CN113449769A - Power transmission line icing identification model training method, identification method and storage medium - Google Patents
Power transmission line icing identification model training method, identification method and storage medium Download PDFInfo
- Publication number
- CN113449769A CN113449769A CN202110539796.8A CN202110539796A CN113449769A CN 113449769 A CN113449769 A CN 113449769A CN 202110539796 A CN202110539796 A CN 202110539796A CN 113449769 A CN113449769 A CN 113449769A
- Authority
- CN
- China
- Prior art keywords
- icing
- transmission line
- power transmission
- training
- image
- 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
- 230000005540 biological transmission Effects 0.000 title claims abstract description 154
- 238000012549 training Methods 0.000 title claims abstract description 82
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000002372 labelling Methods 0.000 claims abstract description 9
- 239000012212 insulator Substances 0.000 claims description 37
- 239000011248 coating agent Substances 0.000 claims description 24
- 238000000576 coating method Methods 0.000 claims description 24
- 239000004020 conductor Substances 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 13
- 238000001514 detection method Methods 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 11
- 230000000694 effects Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 230000008685 targeting Effects 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 7
- 238000012545 processing Methods 0.000 description 12
- 230000015556 catabolic process Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000007689 inspection Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000006731 degradation reaction Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a training method, a recognition method and a storage medium for an icing recognition model of a power transmission line, wherein the training method comprises the following steps: acquiring training data, wherein the training data comprises a positive sample and a negative sample, the positive sample is a power transmission line icing image containing an icing target, and the negative sample is a power transmission line icing image not containing the icing target; labeling the training data based on a preset classification category; constructing a fast RCNN recognition model network architecture; training a fast RCNN recognition model network architecture by using the training data to obtain a fast RCNN recognition model for recognizing the icing state of the power transmission line. The method and the device for identifying the icing of the power transmission line based on the fast RCNN deep convolution neural network training can obtain the icing identification model of the power transmission line, can quickly and accurately identify the icing fault in the image of the power transmission line, effectively improve the identification accuracy and robustness of the model, and improve the generalization capability of the model.
Description
Technical Field
The application relates to the technical field of power equipment detection, in particular to a training method and an identification method for an icing identification model of a power transmission line and a storage medium.
Background
With the rapid development of economy and power technology in China, overhead transmission lines are distributed throughout provinces and cities in China, and electric energy also becomes an indispensable energy source for daily life and industrial development of people. In a transmission line, insulators play a role in mechanical connection and electrical insulation and are important components of a power system. Because part of transmission lines of the power system operate in the field all the year round and are eroded by ice, frost, snow and rain, the conductors and insulators on the lines are easy to be coated with ice, the ice coating not only affects the stability of transmission voltage of the transmission lines and causes great reduction of transmission efficiency, but also insulator flashover or breakdown can occur when the ice coating is accumulated too thick to cause line tripping, even breakage of transmission conductors after galloping and collapse of transmission towers occur, and the safe and stable operation of the power system is seriously threatened. Therefore, timely and effective monitoring of the icing condition of the insulator of the power transmission line becomes an important means for maintaining safe operation of the power transmission line, is also an important guarantee for ensuring stable operation of a power grid, and is of great importance for developing research on an automatic identification method of the icing image of the insulator of the power transmission line.
Because the breadth of our country is vast, and a part of the transmission line must pass through high altitude and open-air areas, the transmission line has higher requirements on the stable operation of the power system under the weather influence of low temperature, strong wind and snow all the year round. At present, the ice coating identification of the insulator of the power transmission line by domestic and foreign network companies is mainly carried out by manual inspection and helicopter inspection on the site; the other method is to collect the image of the power transmission line through a shooting device fixed on a power transmission tower pole, utilize the traditional image processing technology, firstly preprocess the collected image, including graying and image enhancement of the collected image, then extract the edge of the target object by using an edge extraction algorithm, finally judge whether to cover ice by calculating and comparing the change of pixel values of the image which is not covered with ice and the image which is covered with ice, and further judge the ice covering thickness of the insulator of the power transmission line. However, such methods are often only suitable for images shot at specific positions and specific angles, and have poor generalization capability for complex background environments of the power transmission line site and poor recognition effect for power transmission line pictures obtained on site. At present, some power companies also use unmanned aerial vehicles to shoot images, however, the identification method still distinguishes images through manual identification or traditional image processing technology, the workload is still large, and the method is not suitable for mass image information acquired by a power system during routing inspection. In the prior art, the icing state of the power transmission line needs to be manually distinguished, so that the influence of subjective factors is large, and results are possibly different; moreover, because the icing of the power transmission line frequently occurs in the field, the manual inspection is dangerous, and the input labor cost is high. The method is weak in capability of identifying the icing state based on the image processing technology, low in precision and not suitable for processing pictures of complex backgrounds on the power transmission line site.
Disclosure of Invention
In view of the technical problems in the prior art, embodiments of the present application provide a training method, an identification method, and a storage medium for an ice coating identification model of a power transmission line, which can solve the problems in the prior art that the speed and the accuracy of identifying faults such as ice coating of a power transmission conductor and an insulator from an image of the power transmission line are low; meanwhile, the problems that the icing identification model of the power transmission line in the prior art is weak in generalization capability and insufficient in picture processing capability can be solved.
In order to solve the technical problem, the embodiment of the application adopts the following technical scheme:
the embodiment of the application provides a training method of an icing identification model of a power transmission line, which comprises the following steps:
acquiring training data, wherein the training data comprises a positive sample and a negative sample, the positive sample is a power transmission line icing image containing an icing target, and the negative sample is a power transmission line icing image not containing the icing target;
labeling the training data based on a preset classification category;
constructing a fast RCNN recognition model network architecture;
and training the network architecture of the fast RCNN recognition model by using the training data to obtain the fast RCNN recognition model for recognizing the icing state of the power transmission line.
In some embodiments, after acquiring the training data, the method further comprises:
and performing data preprocessing on the acquired ice coating image of the power transmission line, wherein the data preprocessing comprises performing data enhancement on the ice coating image of the power transmission line in at least one of translation, rotation, mirror image or expansion.
In some embodiments, constructing a fast RCNN recognition model network architecture comprises:
inputting the icing image of the power transmission line into a feature extraction network to generate a feature map;
extracting candidate areas containing icing targets from the feature map by using an area suggestion network to generate target candidate frames;
mapping the target candidate box to the feature map to obtain a feature matrix;
constructing an ROI posing layer, and inputting the feature matrix into the ROI posing layer to obtain a target feature map with a fixed size;
and constructing a full connection layer, and performing regression and classification on the target feature map by using the full connection layer to obtain frame coordinate parameters and classification results of the target candidate frame.
In some embodiments, the feature extraction network is a residual extraction network resnet50, including 49 convolutional layers and 1 fully-connected layer, the convolutional layers including 16 residual units, each residual unit including a 1 × 1 convolution kernel and a 3 × 3 convolution kernel, the 1 × 1 convolution kernel being used to reduce dimensionality, the 3 × 3 convolution kernel being used to extract features.
In some embodiments, training the fast RCNN recognition model network architecture with the training data comprises:
training the fast RCNN recognition model network architecture by adopting a mini-batch stochastic gradient descent method;
wherein the initial learning rate of the fast RCNN recognition model network architecture training is 0.005, the momentum parameter is 0.9, and the learning attenuation rate is 0.0005.
In some embodiments, the method further comprises:
calculating a loss value of a loss function;
updating the fast RCNN recognition model based on the loss value of the loss function;
wherein the loss function is:
wherein i represents the serial number of the detection box in each small batch; p is a radical ofiRepresenting the prediction probability of targeting the detection box, p when the prediction is targeted* i1, otherwise 0; t is ti4 parametric coordinate vectors, t, representing target candidate boxes* iRepresenting a real box coordinate associated with the positive sample; n is a radical ofclsIndicates the number of all samples in a mini-batch, NregNumber L of detection frame positionsreg(ti，t* i) Represents the regression loss function, Lcls(pi，p* i) Representing a classification loss function; λ is the equilibrium parameter, λ 10.
In some embodiments, the method further comprises:
testing the recognition effect of the recognition model by using the accuracy, the recall rate and the average accuracy mean value;
and if the test result is lower than a preset threshold value, modifying the training parameters of the Faster RCNN recognition model, and training the Faster RCNN recognition model until the test result reaches the preset threshold value.
In some embodiments, the training data further includes non-iced transmission line images, and labeling the training data based on a preset classification category includes:
and marking the non-icing insulator and the non-icing conductor in the non-icing power transmission line image.
The embodiment of the application further provides a method for identifying the icing state of the power transmission line, which comprises the following steps:
collecting an icing image of the power transmission line;
inputting the icing image of the power transmission line into a preset fast RCNN recognition model, and recognizing an icing target in the icing image of the power transmission line, wherein the icing target comprises an icing insulator or an icing conductor.
The embodiment of the application also provides a computer-readable storage medium, on which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the steps of the training method of the electric transmission line icing identification model or the identification method of the electric transmission line icing state are realized.
Compared with the prior art, the training method of the power transmission line icing identification model and the identification method of the power transmission line icing state provided by the embodiment of the application are based on the power transmission line icing identification model obtained by fast RCNN deep convolution neural network training, can quickly and accurately identify whether icing faults exist in a power transmission line image, accurately identify the specific classification types of the icing faults, do not need a maintainer of the power transmission line to monitor the power transmission line faults at the background for a long time, and maximize the working efficiency of the maintainer of the power transmission line and reduce unnecessary workload. In addition, the residual error neural network resnet50 is adopted to extract the features of the input image, so that the problem of network degradation caused by the number of network layers is solved, and the identification accuracy and the identification efficiency are improved. In addition, data preprocessing is carried out after the power transmission line image is collected, and features of each angle of the power transmission line image can be extracted in modes of rotating different angles, mirroring, stretching and the like, so that multi-angle and multi-posture feature information of an icing target can be obtained, icing faults of the power transmission line can be accurately identified, the icing target identification is more objective, accurate, rapid and efficient, and the identification accuracy and robustness of a model are effectively improved; meanwhile, the generalization capability of the ice coating recognition model of the power transmission line and the processing capability of massive pictures can be improved.
Drawings
Fig. 1 is a flowchart of a training method of an ice coating recognition model of a power transmission line according to an embodiment of the present application;
FIG. 2 is a schematic structural diagram of a feature extraction network according to an embodiment of the present application;
FIG. 3 is a flow chart illustrating the construction of a fast RCNN recognition model network architecture according to an embodiment of the present disclosure;
FIG. 4 is a graph of training loss versus learning rate for the fast RCNN recognition model according to an embodiment of the present application;
FIG. 5 is a graph illustrating the mean average accuracy of the fast RCNN recognition model according to an embodiment of the present application;
fig. 6(a) is an identification effect diagram of the identification method of the icing state of the power transmission line according to the embodiment of the present application;
fig. 6(b) is another recognition effect diagram of the method for recognizing the icing state of the power transmission line according to the embodiment of the application.
Detailed Description
In order to make the technical solutions of the present application better understood, the present application is described in detail below with reference to the accompanying drawings and the detailed description.
It will be understood that various modifications may be made to the embodiments disclosed herein. Accordingly, the foregoing description should not be construed as limiting, but merely as exemplifications of embodiments. Those skilled in the art will envision other modifications within the scope and spirit of the application.
The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the application and, together with a general description of the application given above and the detailed description of the embodiments given below, serve to explain the principles of the application.
These and other characteristics of the present application will become apparent from the following description of preferred forms of embodiment, given as non-limiting examples, with reference to the attached drawings.
It should also be understood that, although the present application has been described with reference to some specific examples, a person of skill in the art shall certainly be able to achieve many other equivalent forms of application, having the characteristics as set forth in the claims and hence all coming within the field of protection defined thereby.
The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.
Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it is to be understood that the disclosed embodiments are merely examples of the application, which can be embodied in various forms. Well-known and/or repeated functions and constructions are not described in detail to avoid obscuring the application of unnecessary or unnecessary detail. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a basis for the claims and as a representative basis for teaching one skilled in the art to variously employ the present application in virtually any appropriately detailed structure.
Fig. 1 is a flowchart of a training method of a power transmission line icing identification model according to an embodiment of the present application. As shown in fig. 1, the training method for the power transmission line icing identification model provided by the embodiment of the present application includes:
s101, training data are obtained, wherein the training data comprise a positive sample and a negative sample, the positive sample is an ice coating image of the power transmission line containing an ice coating target, and the negative sample is an ice coating image of the power transmission line not containing the ice coating target.
The ice coating image of the power transmission line is the image of the power transmission line containing ice coating, and can be acquired on the site of the power transmission line through equipment such as a mobile phone and the like and also can be acquired remotely in a phytotron laboratory. After the icing image of the power transmission line is collected, the icing image of the power transmission line is preprocessed, and the icing image of the power transmission line with all pixels of 500 x 375 is obtained.
In some embodiments, the preprocessing of the acquired transmission line image further comprises:
and performing data enhancement on the ice coating image of the power transmission line in at least one mode of translation, rotation, mirror image or expansion.
For example, the acquired power transmission line icing image can be rotated by different angles to obtain power transmission line icing images at different angles so as to expand the number of the power transmission line icing image data sets, thereby enhancing the robustness of the subsequently obtained power transmission line icing identification model.
Optionally, the icing target comprises an icing insulator or an icing conductor.
In this embodiment, the icing insulators in the icing transmission line are mainly identified, so that the icing target is the icing insulator, that is, the icing image of the transmission line including the insulator is a positive sample, and the icing image of the transmission line not including the insulator (for example, the icing image of the transmission line including the wire and the icing image of the transmission line including other components) is a negative sample.
In other embodiments, when the ice-coated conductor is used as the ice-coated target, the ice-coated image of the power transmission line including the ice-coated conductor is a positive sample, and the ice-coated image of the power transmission line not including the ice-coated conductor (for example, the ice-coated image of the power transmission line including the ice-coated insulator and the ice-coated image of the power transmission line including other components) is a negative sample.
In some embodiments, the training data may further include non-icing transmission line images to facilitate identifying transmission line icing images from a mass of transmission line images to further identify icing insulators or icing conductors from the transmission line icing images.
And S102, marking the training data based on preset classification categories.
After training data including positive samples and negative samples are obtained, an image labeling tool (Label Img) is adopted to Label and classify the ice-coated images of the power transmission line. The image annotation tool may be a QT5 graphical interface annotation tool written using Python.
In this embodiment, the labeling may be performed based on the acquired training data, and when the training data is only the transmission line icing image, the labeling may be performed based on whether the transmission line icing image includes an insulator or a conductor, to obtain respective category labels of the transmission line icing image, where the respective category labels include two respective category labels of an icing insulator and an icing conductor.
When labeling is carried out, regions containing the icing targets in the icing images of the power transmission line are also required to be labeled to obtain real frame coordinates of the icing targets, so that target candidate frames obtained through subsequent prediction can be compared with the real frames.
In some embodiments, when the training data includes an ice-free power transmission line image, the ice-free insulator and the ice-free conductor in the ice-free power transmission line image may also be labeled to obtain two respective category labels of the ice-free insulator or the ice-free conductor. Similar to the marking of the ice-coated image of the power transmission line, the real frame coordinates of the non-ice-coated insulator and the non-ice-coated conductor are marked and positioned. Namely, the electric transmission line icing identification model of the embodiment of the application can also identify insulators or wires in the electric transmission line image without icing.
In this embodiment, after a certain number of power transmission line images are acquired, the power transmission line images may be labeled based on the different classification category labels, and then a preset number of labeled power transmission line images are selected as training data. Or as in this embodiment, a certain number of power transmission line icing images are obtained first, and then the training data are labeled after the positive and negative samples are determined based on the icing target to be identified.
In a specific implementation, a part of the training data may be used as test data, so as to test the model after the subsequent model training is completed, and evaluate the performance of the model. For example, the collected data can be divided into a training data set and a testing data set according to the proportion of 1:1, so that the generalization capability of the trained model can be objectively measured.
S103, constructing a fast RCNN model network architecture.
After the training data are determined and the classification type labels and the real frame coordinates of the training data are labeled, a fast RCNN recognition model network architecture is constructed.
As shown in fig. 3, step S103 specifically includes:
s1031: and inputting the icing image of the power transmission line into a feature extraction network to generate a feature map.
The characteristic extraction network is a Residual error neural network Resnet50, the Residual error neural network (Resnet) adds a Residual error Learning (training) idea in the traditional convolutional neural network, the problems of gradient dispersion and accuracy reduction (training set) in a deep network are solved, the network can be deeper and deeper, the accuracy is guaranteed, and the speed is controlled. In this embodiment, as shown in fig. 2, the Resnet50 includes 49 convolutional layers and 1 fully-connected layer, the convolutional layers are mainly composed of 16 residual units, each residual unit is composed of a 1 × 1 convolution kernel and a 3 × 3 convolution kernel, the 1 × 1 convolution kernel is used to reduce the dimensionality, and the 3 × 3 convolution kernel is used to extract features. The residual unit may be implemented in the form of a skip-level connection, i.e. the input of the unit is added directly to the output of the unit and then reactivated. Therefore, the residual neural network can be easily realized by a mainstream automatic differential deep learning framework, and parameters are directly updated by using a BP algorithm, so that the problem of network degradation caused by the number of network layers is solved.
S1032: and extracting candidate areas containing icing targets from the feature map by using an area suggestion network to generate a target candidate box.
And generating an accurate target candidate box containing the icing target by a regional suggestion Network (RPN) through a classification function, and further accurately positioning the icing target.
S1033: and mapping the target candidate box to the feature map to obtain a feature matrix.
S1034: and constructing an ROI posing layer, and inputting the feature matrix into the ROI posing layer to obtain a target feature map with a fixed size.
Specifically, scaling each feature matrix through the ROI posing layer yields a target feature map of 7 × 7 size.
S1035: and constructing a full connection layer, and performing regression and classification on the target feature map by using the full connection layer to obtain frame coordinate parameters and classification results of the target candidate frame.
Flattening the target feature graph, performing regression and classification through a series of full connection layers, and calculating the class of a target candidate frame, wherein the target candidate frame can circle the icing target, so that the classification class of the icing target can be determined. Further, the frame coordinate parameters of the target candidate frame can be obtained through the regression calculation of the boundary frame, so that the final accurate position of the icing target is obtained.
And S104, training the fast RCNN model network architecture by using the training data to obtain a fast RCNN recognition model for recognizing the icing state of the power transmission line.
After the fast RCNN model network architecture is constructed, training data are input into the fast RCNN model network architecture, and the fast RCNN model network architecture can be trained until the value of a loss function reaches a loss threshold value, so that a fast RCNN recognition model is obtained.
In the embodiment, a mini-batch stochastic gradient descent method is adopted to train the fast RCNN recognition model network architecture. The batch value is 32, and the batch value is the number of the pictures processed in batch.
The initial learning rate of the fast RCNN recognition model network architecture training is 0.005, the momentum parameter (momentum) is 0.9, and the learning attenuation rate is 0.0005.
In some embodiments, the method further comprises:
s1051: calculating a loss value of a loss function;
s1052: updating the fast RCNN recognition model based on the loss value of the loss function.
The training loss comprises classification loss and regression loss, and the loss function is as follows:
wherein i represents the number of detection boxes (input images) in each small lot; p is a radical ofiIs shown with a detection box (Anchor)[i]) Prediction probability of target when predicting target (Anchor)[i]Is a positive sample) of p* i1, otherwise 0, i.e. Anchor[i]When being negative, p* i＝0；ti4 parametric coordinate vectors, t, representing target candidate boxes* iIndicating a correlation with positive samplesReal frame coordinates; n is a radical ofclsIndicates the number of all samples in a mini-batch, NregNumber L of detection frame positionsreg(ti，t* i) Represents the regression loss function, Lcls(pi，p* i) Representing a classification loss function; λ is weighted as one balance parameter, and λ is set to 10 by default, and the weights of the classification layer and the target regression layer are made substantially the same.
When the calculated loss value reaches a loss threshold (e.g., below 0.2), it may be determined that the Faster RCNN recognition model, which may be used for recognition of the icing state of the power transmission line, converges.
Fig. 4 is a graph of training loss and learning rate of the fast RCNN identification model according to the embodiment of the present disclosure, as shown in fig. 4, the abscissa represents the number of iterations, and the ordinate represents the learning rate (lr) and the loss rate (loss), respectively.
In this embodiment, after the classification result is obtained, the fast RCNN identification model can be tested and verified by using the test data.
After the training of the fast RCNN recognition model is completed, the method further comprises the following steps:
s1061: testing the recognition effect of the recognition model by using the accuracy, the recall rate and the average accuracy mean value;
s1062: and if the test result is lower than a preset threshold value, modifying the training parameters of the Faster RCNN recognition model, and training the Faster RCNN recognition model until the test result reaches the preset threshold value.
Wherein, the accuracy rate represents the ratio of a certain class really belonging to the class identified by the fast RCNN identification model; the recall rate represents the ratio of the number of a certain class identified by the fast RCNN identification model to the total number of the true class; both indexes reflect the recognition capability of the fast RCNN recognition model to a single classification category in the training data, for example, the recognition capability to the ice-coated insulator.
The average accuracy mean means that the average value of the accuracy mean values identified by all the classification categories is calculated, the identification capability of the fast RCNN identification model on all the classification categories in the whole training data can be reflected, and the method is a very useful index for evaluating the fast RCNN identification model (including an icing target positioning model and an icing target detection model).
Specifically, in this embodiment, the icing target is an icing insulator, and the Faster RCNN identification model includes the following four identification results:
(1) true Posives (TP): judging the positive type as a positive type, and identifying the icing insulator as an icing insulator;
(2) false Positives (FP): the negative type is judged to be the positive type, and the ice-coated wire is identified as an ice-coated insulator;
(3) false Negatives (FN): the positive type is judged to be the negative type, and the ice-coated insulator is identified to be an ice-coated wire;
(4) true Negatives (TN): the negative type is judged as the negative type, and the ice-coated wire is identified as not being an ice-coated insulator.
The accuracy (P) can be expressed as:
the recall ratio (R) may be expressed as:
the average accuracy mean can then be expressed as:
wherein, P is accuracy, R is recall, Q represents the number of classification categories, the integral of P to R is the area of the PR curve of a single classification category, which reflects the average of accuracy of a single classification category, and then the average is summed up for all classification categories to be the map average precision. The mAP is a percentage less than 1, and the value is better as being closer to 1, as shown in FIG. 4, in the embodiment, the mAP of the Faster RCNN recognition model is close to 95%, and the recognition capability is good.
According to the training method for the icing identification model of the power transmission line, the icing identification model of the power transmission line is obtained based on the fast RCNN deep convolutional neural network training, whether icing faults exist in images of the power transmission line can be rapidly and accurately identified, specific classification categories of the icing faults are accurately identified, the condition that maintenance personnel of the power transmission line monitor the faults of the power transmission line for a long time at the background is not needed, the working efficiency of the maintenance personnel of the power transmission line is improved to the maximum extent, and unnecessary workload is reduced. In addition, the residual error neural network resnet50 is adopted to extract the features of the input image, so that the problem of network degradation caused by the number of network layers is solved, and the identification accuracy and the identification efficiency are improved. In addition, data preprocessing is carried out after the power transmission line image is collected, and features of each angle of the power transmission line image can be extracted in modes of rotating different angles, mirroring, stretching and the like, so that multi-angle and multi-posture feature information of an icing target can be obtained, icing faults of the power transmission line can be accurately identified, the icing target identification is more objective, accurate, rapid and efficient, and the identification accuracy and robustness of a model are effectively improved; meanwhile, the generalization capability of the ice coating recognition model of the power transmission line and the processing capability of massive pictures can be improved.
The embodiment of the application further provides a method for identifying the icing state of the power transmission line, which comprises the following steps:
s201: collecting an icing image of the power transmission line;
s202: inputting the icing image of the power transmission line into a preset fast RCNN recognition model, and recognizing an icing target in the icing image of the power transmission line, wherein the icing target comprises an icing insulator or an icing conductor.
Similar to the training method of the power transmission line icing identification model, after the power transmission line icing image is collected, the power transmission line icing image is preprocessed to obtain a power transmission line icing image with uniform pixels, and data enhancement is performed so that different features in the power transmission line icing image can be extracted by using Resnet50 subsequently.
Inputting the preprocessed ice coating image of the power transmission line into the fast RCNN recognition model obtained in the steps from S101 to S104, so that the recognition result of the ice coating image of the power transmission line can be obtained, the ice coating target in the ice coating image can be accurately recognized, and the ice coating target can be accurately positioned.
Fig. 6(a) and 6(b) show recognition effect diagrams of the method for recognizing the icing state of the power transmission line according to the embodiment of the present application, and as shown in fig. 6(a), by using the method for recognizing the icing state of the power transmission line, insulators (jyz) in an image of the power transmission line without icing can be effectively recognized, and recognition accuracy of each insulator is high (most of the recognition accuracy can reach 95% or more, and part of the recognition accuracy can reach 99%). As shown in fig. 6(b), by the above-described method for identifying the icing state of the power transmission line, the icing conductor (fbdx) and the icing insulator (fbjyz) in the icing image of the power transmission line are effectively identified, and since the icing conductor is mainly identified (the portion enclosed by the target candidate frame) in fig. 6(b), the identification accuracy of the icing conductor is high (95% or more), and the identification accuracy of the icing insulator which is not the icing target can be low.
The embodiment of the application also provides a computer-readable storage medium, on which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the training method of the electric transmission line icing identification model and the identification method of the electric transmission line icing state in the embodiment are realized.
The processor executing computer-executable instructions described above may be a processing device, such as a microprocessor, Central Processing Unit (CPU), Graphics Processing Unit (GPU), etc., including one or more general purpose processing devices. More specifically, the processor may be a Complex Instruction Set Computing (CISC) microprocessor, Reduced Instruction Set Computing (RISC) microprocessor, Very Long Instruction Word (VLIW) microprocessor, processor running other instruction sets, or processors running a combination of instruction sets. The processor may also be one or more special-purpose processing devices such as an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), a system on a chip (SoC), or the like.
The computer-readable storage medium may be a memory, such as read-only memory (ROM), random-access memory (RAM), phase-change random-access memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), electrically erasable programmable read-only memory (EEPROM), other types of random-access memory (RAM), flash disk or other forms of flash memory, cache, registers, static memory, compact disk read-only memory (CD-ROM), Digital Versatile Disks (DVD) or other optical storage, magnetic cassettes or other magnetic storage devices, or any other possible non-transitory medium that may be used to store information or instructions that may be accessed by a computer device, and so forth.
The above embodiments are only exemplary embodiments of the present application, and are not intended to limit the present application, and the protection scope of the present application is defined by the claims. Various modifications and equivalents may be made by those skilled in the art within the spirit and scope of the present application and such modifications and equivalents should also be considered to be within the scope of the present application.
Claims (10)
1. A training method of an icing identification model of a power transmission line is characterized by comprising the following steps:
acquiring training data, wherein the training data comprises a positive sample and a negative sample, the positive sample is a power transmission line icing image containing an icing target, and the negative sample is a power transmission line icing image not containing the icing target;
labeling the training data based on a preset classification category;
constructing a fast RCNN recognition model network architecture;
and training the network architecture of the fast RCNN recognition model by using the training data to obtain the fast RCNN recognition model for recognizing the icing state of the power transmission line.
2. The method of claim 1, wherein after obtaining training data, the method further comprises:
and performing data preprocessing on the acquired ice coating image of the power transmission line, wherein the data preprocessing comprises performing data enhancement on the ice coating image of the power transmission line in at least one of translation, rotation, mirror image or expansion.
3. The method according to claim 1, wherein constructing a fast RCNN recognition model network architecture comprises:
inputting the icing image of the power transmission line into a feature extraction network to generate a feature map;
extracting candidate areas containing icing targets from the feature map by using an area suggestion network to generate target candidate frames;
mapping the target candidate box to the feature map to obtain a feature matrix;
constructing an ROI posing layer, and inputting the feature matrix into the ROI posing layer to obtain a target feature map with a fixed size;
and constructing a full connection layer, and performing regression and classification on the target feature map by using the full connection layer to obtain frame coordinate parameters and classification results of the target candidate frame.
4. The method of claim 3, wherein the feature extraction network is a residual extraction network resnet50, and comprises 49 convolutional layers and 1 fully-connected layer, the convolutional layers comprise 16 residual units, each residual unit comprises a 1 x 1 convolution kernel and a 3 x 3 convolution kernel, the 1 x 1 convolution kernel is used for dimensionality reduction, and the 3 x 3 convolution kernel is used for feature extraction.
5. The method according to claim 1, wherein training the fast RCNN recognition model network architecture using the training data comprises:
training the fast RCNN recognition model network architecture by adopting a mini-batch stochastic gradient descent method;
wherein the initial learning rate of the fast RCNN recognition model network architecture training is 0.005, the momentum parameter is 0.9, and the learning attenuation rate is 0.0005.
6. The method of claim 3, further comprising:
calculating a loss value of a loss function;
updating the fast RCNN recognition model based on the loss value of the loss function;
wherein the loss function is:
wherein i represents the serial number of the detection box in each small batch; p is a radical ofiRepresenting the prediction probability of targeting the detection box, p when the prediction is targeted* i1, otherwise 0; t is ti4 parametric coordinate vectors, t, representing target candidate boxes* iRepresenting a real box coordinate associated with the positive sample; n is a radical ofclsIndicates the number of all samples in a mini-batch, NregIndicating the number of detection frame positions
Lreg(ti,t* i) Represents the regression loss function, Lcls(pi,p* i) Representing a classification loss function; λ is the equilibrium parameter, λ 10.
7. The method of claim 1, further comprising:
testing the recognition effect of the recognition model by using the accuracy, the recall rate and the average accuracy mean value;
and if the test result is lower than a preset threshold value, modifying the training parameters of the Faster RCNN recognition model, and training the Faster RCNN recognition model until the test result reaches the preset threshold value.
8. The method of claim 1, wherein the training data further comprises non-iced transmission line images, and wherein labeling the training data based on a preset classification category comprises:
and marking the non-icing insulator and the non-icing conductor in the non-icing power transmission line image.
9. A method for identifying the icing state of a power transmission line is characterized by comprising the following steps:
collecting an icing image of the power transmission line;
inputting the icing image of the power transmission line into a preset fast RCNN recognition model, and recognizing an icing target in the icing image of the power transmission line, wherein the icing target comprises an icing insulator or an icing conductor.
10. A computer-readable storage medium having stored thereon computer-executable instructions which, when executed by a processor, implement a method of training a power transmission line icing identification model according to any one of claims 1-8 or a method of identifying a power transmission line icing condition according to claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110539796.8A CN113449769A (en) | 2021-05-18 | 2021-05-18 | Power transmission line icing identification model training method, identification method and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110539796.8A CN113449769A (en) | 2021-05-18 | 2021-05-18 | Power transmission line icing identification model training method, identification method and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113449769A true CN113449769A (en) | 2021-09-28 |
Family
ID=77810063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110539796.8A Pending CN113449769A (en) | 2021-05-18 | 2021-05-18 | Power transmission line icing identification model training method, identification method and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113449769A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113744267A (en) * | 2021-11-04 | 2021-12-03 | 智洋创新科技股份有限公司 | Method for detecting icing and estimating thickness of transmission conductor based on deep learning |
| CN114549914A (en) * | 2022-04-24 | 2022-05-27 | 中国民航大学 | Airplane large-area icing detection method and detection system |
| CN114723678A (en) * | 2022-03-21 | 2022-07-08 | 盛视科技股份有限公司 | High-voltage wire foreign matter detection method and detection system based on video image |
| CN114926723A (en) * | 2022-06-10 | 2022-08-19 | 国网江苏省电力有限公司电力科学研究院 | Method, terminal and storage medium for identifying and triggering alarm of interference objects around power transmission line |
| CN117494026A (en) * | 2023-12-28 | 2024-02-02 | 国网浙江省电力有限公司金华供电公司 | Method, system and storage medium for positioning icing fault of power transmission line under cold and tidal weather |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070530A (en) * | 2019-04-19 | 2019-07-30 | 山东大学 | A kind of powerline ice-covering detection method based on deep neural network |
| US20200075019A1 (en) * | 2017-08-02 | 2020-03-05 | Veritone, Inc. | System and method for neural network orchestration |
| CN111738307A (en) * | 2020-06-02 | 2020-10-02 | 国网宁夏电力有限公司检修公司 | Foreign matter identification method and system in power transmission line environment based on fast RCNN and computer readable storage medium |
| CN112183182A (en) * | 2020-08-11 | 2021-01-05 | 成都卡普数据服务有限责任公司 | Defect identification method for power transmission line |
| CN113221614A (en) * | 2020-12-23 | 2021-08-06 | 南方电网电力科技股份有限公司 | Power grid insulator damage image identification method based on hybrid neural network |
-
2021
- 2021-05-18 CN CN202110539796.8A patent/CN113449769A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200075019A1 (en) * | 2017-08-02 | 2020-03-05 | Veritone, Inc. | System and method for neural network orchestration |
| CN110070530A (en) * | 2019-04-19 | 2019-07-30 | 山东大学 | A kind of powerline ice-covering detection method based on deep neural network |
| CN111738307A (en) * | 2020-06-02 | 2020-10-02 | 国网宁夏电力有限公司检修公司 | Foreign matter identification method and system in power transmission line environment based on fast RCNN and computer readable storage medium |
| CN112183182A (en) * | 2020-08-11 | 2021-01-05 | 成都卡普数据服务有限责任公司 | Defect identification method for power transmission line |
| CN113221614A (en) * | 2020-12-23 | 2021-08-06 | 南方电网电力科技股份有限公司 | Power grid insulator damage image identification method based on hybrid neural network |
Non-Patent Citations (3)
| Title |
|---|
| 何拓 等: "木材树种计算机视觉识别技术发展与应用", 《林业工程学报》 * |
| 彭明霞 等: "融合FPN 的Faster R-CNN 复杂背景下棉田杂草高效识别方法", 《农业工程学报》 * |
| 潘卫国 等: "基于Faster-RCNN 的交通信号灯检测与识别", 《传感器与微系统》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113744267A (en) * | 2021-11-04 | 2021-12-03 | 智洋创新科技股份有限公司 | Method for detecting icing and estimating thickness of transmission conductor based on deep learning |
| CN114723678A (en) * | 2022-03-21 | 2022-07-08 | 盛视科技股份有限公司 | High-voltage wire foreign matter detection method and detection system based on video image |
| CN114549914A (en) * | 2022-04-24 | 2022-05-27 | 中国民航大学 | Airplane large-area icing detection method and detection system |
| CN114549914B (en) * | 2022-04-24 | 2022-07-12 | 中国民航大学 | Airplane large-area icing detection method and detection system |
| CN114926723A (en) * | 2022-06-10 | 2022-08-19 | 国网江苏省电力有限公司电力科学研究院 | Method, terminal and storage medium for identifying and triggering alarm of interference objects around power transmission line |
| CN114926723B (en) * | 2022-06-10 | 2024-05-28 | 国网江苏省电力有限公司电力科学研究院 | Method, terminal and storage medium for identifying and alarming and triggering peripheral interferents of power transmission line |
| CN117494026A (en) * | 2023-12-28 | 2024-02-02 | 国网浙江省电力有限公司金华供电公司 | Method, system and storage medium for positioning icing fault of power transmission line under cold and tidal weather |
| CN117494026B (en) * | 2023-12-28 | 2024-04-05 | 国网浙江省电力有限公司金华供电公司 | Method, system and storage medium for positioning icing fault of power transmission line under cold and tidal weather |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113449769A (en) | Power transmission line icing identification model training method, identification method and storage medium | |
| CN113592822B (en) | Insulator defect positioning method for electric power inspection image | |
| Ou et al. | Infrared image target detection of substation electrical equipment using an improved faster R-CNN | |
| Adou et al. | Insulator faults detection based on deep learning | |
| CN115909093A (en) | Power equipment fault detection method based on unmanned aerial vehicle inspection and infrared image semantic segmentation | |
| Zheng et al. | Component identification and defect detection in transmission lines based on deep learning | |
| CN111462109A (en) | Defect detection method, device and equipment for strain clamp and storage medium | |
| CN111862013A (en) | Insulator detection method, device and equipment based on deep convolutional neural network | |
| CN113252701B (en) | Cloud edge cooperation-based power transmission line insulator self-explosion defect detection system and method | |
| CN112380944B (en) | Method and system for evaluating structural state of transmission tower based on satellite remote sensing | |
| CN115205256A (en) | Power transmission line insulator defect detection method and system based on fusion of transfer learning | |
| Xu et al. | Detection method of insulator based on single shot multibox detector | |
| CN115330718A (en) | Insulator defect detection method based on improved YOLOv4 network | |
| Bai et al. | Insulator fault recognition based on spatial pyramid pooling networks with transfer learning (match 2018) | |
| CN113012107B (en) | Power grid defect detection method and system | |
| Zhang et al. | Improved YOLOv8 Insulator Fault Detection Algorithm Based on BiFormer | |
| CN116503612B (en) | Fan blade damage identification method and system based on multitasking association | |
| CN115565118B (en) | Method for identifying single hanging point and single string of cross crossing point of power transmission line | |
| Chen et al. | Real-time detection of UAV detection image of power line insulator bursting based on YOLOV3 | |
| CN115100546A (en) | Mobile-based small target defect identification method and system for power equipment | |
| CN114418968A (en) | Power transmission line small target defect detection method based on deep learning | |
| CN113033556A (en) | Insulator rapid distinguishing and positioning method and system based on machine vision | |
| CN112651954A (en) | Method and device for detecting insulator string dropping area | |
| Wang et al. | High-Voltage Transmission Line Foreign Object and Power Component Defect Detection Based on Improved YOLOv5 | |
| Luo et al. | SOLOv2-cable: A Power Cable Segmentation Algorithm in Complex Scenarios |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210928 |