CN114330548A - Insulator detection method based on background classification and transfer learning - Google Patents
Insulator detection method based on background classification and transfer learning Download PDFInfo
- Publication number
- CN114330548A CN114330548A CN202111633194.5A CN202111633194A CN114330548A CN 114330548 A CN114330548 A CN 114330548A CN 202111633194 A CN202111633194 A CN 202111633194A CN 114330548 A CN114330548 A CN 114330548A
- Authority
- CN
- China
- Prior art keywords
- insulator
- distribution network
- inspection line
- line image
- data set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of electric power inspection image processing, and discloses an insulator detection method based on background classification and transfer learning. The method comprises the steps of performing migration learning training on preprocessed insulator detection data sets by using a single-stage target detector to obtain an initial insulator detection model, dividing the insulator detection data sets into a plurality of data sets according to background categories, marking a corresponding background category on each data set, performing migration learning training on each data set by using the initial insulator detection model respectively to obtain insulator sub-target detection models under different background categories, calculating the detection accuracy of the insulator target detection models under different background categories for detecting the corresponding data sets until the detection accuracy of the insulators under the backgrounds reaches a set value, and accordingly improving the detection accuracy of the insulators.
Description
Technical Field
The invention relates to the technical field of electric power inspection image processing, in particular to an insulator detection method based on background classification and transfer learning.
Background
The routing inspection of the distribution network line is an effective means for ensuring the safe and reliable operation of the power system. In a distribution network line, an insulator is widely used equipment, has double functions of electrical insulation and mechanical support, and the state monitoring of the insulator is one of the most important and difficult tasks in power transmission line inspection. Traditional manual work is patrolled and examined and is wasted time and energy, can patrol and examine the replacement with unmanned aerial vehicle, realizes more automatic, the efficient patrols and examines. However, the aerial images of the unmanned aerial vehicle contain cluttered backgrounds and various types of insulators, and the change of the visual angle, different lighting conditions, partial shielding and other external interference factors make the insulator detection very difficult. The existing detection method mainly extracts the features of the aerial images by means of image processing, and distinguishes insulators from complex backgrounds such as color, shape, texture features and the like, but the accuracy of insulator detection is greatly reduced.
Disclosure of Invention
The invention provides an insulator detection method based on background classification and transfer learning, and solves the technical problem of reducing the accuracy of insulator detection.
In view of this, the present invention provides an insulator detection method based on background classification and transfer learning, including the following steps:
the method comprises the steps of firstly, obtaining a plurality of distribution network routing inspection line images, marking the positions of insulators on each distribution network routing inspection line image by using a rectangular frame through labelme software, and generating a marking file to form an insulator detection data set, wherein the marking file comprises target position coordinates and category information;
secondly, preprocessing the insulator detection data set, wherein the preprocessing mode comprises a histogram equalization algorithm, an image filtering algorithm and image sharpening;
performing migration learning training on the preprocessed insulator detection data set by using a single-stage target detector to obtain an initial insulator detection model, wherein the single-stage target detector takes a ResNeSt convolutional network as a backbone network and takes a characteristic pyramid BiFPN as a characteristic extraction network, and the ResNeSt convolutional network is obtained by performing pre-training in an ImageNet deep learning network;
dividing the insulator detection data set into a plurality of data sets according to the background category, and labeling a corresponding background category on each data set;
step five, performing transfer learning training on each data set by using the initial insulator detection model to obtain insulator sub-target detection models under different background categories;
and step six, calculating the detection accuracy of the insulator target detection models in different background categories for detecting corresponding data sets, if the accuracy is lower than a preset accuracy threshold, updating the network parameters of the insulator target detection models in the corresponding background categories, executing step five until the accuracy of the insulator sub-target detection models reaches the preset accuracy threshold, and outputting the optimal insulator target detection model.
Preferably, the step one specifically includes:
acquiring a plurality of distribution network inspection line images based on an unmanned aerial vehicle;
and setting and loading a distribution network inspection line image by using labelme software, and performing frame selection marking on the insulators in the distribution network inspection line image by using a rectangular frame to generate a marking file to form an insulator detection data set.
Preferably, the preprocessing of the insulator detection data set by using a histogram equalization algorithm specifically includes:
calculating the frequency density of the distribution network routing inspection line image after histogram equalization according to the pixel number and the gray level depth of the distribution network routing inspection line image before histogram equalization,
in the formula, HB(D) Representing the frequency density, A, of the distribution network patrol line image after histogram equalization0The number of pixels of the distribution network routing inspection line image before histogram equalization is represented, and L represents the gray level depth of the distribution network routing inspection line image before histogram equalization;
gray value D of distribution network routing inspection line image A before histogram equalizationAGray value D in distribution network routing inspection line image after histogram equalization mapping through gray value transformation functionBI.e. DB＝f(DA)，f(DA) Representing a grey value transformation function, the grey value DAIn [0, D ]A]Inner pixel number and gray value DBIn [0, D ]B]The number of pixels in the same, i.e.
Solving the above formula to obtain a gray value transformation function f (D)A) In order to realize the purpose,
transforming the gray value into a function f (D)A) Discretizing to obtain a discretized gray value transformation function f' (D)A) Inputting each pixel in the distribution network patrol inspection line image A before histogram equalization into the discretization gray value transformation function f' (D)A) Obtaining a gray value D of a distribution network routing inspection line image B after histogram equalizationBThereby realizing histogram equalization of each distribution network routing inspection line image in the insulator detection data set, wherein the gray value D isBIn order to realize the purpose,
preferably, the process of preprocessing the insulator detection data set by using an image filtering algorithm specifically comprises:
filtering the distribution network inspection line image in the insulator detection data set based on a median filter, setting the pixel value of the ith row and the jth column in the distribution network inspection line image as p (i, j), filtering by the median filter to obtain the pixel value of the position as p (i, j),
med({p(i+x,j+y)|x∈{-1,0,1},y∈{-1,0,1}})
wherein med (-) is a function of median of elements in the set, and p (i + x, j + y) is a pixel value filtered by a median filter;
filtering and denoising the distribution network patrol inspection line image after filtering by using a Gaussian filter, wherein the pixel value after Gaussian filtering is
Where σ is the standard deviation of the given pixel value.
Preferably, the process of preprocessing the insulator detection data set by using image sharpening specifically includes:
calculating the second-order partial derivative of the pixel value of the distribution network routing inspection line image in the insulator detection data set as,
in the formula, p0(i, j) is the gray value of the ith row and the jth column pixel point of the image before the Laplace operator acts on the pixel point of the distribution network routing inspection line image;
the gray value of the Laplace operator after acting on the pixel points of the distribution network routing inspection line image is calculated as,
calculating the pixel value of the sharpened distribution network routing inspection line image as
Where k is the coefficient of the diffusion effect.
Preferably, the method further comprises:
dividing input X of the ResNeSt convolutional network into K groups of cardinal number units along the dimension of an input channel, dividing each group of cardinal number units into R groups of sub cardinal numbers, and dividing the input channel into G groups of sub channels, wherein G is KR;
based on ResNeSt convolution network, each group of sub-channels sequentially pass through 1 × 1 convolution layer and 3 × 3 convolution layer for feature extraction to obtain features of Uz，Uz∈RH×W×CZ is 1,2, …, G, H, W, C are each UzThree dimensions of (a);
the sum of the extracted features for each group of sub-cardinalities in the kth group of cardinality units is calculated as
calculating the weight of the c component of the ith group of sub-bases in the kth group of base unit
In order to realize the purpose,
in the formula (I), the compound is shown in the specification,
representing features s according to the kth group of cardinal unitskThe constructed ith group divides the weight of the c component;
weighting and summing the extracted features of each group of sub-cardinalities in each cardinality unit to obtain the features of the cardinality unit, wherein the c component of the features of the k-th cardinality unit is,
characterizing each radix unit
Splicing, and inputtingAdding X to obtain the characteristic Y extracted by the ResNeSt convolution network,
V＝Concat{V1,V2,…，VK}
wherein Concat {. denotes a splicing operation,
the conversion of the convolutional layer and the pooling layer is shown so that the number of channels of the input X coincides with V.
Preferably, the method further comprises:
constructing a loss function of an insulation sub-target detection model by using the label file of the distribution network inspection line image,
FL(pt)＝-(1-pt)γlog(pt)
in the formula, a is a type label of a detection target, a ═ 1 indicates that a detection object is an insulator, a ═ 0 indicates a non-insulator, p indicates a probability that the detection object is an insulator, and γ is a given focusing parameter.
According to the technical scheme, the invention has the following advantages:
the insulator detection data set is preprocessed by constructing the insulator detection data set. Performing migration learning training on the preprocessed insulator detection data set by using a single-stage target detector to obtain an initial insulator detection model, dividing the insulator detection data set into a plurality of data sets according to background categories, labeling corresponding background categories on each data set, performing transfer learning training on each data set by using an initial insulator detection model, to obtain insulator target detection models under different background categories, to realize accurate detection of insulators under each background, to calculate the accuracy of detection of corresponding data sets by the insulator target detection models under different background categories, if the accuracy is lower than a preset accuracy threshold, network parameters of the insulator target detection model under the corresponding background category are updated, and migration learning is carried out until the accuracy of insulator detection under each background reaches a set value, so that the accuracy of insulator detection is improved.
Drawings
Fig. 1 is a flowchart of an insulator detection method based on background classification and transfer learning according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, 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 derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
For easy understanding, please refer to fig. 1, the method for detecting an insulator based on background classification and transfer learning provided by the present invention includes the following steps:
the method comprises the steps of firstly, obtaining a plurality of distribution network routing inspection line images, marking the positions of insulators on each distribution network routing inspection line image by using a rectangular frame through labelme software, and generating a marking file to form an insulator detection data set, wherein the marking file comprises target position coordinates and category information.
And secondly, preprocessing the insulator detection data set, wherein the preprocessing mode comprises a histogram equalization algorithm, an image filtering algorithm and image sharpening.
And step three, performing migration learning training on the preprocessed insulator detection data set by using a single-stage target detector to obtain an initial insulator detection model, wherein the single-stage target detector is obtained by using a ResNeSt convolution network as a backbone network and a characteristic pyramid BiFPN as a characteristic extraction network, and the ResNeSt convolution network is obtained by performing pre-training in an ImageNet deep learning network.
It can be understood that the characteristic pyramid BiFPN is used as a characteristic extraction network, so that the multi-scale characteristics of the insulator can be extracted, and the accuracy of insulator detection is improved.
And fourthly, dividing the insulator detection data set into a plurality of data sets according to the background category, and marking the corresponding background category on each data set.
The background category can be divided according to the intensity of brightness to simulate a plurality of scenes, such as night, rainy day, daytime and the like.
And step five, performing transfer learning training on each data set by using the initial insulator detection model to obtain insulator target detection models under different background categories.
In this embodiment, by performing migration learning training on each data set, insulator target detection models for different background categories can be obtained, so as to improve target detection accuracy under the corresponding background category.
And step six, calculating the detection accuracy of the insulator target detection models in different background categories for detecting corresponding data sets, if the accuracy is lower than a preset accuracy threshold, updating the network parameters of the insulator target detection models in the corresponding background categories, executing step five until the accuracy of the insulator sub-target detection models reaches the preset accuracy threshold, and outputting the optimal insulator target detection model.
In one embodiment, the first step specifically includes:
acquiring a plurality of distribution network inspection line images based on an unmanned aerial vehicle;
and setting and loading the distribution network inspection line image by using labelme software, and performing frame selection marking on the insulators in the distribution network inspection line image by using a rectangular frame to generate a marking file to form an insulator detection data set.
In a specific embodiment, the process of preprocessing the insulator detection data set by using the histogram equalization algorithm specifically includes:
calculating the frequency density of the distribution network routing inspection line image after histogram equalization according to the pixel number and the gray level depth of the distribution network routing inspection line image before histogram equalization,
in the formula, HB(D) Representing the frequency density, A, of the distribution network patrol line image after histogram equalization0The number of pixels of the distribution network routing inspection line image before histogram equalization is represented, and L represents the gray level depth of the distribution network routing inspection line image before histogram equalization;
gray value D of distribution network routing inspection line image A before histogram equalizationAGray value D in distribution network routing inspection line image after histogram equalization mapping through gray value transformation functionBI.e. DB＝f(DA)，f(DA) Representing a grey value transformation function, the grey value DAIn [0, D ]A]Inner pixel number and gray value DBIn [0, D ]B]The number of pixels in the same, i.e.
Solving the above formula to obtain a gray value transformation function f (D)A) In order to realize the purpose,
transforming the gray value into a function f (D)A) Discretizing to obtain a discretized gray value transformation function f' (D)A) Inputting each pixel in the distribution network patrol inspection line image A before histogram equalization into a discretization gray value transformation function f' (D)A) Obtaining a gray value D of a distribution network routing inspection line image B after histogram equalizationBThereby realizing histogram equalization of each distribution network routing inspection line image in the insulator detection data set, wherein the gray valueDBIn order to realize the purpose,
it will be appreciated that the gray value is transformed by the function f (D)A) Discretizing to rounding up the gray value transformation function and conveniently processing each pixel in the distribution network routing inspection line image A.
In a specific embodiment, the process of preprocessing the insulator detection data set by using the image filtering algorithm specifically includes:
filtering the distribution network inspection line image in the insulator detection data set based on a median filter, setting the pixel value of the ith row and the jth column in the distribution network inspection line image as p (i, j), filtering by the median filter to obtain the pixel value of the position as p (i, j),
med({p(i+x,j+y)|x∈{-1,0,1},y∈{-1,0,1}})
wherein med (-) is a function of median of elements in the set, and p (i + x, j + y) is a pixel value filtered by a median filter;
filtering and denoising the distribution network patrol inspection line image after filtering by using a Gaussian filter, wherein the pixel value after Gaussian filtering is
Where σ is the standard deviation of the given pixel value.
In a specific embodiment, the process of preprocessing the insulator detection data set by using image sharpening specifically includes:
calculating the second-order partial derivative of the pixel value of the distribution network routing inspection line image in the insulator detection data set as,
in the formula, p0(i, j) is the gray value of the ith row and the jth column pixel point of the image before the Laplace operator acts on the pixel point of the distribution network routing inspection line image;
the gray value of the Laplace operator after acting on the pixel points of the distribution network routing inspection line image is calculated as,
calculating the pixel value of the sharpened distribution network routing inspection line image as
Where k is the coefficient of the diffusion effect.
In one embodiment, the method further comprises:
dividing input X of the ResNeSt convolutional network into K groups of cardinal number units along the dimension of an input channel, dividing each group of cardinal number units into R groups of sub cardinal numbers, and dividing the input channel into G groups of sub channels, wherein G is KR;
based on ResNeSt convolution network, each group of sub-channels sequentially pass through 1 × 1 convolution layer and 3 × 3 convolution layer for feature extraction to obtain features of Uz，Uz∈RH×W×CZ is 1,2, …, G, H, W, C are each UzThree dimensions of (a);
the sum of the extracted features for each group of sub-cardinalities in the kth group of cardinality units is calculated as
calculating the weight of the c component of the ith group of sub-bases in the kth group of base unit
In order to realize the purpose,
in the formula (I), the compound is shown in the specification,
representing features s according to the kth group of cardinal unitskThe constructed ith group divides the weight of the c component;
weighting and summing the extracted features of each group of sub-cardinalities in each cardinality unit to obtain the features of the cardinality unit, wherein the c component of the features of the k-th cardinality unit is,
characterizing each radix unit
Splicing and adding the obtained result and the input X to obtain the characteristic Y extracted by the ResNeSt convolution network,
V＝Concat{V1,V2,…，VK}
wherein Concat {. denotes a splicing operation,
the conversion of the convolutional layer and the pooling layer is shown so that the number of channels of the input X coincides with V.
In one embodiment, the method further comprises:
constructing a loss function of an insulation sub-target detection model by using a label file of a distribution network inspection line image,
FL(Pt)＝-(1-Pt)γlog(Pt)
in the formula, a is a type label of a detection target, a ═ 1 indicates that a detection object is an insulator, a ═ 0 indicates a non-insulator, p indicates a probability that the detection object is an insulator, and γ is a given focusing parameter.
The method comprises the steps that Adam algorithm optimization is used, multi-scale features are input into a fully-connected neural network, an output layer of the neural network is provided with only one unit, and the value of the unit is the probability that a detected object corresponding to an image is an insulator.
The insulator detection method based on background classification and transfer learning provided by the invention is characterized in that an insulator detection data set is constructed and is preprocessed. Performing migration learning training on the preprocessed insulator detection data set by using a single-stage target detector to obtain an initial insulator detection model, dividing the insulator detection data set into a plurality of data sets according to background categories, labeling corresponding background categories on each data set, performing transfer learning training on each data set by using an initial insulator detection model, to obtain insulator target detection models under different background categories, to realize accurate detection of insulators under each background, to calculate the accuracy of detection of corresponding data sets by the insulator target detection models under different background categories, if the accuracy is lower than a preset accuracy threshold, network parameters of the insulator target detection model under the corresponding background category are updated, and migration learning is carried out until the accuracy of insulator detection under each background reaches a set value, so that the accuracy of insulator detection is improved.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, a division of a unit is merely a logical division, and an actual implementation may have another division, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
Units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (7)
1. An insulator detection method based on background classification and transfer learning is characterized by comprising the following steps:
the method comprises the steps of firstly, obtaining a plurality of distribution network routing inspection line images, marking the positions of insulators on each distribution network routing inspection line image by using a rectangular frame through labelme software, and generating a marking file to form an insulator detection data set, wherein the marking file comprises target position coordinates and category information;
secondly, preprocessing the insulator detection data set, wherein the preprocessing mode comprises a histogram equalization algorithm, an image filtering algorithm and image sharpening;
performing migration learning training on the preprocessed insulator detection data set by using a single-stage target detector to obtain an initial insulator detection model, wherein the single-stage target detector takes a ResNeSt convolutional network as a backbone network and takes a characteristic pyramid BiFPN as a characteristic extraction network, and the ResNeSt convolutional network is obtained by performing pre-training in an ImageNet deep learning network;
dividing the insulator detection data set into a plurality of data sets according to the background category, and labeling a corresponding background category on each data set;
step five, performing transfer learning training on each data set by using the initial insulator detection model to obtain insulator sub-target detection models under different background categories;
and step six, calculating the detection accuracy of the insulator target detection models in different background categories for detecting corresponding data sets, if the accuracy is lower than a preset accuracy threshold, updating the network parameters of the insulator target detection models in the corresponding background categories, executing step five until the accuracy of the insulator sub-target detection models reaches the preset accuracy threshold, and outputting the optimal insulator target detection model.
2. The background classification and transfer learning-based insulator detection method according to claim 1, wherein the first step specifically comprises:
acquiring a plurality of distribution network inspection line images based on an unmanned aerial vehicle;
and setting and loading a distribution network inspection line image by using labelme software, and performing frame selection marking on the insulators in the distribution network inspection line image by using a rectangular frame to generate a marking file to form an insulator detection data set.
3. The background classification and transfer learning-based insulator detection method according to claim 1, wherein the process of preprocessing the insulator detection data set by using a histogram equalization algorithm specifically comprises:
calculating the frequency density of the distribution network routing inspection line image after histogram equalization according to the pixel number and the gray level depth of the distribution network routing inspection line image before histogram equalization,
in the formula, HB(D) Representing the frequency density, A, of the distribution network patrol line image after histogram equalization0The number of pixels of the distribution network routing inspection line image before histogram equalization is represented, and L represents the gray level depth of the distribution network routing inspection line image before histogram equalization;
gray value D of distribution network routing inspection line image A before histogram equalizationAGray value D in distribution network routing inspection line image after histogram equalization mapping through gray value transformation functionBI.e. DB＝f(DA)，f(DA) Representing a grey value transformation function, the grey value DAIn [0, D ]A]Inner pixel number and gray value DBIn [0, D ]B]The number of pixels in the same, i.e.
Solving the above formula to obtain a gray value transformation function f (D)A) In order to realize the purpose,
transforming the gray value into a function f (D)A) Discretizing to obtain a discretized gray value transformation function f' (D)A) Inputting each pixel in the distribution network patrol inspection line image A before histogram equalization into the discretization gray value transformation function f' (D)A) Obtaining a gray value D of a distribution network routing inspection line image B after histogram equalizationBThereby realizing histogram equalization of each distribution network routing inspection line image in the insulator detection data set, wherein the gray value D isBIn order to realize the purpose,
4. the background classification and transfer learning-based insulator detection method according to claim 1, wherein the process of preprocessing the insulator detection data set by using an image filtering algorithm specifically comprises:
filtering the distribution network inspection line image in the insulator detection data set based on a median filter, setting the pixel value of the ith row and the jth column in the distribution network inspection line image as p (i, j), filtering by the median filter to obtain the pixel value of the position as p (i, j),
med({p(i+x,j+y)|x∈{-1,0,1},y∈{-1,0,1}})
wherein med (-) is a function of median of elements in the set, and p (i + x, j + y) is a pixel value filtered by a median filter;
filtering and denoising the distribution network patrol inspection line image after filtering by using a Gaussian filter, wherein the pixel value after Gaussian filtering is
Where σ is the standard deviation of the given pixel value.
5. The background classification and transfer learning-based insulator detection method according to claim 1, wherein the process of preprocessing the insulator detection data set by using image sharpening specifically comprises:
calculating the second-order partial derivative of the pixel value of the distribution network routing inspection line image in the insulator detection data set as,
in the formula, p0(i, j) is the gray value of the ith row and the jth column pixel point of the image before the Laplace operator acts on the pixel point of the distribution network routing inspection line image;
the gray value of the Laplace operator after acting on the pixel points of the distribution network routing inspection line image is calculated as,
calculating the pixel value of the sharpened distribution network routing inspection line image as
Where k is the coefficient of the diffusion effect.
6. The background classification and transfer learning-based insulator detection method according to claim 1, further comprising:
dividing input X of the ResNeSt convolutional network into K groups of cardinal number units along the dimension of an input channel, dividing each group of cardinal number units into R groups of sub cardinal numbers, and dividing the input channel into G groups of sub channels, wherein G is KR;
based on ResNeSt convolution network, each group of sub-channels sequentially pass through 1 × 1 convolution layer and 3 × 3 convolution layer for feature extraction to obtain features of Uz，Uz∈RH×W×CZ is 1,2, …, G, H, W, C are each UzThree dimensions of (a);
the sum of the extracted features for each group of sub-cardinalities in the kth group of cardinality units is calculated as
calculating the weight of the c component of the ith group of sub-bases in the kth group of base unit
In order to realize the purpose,
in the formula (I), the compound is shown in the specification,
representing features s according to the kth group of cardinal unitskThe constructed ith group divides the weight of the c component;
weighting and summing the extracted features of each group of sub-cardinalities in each cardinality unit to obtain the features of the cardinality unit, wherein the c component of the features of the k-th cardinality unit is,
characterizing each radix unit
Splicing and adding the obtained result and the input X to obtain the characteristic Y extracted by the ResNeSt convolution network,
V＝Concat{V1,V2,…,VK}
wherein Concat {. denotes a splicing operation,
the conversion of the convolutional layer and the pooling layer is shown so that the number of channels of the input X coincides with V.
7. The background classification and transfer learning-based insulator detection method according to claim 1, further comprising:
constructing a loss function of an insulation sub-target detection model by using the label file of the distribution network inspection line image,
FL(pt)＝-(1-pt)γlog(pt)
in the formula, a is a type label of a detection target, a ═ 1 indicates that a detection object is an insulator, a ═ 0 indicates a non-insulator, p indicates a probability that the detection object is an insulator, and γ is a given focusing parameter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111633194.5A CN114330548A (en) | 2021-12-28 | 2021-12-28 | Insulator detection method based on background classification and transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111633194.5A CN114330548A (en) | 2021-12-28 | 2021-12-28 | Insulator detection method based on background classification and transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114330548A true CN114330548A (en) | 2022-04-12 |
Family
ID=81014110
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111633194.5A Pending CN114330548A (en) | 2021-12-28 | 2021-12-28 | Insulator detection method based on background classification and transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114330548A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117274723A (en) * | 2023-11-22 | 2023-12-22 | 国网智能科技股份有限公司 | Target identification method, system, medium and equipment for power transmission inspection |
-
2021
- 2021-12-28 CN CN202111633194.5A patent/CN114330548A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117274723A (en) * | 2023-11-22 | 2023-12-22 | 国网智能科技股份有限公司 | Target identification method, system, medium and equipment for power transmission inspection |
| CN117274723B (en) * | 2023-11-22 | 2024-03-26 | 国网智能科技股份有限公司 | Target identification method, system, medium and equipment for power transmission inspection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111428748B (en) | HOG feature and SVM-based infrared image insulator identification detection method | |
| CN110059694B (en) | Intelligent identification method for character data in complex scene of power industry | |
| CN104483326B (en) | High-voltage line defects of insulator detection method and system based on depth belief network | |
| CN108681692B (en) | Method for identifying newly added buildings in remote sensing image based on deep learning | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| CN108960245B (en) | Tire mold character detection and recognition method, device, equipment and storage medium | |
| CN109272455B (en) | Image defogging method based on weak supervision generation countermeasure network | |
| CN106971152B (en) | Method for detecting bird nest in power transmission line based on aerial images | |
| CN108038846A (en) | Transmission line equipment image defect detection method and system based on multilayer convolutional neural networks | |
| CN107481292A (en) | The attitude error method of estimation and device of vehicle-mounted camera | |
| CN113205063A (en) | Visual identification and positioning method for defects of power transmission conductor | |
| CN109858480A (en) | A kind of digital instrument recognition methods | |
| CN110197116B (en) | Human behavior recognition method, device and computer readable storage medium | |
| CN108875645B (en) | Face recognition method under complex illumination condition of underground coal mine | |
| CN110133443B (en) | Power transmission line component detection method, system and device based on parallel vision | |
| CN113205507B (en) | Visual question answering method, system and server | |
| CN109033975A (en) | Birds detection, identification and method for tracing and device in a kind of monitoring of seashore | |
| CN112561899A (en) | Electric power inspection image identification method | |
| CN110969171A (en) | Image classification model, method and application based on improved convolutional neural network | |
| CN112686120B (en) | Power transmission line abnormity detection method based on unmanned aerial vehicle aerial image | |
| CN113327255A (en) | Power transmission line inspection image processing method based on YOLOv3 detection, positioning and cutting and fine-tune | |
| CN108615401B (en) | Deep learning-based indoor non-uniform light parking space condition identification method | |
| CN108345835B (en) | Target identification method based on compound eye imitation perception | |
| CN115841633A (en) | Power tower and power line associated correction power tower and power line detection method | |
| CN113435407A (en) | Small target identification method and device for power transmission system |
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 |