CN113838056A - Power equipment joint detection and identification method, system, equipment and storage medium - Google Patents

Power equipment joint detection and identification method, system, equipment and storage medium Download PDF

Info

Publication number
CN113838056A
CN113838056A CN202111429591.0A CN202111429591A CN113838056A CN 113838056 A CN113838056 A CN 113838056A CN 202111429591 A CN202111429591 A CN 202111429591A CN 113838056 A CN113838056 A CN 113838056A
Authority
CN
China
Prior art keywords
power equipment
electric power
auxiliary
main
power device
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111429591.0A
Other languages
Chinese (zh)
Other versions
CN113838056B (en
Inventor
焦飞
莫文昊
谈元鹏
蒲天骄
刘海莹
蔡常雨
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Electric Power Research Institute Co Ltd CEPRI
Original Assignee
China Electric Power Research Institute Co Ltd CEPRI
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China Electric Power Research Institute Co Ltd CEPRI filed Critical China Electric Power Research Institute Co Ltd CEPRI
Priority to CN202111429591.0A priority Critical patent/CN113838056B/en
Publication of CN113838056A publication Critical patent/CN113838056A/en
Application granted granted Critical
Publication of CN113838056B publication Critical patent/CN113838056B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method, a system, equipment and a storage medium for joint detection and identification of power equipment, wherein the method comprises the following steps: acquiring image information of the power equipment, and labeling each power equipment in the image information of the power equipment by using a rotary regression frame; inputting the image information of the marked power equipment into the trained network model to obtain the position information and the angle information of the rotary regression frame, the main power equipment set and the auxiliary power equipment set; according to the position information and the angle information of the rotating regression frame, each main power device in the main power device set is paired with an auxiliary power device in the auxiliary power device set to determine the auxiliary power device corresponding to the main power device, and the power device joint detection and identification facing to image data are completed.

Description

Power equipment joint detection and identification method, system, equipment and storage medium
Technical Field
The invention belongs to the technical field of automatic detection, and relates to a power equipment joint detection and identification method, a system, equipment and a storage medium.
Background
China is currently in a key stage of clean energy transformation, and the access of large-scale clean energy puts higher requirements on the stability of a power system. The stable operation of the power equipment is the basis and guarantee of the safe operation of the power network, and the real-time sensing and monitoring of the state of the power equipment have important significance for equipment state evaluation, potential hidden danger troubleshooting and real-time fault diagnosis. In recent years, the collection means of characteristic information of power equipment is continuously abundant, and the traditional sensing technology and equipment fault diagnosis method are difficult to adapt to the intelligent requirements of power grids.
The target detection is a key research problem in the field of pattern recognition, in recent years, the performance of a target detection model is remarkably improved based on the rapid development of the artificial intelligence technology, particularly the image recognition and the target detection of the deep learning technology, great breakthrough is made in the fields of power equipment recognition, defect detection and the like, and the development of the intelligent operation and detection technology is promoted. Currently, target detection algorithms are mainly classified into three categories: the method comprises the following steps that (1) a two-stage network represented by Faster R-CNN, Cascade R-CNN and the like adopts RPN to screen default frames, so that the network detection precision is high, but the model computation amount is large; the single-stage network represented by SSD, YOLO and the like directly carries out prediction based on a default frame on a feature map, and the model has a simple structure and high calculation speed, but has poor detection precision, particularly small target detection effect; the anchor-frame-free network represented by FCOS, CornerNet and the like does not preset a default frame, and directly regresses information such as a central point position, a detection width and a detection frame key point on a feature map.
The small-sized device and the shape characteristic fuzzy device are one of the difficulties in power equipment detection, are easy to miss detection and error detection, and are more difficult to detect faults. The operation state of the small device directly influences the operation safety of the power equipment, such as: small-size gold utensil of transmission line and substation equipment. The existing image-based target detection technology can independently detect each type, does not consider the logic relevance of adjacent detection targets, is difficult to combine with expert knowledge, and has low learning efficiency. At present, horizontal marking frames are adopted for equipment detection, and the horizontal rectangular frames have poor effect on detection of long and narrow objects and rotating objects. Along with the change of the angle of the target in the image, the horizontal detection frame has no rotation invariance, a large amount of noise is introduced into the characteristic and the detection area, the learning effect of the model is influenced, and the detection result is poor.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a power equipment joint detection and identification method, system, equipment and storage medium, which can accurately perform power equipment joint detection.
In order to achieve the purpose, the invention adopts the following technical scheme:
in one aspect, the present invention provides a power device joint detection and identification method, including:
acquiring image information of the power equipment to be identified;
inputting image information of the power equipment to be identified into a trained network model to obtain position information and angle information of a rotary regression frame, a main power equipment set and an auxiliary power equipment set, wherein the network model is trained by image samples of the power equipment marked by the rotary regression frame;
and pairing each main power device in the main power device set and the auxiliary power devices in the auxiliary power device set according to the position information and the angle information of the rotating regression frame to determine the auxiliary power devices corresponding to the main power devices, and finishing the joint detection and identification of the power devices facing the image data.
The further improvement of the power equipment joint detection and identification method of the invention is that:
the method further comprises the following steps: and labeling each electric power device in the image sample of the electric power device by using a rotating regression frame, wherein the labeling direction of the rotating regression frame is consistent with the shape direction of the electric power device, and simultaneously establishing the incidence relation labeling of the main electric power device and the auxiliary electric power devices.
The network model comprises a feature extraction network based on an attention mechanism, a multi-task detection network, a direction vector prediction branch and an angle prediction branch, wherein the angle prediction branch is formed by splicing a deformable convolution module and a first convolution module, and the direction vector prediction branch comprises a second convolution module.
The penalty function for an angle predicted branch is:
Figure 883788DEST_PATH_IMAGE001
wherein the content of the first and second substances,
Figure 586165DEST_PATH_IMAGE002
the labeled rotation angle of the ith rotation regression box,
Figure 416718DEST_PATH_IMAGE003
for the predicted angle of the i-th rotating regression box,
Figure 597163DEST_PATH_IMAGE004
in order to rotate the aspect ratio of the regression box,
Figure 462351DEST_PATH_IMAGE005
and
Figure 586163DEST_PATH_IMAGE006
is a hyper-parameter.
According to the position information and the angle information of the rotating regression frame, pairing each main electric power device in the main electric power device set with an auxiliary electric power device in the auxiliary electric power device set to determine the auxiliary electric power device corresponding to the main electric power device comprises the following specific processes:
and pairing each main electric power device in the main electric power device set and auxiliary electric power devices in the auxiliary electric power device set according to the position information and the angle information of the rotating regression frame, and determining the auxiliary electric power device corresponding to the main electric power device based on an electric power professional knowledge association distance electric power device pairing algorithm.
According to the position information and the angle information of the rotating regression frame, pairing each main electric power device in the main electric power device set with an auxiliary electric power device in the auxiliary electric power device set to determine the auxiliary electric power device corresponding to the main electric power device comprises the following specific processes:
calculating the association distance between the main power equipment and each auxiliary power equipment in the auxiliary power equipment set according to the position information and the angle information of the rotary regression frame;
selecting the one with the closest correlation distance
Figure 904012DEST_PATH_IMAGE007
The auxiliary power device is used as the auxiliary power device corresponding to the main power device.
Main power equipment
Figure 888149DEST_PATH_IMAGE008
And auxiliary power equipment
Figure 607843DEST_PATH_IMAGE009
Associated distance of
Figure 652022DEST_PATH_IMAGE010
Comprises the following steps:
Figure 457167DEST_PATH_IMAGE011
Figure 244995DEST_PATH_IMAGE012
wherein the content of the first and second substances,
Figure 819196DEST_PATH_IMAGE013
as main power equipmentMAnd auxiliary power equipmentDThe correlation bias of (a) is set,
Figure 34276DEST_PATH_IMAGE014
as main power equipment
Figure 61138DEST_PATH_IMAGE015
Center and accessory power equipment corresponding to rotary regression frame
Figure 652657DEST_PATH_IMAGE016
Corresponding to the euclidean distance between the center points of the rotation regression boxes,
Figure 81364DEST_PATH_IMAGE017
as main power equipment
Figure 467346DEST_PATH_IMAGE018
Center-to-accessory power equipment corresponding to rotating regression frame
Figure 247083DEST_PATH_IMAGE019
The perpendicular distance of the direction vector corresponding to the center of the rotating regression box,
Figure 376713DEST_PATH_IMAGE020
the weight corresponding to the distance between the two exceeds the parameter,
Figure 158462DEST_PATH_IMAGE021
is a master device
Figure 980924DEST_PATH_IMAGE022
The width in the current image is such that,
Figure 982379DEST_PATH_IMAGE023
is a master device
Figure 650120DEST_PATH_IMAGE024
Height in the current image.
In a second aspect of the present invention, the present invention provides a power device joint detection and identification system, including:
the marking module is used for acquiring image information of the power equipment to be identified;
the calculation module is used for inputting image information of the electric power equipment to be identified into the trained network model to obtain position information and angle information of the rotary regression frame, a main electric power equipment set and an auxiliary electric power equipment set, wherein the network model is formed by training image samples of the electric power equipment marked by the rotary regression frame;
and the pairing module is used for pairing each main electric power device in the main electric power device set with an auxiliary electric power device in the auxiliary electric power device set according to the position information and the angle information of the rotary regression frame so as to determine the auxiliary electric power device corresponding to the main electric power device and finish the joint detection and identification of the electric power devices facing the image data.
In another aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the power device joint detection and identification method when executing the computer program.
In a fourth aspect of the present invention, the present invention provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the computer program implements the steps of the power device joint detection and identification method.
The invention has the following beneficial effects:
the method, the system, the equipment and the storage medium for joint detection and identification of the power equipment mark each power equipment in an image sample of the power equipment by using the rotary regression frame during specific operation, wherein the rotary regression frame has rotary variability, so that noise is prevented from being introduced into characteristics and a detection area, and the problem that a traditional horizontal detection frame is poor in adaptability to a detection target shape is avoided, so that the accuracy of detection and identification of a network model is improved, the accuracy of a detection result is improved, and each main power equipment in the main power equipment set and each auxiliary power equipment in the auxiliary power equipment set are paired according to the position information and the angle information of the rotary regression frame to determine the auxiliary power equipment corresponding to the main power equipment, so that joint detection of the power equipment is realized, and the method, the system, the equipment and the storage medium are convenient and simple and have extremely high practicability.
Furthermore, the length-width ratio of the rotation regression frame is considered in the loss function corresponding to the angle prediction branch, so that the accuracy of the angle prediction of the rotation regression frame is improved.
Further, the auxiliary power equipment corresponding to the main power equipment is determined based on a power equipment pairing algorithm of the power professional knowledge correlation distance, so that the recall rate is improved.
Drawings
FIG. 1 is a schematic diagram of the present invention;
FIG. 2 is a flow chart of a method of the present invention;
FIG. 3 is a system block diagram of the present invention;
FIG. 4 is a schematic diagram of a first simulation experiment after a rotation regression box is marked;
FIG. 5 is a diagram of the test results of simulation experiment one;
FIG. 6a is a diagram of an effect of a second simulation experiment;
FIG. 6b is another effect diagram of the second simulation experiment;
FIG. 6c is another effect diagram of the second simulation experiment;
FIG. 6d is another effect diagram of the second simulation experiment;
FIG. 6e is another effect diagram of the second simulation experiment;
FIG. 6f is another effect diagram of the second simulation experiment;
FIG. 6g is another effect diagram of the second simulation experiment;
FIG. 6h is another effect diagram of the second simulation experiment;
FIG. 6i is another effect diagram of the second simulation experiment;
fig. 6j is another effect diagram of the second simulation experiment.
Wherein, 1 is a marking module, 2 is a calculating module, and 3 is a matching module.
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, not all of the embodiments, and are not intended to limit the scope of the present disclosure. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure. 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.
There is shown in the drawings a schematic block diagram of a disclosed embodiment in accordance with the invention. The figures are not drawn to scale, wherein certain details are exaggerated and possibly omitted for clarity of presentation. The shapes of various regions, layers and their relative sizes and positional relationships shown in the drawings are merely exemplary, and deviations may occur in practice due to manufacturing tolerances or technical limitations, and a person skilled in the art may additionally design regions/layers having different shapes, sizes, relative positions, according to actual needs.
Referring to fig. 1 and 2, the method for jointly detecting and identifying power equipment oriented to image data according to the present invention includes the following steps:
1) acquiring an image sample of the power equipment, labeling the power equipment by using a rotary regression frame, and training a network model by using the labeled image sample of the power equipment;
11) considering the resolution requirements of the network model and the size of the detection target, the image is scaled by a smaller direction pixel size 608 to form a plurality of image samples of a predetermined size while keeping the original image size unchanged.
The labeling of the rotation regression frame adopts a five-parameter labeling method, namely, for any rotation regression frame, a five-parameter labeling method is used
Figure 53420DEST_PATH_IMAGE025
Labeling and storing, wherein,
Figure 46784DEST_PATH_IMAGE026
the abscissa of the center point of the regression frame is rotated,
Figure 269955DEST_PATH_IMAGE027
is the ordinate of the center point of the rotating regression frame,
Figure 6966DEST_PATH_IMAGE028
in order to rotate the width of the regression box,
Figure 264772DEST_PATH_IMAGE029
in order to rotate the height of the regression box,
Figure 163458DEST_PATH_IMAGE030
for rotating the rotation angle of the regression frame, considering the periodicity of the angle and the replaceability of the width and the height, setting: width of the rotating regression frame
Figure 139504DEST_PATH_IMAGE031
The height of the rotating return frame is the long side of the rectangle
Figure 414628DEST_PATH_IMAGE032
Is the short side of the rectangle; rotation angle of rotating regression frame
Figure 526941DEST_PATH_IMAGE030
The positive direction of the longitudinal axis of the image forms an included angle with any long side of the rectangle clockwise
Figure 862107DEST_PATH_IMAGE033
(ii) a And the marking direction of the rotary regression frame is consistent with the shape direction of the electric power equipment, and meanwhile, the incidence relation marking of the main electric power equipment and the auxiliary electric power equipment is established, and the targets which cannot embody the characteristics due to shielding are completely marked according to the shape structure of the targets.
12) Training the network model by using the image sample;
the specific process is as follows: preprocessing an image sample in a manner that includes the imageSize mapping, random overturning and data splicing fusion, and carrying out normalization processing on the image sample according to a preset mean value and variance to obtain the size
Figure 59870DEST_PATH_IMAGE034
The input data is input into a network model for training to obtain the trained network model, and the network model comprises a feature extraction network, a multi-task detection network, a direction vector prediction branch and an angle prediction branch.
The feature extraction network uses a Hourglass network introducing attention mechanism, the feature extraction network is composed of a group of Hourglass modules with different feature sizes connected in series, the resolution of the next Hourglass module is half of that of the last Hourglass module, the Hourglass module with higher resolution tends to extract features of smaller samples, the Hourglass module with lower resolution tends to extract features of larger targets, and meanwhile, the automatic generation and feature information enhancement of a focus area are realized by combining spatial attention network branches with different sizes.
For the feature layer to be extracted
Figure 138685DEST_PATH_IMAGE035
Attention layer of
Figure 604039DEST_PATH_IMAGE036
Comprises the following steps:
Figure 844527DEST_PATH_IMAGE037
(1)
wherein the content of the first and second substances,
Figure 795166DEST_PATH_IMAGE038
for batch standardization layers, i.e.
Figure 677671DEST_PATH_IMAGE039
(2)
Figure 764576DEST_PATH_IMAGE040
(3)
Figure 175965DEST_PATH_IMAGE041
(4)
Figure 348321DEST_PATH_IMAGE042
(5)
Figure 768938DEST_PATH_IMAGE043
Standard convolutional layers with convolutional kernels of 1x1 size;
Figure 710349DEST_PATH_IMAGE044
is a void convolution layer with a convolution kernel of 3x3, i.e. a
Figure 558219DEST_PATH_IMAGE045
(6)
Attention layer results and feature layers
Figure 217871DEST_PATH_IMAGE046
After point-by-point multiplication, a group of feature map sets with different resolutions are obtained.
The multi-tasking detection network includes a center point and category branch, a center point offset branch, a width and height dimension branch, and a regression angle branch.
The length and width of the central point and the class branch are the same as those of the feature map, the depth of the central point and the class branch is a class to be detected, for the extracted feature map, detection results are obtained through 2 convolution network regressions of 3x3 and a Softmax function, the detection results are sorted according to scores of all points through non-maximum suppression calculation of 3x3, the sorted results are used as finally output rotation regression frames, and in order to establish the corresponding relation between a real detection frame and the feature map, a Gaussian kernel is used for establishing a heat map of the real detection frame
Figure 911020DEST_PATH_IMAGE047
;
Figure 972517DEST_PATH_IMAGE048
(7)
Wherein the content of the first and second substances,
Figure 991289DEST_PATH_IMAGE049
in order to be the size of the input image,Rin order to achieve the shrinkage ratio of the final feature layer,
Figure 872657DEST_PATH_IMAGE050
the hyper-parameters are used for adjusting the marked areas on the feature map; when the same feature point is included in a plurality of gaussian kernels, the highest value of the feature points is used as the feature value of the feature point.
The central point offset and the width and height prediction network depth are both 2, the prediction results are obtained through 2 convolution network regression of 3x3, all categories share the prediction results, in the training, the width and height prediction values are set only at the position of the central point of the real target, and the rest positions are filled with zero.
The direction vector prediction branch comprises a group of second convolution modules of 3x3, and is used for predicting a direction vector from a central point of a corresponding rotation regression frame of each auxiliary electric device to a central point of a corresponding rotation regression frame of the main electric device, wherein the direction vector is used for judging the dependency of the auxiliary electric devices, and the direction vectors of the main electric device and the auxiliary electric devices without the corresponding main electric device are zero.
The angle prediction branch comprises a combination of a deformable convolution module of 3x3 and a first convolution module of 3x3 and is used for enhancing the adaptability to a rotating target, the deformable convolution model enables a convolution kernel to freely extract any feature on a feature map by predicting the position offset of all convolution points corresponding to each feature point, and the angle prediction branch has better adaptability to complex detection results of heterogeneous objects, rotating targets and the like.
In consideration of the problems that the influence weight of the angle on the target detection precision is large and the linear angle regression is not uniform with the detection frame, the invention provides an angle prediction Loss function AR L1 Loss based on the aspect ratio of the target frame, namely,
Figure 133612DEST_PATH_IMAGE051
Figure 49615DEST_PATH_IMAGE051
(8)
wherein the content of the first and second substances,
Figure 239288DEST_PATH_IMAGE052
the labeled rotation angle of the ith rotation regression box,
Figure 873532DEST_PATH_IMAGE053
for the predicted angle of the i-th rotating regression box,
Figure 174063DEST_PATH_IMAGE054
for rotating the aspect ratio of the regression box, b and
Figure 944573DEST_PATH_IMAGE055
in order to be a hyper-parameter,
Figure 305147DEST_PATH_IMAGE056
。
2) acquiring image information of the power equipment to be identified;
3) inputting image information of the electric power equipment to be identified into the trained network model to obtain position information and angle information of the rotary regression frame, a main electric power equipment set and an auxiliary electric power equipment set;
4) and pairing each main power device in the main power device set and the auxiliary power devices in the auxiliary power device set according to the position information and the angle information of the rotating regression frame to determine the auxiliary power devices corresponding to the main power devices, and finishing the joint detection and identification of the power devices facing the image data.
The specific process of the step 4) is as follows:
and pairing each main power device in the main power device set and the auxiliary power devices in the auxiliary power device set according to the position information and the angle information of the rotating regression frame, and determining the auxiliary power devices corresponding to the main power devices based on a power device pairing algorithm of Correlation Distance (CoD) based on power professional knowledge.
According to the electric power professional knowledge, a distribution rule of each main electric power device and auxiliary electric power devices thereof is constructed, wherein the distribution rule comprises the number of the auxiliary devices
Figure 161108DEST_PATH_IMAGE057
And correlation bias
Figure 265330DEST_PATH_IMAGE058
And the like.
For main power equipmentMAnd its auxiliary power equipmentDBuilding a set
Figure 155926DEST_PATH_IMAGE059
、
Figure 687401DEST_PATH_IMAGE060
And
Figure 30658DEST_PATH_IMAGE061
wherein, in the step (A),
Figure 672992DEST_PATH_IMAGE062
is as followspThe main electric power device corresponds to the center point of the rotating regression box,
Figure 418094DEST_PATH_IMAGE063
is as followsqEach auxiliary power device corresponds to the central point of the rotating regression frame,
Figure 120471DEST_PATH_IMAGE064
is as followsqThe direction vector of the auxiliary electric power equipment is predicted, and the main electric power equipment
Figure 951023DEST_PATH_IMAGE065
And auxiliary power equipment
Figure 635864DEST_PATH_IMAGE066
Associated distance of
Figure 501051DEST_PATH_IMAGE067
Comprises the following steps:
Figure 374329DEST_PATH_IMAGE068
（9）
Figure 426599DEST_PATH_IMAGE069
（10）
Figure 676315DEST_PATH_IMAGE070
（11）
Figure 396009DEST_PATH_IMAGE071
（12）
wherein the content of the first and second substances,
Figure 440189DEST_PATH_IMAGE072
as main power equipmentMAnd auxiliary power equipmentDIs biased when
Figure 979754DEST_PATH_IMAGE072
The larger, the main power plantMAnd auxiliary power equipmentDThe weaker the association of (a);
Figure 33161DEST_PATH_IMAGE073
as main power equipmentMCorresponding to the center point of the rotary regression frame and the auxiliary power equipmentDCorresponding to Euclidean distance between the central points of the rotation regression frames;
Figure 607362DEST_PATH_IMAGE074
as main power equipmentMCorresponding to the central point of the rotary regression frame to the accessory power equipmentDThe vertical distance of the direction vector corresponding to the center point of the rotation regression box,
Figure 556863DEST_PATH_IMAGE075
the weight corresponding to the distance between the two exceeds the parameter,
Figure 849304DEST_PATH_IMAGE076
is a master device
Figure 175243DEST_PATH_IMAGE077
The width in the current image is such that,
Figure 869530DEST_PATH_IMAGE078
is a master device
Figure 255512DEST_PATH_IMAGE079
Height in the current image.
Determining a primary power device
Figure 268205DEST_PATH_IMAGE077
Corresponding to closest correlation distance
Figure 397835DEST_PATH_IMAGE080
An auxiliary power equipmentDWill associate the nearest
Figure 946628DEST_PATH_IMAGE080
An auxiliary power equipmentDAnd the auxiliary power equipment serving as the main power equipment completes the joint detection and identification of the power equipment for the image data.
Simulation experiment I
The transformer is widely arranged in the power system, whether the transformer can stably operate directly influences the operation safety of the power system, and the transformer operation state detection and fault analysis prejudgment have important practical values. In an actual working environment, due to factors such as background confusion and interference between adjacent equipment and a line, the detection difficulty of the transformer and the auxiliary equipment thereof is high, and the conditions of missing detection and false detection are easy to occur.
Referring to fig. 4, the method includes the steps of firstly acquiring images of a transformer and accessory equipment of the transformer, carrying out rotary regression frame labeling on the transformer, an insulating sleeve, an oil conservator, a breather, a radiator and other equipment, and establishing incidence relation labeling of relevant accessory equipment and the transformer, wherein the labeling of the transformer needs to cover all accessory equipment areas including the oil conservator and the cooler, the labeling direction of the insulating sleeve is the direction of the head of the equipment, then, the integrated and labeled data are sent to a network model for training, three Hourglass modules connected in series are arranged as feature extractors according to the size of the equipment in a sample, the three Hourglass modules are respectively connected with an attention network layer, and each attention layer is independently trained and predicted. And (4) transmitting the feature extraction result to output the detection results of all types of equipment. And finally, carrying out joint detection on the transformer body and related equipment thereof, considering the structural characteristics of the transformer body, identifying that no more than one oil conservator, radiator and respirator and no more than 8 sleeves are needed by one transformer, and further screening detection results based on the characteristics, referring to fig. 5.
Through experiments, the method effectively improves the identification accuracy of various devices through the combined learning of the related devices of the transformer, realizes the fine detection of the devices such as the sleeve, the oil conservator and the like through the direction self-adaptive detection, and can effectively reduce the interference of background information.
Simulation experiment two
The simulation experiment carries out joint detection on the transmission tower and the shadow, and the specific process is as follows:
firstly, collecting image data of a transmission tower and shadows thereof, labeling the image data by using a rotating regression frame, considering the requirement of a subsequent model on resolution and the size of a detection target, and carrying out image scaling by using a pixel size in a smaller direction as 608 under the condition of keeping the size of an original image unchanged to form a plurality of image samples with specified sizes.
The labeling of the rotation regression frame adopts a five-parameter labeling method, namely, for any rotation regression frame, a five-parameter labeling method is used
Figure 706774DEST_PATH_IMAGE081
Labeling and storing, wherein,
Figure 708228DEST_PATH_IMAGE082
is the abscissa of the center point of the rotating frame,
Figure 641549DEST_PATH_IMAGE083
Is a longitudinal coordinate of the center point of the rotating frame,
Figure 44848DEST_PATH_IMAGE084
Is the width of the rotating frame,
Figure 38212DEST_PATH_IMAGE085
Is the height of the rotating frame,
Figure 261383DEST_PATH_IMAGE086
The rotation angle of the rotating frame. Considering the periodicity and the width and height replaceability of the angle, the following rules are set to ensure that the labeling result is unique and the training result does not generate ambiguity: setting the width of the rotary regression frame
Figure 998395DEST_PATH_IMAGE084
The height of the rotating return frame is the long side of the rectangle
Figure 256201DEST_PATH_IMAGE085
Is the short side of the rectangle; rotation angle of rotating regression frame
Figure 154887DEST_PATH_IMAGE086
The positive direction of the longitudinal axis of the image forms an included angle with any long side of the rectangle clockwise,
Figure 130933DEST_PATH_IMAGE087
(ii) a The direction of the marking of the rotating regression frame is consistent with the top orientation of the transmission tower and the shadow thereof, and the target which cannot reflect the characteristics due to shielding is completely marked according to the shape structure of the target.
Training the model according to the labeling data, and preprocessing the original picture, wherein the method comprises the following steps: image size mapping, random turning, data splicing and fusion and the like. Further normalizing the data according to the given mean value and variance to obtain the size of
Figure 904592DEST_PATH_IMAGE088
The input data is sent to a network model for training, as shown in FIG. 1The model is divided into a feature extraction network and a multi-task detection network, which are respectively responsible for high-dimensional feature extraction and regression of detection parameters.
The feature extraction network uses a Hourglass network with attention-drawing mechanism and consists of two Hourglass modules with different feature sizes connected in series. Considering that data distribution of a transmission tower and shadow has obvious characteristics, the shadow area and the length-width ratio are larger than those of a tower body, two characteristic networks with different sizes are established, a network branch with higher resolution tends to perform characteristic extraction on a smaller sample in the tower body, a network branch with lower resolution tends to perform characteristic extraction on a complex larger target, and automatic generation and characteristic information enhancement of an attention area are realized by combining spatial attention network branches with different sizes.
For the feature layer to be extracted
Figure 16904DEST_PATH_IMAGE035
Attention layer of
Figure 352070DEST_PATH_IMAGE089
Comprises the following steps:
Figure 549834DEST_PATH_IMAGE037
(1)
wherein the content of the first and second substances,
Figure 628648DEST_PATH_IMAGE038
for batch standardization layers, i.e.
Figure 595467DEST_PATH_IMAGE090
(2)
Figure 101535DEST_PATH_IMAGE091
(3)
Figure 786594DEST_PATH_IMAGE092
(4)
Figure 403520DEST_PATH_IMAGE093
(5)
Wherein the content of the first and second substances,
Figure 490425DEST_PATH_IMAGE094
standard convolutional layers with convolutional kernels of 1x1 size;
Figure 167394DEST_PATH_IMAGE095
is a void convolution layer with a convolution kernel of 3x3, i.e. a
Figure 339749DEST_PATH_IMAGE096
(6)
Results and feature layers of the attention layer
Figure 760366DEST_PATH_IMAGE046
After point-by-point multiplication, a group of feature map sets with different resolutions are output for the multi-task branch regression. The multitask branch network comprises four parts: a center point and category branch, a center point offset branch, a width and height dimension branch, and a regression angle branch. Each part is independently calculated on the multi-scale characteristic diagram, and a small target and a large-size target are respectively detected.
The length and width of the central point and the branch of the category size are the same as those of the feature map, and the depth is the category to be detected. For the extracted feature map, the detection results are obtained through 2 convolution network regressions of 3 × 3 and a Softmax function, and are sorted according to scores of points through a 3 × 3 non-maximum suppression calculation to serve as a regression frame of final output. In order to establish the corresponding relation between the real detection frame and the characteristic diagram, a Gaussian kernel is used for establishing the heat map of the real detection frame
Figure 701777DEST_PATH_IMAGE097
Wherein, in the step (A),
Figure 549648DEST_PATH_IMAGE098
(7)
wherein, among others,
Figure 209299DEST_PATH_IMAGE099
in order to be the size of the input image,Rin order to achieve the shrinkage ratio of the final feature layer,
Figure 666563DEST_PATH_IMAGE100
in order to adjust the hyper-parameters of the marked areas on the feature map, if the same feature point is included in a plurality of Gaussian kernels, the highest value of the feature points is used as the feature value of the feature point.
The central point offset and the width and height prediction network depth are both 2, the prediction results are obtained through 2 convolution network regression of 3x3, all categories share the prediction results, in the training, the width and height and the offset prediction values are set only at the position of the central point of the real target, and the rest positions are filled with zero and do not participate in the training.
The angle prediction adopts a deformable convolution module of 3x3 and a convolution module of 3x3 to be combined to enhance the adaptability to the rotating target. The deformable convolution can freely extract any feature on the feature map by predicting the position offset of all the convolution points corresponding to each feature point, and has better adaptability to complex detection results of heterogeneous objects, rotating targets and the like.
Considering the problems that the influence weight of an angle on the target detection precision is large and the intersection ratio of linear angle regression and a detection frame is not uniform, the invention provides an angle prediction Loss function AR L1 Loss based on the length-width ratio of a target frame:
Figure 728060DEST_PATH_IMAGE101
(8)
wherein the content of the first and second substances,
Figure 481252DEST_PATH_IMAGE102
the labeled rotation angle of the ith rotation regression box,
Figure 628200DEST_PATH_IMAGE103
for the predicted angle of the i-th rotating regression box,
Figure 390620DEST_PATH_IMAGE104
for rotating the aspect ratio of the regression box, b and
Figure 306623DEST_PATH_IMAGE055
in order to be a hyper-parameter,
Figure 496296DEST_PATH_IMAGE105
。
finally, according to the one-to-one correspondence relation and the spatial proximity of the transmission tower and the shadow, a transmission tower and shadow pairing algorithm based on the central point distance is provided, and for two kinds of electric power equipment with determined relevance
Figure 864960DEST_PATH_IMAGE106
Dividing the network detection results into sets
Figure 165492DEST_PATH_IMAGE107
Wherein, in the step (A),
Figure 201581DEST_PATH_IMAGE108
is as followsiAn
Figure 562155DEST_PATH_IMAGE109
The center point of the device-like device,
Figure 418115DEST_PATH_IMAGE110
is as followsjAn
Figure 522338DEST_PATH_IMAGE111
A device-like center point. Finding collections in turnTAndSmiddle closest set of points
Figure 412933DEST_PATH_IMAGE112
And taking it out of the set, and keeping it as a pair of targets until the setTOrSHas no element, orTAndSthe distance between the middle and the nearest points exceeds a set threshold value and exceeds a parameter.
Referring to fig. 6a to 6j, the method can effectively identify the transmission tower and the type thereof under the remote sensing image, the detection accuracy and the AP are both over 90%, and the problems of difficult foreground and background separation and fuzzy type information are solved well. Meanwhile, the method can play an important role in the scenes of regional power transmission line statistics, remote tower anti-toppling and operation and maintenance management, power transmission line and tower construction operation and the like, and has practical application value.
Example two
Referring to fig. 3, the power equipment joint detection and identification system of the present invention includes:
the marking module 1 is used for acquiring image information of the electric power equipment to be identified;
the calculation module 2 is used for inputting image information of the electric power equipment to be identified into the trained network model to obtain position information and angle information of the rotary regression frame, a main electric power equipment set and an auxiliary electric power equipment set, wherein the network model is trained by image samples of the electric power equipment marked by the rotary regression frame;
and the pairing module 3 is used for pairing each main power device in the main power device set and the auxiliary power device in the auxiliary power device set according to the position information and the angle information of the rotating regression frame so as to determine the auxiliary power device corresponding to the main power device and complete the power device joint detection and identification facing the image data.
EXAMPLE III
A computer device comprising a storage, a processor and a computer program stored in the storage and executable on the processor, wherein the processor implements the steps of the power device joint detection and identification method when executing the computer program, wherein the storage may include a memory such as a high-speed random access memory, and may further include a nonvolatile memory such as at least one disk storage; the processor, the network interface and the memory are connected with each other through an internal bus, wherein the internal bus can be an industrial standard system structure bus, a peripheral component interconnection standard bus, an extended industrial standard structure bus and the like, and the bus can be divided into an address bus, a data bus, a control bus and the like. The memory is used for storing programs, and particularly, the programs can comprise program codes which comprise computer operation instructions. The memory may include both memory and non-volatile storage and provides instructions and data to the processor.
Example four
A computer-readable storage medium, in which a computer program is stored, which, when being executed by a processor, carries out the steps of the power device joint detection identification method, in particular, but not exclusively, volatile memory and/or non-volatile memory, for example. The volatile memory may include Random Access Memory (RAM) and/or cache memory (cache), among others. The non-volatile memory may include a Read Only Memory (ROM), hard disk, flash memory, optical disk, magnetic disk, and the like.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.

Claims (10)

1. A power equipment joint detection and identification method is characterized by comprising the following steps:
acquiring image information of the power equipment to be identified;
inputting image information of the power equipment to be identified into a trained network model, and obtaining position information and angle information of a rotary regression frame, a main power equipment set and an auxiliary power equipment set, wherein the network model is trained by image samples of the power equipment marked by the rotary regression frame;
and pairing each main power device in the main power device set and the auxiliary power devices in the auxiliary power device set according to the position information and the angle information of the rotating regression frame to determine the auxiliary power devices corresponding to the main power devices, and finishing the joint detection and identification of the power devices facing the image data.
2. The power equipment joint detection and identification method according to claim 1, further comprising:
and labeling each electric power device in the image sample of the electric power device by using a rotating regression frame, wherein the labeling direction of the rotating regression frame is consistent with the shape direction of the electric power device, and simultaneously establishing the incidence relation labeling of the main electric power device and the auxiliary electric power devices.
3. The power equipment joint detection and identification method according to claim 1, wherein the network model comprises an attention-based feature extraction network, a multitask detection network, a direction vector prediction branch and an angle prediction branch, wherein the angle prediction branch is spliced by a deformable convolution module and a first convolution module, and the direction vector prediction branch comprises a second convolution module.
4. The power equipment joint detection and identification method according to claim 3, wherein the loss function corresponding to the angle prediction branch is:
Figure 677490DEST_PATH_IMAGE001
wherein the content of the first and second substances,
Figure 88880DEST_PATH_IMAGE002
the labeled rotation angle of the ith rotation regression box,
Figure 261235DEST_PATH_IMAGE003
for the predicted angle of the i-th rotating regression box,
Figure 416273DEST_PATH_IMAGE004
for rotating the length of the return frameThe width ratio of the width of the film,
Figure 623263DEST_PATH_IMAGE005
and
Figure 205554DEST_PATH_IMAGE006
is a hyper-parameter.
5. The method for jointly detecting and identifying electric power equipment according to claim 1, wherein the specific process of pairing each main electric power equipment in the main electric power equipment set with an auxiliary electric power equipment in the auxiliary electric power equipment set according to the position information and the angle information of the rotating regression frame to determine the auxiliary electric power equipment corresponding to the main electric power equipment is as follows:
and matching each main electric power device in the main electric power device set and auxiliary electric power devices in the auxiliary electric power device set according to the position information and the angle information of the rotary regression frame based on the electric power professional knowledge correlation distance to determine the auxiliary electric power devices corresponding to the main electric power devices.
6. The method for jointly detecting and identifying electric power equipment according to claim 1, wherein the specific process of pairing each main electric power equipment in the main electric power equipment set with an auxiliary electric power equipment in the auxiliary electric power equipment set according to the position information and the angle information of the rotating regression frame to determine the auxiliary electric power equipment corresponding to the main electric power equipment is as follows:
calculating the association distance between the main power equipment and each auxiliary power equipment in the auxiliary power equipment set according to the position information and the angle information of the rotary regression frame;
selecting the one with the closest correlation distance
Figure 130785DEST_PATH_IMAGE007
The auxiliary power device is used as the auxiliary power device corresponding to the main power device.
7. The method of claim 6The power equipment joint detection and identification method is characterized in that the main power equipment
Figure 89514DEST_PATH_IMAGE008
And auxiliary power equipment
Figure 151011DEST_PATH_IMAGE009
Associated distance of
Figure 169782DEST_PATH_IMAGE010
Comprises the following steps:
Figure 316730DEST_PATH_IMAGE011
Figure 547991DEST_PATH_IMAGE012
wherein the content of the first and second substances,
Figure 729574DEST_PATH_IMAGE013
as main power equipmentMAnd auxiliary power equipmentDThe correlation bias of (a) is set,
Figure 919247DEST_PATH_IMAGE014
as main power equipment
Figure 553490DEST_PATH_IMAGE015
Center and accessory power equipment corresponding to rotary regression frame
Figure 854022DEST_PATH_IMAGE016
Corresponding to the euclidean distance between the center points of the rotation regression boxes,
Figure 123067DEST_PATH_IMAGE017
as main power equipment
Figure 483641DEST_PATH_IMAGE018
Center-to-accessory power equipment corresponding to rotating regression frame
Figure 605181DEST_PATH_IMAGE019
The perpendicular distance of the direction vector corresponding to the center of the rotating regression box,
Figure 443824DEST_PATH_IMAGE020
the weight corresponding to the distance between the two exceeds the parameter,
Figure 334419DEST_PATH_IMAGE021
is a master device
Figure 865895DEST_PATH_IMAGE022
The width in the current image is such that,
Figure 209151DEST_PATH_IMAGE023
is a master device
Figure 117064DEST_PATH_IMAGE024
Height in the current image.
8. A power equipment joint detection and identification system is characterized by comprising:
the marking module is used for acquiring image information of the power equipment to be identified;
the calculation module is used for inputting image information of the electric power equipment to be identified into the trained network model to obtain position information and angle information of the rotary regression frame, a main electric power equipment set and an auxiliary electric power equipment set, wherein the network model is formed by training image samples of the electric power equipment marked by the rotary regression frame;
and the pairing module is used for pairing each main electric power device in the main electric power device set with an auxiliary electric power device in the auxiliary electric power device set according to the position information and the angle information of the rotary regression frame so as to determine the auxiliary electric power device corresponding to the main electric power device and finish the joint detection and identification of the electric power devices facing the image data.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the power device joint detection and identification method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, in which a computer program is stored, which, when being executed by a processor, carries out the steps of the power device joint detection and identification method according to any one of claims 1 to 7.
CN202111429591.0A 2021-11-29 2021-11-29 Power equipment joint detection and identification method, system, equipment and storage medium Active CN113838056B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111429591.0A CN113838056B (en) 2021-11-29 2021-11-29 Power equipment joint detection and identification method, system, equipment and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111429591.0A CN113838056B (en) 2021-11-29 2021-11-29 Power equipment joint detection and identification method, system, equipment and storage medium

Publications (2)

Publication Number Publication Date
CN113838056A true CN113838056A (en) 2021-12-24
CN113838056B CN113838056B (en) 2022-03-01

Family

ID=78971866

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111429591.0A Active CN113838056B (en) 2021-11-29 2021-11-29 Power equipment joint detection and identification method, system, equipment and storage medium

Country Status (1)

Country Link
CN (1) CN113838056B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110458005A (en) * 2019-07-02 2019-11-15 重庆邮电大学 It is a kind of based on the progressive invariable rotary method for detecting human face with pseudo-crystalline lattice of multitask
EP3623882A1 (en) * 2018-09-13 2020-03-18 Siemens Aktiengesellschaft Identifying type and alignment of a workpiece
CN111382687A (en) * 2020-03-05 2020-07-07 平安科技（深圳）有限公司 Face detection method and system
CN111414931A (en) * 2019-12-31 2020-07-14 杭州电子科技大学 Multi-branch multi-scale small target detection method based on image depth
CN112542800A (en) * 2020-11-24 2021-03-23 广西电网有限责任公司电力科学研究院 Method and system for identifying transmission line fault
CN113158789A (en) * 2021-03-15 2021-07-23 华南理工大学 Target detection method, system, device and medium for remote sensing image

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3623882A1 (en) * 2018-09-13 2020-03-18 Siemens Aktiengesellschaft Identifying type and alignment of a workpiece
CN110458005A (en) * 2019-07-02 2019-11-15 重庆邮电大学 It is a kind of based on the progressive invariable rotary method for detecting human face with pseudo-crystalline lattice of multitask
CN111414931A (en) * 2019-12-31 2020-07-14 杭州电子科技大学 Multi-branch multi-scale small target detection method based on image depth
CN111382687A (en) * 2020-03-05 2020-07-07 平安科技（深圳）有限公司 Face detection method and system
CN112542800A (en) * 2020-11-24 2021-03-23 广西电网有限责任公司电力科学研究院 Method and system for identifying transmission line fault
CN113158789A (en) * 2021-03-15 2021-07-23 华南理工大学 Target detection method, system, device and medium for remote sensing image

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
刘钦: "基于大数据的电力设备数据融合与应用", 《信息化建设》 *

Also Published As

Publication number Publication date
CN113838056B (en) 2022-03-01

Similar Documents

Publication Publication Date Title
CN106127204B (en) A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks
CN109712118A (en) A kind of substation isolating-switch detection recognition method based on Mask RCNN
CN108734143A (en) A kind of transmission line of electricity online test method based on binocular vision of crusing robot
Liang et al. A real-time detection framework for on-tree mango based on SSD network
CN108710913A (en) A kind of switchgear presentation switch state automatic identification method based on deep learning
CN111898432B (en) Pedestrian detection system and method based on improved YOLOv3 algorithm
US11810348B2 (en) Method for identifying power equipment targets based on human-level concept learning
CN114821014B (en) Multi-mode and countermeasure learning-based multi-task target detection and identification method and device
CN109784375A (en) Adaptive transformer part detection recognition method based on Faster RCNN
Tian et al. Research on flower image classification method based on YOLOv5
Yi et al. Detection and localization for lake floating objects based on CA-faster R-CNN
Zhao et al. G3DOA: Generalizable 3D descriptor with overlap attention for point cloud registration
CN114219753A (en) Power equipment surface defect detection method based on deep learning and terminal
CN113838056B (en) Power equipment joint detection and identification method, system, equipment and storage medium
CN116205905B (en) Power distribution network construction safety and quality image detection method and system based on mobile terminal
Zhang et al. Recognition and detection of wolfberry in the natural background based on improved YOLOv5 network
CN116912774A (en) Infrared image target identification method, electronic device and storage medium of power transmission and transformation equipment based on edge calculation
CN112818836B (en) Method and system for detecting personnel target of transformer substation scene
Wang et al. Research on surface defect detection of disc harrow based on YOLOv5
Feng et al. Recognition of insulator based on yolov5 algorithm
Lu et al. MStrans: Multiscale Vision Transformer for Aerial Objects Detection
Wu et al. Liquid Drip Detection in Power Plant based on Machine Vision
Liao Road Damage Intelligent Detection with Deep Learning Techniques
Chen et al. Smart Grid Image Recognition Based on Neural Network and SIFT Algorithm
Hong et al. Target Registration of Electric Power Installation Based on Hybrid Network Model Under the Application of Intelligent Inspection Application

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant