CN114187505A - Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment - Google Patents
Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment Download PDFInfo
- Publication number
- CN114187505A CN114187505A CN202111349907.5A CN202111349907A CN114187505A CN 114187505 A CN114187505 A CN 114187505A CN 202111349907 A CN202111349907 A CN 202111349907A CN 114187505 A CN114187505 A CN 114187505A
- Authority
- CN
- China
- Prior art keywords
- damper
- defect sample
- defect
- labeled
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
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
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
Abstract
The invention discloses a method for detecting falling of a damper of a power transmission line, which comprises the following steps: acquiring a damper image sample set; preprocessing an anti-vibration hammer image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set; training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through a normal sample training set and a defect sample training set to obtain an initial detection model; testing and analyzing the initial detection model through a defect sample test set and average precision to obtain a vibration damper falling-off detection model; and inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain the falling detection result of the damper. The method can effectively transfer the knowledge of a large number of non-defective samples to a small number of defective samples, realizes the construction of the vibration damper falling-off detection model by using a small number of defective samples, and has high adaptation speed and high detection precision compared with the traditional vibration damper falling-off detection method.
Description
Technical Field
The invention relates to the technical field of remote sensing images, in particular to a method and a device for detecting falling of a damper of a power transmission line, a computer readable storage medium and terminal equipment.
Background
Ensuring the reliability and stability of the power transmission line is an important content for the construction of the smart power grid. The damper is used as a key device for protecting the aging and damage of the transmission line, has the function of reducing the vibration frequency and amplitude of the lead, and is widely applied to the high-voltage transmission line. The fixed wire clamp of the damper is easily loosened due to the influence of natural environments such as insolation, rainwater, lightning stroke and the like all the year round, so that the defects of falling, displacement and the like can be caused, and accidents such as strand breakage of a wire, strand scattering of the wire, short circuit between wires, hardware damage and the like can be caused over time. The safe and stable operation of the transmission line can be seriously influenced by the performance failure of the damper, so that the timely detection of various defects of the damper and the rapid fault diagnosis are very important.
The territory of China is vast, and there are many different topography, and in the circuit field of patrolling and examining, because transmission line corridor terrain environment is complicated, can't in time feed back the operation condition of transmission line, under some harsh environmental conditions such as high altitude area and the area of environment abominable, there are not patrolling roads basically in some areas along this district, and the electric power construction in this kind of area has many difficulties. Therefore, the manual inspection method is time-consuming and labor-consuming, the working conditions are extremely difficult, and the subsequent maintenance work is difficult. In recent years, electric power robots develop rapidly and exhibit the advantages of exquisite structure, flexible performance, strong concealment, clear collected images and the like. The inspection by the aid of the electric robot is efficient, and inspection quality and efficiency can be improved. The electric power robot firstly collects the power transmission line inspection image, and then realizes fault detection of the damper through technologies such as later-stage image processing and deep learning, and further promotes intelligent construction of power transmission line detection.
At present, few researches are conducted on detection of defects of the damper in the power transmission line, and the existing damper defect detection algorithm mainly adopts a sliding window-based region selection strategy and a combination characteristic-based cascading strategy. However, the algorithm is difficult to solve the problems that the intelligent inspection equipment such as the power robot detects the damper in multiple angles, the target size is small, the background is complex and changeable, and the like, so that the detection accuracy is low. With the occurrence of a deep learning algorithm, the vibration damper defect detection technology is in breakthrough progress, and the bottleneck that only shallow layer features can be extracted by the conventional target detection algorithm is broken through by utilizing a deep convolutional neural network, so that the vibration damper defect detection performance under a complex background is improved.
Disclosure of Invention
The embodiment of the invention provides an optimization method and device of a vibration damper falling detection model, terminal equipment and a storage medium, which can improve the adaptation speed and detection precision of vibration damper falling detection.
In order to achieve the above object, an embodiment of the present invention provides a method for detecting falling of a damper of a power transmission line, including:
acquiring a damper image sample set;
preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set;
training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model;
testing and analyzing the initial detection model through the defect sample test set and the average precision to obtain a vibration damper falling-off detection model;
and inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain a falling detection result of the damper.
As an improvement of the above scheme, the obtaining of the normal sample training set, the defect sample training set, and the defect sample testing set by preprocessing the damper image sample set specifically includes:
dividing the damper image sample set into a normal sample set, a first defect sample set and a second defect sample set;
respectively labeling the normal sample set, the first defect sample set and the second defect sample set by an image calibration tool to obtain a labeled normal sample set, a labeled first defect sample set and a labeled second defect sample set;
and respectively carrying out image enhancement on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set.
As an improvement of the above scheme, the performing image enhancement on the labeled normal sample set, the labeled first defect sample set, and the labeled second defect sample set respectively specifically includes:
horizontally overturning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and clockwise rotating according to a preset angle; or the like, or, alternatively,
horizontally overturning the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and introducing noise;
horizontally turning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and introducing weather factors; or the like, or, alternatively,
performing defocusing blurring on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and performing clockwise rotation according to a preset angle; or the like, or, alternatively,
performing defocus blur and introducing noise on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set; or the like, or, alternatively,
and performing out-of-focus blur and introducing weather factors on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set.
As an improvement of the above scheme, the FasterR-CNN network established in advance based on the transfer learning algorithm is trained through the normal sample training set and the defect sample training set to obtain an initial detection model, which specifically includes:
inputting the normal sample training set into a first feature extraction network to obtain a normal sample feature map;
inputting the defect sample training set into a second feature extraction network, and migrating the normal sample feature map into the feature space of the defect sample based on migration learning to obtain a defect sample feature map;
inputting the normal sample feature map and the defect sample feature map into a regional suggestion network to generate a plurality of suggestion windows;
inputting the normal sample feature map, the defect sample feature map and a plurality of the suggestion windows into a region of interest, and outputting a feature map of a fixed size of each suggestion window;
inputting the feature map with the fixed size of each suggestion window into a classifier of the FasterR-CNN network, and outputting a maximum probability estimation and a target bounding box position;
and taking the suggestion window corresponding to the maximum probability estimation as an initial detection model.
As an improvement of the above scheme, in the process of training the faster r-CNN network built in advance based on the transfer learning algorithm by using the normal sample training set and the defect sample training set, the method further includes:
and constructing a loss function of the FasterR-CNN network, and iteratively updating the network weight of the FasterR-CNN network through the loss function and a random gradient descent algorithm.
As an improvement of the above scheme, the expression of the loss function L is:
L＝LS+LT；
wherein L isSAs a function of the source domain loss of the FasterR-CNN network, LTIs a target domain loss function of the FasterR-CNN network;
source domain loss function L of the FasterR-CNN networkSThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification loss of the source domain,
as a regression loss function of the source domain, NclsNumber of samples lost for classification, NregNumber of samples lost to regression, l is the equilibrium parameter, piFor the prediction probability that the ith anchor frame belongs to the target in the ith true annotated image,
for the ith real annotation image, tiFor the four-dimensional coordinates of the ith prediction bounding box,
labeling the ith real four-dimensional coordinate, u, of the bounding boxiA set of four-dimensional coordinates of the ith prediction bounding box and the four-dimensional coordinates of the real labeling bounding box;
target domain loss function L of the FasterR-CNN networkTThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification error of the target domain,
is the regression error function of the target domain.
As an improvement of the above scheme, the testing and analyzing of the initial detection model through the defect sample test set and the average precision to obtain the vibration damper falling-off detection model specifically includes:
inputting the defect sample test set into the initial detection model to obtain a detection result of the defect sample test set;
and when the average precision of the detection results of the defect sample test set is greater than a preset threshold value, taking the initial detection model as a vibration damper falling-off detection model.
In order to achieve the above object, an embodiment of the present invention correspondingly provides a device for detecting falling of a damper of a power transmission line, including:
the data receiving module is used for acquiring a damper image sample set;
the data processing module is used for preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set;
the model training module is used for training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model;
the sample testing module is used for testing and analyzing the initial detection model through the defect sample testing set and the average precision to obtain a vibration damper falling-off detection model;
and the falling detection module is used for inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain the falling detection result of the damper.
In order to achieve the above object, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program, and when the computer program runs, the apparatus where the computer-readable storage medium is located is controlled to execute the method for detecting the drop of the damper of the power transmission line according to the above embodiment of the present invention.
In order to achieve the above object, an embodiment of the present invention further provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, where the processor executes the computer program to implement the method for detecting the drop of the damper of the power transmission line according to the above embodiment of the present invention.
Compared with the prior art, the detection method and device for the vibration damper falling off of the power transmission line, the computer readable storage medium and the terminal device disclosed by the embodiment of the invention have the advantages that firstly, a vibration damper image sample set is obtained; secondly, preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set; then, training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model; further, the initial detection model is subjected to test analysis through the defect sample test set and the average precision, and a vibration damper falling-off detection model is obtained; and finally, inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain a falling detection result of the damper. Therefore, knowledge of a large number of non-defective samples can be effectively transferred to a small number of defective samples, a vibration damper falling detection model is constructed by the small number of defective samples, and compared with the traditional vibration damper falling detection method, the vibration damper falling detection method is high in adaptation speed and detection precision.
Drawings
Fig. 1 is a schematic flow chart of a method for detecting falling of a damper of a power transmission line according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a vibration damper falling-off detection model according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a detection apparatus for detecting falling of a damper of a power transmission line according to an embodiment of the present invention;
fig. 4 is a block diagram of a terminal device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of a method for detecting falling of a damper of a power transmission line according to an embodiment of the present invention.
The method for detecting the falling of the damper of the power transmission line provided by the embodiment of the invention comprises the following steps:
s11, obtaining a damper image sample set;
s12, preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set;
s13, training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model;
s14, carrying out test analysis on the initial detection model through the defect sample test set and the average precision to obtain a vibration damper falling-off detection model;
and S15, inputting the pre-acquired damper image to be detected into the damper falling detection model to obtain the falling detection result of the damper.
In some preferred embodiments, the step S12 specifically includes:
dividing the damper image sample set into a normal sample set, a first defect sample set and a second defect sample set;
respectively labeling the normal sample set, the first defect sample set and the second defect sample set by an image calibration tool to obtain a labeled normal sample set, a labeled first defect sample set and a labeled second defect sample set;
and respectively carrying out image enhancement on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set.
Preferably, the sample ratio of the first defect sample set to the second defect sample set is 4: 1.
the first defect sample set is used for training a defect sample detection model, and a defect sample training set is obtained by labeling and image enhancing the first defect sample set; and the second defect sample set is used for testing a detection model, and a defect sample test set is obtained by labeling and image enhancing the second defect sample set.
Preferably, the image calibration tool is a LabelImg tool.
It should be noted that, after the normal sample set, the first defect sample set, and the second defect sample set are labeled by the label img tool, the obtained labeled normal sample set, the labeled first defect sample set, and the labeled second defect sample set are xml files. Due to the complexity of the defect condition of falling off of the damper, the labeling object is divided into the following two conditions during labeling: (1) if the damper falls off locally, only marking a defect target of a single damper; (2) if the damper falls off completely, a reference target, namely a defect target and a matching target thereof, needs to be set.
In a specific embodiment, the performing image enhancement on the labeled normal sample set, the labeled first defect sample set, and the labeled second defect sample set respectively specifically includes:
horizontally overturning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and clockwise rotating according to a preset angle; or the like, or, alternatively,
horizontally overturning the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and introducing noise;
horizontally turning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and introducing weather factors; or the like, or, alternatively,
performing defocusing blurring on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and performing clockwise rotation according to a preset angle; or the like, or, alternatively,
performing defocus blur and introducing noise on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set; or the like, or, alternatively,
and performing out-of-focus blur and introducing weather factors on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set.
Preferably, the angle of clockwise rotation comprises at least one of: 30 °, 60 °, 270 °, and 330 °.
Preferably, the noise includes: gaussian noise, shot noise and impulse noise.
Preferably, the weather factors include: rain, snow and fog.
It can be understood that because the data volume of the damper of the power transmission line is small, image enhancement needs to be performed on the training sample and the test sample to improve the robustness and the detection accuracy of the system,
it should be noted that, in the step S13, the FasterR-CNN network constructed in advance based on the migration learning algorithm is mainly composed of four parts, namely, a feature extraction network, a feature adaptive migration module, an area suggestion network, and an area of interest. The feature extraction network learns the features of the normal sample and the defect sample, the feature self-adaptive migration module effectively migrates the normal sample information into the defect sample, and then the defect detection result is output through the area suggestion network and the region of interest. It should be noted that in fig. 2, the feature extraction network is also called a feature extractor, the feature adaptive migration module is also called a feature migration module, and the region of interest is also called a prediction layer.
Referring to fig. 2, fig. 2 is a schematic diagram of a damper drop-out detection model according to an embodiment of the present invention.
As can be seen from fig. 2, two feature extraction networks are constructed in the present invention, which are hereinafter referred to as a first feature extraction network and a second feature extraction network, and the two feature extraction networks are constructed because a strongly shared feature extraction structure shows a feature support for a specific domain to a large extent, and because the distribution of the source domain and the target domain of the FasterR-CNN network is inconsistent, when a normal sample feature map is migrated to a defective sample feature space, the source domain has harmful information for the target domain, which causes the structure of the target domain to be damaged, which affects the migration effect, so that the feature extraction networks are not shared in the training process of the source domain and the target domain in the present invention.
Preferably, the FasterR-CNN network established in advance based on the migration learning algorithm includes: the system comprises a first feature extraction network, a second feature extraction network, a region suggestion network, a feature adaptive migration module, a region of interest and a classifier of a FasterR-CNN network.
In a more preferred embodiment, the step S13 specifically includes:
inputting the normal sample training set into a first feature extraction network to obtain a normal sample feature map;
inputting the defect sample training set into a second feature extraction network, and migrating the normal sample feature map into the feature space of the defect sample based on migration learning to obtain a defect sample feature map;
inputting the normal sample feature map and the defect sample feature map into a regional suggestion network to generate a plurality of suggestion windows;
inputting the normal sample feature map, the defect sample feature map and a plurality of the suggestion windows into a region of interest, and outputting a feature map of a fixed size of each suggestion window;
inputting the feature map with the fixed size of each suggested window into a classifier of a FasterR-CNN network, and outputting a maximum probability estimation and a target bounding box position;
and taking the suggestion window corresponding to the maximum probability estimation as an initial detection model.
It should be noted that the first feature extraction network and the second feature extraction network both adopt a structure combining ResNet50 and a feature pyramid, and can effectively merge features of each layer through a multidirectional information transfer manner; and inputting the learned normal sample feature map into a feature adaptive migration module to obtain a re-weighting vector, and then effectively migrating the re-weighting vector into the feature space of the defect sample so as to improve the feature expression capability of the defect sample.
The characteristic pyramid structure comprises three parts of bottom-to-top, top-to-bottom and transverse connection, and the characteristic pyramid structure can fuse the characteristics of all levels, simultaneously has strong spatial information and strong context information, and can enhance the characteristic learning capability. The ResNet50 is used as a basic backbone network from bottom to top, and is divided into 5 stages C1-C5 according to the size of a feature diagram; from top to bottom, nearest neighbor upsampling is carried out from the highest layer; the cross-linking is to fuse the bottom-up captured feature map with the same size results generated by the upsampling. Illustratively, each layer in C2-C5 is subjected to a 1x1 convolution operation and then summed with the upsampled feature map; to eliminate aliasing effects caused by upsampling, the feature map after fusion is operated on using a convolution kernel of 3x 3.
The method comprises the steps that a source domain image (namely an image of a normal sample training set) passes through a first feature extraction network to obtain a normal sample feature map of a corresponding layer, the normal sample feature map and the normal sample feature map are input into a lightweight feature adaptive migration module together to obtain a re-weighting vector of the normal sample feature map corresponding to a source domain, then the re-weighting vector is used as a convolution kernel, convolution operation is respectively carried out on the feature maps corresponding to a target domain, and a re-weighted target domain feature map, namely a defect sample feature map, is obtained; then, inputting the normal sample feature map and the defect sample feature map into a regional recommendation network (RPN) to generate a plurality of recommendation windows; then mapping a plurality of suggestion windows onto the last layer of convolution feature map of the convolution neural network of the first feature extraction network and the second feature extraction network, and generating a plurality of feature maps with fixed sizes of the suggestion windows through the pooling layer of the ROI so that the RPN generates a batch of interest regions with features of different proportions according to the anchor frame; and finally, inputting the feature maps with fixed sizes of the plurality of the suggested windows into a classifier of the Faster R-CNN network, wherein the classifier can finally generate the maximum probability estimation of the static target, optimize the position of a boundary box of the target and output a detection result.
As an optional embodiment, in step S14, in the training process of the FasterR-CNN network constructed in advance based on the migration learning algorithm by using the normal sample training set and the defect sample training set, the method further includes:
and constructing a loss function of the FasterR-CNN network, and iteratively updating the network weight of the FasterR-CNN network through the loss function and a random gradient descent algorithm.
Specifically, the expression of the loss function L is:
L＝LS+LT；
wherein L isSAs a function of the source domain loss of the FasterR-CNN network, LTIs a target domain loss function of the FasterR-CNN network;
source domain loss function L of the FasterR-CNN networkSThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification loss of the source domain,
as a regression loss function of the source domain, NclsNumber of samples lost for classification, NregNumber of samples lost to regression, l is the equilibrium parameter, piFor the prediction probability that the ith anchor frame belongs to the target in the ith true annotated image,
for the ith real annotation image, tiFor the four-dimensional coordinates of the ith prediction bounding box,
labeling the ith real four-dimensional coordinate, u, of the bounding boxiIs the ithPredicting a set of four-dimensional coordinates of the bounding box and the four-dimensional coordinates of the real labeling bounding box;
target domain loss function L of the FasterR-CNN networkTThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification error of the target domain,
is the regression error function of the target domain.
Further, the classification loss function and the regression loss function of the source domain or the target domain are calculated according to the following formulas:
wherein L isclsAs a function of classification loss, LregAs a function of the regression loss, SL1For the smoothing formula, x is the difference between the four-dimensional coordinates of the predicted bounding box and the four-dimensional coordinates of the true labeled bounding box.
Further, the network weight of the FasterR-CNN network is iteratively updated through the loss function and a random gradient descent algorithm, specifically:
updating the network weights of the FasterR-CNN network according to the following formula:
α＝base_α×0.1floor(iter/5000)；
wherein, the Wt+1Network weight of round t +1, WtIs the network weight of the t-th round, mu is the weight of the gradient value of the t-1 th round, VtFor the updated value of the weight of the t-th round,
and (4) carrying out derivation on the network weight of the t-th round for a loss function, wherein alpha is a learning rate, base _ alpha is an initial value of the learning rate, and iter is the current iteration number.
Preferably, the initial value of the learning rate is 0.001.
Note that the four-dimensional coordinate t of the ith prediction bounding boxi＝{tx,ty,tw,thI.e. four parameterized coordinates.
In some preferred embodiments, the step S15 specifically includes:
inputting the defect sample test set into the initial detection model to obtain a detection result of the defect sample test set;
and when the average precision of the detection results of the defect sample test set is greater than a preset threshold value, taking the initial detection model as a vibration damper falling-off detection model.
Specifically, the detection result of the defect sample test set includes: the defect sample test set is centered on the detection regression box and the classification accuracy for each image.
It is required to be noted that before calculating the average precision of the detection results of the defect sample test set, the accuracy and recall rate of the detection results need to be calculated; wherein the content of the first and second substances,
calculating the accuracy of the detection result Pre according to the following formula:
calculating the Recall rate Recall of the detection results according to the following formula:
wherein N isTPNumber of samples actually true and predicted to be true, NFPNumber of samples that are actually false but predicted to be true, NFNIs the number of samples that are actually true but predicted to be false.
It can be understood that the accuracy can reflect the false detection condition of the detection model to a certain extent, and the recall rate can reflect the missed detection condition of the detection model to a certain extent. And constructing a PR curve by taking the recall rate as a horizontal axis and the accuracy rate as a vertical axis, wherein the average precision AP is the average value of the accuracy rates on the PR curve.
Correspondingly, the embodiment of the invention also provides a device for detecting the falling of the damper of the power transmission line, which can realize all the processes of the method for detecting the falling of the damper of the power transmission line.
Fig. 3 is a schematic structural diagram of a detection apparatus for detecting falling of a damper of a power transmission line according to an embodiment of the present invention.
The detection device for the falling of the damper of the power transmission line provided by the embodiment of the invention comprises:
the data receiving module 21 is configured to obtain a damper image sample set;
the data processing module 22 is configured to perform preprocessing on the damper image sample set to obtain a normal sample training set, a defect sample training set, and a defect sample testing set;
the model training module 23 is configured to train a FasterR-CNN network, which is established in advance based on a transfer learning algorithm, through the normal sample training set and the defect sample training set to obtain an initial detection model;
the sample testing module 24 is used for testing and analyzing the initial detection model through the defect sample testing set and the average precision to obtain a vibration damper falling-off detection model;
and the falling detection module 25 is configured to input a pre-acquired damper image to be detected to the damper falling detection model to obtain a falling detection result of the damper.
As one optional implementation, the data processing module 22 includes:
the sample dividing unit is used for dividing the damper image sample set into a normal sample set, a first defect sample set and a second defect sample set;
the sample labeling unit is used for labeling the normal sample set, the first defect sample set and the second defect sample set respectively through an image calibration tool to obtain a labeled normal sample set, a labeled first defect sample set and a labeled second defect sample set;
and the image enhancement unit is used for respectively carrying out image enhancement on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set.
Further, the image enhancement unit is specifically configured to:
horizontally overturning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and clockwise rotating according to a preset angle; or the like, or, alternatively,
horizontally overturning the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and introducing noise;
horizontally turning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and introducing weather factors; or the like, or, alternatively,
performing defocusing blurring on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and performing clockwise rotation according to a preset angle; or the like, or, alternatively,
performing defocus blur and introducing noise on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set; or the like, or, alternatively,
and performing out-of-focus blur and introducing weather factors on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set.
As a specific embodiment, the model training module 23 is specifically configured to:
inputting the normal sample training set into a first feature extraction network to obtain a normal sample feature map;
inputting the defect sample training set into a second feature extraction network, and migrating the normal sample feature map into the feature space of the defect sample based on migration learning to obtain a defect sample feature map;
inputting the normal sample feature map and the defect sample feature map into a regional suggestion network to generate a plurality of suggestion windows;
inputting the normal sample feature map, the defect sample feature map and a plurality of the suggestion windows into a region of interest, and outputting a feature map of a fixed size of each suggestion window;
inputting the feature map with the fixed size of each suggested window into a classifier of a FasterR-CNN network, and outputting a maximum probability estimation and a target bounding box position;
and taking the suggestion window corresponding to the maximum probability estimation as an initial detection model.
As one preferred embodiment, in the model training module 23, in the process of training the fasterrr-CNN network built in advance based on the migration learning algorithm through the normal sample training set and the defect sample training set, the method further includes:
and constructing a loss function of the FasterR-CNN network, and iteratively updating the network weight of the FasterR-CNN network through the loss function and a random gradient descent algorithm.
Specifically, the expression of the loss function L is:
L＝LS+LT；
wherein L isSAs a function of the source domain loss of the FasterR-CNN network, LTIs a target domain loss function of the FasterR-CNN network;
source domain loss function L of the FasterR-CNN networkSThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification loss of the source domain,
as a regression loss function of the source domain, NclsNumber of samples lost for classification, NregNumber of samples lost to regression, l is the equilibrium parameter, piFor the prediction probability that the ith anchor frame belongs to the target in the ith true annotated image,
for the ith real annotation image, tiTo predict the four-dimensional coordinates of the bounding box,
four-dimensional coordinates, u, of the bounding box of the ith real annotation imageiA set of four-dimensional coordinates of the ith prediction bounding box and the four-dimensional coordinates of the real labeling bounding box;
target domain loss function L of the FasterR-CNN networkTThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification error of the target domain,
is the regression error function of the target domain.
As an optional embodiment, the sample testing module 24 includes:
inputting the defect sample test set into the initial detection model to obtain a detection result of the defect sample test set;
and when the average precision of the detection results of the defect sample test set is greater than a preset threshold value, taking the initial detection model as a vibration damper falling-off detection model.
It should be noted that, for the specific description and the beneficial effects related to each embodiment of the detection apparatus for detecting the falling-off of the damper of the power transmission line of the present embodiment, reference may be made to the specific description and the beneficial effects related to each embodiment of the detection method for detecting the falling-off of the damper of the power transmission line, and details are not repeated herein.
It should be noted that the above-described device embodiments are merely illustrative, where the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. In addition, in the drawings of the embodiment of the apparatus provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, and may be specifically implemented as one or more communication buses or signal lines. One of ordinary skill in the art can understand and implement it without inventive effort.
Accordingly, an embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program; when the computer program runs, the device where the computer readable storage medium is located is controlled to execute the method for detecting the falling-off of the damper of the power transmission line according to any one of the above embodiments.
An embodiment of the present invention further provides a terminal device, referring to fig. 4, which is a block diagram of a structure of a terminal device according to an embodiment of the present invention, and the terminal device includes a processor 10, a memory 20, and a computer program stored in the memory 20 and configured to be executed by the processor 10, where the processor 10, when executing the computer program, implements the method for detecting the drop of the damper of the power transmission line according to any of the embodiments.
Preferably, the computer program may be divided into one or more modules/units (e.g., computer program 1, computer program 2, … …) that are stored in the memory 20 and executed by the processor 10 to implement the present invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used for describing the execution process of the computer program in the terminal device.
The Processor 10 may be a Central Processing Unit (CPU), other general-purpose Processor, a Digital Signal Processor (DSP), an application specific integrated circuit (AStC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, a discrete hardware component, etc., the general-purpose Processor may be a microprocessor, or the Processor 10 may be any conventional Processor, the Processor 10 is a control center of the terminal device, and various interfaces and lines are used to connect various parts of the terminal device.
The memory 20 mainly includes a program storage area that may store an operating system, an application program required for at least one function, and the like, and a data storage area that may store related data and the like. In addition, the memory 20 may be a high speed random access memory, a non-volatile memory such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card), and the like, or the memory 20 may be other volatile solid state memory devices.
It should be noted that the terminal device may include, but is not limited to, a processor and a memory, and those skilled in the art will understand that the structural block diagram of fig. 4 is only an example of the terminal device, and does not constitute a limitation to the terminal device, and may include more or less components than those shown, or combine some components, or different components.
To sum up, according to the method and the device for detecting falling of the damper of the power transmission line, the computer-readable storage medium and the terminal device provided by the embodiment of the invention, firstly, a damper image sample set is obtained; secondly, preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set; then, training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model; further, the initial detection model is subjected to test analysis through the defect sample test set and the average precision, and a vibration damper falling-off detection model is obtained; and finally, inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain a falling detection result of the damper. Therefore, knowledge of a large number of non-defective samples can be effectively transferred to a small number of defective samples, a vibration damper falling detection model is constructed by the small number of defective samples, and compared with the traditional vibration damper falling detection method, the vibration damper falling detection method is high in adaptation speed and detection precision.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (10)
1. A method for detecting the falling of a damper of a power transmission line is characterized by comprising the following steps:
acquiring a damper image sample set;
preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set;
training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model;
testing and analyzing the initial detection model through the defect sample test set and the average precision to obtain a vibration damper falling-off detection model;
and inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain a falling detection result of the damper.
2. The method for detecting the falling of the damper of the power transmission line according to claim 1, wherein the method comprises the steps of preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set, and specifically comprises the following steps:
dividing the damper image sample set into a normal sample set, a first defect sample set and a second defect sample set;
respectively labeling the normal sample set, the first defect sample set and the second defect sample set by an image calibration tool to obtain a labeled normal sample set, a labeled first defect sample set and a labeled second defect sample set;
and respectively carrying out image enhancement on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set.
3. The method for detecting the falling of the damper of the power transmission line according to claim 2, wherein the image enhancement is performed on the labeled normal sample set, the labeled first defect sample set, and the labeled second defect sample set, respectively, specifically:
horizontally overturning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and clockwise rotating according to a preset angle; or the like, or, alternatively,
horizontally overturning the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and introducing noise;
horizontally turning the marked normal sample set, the marked first defect sample set and the marked second defect sample set and introducing weather factors; or the like, or, alternatively,
performing defocusing blurring on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set and performing clockwise rotation according to a preset angle; or the like, or, alternatively,
performing defocus blur and introducing noise on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set; or the like, or, alternatively,
and performing out-of-focus blur and introducing weather factors on the labeled normal sample set, the labeled first defect sample set and the labeled second defect sample set.
4. The method for detecting the drop of the damper of the power transmission line according to claim 1, wherein an initial detection model is obtained by training a FasterR-CNN network which is constructed in advance based on a migration learning algorithm through the normal sample training set and the defect sample training set, and specifically comprises:
inputting the normal sample training set into a first feature extraction network to obtain a normal sample feature map;
inputting the defect sample training set into a second feature extraction network, and migrating the normal sample feature map into the feature space of the defect sample based on migration learning to obtain a defect sample feature map;
inputting the normal sample feature map and the defect sample feature map into a regional suggestion network to generate a plurality of suggestion windows;
inputting the normal sample feature map, the defect sample feature map and a plurality of the suggestion windows into a region of interest, and outputting a feature map of a fixed size of each suggestion window;
inputting the feature map with the fixed size of each suggestion window into a classifier of the FasterR-CNN network, and outputting a maximum probability estimation and a target bounding box position;
and taking the suggestion window corresponding to the maximum probability estimation as an initial detection model.
5. The method for detecting the drop-off of the damper of the power transmission line according to claim 1, wherein in the process of training the fasterrr-CNN network built in advance based on the transfer learning algorithm through the normal sample training set and the defect sample training set, the method further comprises:
and constructing a loss function of the FasterR-CNN network, and iteratively updating the network weight of the FasterR-CNN network through the loss function and a random gradient descent algorithm.
6. The method for detecting the falling of the damper of the power transmission line according to claim 5, wherein the expression of the loss function L is as follows:
L＝LS+LT；
wherein L isSAs a function of the source domain loss of the FasterR-CNN network, LTIs a target domain loss function of the FasterR-CNN network;
source domain loss function L of the FasterR-CNN networkSThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification loss of the source domain,
as a regression loss function of the source domain, NclsNumber of samples lost for classification, NregNumber of samples lost to regression, l is the equilibrium parameter, piFor the prediction probability that the ith anchor frame belongs to the target in the ith true annotated image,
for the ith real annotation image, tiFor the four-dimensional coordinates of the ith prediction bounding box,
labeling the ith real four-dimensional coordinate, u, of the bounding boxiA set of four-dimensional coordinates of the ith prediction bounding box and the four-dimensional coordinates of the real labeling bounding box;
target domain loss function L of the FasterR-CNN networkTThe expression of (a) is:
wherein the content of the first and second substances,
is a function of the classification error of the target domain,
is the regression error function of the target domain.
7. The method for detecting the drop of the damper of the power transmission line according to claim 1, wherein the initial detection model is subjected to test analysis through the defect sample test set and the average precision to obtain the drop detection model of the damper, and specifically comprises:
inputting the defect sample test set into the initial detection model to obtain a detection result of the defect sample test set;
and when the average precision of the detection results of the defect sample test set is greater than a preset threshold value, taking the initial detection model as a vibration damper falling-off detection model.
8. The utility model provides a detection apparatus that transmission line damper drops which characterized in that includes:
the data receiving module is used for acquiring a damper image sample set;
the data processing module is used for preprocessing the damper image sample set to obtain a normal sample training set, a defect sample training set and a defect sample testing set;
the model training module is used for training a FasterR-CNN network which is built in advance based on a transfer learning algorithm through the normal sample training set and the defect sample training set to obtain an initial detection model;
the sample testing module is used for testing and analyzing the initial detection model through the defect sample testing set and the average precision to obtain a vibration damper falling-off detection model;
and the falling detection module is used for inputting the pre-acquired damper image to be detected to the damper falling detection model to obtain the falling detection result of the damper.
9. A computer-readable storage medium, comprising a stored computer program, wherein when the computer program runs, the computer-readable storage medium controls a device to execute the method for detecting the drop of the damper on the transmission line according to any one of claims 1 to 7.
10. A terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method for detecting drop-off of a damper on a power transmission line according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111349907.5A CN114187505A (en) | 2021-11-15 | 2021-11-15 | Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111349907.5A CN114187505A (en) | 2021-11-15 | 2021-11-15 | Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114187505A true CN114187505A (en) | 2022-03-15 |
Family
ID=80540078
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111349907.5A Pending CN114187505A (en) | 2021-11-15 | 2021-11-15 | Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114187505A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114066848A (en) * | 2021-11-16 | 2022-02-18 | 苏州俪濠智能科技有限公司 | FPCA appearance defect visual inspection system |
| CN114863221A (en) * | 2022-05-31 | 2022-08-05 | 商汤人工智能研究中心(深圳)有限公司 | Training method, device, system, equipment and storage medium for detection model |
| CN116242609A (en) * | 2022-11-23 | 2023-06-09 | 广东石油化工学院 | Variable working condition bearing fault diagnosis method, system, medium, equipment and terminal |
| CN116242609B (en) * | 2022-11-23 | 2024-05-14 | 广东石油化工学院 | Variable working condition bearing fault diagnosis method, system, medium, equipment and terminal |
-
2021
- 2021-11-15 CN CN202111349907.5A patent/CN114187505A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114066848A (en) * | 2021-11-16 | 2022-02-18 | 苏州俪濠智能科技有限公司 | FPCA appearance defect visual inspection system |
| CN114066848B (en) * | 2021-11-16 | 2024-03-22 | 苏州极速光学科技有限公司 | FPCA appearance defect visual detection system |
| CN114863221A (en) * | 2022-05-31 | 2022-08-05 | 商汤人工智能研究中心(深圳)有限公司 | Training method, device, system, equipment and storage medium for detection model |
| CN116242609A (en) * | 2022-11-23 | 2023-06-09 | 广东石油化工学院 | Variable working condition bearing fault diagnosis method, system, medium, equipment and terminal |
| CN116242609B (en) * | 2022-11-23 | 2024-05-14 | 广东石油化工学院 | Variable working condition bearing fault diagnosis method, system, medium, equipment and terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059694B (en) | Intelligent identification method for character data in complex scene of power industry | |
| CN111784685B (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN107742093B (en) | Real-time detection method, server and system for infrared image power equipment components | |
| CN108009591B (en) | Contact network key component identification method based on deep learning | |
| CN110458798B (en) | Vibration damper defect visual detection method, system and medium based on key point detection | |
| CN111179249A (en) | Power equipment detection method and device based on deep convolutional neural network | |
| CN108038846A (en) | Transmission line equipment image defect detection method and system based on multilayer convolutional neural networks | |
| CN114743119B (en) | High-speed rail contact net hanger nut defect detection method based on unmanned aerial vehicle | |
| CN114187505A (en) | Detection method and device for falling-off of damper of power transmission line, medium and terminal equipment | |
| CN111862013B (en) | Insulator detection method, device and equipment based on deep convolutional neural network | |
| CN112668696A (en) | Unmanned aerial vehicle power grid inspection method and system based on embedded deep learning | |
| CN111914720B (en) | Method and device for identifying insulator burst of power transmission line | |
| CN109815800A (en) | Object detection method and system based on regression algorithm | |
| CN112070135A (en) | Power equipment image detection method and device, power equipment and storage medium | |
| CN114943689A (en) | Method for detecting components of steel cold-rolling annealing furnace based on semi-supervised learning | |
| CN113627299A (en) | Intelligent wire floater identification method and device based on deep learning | |
| CN114694130A (en) | Method and device for detecting telegraph poles and pole numbers along railway based on deep learning | |
| CN113963314A (en) | Rainfall monitoring method and device, computer equipment and storage medium | |
| CN112561885B (en) | YOLOv 4-tiny-based gate valve opening detection method | |
| CN113487610A (en) | Herpes image recognition method and device, computer equipment and storage medium | |
| CN115294392B (en) | Visible light remote sensing image cloud removal method and system based on network model generation | |
| CN116824135A (en) | Atmospheric natural environment test industrial product identification and segmentation method based on machine vision | |
| CN116071299A (en) | Insulator RTV spraying defect detection method and system | |
| CN115908786A (en) | Electrical cabinet grounding cable abnormity identification method and system based on deep learning | |
| CN114037840A (en) | Power transmission line visual object extraction method and system based on multi-feature fusion |
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 |