CN106595551B - Ice covering thickness detection method in powerline ice-covering image based on deep learning - Google Patents
Ice covering thickness detection method in powerline ice-covering image based on deep learning Download PDFInfo
- Publication number
- CN106595551B CN106595551B CN201611145878.XA CN201611145878A CN106595551B CN 106595551 B CN106595551 B CN 106595551B CN 201611145878 A CN201611145878 A CN 201611145878A CN 106595551 B CN106595551 B CN 106595551B
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- icing
- output
- ice
- 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.)
- Expired - Fee Related
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 238000001514 detection method Methods 0.000 title abstract description 4
- 238000000034 method Methods 0.000 claims abstract description 45
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000012544 monitoring process Methods 0.000 claims abstract description 24
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 23
- 230000005540 biological transmission Effects 0.000 claims abstract description 20
- 230000035945 sensitivity Effects 0.000 claims abstract description 10
- 238000012360 testing method Methods 0.000 claims abstract description 3
- 239000011248 coating agent Substances 0.000 claims description 26
- 238000000576 coating method Methods 0.000 claims description 26
- 238000000605 extraction Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 15
- 230000004913 activation Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 11
- 238000003709 image segmentation Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 238000012937 correction Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000013136 deep learning model Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims 1
- 238000012423 maintenance Methods 0.000 abstract description 2
- 230000000877 morphologic effect Effects 0.000 abstract description 2
- 230000000694 effects Effects 0.000 description 10
- 230000008569 process Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 8
- 238000003708 edge detection Methods 0.000 description 8
- 238000011176 pooling Methods 0.000 description 6
- 230000011218 segmentation Effects 0.000 description 5
- 238000007689 inspection Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000002902 bimodal effect Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000005672 electromagnetic field Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B21/00—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant
- G01B21/02—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant for measuring length, width, or thickness
- G01B21/08—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant for measuring length, width, or thickness for measuring thickness
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
Ice covering thickness detection method in the invention discloses a kind of powerline ice-covering image based on deep learning, belong to digital image understanding field, it aims to overcome that and existing answers pulling force monitoring medium sensitivity and the not high problem of reliability, improve the accuracy and the degree of automation of electric power line ice-covering thickness monitoring, ice covering thickness monitoring and transfinite alarm of the present invention for power system transmission line, comprising: (1) collect icing image；(2) pretreatment image and data set is established；(3) convolutional neural networks are established；(4) training and test model；(5) ice covering thickness information is extracted and five steps such as be transmitted back to control centre；Digital picture characteristic recognition method is introduced into the ice covering thickness detection of power transmission line and shaft tower by the present invention, thickness information is automatically extracted using the morphological feature of icing in image, deicing plan is formulated for operation maintenance personnel, to guarantee that safe and stable operation of power system provides a kind of new intuitive and intelligentized means.
Description
Technical Field
The invention belongs to the technical field of digital image recognition, and particularly relates to a method for detecting the icing thickness state in an image based on a deep learning algorithm, which can be used for icing monitoring and overrun warning of power system transmission network equipment.
Background
The importance of safe and stable operation of the power grid to national economic development is self-evident, and along with continuous deepening of power grid interconnection and gradual implementation of the power market, the operation environment of the power grid is more complex, and higher requirements are provided for the stability and reliability of the power grid. China has broad members, various climates and complex terrains, power networks which are distributed all over the country are often damaged by various natural disasters, and most of China often has large-area power failure accidents due to extreme low temperature and icing. The icing disaster can cause mechanical faults and electrical faults of the power grid, such as transformer substation shutdown, pole tower collapse, ice flash tripping, line galloping, transformer substation equipment damage and other accidents.
Icing causes great equipment loss and large-area power failure accidents to power grids in China, and icing areas are often bad in weather conditions, interrupted in traffic and communication and high in emergency repair difficulty, large-area power failure accidents are caused, and power supply reliability is seriously affected. The ice coating of the power transmission line and the power equipment is objective and cannot be eliminated fundamentally. In order to reduce the disasters caused by ice coating, the ice coating of the electrical equipment in the power grid must be protected, and the potential safety hazard of ice coating is eliminated in time.
At present, icing protection is mainly achieved through two methods of monitoring and inhibiting, the monitoring means is that a sensor and a camera are installed on electrical equipment to achieve online monitoring of icing states, or a key line is subjected to inspection in a manual inspection mode to find fault hidden dangers; in China, areas frequently affected by icing disasters are all provided with icing monitoring systems for monitoring icing states of key lines and nodes of a power grid, but frequent icing accidents in recent years prove that the current monitoring means cannot meet the safe and stable operation requirements of a power system, for example, in 2014, 597 trips caused by icing are carried out, the tripping rate is 0.103 times/hundred kilometers per year, the coincidence success rate is 46.4%, and compared with 2013 (221 times), the coincidence success rate is increased by 376 times and the amplification is 170.1%. Ice coating in 2014 caused 320 failures, the failure non-failure rate was 0.055 times/hundred kilometer-year, and the failure non-failure rate in 2014 caused 5 times of that in 2013 (63 times). The reason that the current icing monitoring is difficult to meet the power grid operation requirement is analyzed, and the method can be summarized as follows:
(1) the sensor monitors that the ice coating is greatly influenced by the working environment, and the measurement precision of the sensor is reduced under severe weather and meteorological conditions or electromagnetic field interference;
(2) the camera monitoring judges the current icing state by shooting the icing image of the tower or the line at the monitoring point, but reliable icing information cannot be obtained due to the lack of an effective processing and utilizing method for the icing image;
(3) the manual inspection mode or helicopter inspection has high cost and low efficiency, and the whole network is difficult to monitor;
(4) the system can not find the icing hidden trouble in time, so that the icing early warning can not be sent out in time, and the icing can not be eliminated in time due to the relative lag of the deicing operation.
Disclosure of Invention
In order to solve the technical problem, the invention provides a method for identifying the icing image of the power transmission line based on a deep learning algorithm, which can automatically identify the icing thickness according to the input icing image and ensure the timely icing condition of the power department.
The technical scheme adopted by the invention is as follows: a method for detecting the icing thickness in an icing image of a power transmission line based on deep learning is characterized by comprising the following steps:
step 1: acquiring icing image data and corresponding tension monitoring data;
step 2, preprocessing the original ice-coated image, processing the size of the original ice-coated image into an image with the same size, and using the ice-coated thickness measured by a tension sensor as an image label;
and step 3: establishing a deep learning convolutional neural network model, establishing corresponding model parameters according to the number and the size of images, and setting the number of units and an activation function of each layer of the network;
and 4, step 4: adjusting the weight and training a model, extracting and combining the characteristics of the images, and judging and outputting the icing thickness;
and 5: and analyzing the model training classification result and extracting the icing thickness information of the icing image.
The real-time performance and effectiveness of electric icing monitoring in China are insufficient, and the monitoring effect of the sensor is mainly improved from the aspect of hardware as long as the reason is that the monitoring effect of the sensor is improved; starting from the ice coating image data, the invention researches a method for monitoring the ice coating state more quickly and accurately, eliminates the hidden trouble in time, improves the reliability of the power grid ice coating monitoring system, and ensures the safe and stable operation of the power system.
The digital image feature recognition method is introduced into the icing thickness detection of the transmission line and the tower, the thickness information is automatically extracted by utilizing the morphological features of the icing in the image, a deicing plan is made for operation and maintenance personnel, and a new visual and intelligent means is provided for ensuring the safe and stable operation of the power system.
Drawings
FIG. 1 is a schematic flow diagram of an embodiment of the present invention;
FIG. 2 is an original ice image of an embodiment of the present invention;
FIG. 3 is a result of an iterative method of segmenting an image according to an embodiment of the present invention;
FIG. 4 shows an edge detection result of an automatic threshold LoG operator according to an embodiment of the present invention;
FIG. 5 is the Prewitt operator edge detection result of the automatic threshold of the embodiment of the present invention;
FIG. 6 shows the edge detection result of the Sobel operator for the automatic threshold according to the embodiment of the present invention;
FIG. 7 is a schematic diagram of convolutional neural network feature extraction according to an embodiment of the present invention;
FIG. 8 is a schematic diagram of a convolution process according to an embodiment of the present invention;
FIG. 9 is a schematic illustration of a pooling process of an embodiment of the present invention;
FIG. 10 is a comparison of 1 recognition accuracy of convolutional neural network traversal data for structures of an embodiment of the present invention;
FIG. 11 is a comparison of 5 recognition accuracies of convolutional neural network traversal data for structures of an embodiment of the present invention;
FIG. 12 is a comparison of 10 recognition accuracies of convolutional neural network traversal data for structures of an embodiment of the present invention;
FIG. 13 is a comparison of 15 recognition accuracies of convolutional neural network traversal data for structures of an embodiment of the present invention;
FIG. 14 is a graph of 20 comparison of recognition accuracy for convolutional neural network traversal data for structures in accordance with an embodiment of the present invention;
FIG. 15 is a comparison of different model identification errors for a convolutional neural network, in accordance with an embodiment of the present invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail with reference to the accompanying drawings and examples, it is to be understood that the embodiments described herein are merely illustrative and explanatory of the present invention and are not restrictive thereof.
Referring to fig. 1, the method for detecting the thickness of the ice coating in the ice coating image of the power transmission line based on deep learning provided by the invention comprises the following steps:
step 1: collecting icing image data and corresponding tension monitoring data from an icing monitoring system of a power grid department;
the collected icing images comprise icing images of transmission lines, towers, hardware fittings, electrical equipment such as transformers and the like, and the collected icing images are clear as much as possible; the collected stress force data comprises the model and the position of the tower and the equivalent ice coating thickness measured by the stress force sensor.
Step 2: preprocessing an original image, adjusting the size of the image to be the same, and establishing an icing image data set;
preprocessing an original image, including image segmentation and edge extraction;
the image segmentation is a technology and a process for dividing an image into regions with characteristics and extracting an interested target;
the threshold segmentation is a common method in image segmentation, all pixels with gray levels larger than or equal to a threshold are judged to belong to an object, the gray level value is represented by '255', otherwise, the pixel points are excluded from an object region, and the threshold segmentation comprises a double-peak method and an iterative method;
the bimodal method divides the image into a foreground part and a background part, and the gray distribution curve of the image is approximately considered to be formed by two normal distribution functionsAndthe two peaks are separated from each other, and the valley between the two peaks is the threshold of the image;
the iterative method is an improvement of the bimodal method, firstly an approximate threshold value T is selected, and the image is divided into parts R1And R2Calculating the region R1And R2Mean value of (a)1And mu2Selecting new segmentation threshold value T ═ mu1+μ2) And/2, repeating the steps until mu1And mu2Until no change, the invention uses an iteration method to carry out image segmentation processing on the image;
edge extraction firstly detects edge points in an image, then connects the edge points into a contour so as to form a segmentation region, because the edge is a boundary line of an object to be extracted and a background, the object and the background can be separated only by extracting the edge, a gradient modulus operator has the properties of displacement invariance and isotropy and is suitable for edge detection, and the direction of gray level change, namely the direction of the boundary, can be determined by thetag＝arctan(fy/fx) Is determined in which fxAnd fyIn the x and y directions, respectively, thetagIs the direction of the maximum value of the edge detection gradient of the continuous image; the invention expresses the operator in the form of differential operator, and then realizes the operator by using fast convolution function.
And step 3: establishing a deep learning convolutional neural network model, establishing corresponding appropriate model parameters according to the number and the size of pictures, and setting the number of units and an activation function of each layer of the network;
the deep learning convolutional neural network is a deep learning model reconstructed based on the BP neural network, the complexity of the network model can be reduced by the weight sharing network structure, the number of weights is reduced, the advantage is more obvious when the input of the network is a multi-dimensional image, the image can be directly used as the input of the network, and the complex characteristic extraction and data reconstruction process in the traditional recognition algorithm is avoided;
the input to the convolutional layer is deconvolved with a convolution kernel in the convolutional layer. Firstly, convolving the data with the same position of each output characteristic diagram of the previous layer with the convolution kernel of the layer, and then adding all the convolution results at the same position to obtain the output of the corresponding position of the output characteristic diagram of the layer. In order to reduce the number of parameters and reduce the difficulty of model training, a weight sharing mechanism is adopted for deep learning. The same convolution kernel is used for the same output feature graph, each convolution kernel corresponds to a corresponding filter, and only one feature is extracted by one convolution kernel, so that the feature extraction is ensured not to be mixed;
after the features are obtained through convolution, all the extracted features are used for training a classifier, each feature and image convolution can obtain an n-dimensional convolution feature by utilizing a Softmax classifier, and due to the fact that n features exist, learning of a classifier with more than twenty million feature inputs is inconvenient, and overfitting is easy to occur;
and 4, step 4: adjusting the weight and training a model, extracting and extracting the characteristics of the image, judging and outputting the icing thickness, and comparing the thickness information of the image label;
training and testing the established convolutional neural network to obtain a model meeting the precision requirement for detecting the icing thickness in the picture, wherein the training process comprises the following steps:
(1) sensitivity and error correction;
for the class C problem, there are N training samples, and the mean square error of the model can be expressed as:
whereinA class k expected output representing the nth sample,representing the class k actual output of the nth sample.
For the nth sample, the mean square error of the actual output and the ideal output can be expressed as:
suppose L is the output layer, L is the hidden layer, 1 is the input layer; the activation output of layer l is: x is the number ofl＝f(ul) Wherein u isl＝Wlxl-1+blF (-) is an activation function, WlIs the weight of the l-th layer, blIs the bias of the l-th layer; the sensitivity is defined as:
whereinThe sensitivities of the l-th and output layers are expressed as:
the resulting error correction (η for the learning rate) can be expressed as:
(2) forward propagation
1) Convolutional layer
Assuming that the l-th layer is a convolutional layer, the signature and the signature size output by the layer are respectively expressed as:
output.size＝input.size-ker nel.size+1
wherein,is the ith output of the l-1 th layer,is the jth convolution kernel of the ith layer for the ith input,is the jth bias of the ith layer, f (-) is the activation function,is the jth output of the ith layer.
The feature extraction of the convolutional neural network has two characteristics:
I. through convolution, one pixel of the output feature map is used for representing the pixel feature of a local area of the input feature map, which is the feature extraction of the convolution neural network and simultaneously reduces the data dimension;
and II, weight sharing is performed, one feature is extracted by using the same convolution kernel in the same feature graph, the number of parameters can be reduced, and the time complexity is reduced.
2) Sub-sampling layer
The sub-sampling layer can be regarded as a pooling process and also a dimension reduction process, so that the input feature maps are re-represented on the output feature maps without overlapping, namely the combination of features; in addition, pooling can also reduce data dimension and accelerate calculation speed. The expression of the output map can be expressed as:
where down (x) is a sampling operation on a pixel area of the input image n x n,is a control factor, controls the value of the sampling result in the range of the number of color pixels, and reduces the noise interference.
(3) Counter-propagating
1) Convolutional layer
The sensitivity of the convolutional layer can be expressed as follows, considering that there are sub-sampling layers before and after the convolutional layer, it can be rewritten as:
where, β is the weight value,
instead of sigma deltal+1。
The gradient of the basis and convolution kernels can be expressed as:
where u, v are the corresponding parts of the previous layer image deconvolved with a convolution kernel. Wherein,is the sensitivity of the jth convolution kernel at the ith layer,is the momentum factor of layer l-1;
2) sub-sampling layer
The forward propagation has saved:therefore, the weight gradient is:
wherein,is a function of the downsampling operation of the l-th layer.
(4) Feature combination
An important feature of deep learning network is automatic learning of features, when the representation in training process is the combination of learning feature graph, each extracted feature is given weight, forward propagation and backward propagation are repeated to correct error and adjust weight, thus achieving the purpose of optimizing combination of features, α is used for feature weightijRepresenting the weight or contribution of the ith input feature map representing the jth output feature map, is generally represented by the following equation:
wherein, cijRepresents the weight, sigma of the ith input feature map in the jth output feature mapkexp(ckj) All weighted sums representing the jth output signature.
The jth feature output can be rewritten as:
where f (-) is an activation function, αijThe weight values of the output feature map are represented,is the input to the layer(s),is a convolution kernel that is a function of the convolution kernel,is the layer bias.
The above formula satisfies:0≤αij≤1；
for a single output cell, the corner mark j is ignored, since:
and
after the basis gradient, convolution kernel gradient, connection weight gradient and feature weight gradient of the hidden layer are obtained, the error is more regular as shown in table 1 (η is the learning rate).
TABLE 1 convolutional neural network correction values
Where, η is the learning rate,is the gradient of the basis(s),is the gradient of the convolution kernel and,is the gradient of ligation, αiIs the feature weight.
And 5: and analyzing the model training classification result, extracting the icing thickness information of the icing image and transmitting the icing thickness information back to the icing monitoring center.
In the embodiment, step 1, collecting original icing image data of a power grid and data of a tension sensor, wherein an attached diagram 2 shows an icing image of a power transmission line, the icing image requires that an icing shooting part is clear and is not covered by turbid matters, and the tension data comprises the model and position of a tower and the icing thickness measured by the tension sensor so as to find a fault place in time;
in the step 2 of the embodiment, the pre-processed icing image is extracted through image segmentation and edge extraction, the size of the original image is set to be consistent, and the icing thickness measured by a tension sensor is used as an image label; the method comprises the steps of carrying out image segmentation and edge processing on an image to remove noise and improve the characteristic extraction efficiency, wherein the image segmentation and edge extraction are used for eliminating the interference of environment and irrelevant objects on the ice coating characteristic extraction; FIG. 3 shows the image effect of the iterative segmentation, and FIGS. 4 to 6 show the results of the edge detection of the automatic threshold LoG operator, the edge detection of the Prewitt operator and the edge detection of the Sobel operator, respectively;
step 3 of this embodiment is to establish a convolutional neural network, which utilizes a spatial relationship to reduce the number of parameters to be learned so as to improve the training performance of forward and backward propagation algorithms, and fig. 7 is a brief description of a model structure;
the convolutional neural network comprises an input layer, a hidden layer and an output layer, wherein the internal hidden layer of the convolutional neural network is constructed by a convolutional layer and a pooling layer in an iterative manner, the convolutional layer can extract data features, namely a feature extraction layer which is essentially convolution, the attached figure 8 is a schematic diagram of a convolution process, a sub-sampling layer is also called a feature mapping layer, a feature mapping is obtained by weighting and activating a function through pixels, and the attached figure 9 is a schematic diagram of a pooling process;
in the convolutional neural network, a part of an image (a local sensing area) is used as the input of the lowest layer of a hierarchical structure, information is transmitted to different layers in sequence, each layer obtains the most significant characteristics of observed data through a digital filter, therefore, the significant characteristics of the observed data which are unchanged in translation, scaling and rotation can be obtained,
defining model parameters and initializing each parameter in step 3, and setting an activation function as sigmoid;
step 4 of this embodiment trains the established convolutional neural network, and according to the training effect, relevant parameters such as the model structure, the training steps, the batch processing size, the activation function, the number of neurons in each layer, and the number of neurons can be adjusted;
due to the fact that the number of model parameters is large, 15235 icing pictures are used for training, the icing state monitoring mechanism is judged according to a model according to a power grid, the thickness of the icing is divided into six grades including no icing (0cm), slight icing (0-5 cm), moderate icing (5-10 cm), severe icing (10-15 cm), dangerous icing (15-20 cm) and fault early warning icing (more than 20cm), and output and image are compared and judged previously, and errors are adjusted.
Step 5 of the embodiment, analyzing the model training classification result, extracting the icing thickness information of the icing image and transmitting the icing thickness information back to the icing monitoring center;
the invention builds four models of the convolutional neural network, which are respectively (1)6-4-12-2, (2)12-4-24-2, (3)4-2-8-2-16-2-32-2, and (4) 16-2-8-2-4-2-2-2. Wherein the odd layers are convolution layers and the even layers are pooling layers. When the four models traverse data for different times, the mean square error of each model changes. The mean square error of the model is used for measuring the training effect of the model, when the mean square error of the model is smaller, the training effect is better, the parameter setting is more reasonable, and the obtained output is closer to an ideal result. The step number represents the times of adjusting the weight, in the deep learning algorithm, training samples are usually trained in batches, and the weight is adjusted once after the training of a batch of samples is finished, so that the training speed and the training effect are improved;
FIG. 10 shows 4 model configurations: (1)6-4-12-2, (2)12-4-24-2, (3)4-2-8-2-16-2-32-2, and (4)16-2-8-2-4-2-2-2, respectively traversing the data once, wherein the abscissa represents the number of times of adjusting the parameters and the ordinate represents the mean square error in the change situation of the mean square error in the training process; the curve trend can reflect the training effect of the model, and simultaneously, the model parameters are adjusted according to the training effect so as to achieve the optimal training effect. Similarly, fig. 11-14 show the variation of the mean square error when the four model structures traverse the data 5 times, 10 times, 15 times and 20 times, respectively, and generally, the larger the number of traverses, the better the classification performance of the model.
FIG. 15 shows the results for 4 model configurations: (1)6-4-12-2, (2)12-4-24-2, (3)4-2-8-2-16-2-32-2, and (4)16-2-8-2-4-2-2, wherein the abscissa is the number of times of traversing data, and the ordinate is the classification error.
It should be understood that parts of the specification not set forth in detail are well within the prior art.
It should be understood that the above description of the preferred embodiments is given for clarity and not for any purpose of limitation, and that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (8)
1. A method for detecting the icing thickness in an icing image of a power transmission line based on deep learning is characterized by comprising the following steps:
step 1: acquiring icing image data and corresponding tension monitoring data;
step 2, preprocessing the original ice-coated image, processing the size of the original ice-coated image into an image with the same size, and using the ice-coated thickness measured by a tension sensor as an image label;
and step 3: establishing a deep learning convolutional neural network model, establishing corresponding model parameters according to the number and the size of images, and setting the number of units and an activation function of each layer of the network;
and 4, step 4: adjusting the weight and training a model, extracting and combining the characteristics of the images, and judging and outputting the icing thickness;
the training and testing of the established convolutional neural network comprises the following substeps:
step 4.1: calculating sensitivity and error correction;
in step 4.1, for the C classification problem, there are N training samples,a class k expected output representing the nth sample,and representing the k-th actual output of the nth sample, the error of the output layer of the model is represented by the mean square error of the ideal output and the actual output as follows:
for the nth sample, the mean square error of the actual output and the ideal output can be expressed as:
suppose L is the output layer, L is the hidden layer, 1 is the input layer; the activation output of layer l is: x is the number ofl＝f(ul) Wherein u isl＝Wlxl-1+blF (-) is an activation function, WlIs the weight of the l-th layer, blIs the bias of the l-th layer; the sensitivity is defined as:
whereinThe sensitivities of the l-th and output layers are expressed as:
this results in an error correction of:
wherein η is the learning rate;
step 4.2: forward propagation;
step 4.3: backward propagation;
step 4.4: combining the characteristics; giving a weight to each extracted feature, repeatedly performing forward propagation and backward propagation to correct errors and adjusting the weight to achieve the purpose of feature optimization and combination;
and 5: and analyzing the model training classification result and extracting the icing thickness information of the icing image.
2. The method for detecting the thickness of the ice coating in the deeply learned ice coating image of the power transmission line according to claim 1, wherein the method comprises the following steps: the icing image in the step 1 comprises a power transmission line icing image, a tower icing image, a hardware fitting icing image and an electrical equipment icing image; the tension monitoring data comprises the model and position of the tower and the icing equivalent thickness.
3. The method for detecting the thickness of the ice coating in the deeply learned ice coating image of the power transmission line according to claim 1, wherein the method comprises the following steps: preprocessing the original ice-coated image in the step 2, wherein the preprocessing comprises image segmentation and edge extraction; the image segmentation is to use an iteration method to perform image segmentation processing on an image, and a gradient modulus operator in the edge extraction is expressed in the form of a differential operator and is realized by a fast convolution function.
4. The method for detecting the thickness of the ice coating in the deeply learned ice coating image of the power transmission line according to claim 1, wherein the method comprises the following steps: and 3, establishing the deep learning convolutional neural network, namely a deep learning model reconstructed based on the BP neural network, deconvolving the input of the layer by using a convolution kernel in the convolutional layer, firstly convolving the data with the same position of each output feature map of the previous layer with the convolution kernel of the layer, and then adding all the convolution results at the same position to obtain the output of the corresponding position of the output feature map of the layer.
5. The method for detecting the thickness of the ice coating in the deeply learned ice coating image of the power transmission line according to claim 4, wherein the method comprises the following steps: in order to reduce the number of parameters and reduce the difficulty of model training, a weight sharing mechanism is adopted, the same convolution kernel is used for the same output characteristic graph, each convolution kernel corresponds to a corresponding filter, only one characteristic is extracted by one convolution kernel, and the characteristic extraction is ensured not to be mixed.
6. The method for detecting the thickness of the ice coating in the deeply learned ice coating image of the power transmission line according to claim 1, wherein the method comprises the following steps: in step 4.2, for the convolutional layer, assuming that the l-th layer is a convolutional layer, the sizes of the feature map and the feature map output by the layer are respectively:
output.size＝input.size-kernel.size+1
wherein,is the ith output of the l-1 th layer,is the jth convolution kernel of the ith layer for the ith input,is the jth bias of the ith layer, f (-) is the activation function,is the jth output of the ith layer;
for the sub-sampling layer, the output graph is:
wherein down (x) is a sampling operation on a pixel region of the input image n × n,is a control factor, controls the value of the sampling result in the range of the number of color pixels, and reduces the noise interference.
7. The method for detecting the thickness of the ice coating in the deeply learned ice coating image of the power transmission line according to claim 6, wherein the method comprises the following steps: in step 4.3, the sensitivity of the convolutional layer can be expressed as:
where, β is the weight value,instead of sigma deltal+1；
The gradient of the basis and convolution kernels can be expressed as:
wherein u and v are parts corresponding to the previous layer of image deconvoluted by a convolution kernel;
for the sub-sampling layer, the forward propagation has been preservedTherefore, the weight gradient is:
8. the method for detecting the thickness of the ice coating in the deeply learned icing image of the power transmission line according to claim 7, wherein in the step 4.4, the characteristic weight is αijRepresenting the weight or contribution of the ith input feature map of the jth output feature map;
the jth feature output can be rewritten as:
the above formula satisfies:
for a single output cell, the corner mark j is ignored, since:
and
after the base gradient, the convolution kernel gradient, the connection weight gradient and the characteristic weight gradient of the hidden layer are obtained, the errors are respectively corrected as follows: correction value of base isThe more positive value of the convolution kernel isThe more positive value of the connection weight isThe more positive value of the feature weight is
Where, η is the learning rate,is the gradient of the basis(s),is the gradient of the convolution kernel and,is connected toGradient, αiIs the feature weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611145878.XA CN106595551B (en) | 2016-12-13 | 2016-12-13 | Ice covering thickness detection method in powerline ice-covering image based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611145878.XA CN106595551B (en) | 2016-12-13 | 2016-12-13 | Ice covering thickness detection method in powerline ice-covering image based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106595551A CN106595551A (en) | 2017-04-26 |
| CN106595551B true CN106595551B (en) | 2019-01-04 |
Family
ID=58802070
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611145878.XA Expired - Fee Related CN106595551B (en) | 2016-12-13 | 2016-12-13 | Ice covering thickness detection method in powerline ice-covering image based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106595551B (en) |
Families Citing this family (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107230205A (en) * | 2017-05-27 | 2017-10-03 | 国网上海市电力公司 | A kind of transmission line of electricity bolt detection method based on convolutional neural networks |
| CN109003254B (en) * | 2018-05-28 | 2021-04-27 | 南方电网科学研究院有限责任公司 | Method, device, equipment, system and medium for detecting icing based on logistic regression |
| CN109325956A (en) * | 2018-08-15 | 2019-02-12 | 国网河南省电力公司电力科学研究院 | A kind of transmission pressure icing section feature extracting method based on image procossing |
| CN109631794A (en) * | 2018-12-28 | 2019-04-16 | 天津大学 | Target object curvature measurement method based on convolutional neural networks |
| CN109886321B (en) * | 2019-01-31 | 2021-02-12 | 南京大学 | Image feature extraction method and device for fine-grained classification of icing image |
| CN110070530B (en) * | 2019-04-19 | 2020-04-10 | 山东大学 | Transmission line icing detection method based on deep neural network |
| CN110555400A (en) * | 2019-08-23 | 2019-12-10 | 国网山东省电力公司泗水县供电公司 | icing detection method and system based on unmanned aerial vehicle inspection overhead transmission line |
| CN110702015B (en) * | 2019-09-26 | 2021-09-03 | 中国南方电网有限责任公司超高压输电公司曲靖局 | Method and device for measuring icing thickness of power transmission line |
| CN110782431B (en) * | 2019-10-08 | 2022-03-22 | 浙江大学 | High-voltage wire icing area detection method based on deep learning |
| CN110929611A (en) * | 2019-11-12 | 2020-03-27 | 中国地质大学（武汉） | Modeling method of power transmission line icing thickness prediction model based on PR-KELM |
| CN111160424B (en) * | 2019-12-16 | 2022-12-13 | 南方电网科学研究院有限责任公司 | NFC equipment fingerprint authentication method and system based on CNN image identification |
| CN112232591A (en) * | 2020-11-02 | 2021-01-15 | 国网湖南省电力有限公司 | Icing thickness intelligent early warning method based on meteorological factors |
| CN112595276B (en) * | 2020-11-27 | 2022-07-05 | 哈尔滨工程大学 | Power transmission line icing thickness detection method based on deep learning |
| CN112729136A (en) * | 2020-12-29 | 2021-04-30 | 国网江苏省电力有限公司无锡供电分公司 | Distributed optical fiber sensing icing monitoring system and method thereof |
| CN113222892B (en) * | 2021-04-02 | 2024-05-14 | 贵州电网有限责任公司 | Power transmission line icing monitoring and early warning method, system and readable storage medium |
| CN113486936A (en) * | 2021-06-28 | 2021-10-08 | 国网宁夏电力有限公司电力科学研究院 | Icing detection method, device and system for power transmission line equipment and storage medium |
| CN113255239B (en) * | 2021-07-12 | 2021-09-28 | 南京信息工程大学 | Deep learning method for predicting icing thickness of ultrahigh voltage transmission line tower |
| CN114354674A (en) * | 2021-12-06 | 2022-04-15 | 西安交通大学 | Experimental device for water drop impact supercooling surface and application |
| CN115423810B (en) * | 2022-11-04 | 2023-03-14 | 国网江西省电力有限公司电力科学研究院 | Blade icing form analysis method for wind generating set |
| CN116000945B (en) * | 2022-12-05 | 2024-04-12 | 湖北工业大学 | Intelligent control method of cable deicing robot |
| CN115863204B (en) * | 2023-02-27 | 2023-06-02 | 青岛芯康半导体科技有限公司 | On-line thickness monitoring and measuring method and system for wafer processing |
| CN116306893B (en) * | 2023-05-24 | 2023-07-28 | 华东交通大学 | Contact net icing early warning method |
| CN118089854B (en) * | 2024-04-19 | 2024-07-12 | 国网山西省电力公司晋城供电公司 | Overhead transmission line icing observation system and method |
| CN118262183B (en) * | 2024-05-30 | 2024-07-30 | 国网江西省电力有限公司南昌供电分公司 | Overhead conductor icing thickness measuring method and system based on X-ray image |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101793501A (en) * | 2010-04-14 | 2010-08-04 | 华中科技大学 | Transmission line ice coating status detection method based on image |
| CN103400141A (en) * | 2013-07-24 | 2013-11-20 | 华南理工大学 | Method for calculating thickness of ice coated on transmission line on basis of improved image method |
| CN103453867A (en) * | 2013-09-09 | 2013-12-18 | 国家电网公司 | Electric transmission line ice coating thickness monitoring method |
| CN104978580A (en) * | 2015-06-15 | 2015-10-14 | 国网山东省电力公司电力科学研究院 | Insulator identification method for unmanned aerial vehicle polling electric transmission line |
| CN105095670A (en) * | 2015-08-21 | 2015-11-25 | 国家电网公司 | Drawing method of electric transmission line icing-thickness growth rate distribution diagram based on number of continuous freezing rain days |
| CN105138976A (en) * | 2015-08-16 | 2015-12-09 | 东北石油大学 | Power transmission line icing thickness identification method based on genetic wavelet neural network |
-
2016
- 2016-12-13 CN CN201611145878.XA patent/CN106595551B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101793501A (en) * | 2010-04-14 | 2010-08-04 | 华中科技大学 | Transmission line ice coating status detection method based on image |
| CN103400141A (en) * | 2013-07-24 | 2013-11-20 | 华南理工大学 | Method for calculating thickness of ice coated on transmission line on basis of improved image method |
| CN103453867A (en) * | 2013-09-09 | 2013-12-18 | 国家电网公司 | Electric transmission line ice coating thickness monitoring method |
| CN104978580A (en) * | 2015-06-15 | 2015-10-14 | 国网山东省电力公司电力科学研究院 | Insulator identification method for unmanned aerial vehicle polling electric transmission line |
| CN105138976A (en) * | 2015-08-16 | 2015-12-09 | 东北石油大学 | Power transmission line icing thickness identification method based on genetic wavelet neural network |
| CN105095670A (en) * | 2015-08-21 | 2015-11-25 | 国家电网公司 | Drawing method of electric transmission line icing-thickness growth rate distribution diagram based on number of continuous freezing rain days |
Non-Patent Citations (1)
| Title |
|---|
| 基于数字图像处理的输电线路状态智能识别技术;金华等;《微计算机信息》;20120430;第28卷(第4期);第91-92页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106595551A (en) | 2017-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106595551B (en) | Ice covering thickness detection method in powerline ice-covering image based on deep learning | |
| Sohn et al. | Automatic powerline scene classification and reconstruction using airborne lidar data | |
| CN113486865B (en) | Power transmission line suspended foreign object target detection method based on deep learning | |
| CN111723654B (en) | High-altitude parabolic detection method and device based on background modeling, YOLOv3 and self-optimization | |
| CN110827251A (en) | Power transmission line locking pin defect detection method based on aerial image | |
| CN108537154A (en) | Transmission line of electricity Bird's Nest recognition methods based on HOG features and machine learning | |
| CN106198551A (en) | Method and device for detecting defects of power transmission line | |
| CN116385958A (en) | Edge intelligent detection method for power grid inspection and monitoring | |
| CN116258980A (en) | Unmanned aerial vehicle distributed photovoltaic power station inspection method based on vision | |
| CN112380985A (en) | Real-time detection method for intrusion foreign matters in transformer substation | |
| CN115311648A (en) | Visual inseparable bolt defect detection method based on bolt attributes and positions | |
| CN113012107B (en) | Power grid defect detection method and system | |
| Manninen et al. | Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images | |
| CN118038021A (en) | Transformer substation operation site foreign matter intrusion detection method based on improvement yolov4 | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN117058541B (en) | Insulator hyperspectral data acquisition system and method thereof | |
| CN117612044A (en) | Method for inspecting transmission line insulator by unmanned aerial vehicle in complex scene | |
| Yang et al. | Abnormal Object Detection with an Improved YOLOv8 in the Transmission Lines | |
| CN116994161A (en) | Insulator defect detection method based on improved YOLOv5 | |
| CN111539356A (en) | Pavement gap accurate and efficient detection method based on linear array image | |
| CN116385950A (en) | Electric power line hidden danger target detection method under small sample condition | |
| CN113989209B (en) | Power line foreign matter detection method based on Faster R-CNN | |
| Zhouhua et al. | Multi-target defect intelligent recognition of transmission line based on improved Faster-RCNN | |
| Zhao et al. | Transmission Line Fault Identification Method based on Attention Mechanism and Deep Neural Network | |
| Lv et al. | Research on Insulator Defect Detection and Recognition in Overhead Transmission Line Based on Convolutional Neural Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190104 Termination date: 20191213 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |