CN112669264A - Artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of distribution network line - Google Patents
Artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of distribution network line Download PDFInfo
- Publication number
- CN112669264A CN112669264A CN202011497412.2A CN202011497412A CN112669264A CN 112669264 A CN112669264 A CN 112669264A CN 202011497412 A CN202011497412 A CN 202011497412A CN 112669264 A CN112669264 A CN 112669264A
- Authority
- CN
- China
- Prior art keywords
- defect
- distribution network
- image
- fpn
- sample library
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 119
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000007689 inspection Methods 0.000 title claims abstract description 49
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 23
- 230000003416 augmentation Effects 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 7
- 230000003190 augmentative effect Effects 0.000 claims description 4
- 238000004891 communication Methods 0.000 claims 1
- 238000010276 construction Methods 0.000 abstract description 11
- 238000013528 artificial neural network Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 8
- 238000003745 diagnosis Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 239000004020 conductor Substances 0.000 description 2
- 239000012212 insulator Substances 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000001681 protective effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 125000006850 spacer group Chemical group 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides an artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of a distribution network line, and relates to the technical field of distribution networks. According to the method, an original sample library is obtained, wherein the original sample library comprises typical component images and defect images of distribution network lines; carrying out image augmentation on the defect image to obtain an artificial sample, wherein the original sample library and the artificial sample form a typical part of a distribution network line and a defect sample library; acquiring a judging model based on the typical parts of the distribution network lines, the defect sample library and the FPN + Faster-RCNN network; and detecting the to-be-detected patrol video image based on the judging model. The invention expands the defect image data samples with less samples, unobvious characteristics and the like by image augmentation of the defect image, provides enough data for the construction of a study and judgment model, and improves the accuracy of the model.
Description
Technical Field
The invention relates to the technical field of distribution networks, in particular to an artificial intelligence defect identification method and system for distribution network line unmanned aerial vehicle routing inspection.
Background
The distribution network refers to an electric power network which receives electric energy from a transmission network or a regional power plant and distributes the electric energy to various users on site through distribution facilities or step by step according to voltage. The power distribution network consists of overhead lines, cables, towers, distribution transformers, isolating switches, reactive power compensators, accessory facilities and the like, and plays a role in distributing electric energy in a power network. The distribution network line is generally exposed outdoors, and is inevitably influenced by natural environment and social environment during operation. Extreme weather such as thunder and lightning, stormy weather can destroy or damage distribution network equipment or cause accidents such as electric shock. In the urban and rural integrated construction process, unreasonable greening plant construction problems and construction problems of illegal buildings exist; the problems of manual theft, wiring disorder, intentional damage to distribution network lines and the like can influence the normal operation of distribution equipment and even cause the occurrence of safety accidents and legal risks.
Because the requirement of distribution network equipment high reliability, equipment power failure test project is few, and daily inspection is main equipment state management and control means, and the operation and maintenance personnel is comparatively urgent to effective and efficient inspection means demand. In the distribution network inspection process, a large amount of manpower can be wasted aiming at the inspection work which is just needed in daily life, the inspection process is difficult to avoid making mistakes, the inspection environment is severe, and the personal safety of inspection personnel is sometimes injured. Therefore, drones are beginning to be used for distribution network line inspection.
When unmanned aerial vehicle patrols and examines, the automatic identification and the intelligent diagnosis technique of equipment trouble are especially important. At present, the domestic research in the field of power image diagnosis mainly aims at detecting and identifying common line defect types or specific problems, such as strand breakage of a power transmission line, screw falling, insulator damage, vibration damper damage, wire clamp damage and positioning of a spacer, and if other conditions occur, identification errors can occur, so that the identification accuracy of distribution network defects is low.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides an artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of a distribution network line, and solves the technical problem of low distribution network defect identification accuracy in the prior art.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
the invention provides an artificial intelligence defect identification method for distribution network line unmanned aerial vehicle routing inspection, which comprises the following steps:
acquiring an original sample library, wherein the original sample library comprises typical component images and defect images of distribution network lines; carrying out image augmentation on the defect image to obtain an artificial sample, wherein the original sample library and the artificial sample form a typical part of a distribution network line and a defect sample library;
acquiring a judging model based on the typical parts of the distribution network lines, the defect sample library and the FPN + Faster-RCNN network;
and detecting the to-be-detected patrol video image based on the judging model.
Preferably, the image augmenting the defect image comprises:
image augmentation of defect images is performed based on generation of a countermeasure network.
Preferably, the FPN + Faster-RCNN network comprises: the device comprises an FPN layer and a Faster-RCNN layer, wherein the FPN layer is used for extracting routing inspection image defect characteristics of typical parts of the distribution network line, and the Faster-RCNN layer is used for identifying typical equipment parts of the distribution network line.
Preferably, the acquiring and judging model based on the distribution network line typical component, the defect sample library and the FPN + fast-RCNN network includes:
constructing a triple group based on the typical component of the distribution network line and the defect sample library;
and training the FPN + Faster-RCNN network based on the triples to obtain a judging model.
Preferably, the constructing a triplet based on the typical component of the distribution network line and the defect sample library includes:
randomly selecting two image blocks from all classes of typical components of distribution network lines and a defect sample library, and selecting one image block from the other class, wherein each triple contains two similar image blocks and one dissimilar image block.
Preferably, the training of the FPN + fast-RCNN network based on the triples to obtain the judging model comprises the following steps:
putting the training set of the triples into an FPN + Faster-RCNN network, establishing weights of each layer of network according to the FPN + Faster-RCNN network to obtain forward-propagated output, calculating residual errors of actual output and target output, reversely adjusting a weight matrix by using a method of minimizing errors, and adjusting the whole FPN + Faster-RCNN network in a finite iteration process to obtain a studying and judging model.
The invention also provides an artificial intelligence defect identification system for unmanned aerial vehicle routing inspection of a distribution network line, which comprises a computer, wherein the computer comprises:
at least one memory cell;
at least one processing unit;
wherein the at least one memory unit has stored therein at least one instruction that is loaded and executed by the at least one processing unit to perform the steps of:
s1, acquiring an original sample library, wherein the original sample library comprises typical component images and defect images of distribution network lines; carrying out image augmentation on the defect image to obtain an artificial sample, wherein the original sample library and the artificial sample form a typical part of a distribution network line and a defect sample library;
s2, acquiring a judging model based on the typical parts and defect sample library of the distribution network lines and the FPN + Faster-RCNN network;
and S3, detecting the inspection video image to be detected based on the judging model.
Preferably, the image augmenting the defect image comprises:
image augmentation of defect images is performed based on generation of a countermeasure network.
Preferably, the FPN + Faster-RCNN network comprises: the device comprises an FPN layer and a Faster-RCNN layer, wherein the FPN layer is used for extracting routing inspection image defect characteristics of typical parts of the distribution network line, and the Faster-RCNN layer is used for identifying typical equipment parts of the distribution network line.
Preferably, the acquiring and judging model based on the distribution network line typical component, the defect sample library and the FPN + fast-RCNN network includes:
constructing a triple group based on the typical component of the distribution network line and the defect sample library;
and training the FPN + Faster-RCNN network based on the triples to obtain a judging model.
(III) advantageous effects
The invention provides an artificial intelligence defect identification method and system for distribution network line unmanned aerial vehicle routing inspection. Compared with the prior art, the method has the following beneficial effects:
the invention expands the defect image data samples with less samples, unobvious characteristics and the like by image augmentation of the defect image, provides enough data for the construction of a study and judgment model, and improves the accuracy of the model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a block diagram of an artificial intelligence defect identification method for distribution network line unmanned aerial vehicle routing inspection in the embodiment of the invention;
FIG. 2 is a schematic diagram of the construction of triples in an embodiment of the present invention, where (a) is the construction of a triplet before training and (b) is the construction of a triplet after training;
fig. 3 is a schematic diagram illustrating identification of a strand breakage defect of a distribution network line conductor in the embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described, and it is obvious that the described embodiments are a part of the embodiments of the present invention, but 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.
The artificial intelligence defect identification method for the unmanned aerial vehicle inspection of the distribution network line solves the technical problem that the accuracy of identification of the distribution network defect is low in the prior art, improves the accuracy of identification of the distribution network defect, reduces the traditional operation cost of inspection of the distribution network line, and improves inspection efficiency.
In order to solve the technical problems, the general idea of the embodiment of the application is as follows:
with the development of artificial intelligence and industrial automation, the automatic identification of equipment faults and the application of intelligent diagnosis technology become more and more extensive, and the intelligent diagnosis technology based on image identification has the advantages of high identification speed, high accuracy, wide application range and the like, thereby drawing extensive attention. In the distribution network line unmanned aerial vehicle inspection technology, the key step of the distribution network line unmanned aerial vehicle inspection technology is to accurately judge whether the typical equipment parts of the distribution network line have defects, and the key step is the basis of the unmanned aerial vehicle inspection distribution network line. According to the embodiment of the invention, the inspection video image to be detected is detected through the research and judgment model based on the FPN + fast-RCNN network, so that the accuracy of image identification is improved.
In order to better understand the technical solution, the technical solution will be described in detail with reference to the drawings and the specific embodiments.
The invention provides an artificial intelligence defect identification method for distribution network line unmanned aerial vehicle inspection, which comprises the following steps of S1-S3 as shown in figure 1:
s1, obtaining an original sample library, wherein the original sample library comprises typical component images and defect images of the distribution network line, carrying out image amplification on the defect images to obtain artificial samples, and the original sample library and the artificial samples form typical components and defect sample libraries of the distribution network line.
And S2, acquiring a judging model based on the typical parts and defect sample library of the distribution network lines and the FPN + fast-RCNN network.
And S3, detecting the inspection video image to be detected based on the judging model.
The embodiment of the invention expands the defect image data samples with less samples, unobvious characteristics and the like by image augmentation of the defect image, provides enough data for the construction of a study and judgment model, and improves the accuracy of the model.
In an embodiment, S1, an original sample library is obtained, where the original sample library includes typical component images and defect images of the distribution network line, the defect images are subjected to image augmentation to obtain an artificial sample, and the original sample library and the artificial sample constitute a distribution network line typical component and defect sample library. The specific implementation process is as follows:
in the embodiment of the present invention, the distribution network line mainly includes: the urban and rural union department, the hilly area and the plain area are connected with the network line, and the network line is obtained
Typical part images and defect images of distribution network lines. Typical components and defects of distribution network lines are detailed in tables 1-1 to 1-8:
TABLE 1-1 Defect of Foundation and protective Condition
TABLE 1-2 Pole State Defect
TABLE 1-3 wire Condition Defect
TABLE 1-4 insulator chain State Defect
TABLE 1-5 hardware State Defect
TABLE 1-6 lightning protection facilities and earthing device status defects
TABLE 1-7 auxiliary facility Condition Defect
TABLE 1-8 overhead line protective zone State Defect
In the specific implementation process, the proportion difference of the number of the samples of the different types of defect images in the original sample library in the total number of the samples is large, and the original samples of the defect images are expanded by generating a countermeasure network. During the training process of the generated countermeasure network, the discriminator and the generator alternately operate, and the game, the learning and the optimization are continuously performed. In order to prevent the learning speed of the discriminator from being too slow to make the generator difficult to learn, the discriminator is generally trained 1 time again every k times, where k is an integer greater than 0, e.g., k is 3. The method comprises the following specific steps:
in the first step, a fixed generator G is required to train arbiter D. When training starts, a batch of real samples x are randomly selected from defect images in an original database and input into a discriminator D, and the discriminator D outputs a discrimination probability D (x). Comparing the discrimination probability D (x) with the authenticity label of the sample, namely 1 to obtain an error LossD 1; then, the random noise z is input into a generator G to obtain a forged sample G (z), and G (z) is input into a discriminator to obtain a discrimination probability D (G (z)). The discrimination probability D (g (z)) of the counterfeit sample is compared with its authenticity signature, i.e. 0, resulting in an error LossD 2. The total error of discriminator D is denoted LossD1+ LossD 2. And after the error LossD is obtained, the error is reversely propagated to each network node in the discriminator D, and the parameters in the network are updated. At this point, one round of learning by the discriminator D is completed.
Second, fixed arbiter D is required to train generator G. Random noise z is input to the generator G to obtain a fake sample G (z). And then the G (z) is used as an input to be sent to a discriminator D to obtain a probability numerical value D (G (z)) given by the discriminator. Unlike the process of solving the error in the discriminator, the authenticity label of the forged sample g (z) in the discriminator is 0, i.e., labeled "artificial sample". In the generator, the purpose is to train an artificial sample to "trick" the discriminator, so that the discriminator considers the artificial sample as a real sample and gives a result with discrimination probability D (g (z)) close to 1. Therefore, in the generator, the output authenticity flag is typically set to 1, which causes the generator G to fit the random noise z towards a direction similar to the real sample. Similarly, after the authenticity discrimination probability D (G (z)) of the counterfeit sample and the error LossG of the authenticity label 1 are obtained, the errors are propagated back to each node in the generator G, and the parameters in the network are updated. So far, one round of learning of the generator G is completed.
Third, if there are enough samples and the discriminator D can reach the optimum in each game training for the generator G, then finally V (G, D) will reach the global optimum solution. An optimal generator G can be obtained at this time, so that the discrimination probability of the discriminator D at this time for both the real sample x and the fake sample G (z) is 0.5. This means that the artificial samples of the generator G have already been "spurious" and the discriminator D has not been able to distinguish between real and artificial samples. Thereby realizing the enlargement of small defect samples.
And the artificial samples and the original sample library form a typical part of the distribution network line and a defect sample library.
S2, obtaining a judging model based on the typical parts and the defect sample library of the distribution network line and the FPN + Faster-RCNN network, wherein the FPN layer of the FPN + Faster-RCNN network is used for extracting the inspection image defect characteristics of the typical parts of the distribution network line, and the Faster-RCNN layer is used for identifying the typical equipment parts of the distribution network line.
And training the FPN + Faster-RCNN network by adopting typical parts of the distribution network lines and images in a defect sample library. The method specifically comprises the following steps:
s201, constructing a triple based on the typical component of the distribution network line and the defect sample library.
The FPN + fast-RCNN network structure in the embodiment of the invention is designed by training the FPN + fast-RCNN network by using a large number of unlabelled images in typical distribution network line components and a defect sample library and referring to the AlexNet network structure. However, since there is no label on the image, it is not possible to accurately determine the cost function (loss function) and determine a suitable optimization scheme. In the embodiment of the present invention, the euclidean distance is used to determine the similarity between two image blocks, and in an image set, two very similar blocks are also similar in the space of the visual coding mapping, but the similarity feature constraint is far from sufficient, which may result in that all points are mapped to one point in the space. Therefore, to train the FPN + Faster-RCNN network, a third block is introduced, composing a new type of triplet. During the training process, a cost function is employed to ensure that the first block is closer to its similar blocks in the mapping space than the first block and the random block.
Based on similarity comparison of the novel triples, the novel triples are constructed to facilitate inter-class and intra-class distance acquisition in the optimization process, as shown in fig. 2. For each triplet, the following construction strategy is adopted: two image blocks are randomly selected from all classes (defect types) in typical components of distribution network lines and a defect sample library, one image block is selected from the other class, and each triple contains two similar image blocks and one dissimilar image block. According to the strategy, the trained triples are distributed, in the Hamming distance, the matched image blocks are gathered closer (blocks in a dotted circle), the unmatched blocks are separated, and the similar blocks can be guaranteed to have similar Hash codes by the setting.
S202, constructing a triple group based on the typical components of the distribution network line and the defect sample library. The method specifically comprises the following steps:
putting the training set of the triples into an FPN + Faster-RCNN network, establishing weights of each layer of network according to the FPN + Faster-RCNN network to obtain forward-propagated output, calculating residual errors of actual output and target output, reversely adjusting a weight matrix by using a method of minimizing errors, and adjusting the whole FPN + Faster-RCNN network in a finite iteration process to obtain a studying and judging model.
The following description of the FPN + fast-RCNN network:
the method comprises the steps of adopting an FPN layer in a deep neural network FPN + Faster-RCNN network (for convenience of description, hereinafter written as the deep neural network) to extract image characteristics of training images in typical parts of distribution network lines and a defect sample library, wherein the deep neural network aims to establish perfect mapping h (x) and extract p-dimensional characteristic vectors from the network
And mapping to q-dimensional Hash binary code h e to {0,1}qAnd h is a q-dimensional vector with 1 or 0 per dimension. And (3) expressing the image feature extraction and Hash coding process by using a nonlinear conversion function phi (·), inputting an original image, and outputting Hash coding. The expression is as follows:
h＝φ(I)
the objective is that each type of image block has the same hash code, and the difference between the hash codes is calculated by using weighted hamming distance, which is expressed by theta () and expressed as follows:
Θ(h(xi),h(xj))＝h(xi)wh(xj)＝Σkωkhk(xi)hk(xj)
where the matrix w is a diagonal matrix, each dimension of the diagonal matrix being denoted by ωkDenotes that w (k, k) ═ ωk。h(xi),h(xj) The greater the degree of difference, Θ (h (x)i),h(xj) The larger the value). Hereinafter, for the sake of simplicity, hi,hjRepresents h (x)i),h(xj)。
The weighted hash code can be endowed with different weights for each bit of the hash code, and the weights are learned from the deep neural network according to different training sets.
The goal is to maximize inter-class distance and minimize intra-class distance, with similar blocks having similar encodings. In order to make the deep neural network achieve this goal, a special cost function (loss function) is set as follows:
φ(I)＝Φw+Ψw
wherein:
Φwrepresenting a maximum boundary term (Max-MarginTerm) and maximizing the distance between control classes;
Ψwrepresenting a regularization term (regularizationterm) controlling similar blocks to have similar encodings;
Φwmaximum boundary term, maximum control of inter-class distance, control by block-to-block difference, ensuring
Wherein the content of the first and second substances,
the three constitute a triplet, in the ith triplet,
represent the same classThe hash code of (a) of (b),
represents two types of hash codes, satisfy
The difference between the similar hash codes and the heterogeneous hash codes is represented, so the maximum boundary term should be represented as:
regular term Ψ for controlling similar blocks to have similar codingwThe definition is as follows:
wherein h isi,hjAre respectively image blocks Ii,IjHash code (hash code). S represents a similarity matrix, for the element S in the similarity matrixijRepresenting image blocks I in a training seti,IjSimilarity between them, SijThe larger the representative image block, the closer, and vice versa, the farther away. The matrix S is a symmetric matrix, i.e. Sij＝Sji。
Defining a diagonal matrix U, wherein
The laplace matrix L ═ U-S, where L ∈ M × M then the regularization term can be converted into
Wherein, the matrix R ═ R1,r2,…,rM]R ∈ q × M, M is a graphThe total number of images. tr (-) denotes the trace of the matrix. The triplet-based regularization model may be represented as:
next, the objective function is optimized,
tr(RLRT)＝ri T(RLi)+(RLi)Tri-ri TLiiri
wherein L isiRepresenting the ith column of the matrix L. Definition of R-iThe sub-matrix after the i-th column is removed for matrix R, Li,-iIs a vector LiThe vector after the ith value is removed.
The method comprises the steps of extracting features from an input image set by applying a deep neural network, wherein the bottom layer of the network consists of a convolution layer (convolution), a down-sampling layer (boosting) and a full connection layer (full connection), outputting binary hash codes by applying the full connection layer and a tangent function layer after the features are generated, and calculating the weight of each bit of the hash codes on the top layer of the deep neural network, so that deep hash learning and feature learning are jointly optimized through back propagation in the deep neural network. The whole deep neural network adopts a strategy of local connection and weight sharing, different features can be extracted by adopting different convolution kernels, a plurality of weights can be reduced by sharing the weights, and the downsampling layer can perform weighted average on the image through downsampling.
In another embodiment, S3, the inspection video image to be detected is detected based on the judging model. The specific implementation process is as follows:
inputting a to-be-detected inspection video image, and processing the to-be-detected inspection video image through a study and judgment model formed by training FPN + Faster-RCNN. And outputting a device defect detection result. Fig. 3 shows the identification of the strand breakage defect of the distribution network line conductor by adopting a study and judgment model.
The embodiment of the invention also provides an artificial intelligence defect identification system for unmanned aerial vehicle routing inspection of a distribution network line, which comprises a computer, wherein the computer comprises: at least one memory cell; at least one processing unit; wherein the at least one memory unit has at least one instruction stored therein, the at least one instruction being loaded and executed by the at least one processing unit to implement the above steps.
It can be understood that the system for identifying artificial intelligence defects in routing inspection of the distribution network line unmanned aerial vehicle provided by the embodiment of the invention corresponds to the method for identifying artificial intelligence defects in routing inspection of the distribution network line unmanned aerial vehicle, and relevant explanations, examples, beneficial effects and the like can refer to the corresponding contents in the method for identifying artificial intelligence defects in routing inspection of the distribution network line unmanned aerial vehicle, and are not repeated herein.
In summary, compared with the prior art, the method has the following beneficial effects:
1. the embodiment of the invention expands the defect image data samples with less samples, unobvious characteristics and the like by generating the countermeasure network to increase the image of the defect image, provides enough data for the construction of a study and judgment model and improves the accuracy of the model.
2. According to the embodiment of the invention, the inspection image defect characteristics of the typical parts of the distribution network line are extracted based on FPN, and the inspection defects of the distribution network line are detected and identified by combining FPN + Faster-RCNN, so that the defect identification precision is improved. Therefore, the distribution network defect identification accuracy of unmanned aerial vehicle inspection is improved.
It should be noted that, through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
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 (10)
1. An artificial intelligence defect identification method for distribution network line unmanned aerial vehicle routing inspection is characterized by comprising the following steps:
acquiring an original sample library, wherein the original sample library comprises typical component images and defect images of distribution network lines; carrying out image augmentation on the defect image to obtain an artificial sample, wherein the original sample library and the artificial sample form a typical part of a distribution network line and a defect sample library;
acquiring a judging model based on the typical parts of the distribution network lines, the defect sample library and the FPN + Faster-RCNN network;
and detecting the to-be-detected patrol video image based on the judging model.
2. The method for artificial intelligence defect recognition in unmanned aerial vehicle inspection according to claim 1, wherein the image augmenting the defect image comprises:
image augmentation of defect images is performed based on generation of a countermeasure network.
3. The method for identifying artificial intelligence defects in routing inspection of distribution network lines and unmanned aerial vehicles according to claim 1, wherein the FPN + Faster-RCNN network comprises: the device comprises an FPN layer and a Faster-RCNN layer, wherein the FPN layer is used for extracting routing inspection image defect characteristics of typical parts of the distribution network line, and the Faster-RCNN layer is used for identifying typical equipment parts of the distribution network line.
4. The method for artificial intelligence defect recognition in unmanned aerial vehicle inspection according to claim 1, wherein the obtaining of a study and judgment model based on the distribution network line typical component and defect sample library and the FPN + Faster-RCNN network comprises:
constructing a triple group based on the typical component of the distribution network line and the defect sample library;
and training the FPN + Faster-RCNN network based on the triples to obtain a judging model.
5. The method for artificial intelligence defect identification for unmanned aerial vehicle inspection of distribution network lines according to claim 4, wherein the constructing of the triplets based on the typical components of the distribution network lines and the defect sample library comprises:
randomly selecting two image blocks from all classes of typical components of distribution network lines and a defect sample library, and selecting one image block from the other class, wherein each triple contains two similar image blocks and one dissimilar image block.
6. The artificial intelligence defect identification method for distribution network line unmanned aerial vehicle inspection according to claim 4, wherein training the FPN + Faster-RCNN network based on the triples to obtain a study model comprises:
putting the training set of the triples into an FPN + Faster-RCNN network, establishing weights of each layer of network according to the FPN + Faster-RCNN network to obtain forward-propagated output, calculating residual errors of actual output and target output, reversely adjusting a weight matrix by using a method of minimizing errors, and adjusting the whole FPN + Faster-RCNN network in a finite iteration process to obtain a studying and judging model.
7. The utility model provides a join in marriage artificial intelligence defect identification system that net twine way unmanned aerial vehicle patrolled and examined, a serial communication port, the system includes the computer, the computer includes:
at least one memory cell;
at least one processing unit;
wherein the at least one memory unit has stored therein at least one instruction that is loaded and executed by the at least one processing unit to perform the steps of:
s1, acquiring an original sample library, wherein the original sample library comprises typical component images and defect images of distribution network lines; carrying out image augmentation on the defect image to obtain an artificial sample, wherein the original sample library and the artificial sample form a typical part of a distribution network line and a defect sample library;
s2, acquiring a judging model based on the typical parts and defect sample library of the distribution network lines and the FPN + Faster-RCNN network;
and S3, detecting the inspection video image to be detected based on the judging model.
8. The system of claim 7, wherein the image augmenting the defect image comprises:
image augmentation of defect images is performed based on generation of a countermeasure network.
9. The system of claim 7, wherein the FPN + Faster-RCNN network comprises: the device comprises an FPN layer and a Faster-RCNN layer, wherein the FPN layer is used for extracting routing inspection image defect characteristics of typical parts of the distribution network line, and the Faster-RCNN layer is used for identifying typical equipment parts of the distribution network line.
10. The system of claim 7, wherein the acquisition of a study and judgment model based on the distribution network line typical component and defect sample library and the FPN + fast-RCNN network comprises:
constructing a triple group based on the typical component of the distribution network line and the defect sample library;
and training the FPN + Faster-RCNN network based on the triples to obtain a judging model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011497412.2A CN112669264A (en) | 2020-12-17 | 2020-12-17 | Artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of distribution network line |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011497412.2A CN112669264A (en) | 2020-12-17 | 2020-12-17 | Artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of distribution network line |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112669264A true CN112669264A (en) | 2021-04-16 |
Family
ID=75404830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011497412.2A Pending CN112669264A (en) | 2020-12-17 | 2020-12-17 | Artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of distribution network line |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112669264A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361443A (en) * | 2021-06-21 | 2021-09-07 | 广东电网有限责任公司 | Method and system for power transmission line image sample counterstudy augmentation |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019233166A1 (en) * | 2018-06-04 | 2019-12-12 | 杭州海康威视数字技术股份有限公司 | Surface defect detection method and apparatus, and electronic device |
| CN110827251A (en) * | 2019-10-30 | 2020-02-21 | 江苏方天电力技术有限公司 | Power transmission line locking pin defect detection method based on aerial image |
| CN111340787A (en) * | 2020-02-28 | 2020-06-26 | 广东电网有限责任公司 | Method and device for detecting and identifying defects of conducting wire of power transmission line and computer equipment |
| CN111650204A (en) * | 2020-05-11 | 2020-09-11 | 安徽继远软件有限公司 | Transmission line hardware defect detection method and system based on cascade target detection |
| CN111723774A (en) * | 2020-07-03 | 2020-09-29 | 国网山东省电力公司日照供电公司 | Target identification method for power transmission equipment based on unmanned aerial vehicle inspection |
| CN111783590A (en) * | 2020-06-24 | 2020-10-16 | 西北工业大学 | Multi-class small target detection method based on metric learning |
-
2020
- 2020-12-17 CN CN202011497412.2A patent/CN112669264A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019233166A1 (en) * | 2018-06-04 | 2019-12-12 | 杭州海康威视数字技术股份有限公司 | Surface defect detection method and apparatus, and electronic device |
| CN110827251A (en) * | 2019-10-30 | 2020-02-21 | 江苏方天电力技术有限公司 | Power transmission line locking pin defect detection method based on aerial image |
| CN111340787A (en) * | 2020-02-28 | 2020-06-26 | 广东电网有限责任公司 | Method and device for detecting and identifying defects of conducting wire of power transmission line and computer equipment |
| CN111650204A (en) * | 2020-05-11 | 2020-09-11 | 安徽继远软件有限公司 | Transmission line hardware defect detection method and system based on cascade target detection |
| CN111783590A (en) * | 2020-06-24 | 2020-10-16 | 西北工业大学 | Multi-class small target detection method based on metric learning |
| CN111723774A (en) * | 2020-07-03 | 2020-09-29 | 国网山东省电力公司日照供电公司 | Target identification method for power transmission equipment based on unmanned aerial vehicle inspection |
Non-Patent Citations (1)
| Title |
|---|
| 范亚雷: "电力巡检图像中基于深度学习的异物检测方法研究", 中国优秀硕士学位论文全文数据库, no. 11, 15 November 2020 (2020-11-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361443A (en) * | 2021-06-21 | 2021-09-07 | 广东电网有限责任公司 | Method and system for power transmission line image sample counterstudy augmentation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Du et al. | Achieving 100x acceleration for N-1 contingency screening with uncertain scenarios using deep convolutional neural network | |
| CN109190712A (en) | A kind of line walking image automatic classification system of taking photo by plane based on deep learning | |
| CN111832812A (en) | Wind power short-term prediction method based on deep learning | |
| CN113095331A (en) | Visual question answering method, system and equipment for appearance defects of electric equipment and storage medium thereof | |
| CN110008917A (en) | Fault detection method based on the relay protection device constant value forms data that table understands | |
| CN111914484A (en) | Recursive graph convolution network system for power grid transient stability evaluation | |
| CN115271253A (en) | Water-wind power generation power prediction model construction method and device and storage medium | |
| CN112669264A (en) | Artificial intelligence defect identification method and system for unmanned aerial vehicle routing inspection of distribution network line | |
| CN115469184A (en) | New energy transmission line fault identification method based on convolutional network | |
| CN115759365A (en) | Photovoltaic power generation power prediction method and related equipment | |
| CN114137358B (en) | Power transmission line fault diagnosis method based on graph convolution neural network | |
| CN113536944A (en) | Distribution line inspection data identification and analysis method based on image identification | |
| CN117113066B (en) | Transmission line insulator defect detection method based on computer vision | |
| CN112668416A (en) | Real-time analysis and identification method for distribution network inspection video of unmanned aerial vehicle and embedded inspection device | |
| CN116488325A (en) | Smart power grid anomaly detection and classification method, smart power grid anomaly detection and classification equipment and readable storage medium | |
| CN115684786A (en) | Inverter switching tube health diagnosis method, device and system based on gram angular field and parallel CNN | |
| CN115311509A (en) | Power system transient stability evaluation method and system based on imaging data driving | |
| CN114519294A (en) | Power cable fault identification method based on wavelet singular entropy and BP neural network | |
| Sheng et al. | A Method and Implementation of Transmission Line's Key Components and Defects Identification Based on YOLO | |
| CN111950765A (en) | Probabilistic transient stability prediction method based on stacked noise reduction self-encoder | |
| CN114418194B (en) | Tower damage prediction method and device based on data driving and model driving | |
| Poornima | Deep Learning-Based Fault Identification of Microgrid Transformers | |
| Fan et al. | Microgrid Fault Classification Method based on EPSANet56 | |
| Zhao et al. | Lightweight Neural Network for Component Recognition and Fault Detection of Large-Scale Transmission Line | |
| CN116961072A (en) | Multi-DC chain commutation failure recognition method and system based on space-time convolution network |
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 |