | Training of insulator detection network, detection method and equipment thereof, and storage medium | |
| CN116071608B (en) | Target detection method, device, equipment and storage medium | |
| CN113223058B (en) | Training method and device of optical flow estimation model, electronic equipment and storage medium | |
| CN117649683B (en) | Acne grading method, device, equipment and storage medium | |
| CN115168852A (en) | Malicious code detection system training method and device, electronic equipment and storage medium | |
| CN115063810A (en) | Text detection method and device, electronic equipment and storage medium | |
| CN115293260A (en) | Age prediction model training method, device, equipment and storage medium | |
| CN117710792A (en) | Knowledge distillation method, greening region detection method, electronic device, and storage medium | |
| CN115033017A (en) | Method and device for judging reasonability of routing inspection path of unmanned aerial vehicle for photovoltaic power station | |
| CN115344728A (en) | Image retrieval model training method, image retrieval model using method, image retrieval model training device, image retrieval model using device, image retrieval model equipment and image retrieval model medium |
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 |