CN112634216A - Insulator self-explosion detection method based on deep learning model - Google Patents
Insulator self-explosion detection method based on deep learning model Download PDFInfo
- Publication number
- CN112634216A CN112634216A CN202011485662.4A CN202011485662A CN112634216A CN 112634216 A CN112634216 A CN 112634216A CN 202011485662 A CN202011485662 A CN 202011485662A CN 112634216 A CN112634216 A CN 112634216A
- Authority
- CN
- China
- Prior art keywords
- insulator
- model
- training
- image
- cnn
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 239000012212 insulator Substances 0.000 title claims abstract description 106
- 238000004880 explosion Methods 0.000 title claims abstract description 37
- 238000001514 detection method Methods 0.000 title claims abstract description 20
- 238000013136 deep learning model Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 71
- 230000002269 spontaneous effect Effects 0.000 claims abstract description 25
- 238000000034 method Methods 0.000 claims abstract description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 69
- 238000012360 testing method Methods 0.000 claims description 37
- 238000012795 verification Methods 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 24
- 238000011176 pooling Methods 0.000 claims description 19
- 238000002372 labelling Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 229910000831 Steel Inorganic materials 0.000 claims description 2
- 230000009194 climbing Effects 0.000 claims description 2
- 238000005096 rolling process Methods 0.000 claims description 2
- 239000010959 steel Substances 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims 1
- 230000004913 activation Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an insulator self-explosion detection method based on a deep learning model, which comprises the steps of collecting an insulator image, converting the insulator image into a single-channel label graph, constructing a U-Net model and a CNN model, training the U-Net model and the CNN model by using part of the single-channel label graph, improving the pixel precision of the other part of the single-channel label graph by using the trained U-Net model to obtain a mask image of an optimal pixel, inputting the mask image into the trained CNN model, and if the output value of the CNN model is greater than 0.5, determining that the insulator has no self-explosion; otherwise, the insulator is considered to have spontaneous explosion. By adopting the method of the invention to detect the state of the insulator, the manual workload can be effectively reduced, and the identification efficiency and the definition can be improved.
Description
Technical Field
The invention belongs to the technical field of electronic component fault detection, and relates to an insulator self-explosion detection method based on a deep learning model.
Background
Insulator strings are important elements in high-voltage transmission lines and play an important role in electrical insulation and mechanical support. The insulator is exposed to rain, wind or snowfall and other wild animals and plants and meteorological conditions, and the components are easy to crack, pollute and even cause explosion. The self-explosion of the insulator element can cause the serious power failure of the transmission line, and for an electric company, the self-explosion of the insulator element has huge harm and profound influence, and the state of the insulator is detected in time, so that the self-explosion is very necessary to be prevented.
Conventional insulator spontaneous explosion detection requires a professional to examine a video sequence to find potential defects in the power transmission line element, a process that is very time consuming. Spontaneous explosion detection strategies based on traditional computer vision algorithms can provide appropriate results in structured images under controlled lighting and background conditions to improve insulator identification efficiency. However, this method requires manual parameter setting and adjustment, and has a large error.
Disclosure of Invention
The invention aims to provide an insulator self-explosion detection method based on a deep learning model, and solves the problems that parameters need to be set and adjusted manually and the error is large in the existing insulator detection method.
The technical scheme adopted by the invention is that the insulator spontaneous explosion detection method based on the deep learning model is characterized by comprising the steps of collecting an insulator image, converting the insulator image into a single-channel label graph, constructing a U-Net model and a CNN model, training the U-Net model and the CNN model by using a part of single-channel label graphs, improving the pixel precision of the rest part of single-channel label graphs through the trained U-Net model to obtain a mask image of an optimal pixel, inputting the mask image into the trained CNN model, and if the output value of the CNN model is greater than 0.5, determining that the insulator has no spontaneous explosion; otherwise, the insulator is considered to have spontaneous explosion.
The present invention is also technically characterized in that,
the method comprises the following steps:
step 1, collecting a plurality of insulator images of the same insulator, and performing unified adjustment on pixels of the collected insulator images;
step 2, an image marking tool is adopted to mark and format the insulator image to obtain a single-channel marking graph;
step 3, constructing a U-Net model to enable pixels of the images to be consistent before and after the characteristic diagram is subjected to convolution operation; training the U-Net model by adopting the part of the single-channel annotation graphs obtained in the step 2, taking the training parameter with the minimum loss as the final parameter of the U-Net model, and then improving the pixel precision of the rest part of the single-channel annotation graphs by the trained U-Net model to obtain the mask image of the optimal pixel;
step 4, constructing a convolutional neural network CNN model, training the convolutional neural network CNN model by adopting a single-channel label graph of a training U-Net model, and taking a training parameter with minimum loss as a final parameter of the CNN model to obtain the trained CNN model;
step 5, inputting the mask image obtained in the step 3 into the trained CNN model, and if the output value of the CNN model is greater than 0.5, determining that the insulator image is complete, namely the insulator does not have spontaneous explosion; otherwise, the insulator image is considered to have a defect, namely the insulator has spontaneous explosion.
In step 1, an unmanned aerial vehicle, a steel wire rolling robot or a climbing robot is used for collecting an insulator image.
The specific operation process of the step 2 is as follows:
step 2.1, marking the insulator image by using an image marking tool, marking the insulator as disc, and marking the connecting piece as ca;
step 2.2, storing the marked image into a Json format to obtain a three-channel marked image;
and 2.3, carrying out image format conversion on the three-channel labeling graph by using MATLAB to obtain a single-channel labeling graph.
The step 3 specifically comprises the following steps:
step 3.1, constructing a U-Net model, and performing zero filling on the feature map to improve the model before convolution operation of each layer so as to enable pixels of the image to be consistent before and after the convolution operation of the feature map;
step 3.2, dividing the single-channel label graph obtained in the step 2 into three parts, namely a training set, a verification set and a test set;
step 3.3, training the U-Net model obtained in the step 3.1 by adopting an optimizer, inputting a training set into the U-Net model for training, verifying by using a verification set every time the training is finished, and if the training parameter with the minimum loss on the verification set meets an MIoU function, finishing the training of the U-Net model by taking the training parameter with the minimum loss on the verification set as a final parameter of the U-Net model, wherein the MIoU function is as follows:
in the formula, i represents a true value, j represents a predicted value, K +1 is the number of categories (including empty categories), and pijRepresenting the number of pixels, p, that would have been in class i but predicted to be in class jiiRepresenting the true number of pixels, p, of the class ijiRepresents the number of pixels that would have been in category j but predicted to be in category i;
and 3.4, improving the pixel precision of the test set by adopting the trained U-Net model, and obtaining a mask image with the optimal pixel precision.
In step 3.3, when the verification set is used for verification, the cross entropy is used for evaluating the training result, and the loss function adopted for calculating the loss is as follows:
in the formula, i is a verification set sample, namely a single-channel labeled graph in the verification set; c-sequence of classes from 1 to M, M-Total number of classes in the verification set, picSamples i belong to class cPredicted probability of (y)icIndicating a variable, if the sample i belongs to the class c, yicIs 1, otherwise is 0; n-total number of validation set samples.
In step 3.4, the calculation formula of the optimal pixel precision is as follows:
the specific process of step 4 is as follows:
step 4.1, constructing a Convolutional Neural Network (CNN) model;
step 4.1.1, establishing a frame structure of a Convolutional Neural Network (CNN), wherein the whole network structure comprises 3 convolutional layers, 3 complete connection layers and 3 pooling layers;
step 4.1.2, performing feature extraction operation on the convolutional layer and the pooling layer part, and performing classification operation on the complete connection layer part;
4.1.3, after the last layer of convolution layer, vectorizing the characteristic diagram by a Flatten operation, and then following the three complete connection layers to obtain a CNN model;
and 4.2, training the CNN model by adopting an optimizer, inputting the training set into the CNN model for model training, verifying by using a binary cross entropy loss function every time the CNN model is trained, stopping training when the binary cross entropy loss function value does not decrease any more, and storing the current weight value as a trained CNN model parameter to finish the training of the CNN model.
In step 4.1, the number of the neurons of 3 complete connection layers is 128, 64 and 1 respectively; the convolution kernel sizes of the 3 convolutional layers are all 3x3, and the step sizes are all 1x 1; each convolutional layer is followed by a pooling layer, the size of a pooling window of the pooling layer is 2x2, the step length is 2x2, the number of feature maps after pooling is kept unchanged, and the size is reduced to half of the original size.
In step 4.2, the binary cross entropy loss function is:
in the formula, yiLabel of sample i, if sample i is positive, yiIs 1, if sample i is negative, then yiIs 0; p is a radical ofi-probability that sample i is predicted as positive class.
The method has the advantages that the insulator image is collected, the insulator image is converted into the single-channel marking graph, the U-Net model and the CNN model are constructed and trained, the pixel precision of the single-channel marking graph is improved through the trained U-Net model, the mask image of the optimal pixel is obtained, the mask image is input into the trained CNN model, whether the corresponding insulator has spontaneous explosion or not is identified according to the output result, the process does not need to manually set parameters and adjust, the identification precision is improved, the application range is wide, and the practicability is high.
Drawings
FIG. 1 is a schematic flow chart of a deep learning model-based insulator spontaneous explosion detection method of the present invention;
FIG. 2 is a first original insulator image for model testing in an embodiment of the present invention;
FIG. 3 is a second original insulator image for model testing in an embodiment of the present invention;
FIG. 4 is a third original insulator image for model testing in an embodiment of the present invention;
FIG. 5 is an original insulator image four used for model testing in an embodiment of the present invention;
FIG. 6 is a schematic flow diagram of the construction, training and application of the U-net model in an embodiment of the present invention;
FIG. 7 is a mask image of an original insulator image one used in model testing in an embodiment of the present invention;
FIG. 8 is a mask image of an original insulator image two used for model testing in an embodiment of the present invention;
FIG. 9 is a mask image of original insulator image three used for model testing in an embodiment of the present invention;
FIG. 10 is a mask image of an original insulator image four used in model testing in an embodiment of the present invention;
fig. 11 is a schematic flow chart of the CNN model construction, training and application in the embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to an insulator self-explosion detection method based on a deep learning model, which comprises the following steps of:
step 1, an unmanned aerial vehicle is used for collecting 26 insulator images of the same insulator, the original 26 insulator images are divided into two parts, 22 insulator images are used for model training, and 4 insulator images are used for model testing, so that the 22 selected insulator images are further expanded by using a data enhancement technology, 550 expanded data sets are obtained by using operations of image rotation, image left-right interchange, image amplification and reduction and the like in an Augmentor data enhancement tool according to random probability, and the data sets used for model training are expanded to 25 times of the original data sets;
fig. 2 is a first original insulator image for model testing, fig. 3 is a second original insulator image for model testing, fig. 4 is a third original insulator image for model testing, and fig. 5 is a fourth original insulator image for model testing.
The training network uses a limited memory capacity in the computer, and in order to ensure smooth completion of the training, the pixels of the original image need to be adjusted. The pixel size of the original image is different, mostly about 500 × 500 × 3, so the sizes of the above insulator images are uniformly adjusted before training, and the pixels are uniformly adjusted to 256 × 256 × 3;
step 2, labeling and format conversion are carried out on the insulator image by adopting an image labeling tool Lableme, and the specific operation process is as follows:
step 2.1, labeling the insulator image by adopting an image labeling tool Lableme, labeling the insulator as disc, and labeling the connecting piece as cap;
step 2.2, storing the marked image into a Json format to obtain a three-channel marked image;
step 2.3, carrying out image format conversion on the three-channel labeling graph by using MATLAB to obtain a single-channel labeling graph;
step 3, referring to fig. 6, constructing a U-Net model, training the U-Net model by using a part of single-channel labeled graphs, and then improving the pixel precision of the other part of single-channel labeled graphs by using the trained U-Net model to obtain a mask image of an optimal pixel, which specifically comprises the following steps:
step 3.1, constructing a U-Net model, and performing zero filling on the feature map to improve the model before convolution operation of each layer so as to enable pixels of the image to be consistent before and after the convolution operation of the feature map;
step 3.2, dividing the single-channel label graph obtained in the step 2 into three parts, namely a training set, a verification set and a test set, wherein the training set comprises 550 expanded insulator pictures and corresponding label graphs, the verification set comprises 22 initial insulator pictures and corresponding label graphs, and the test set comprises 4 insulator pictures for model testing;
step 3.3, training the U-Net model obtained in the step 3.1 by adopting an Adam optimizer, setting the learning rate to be 0.005, using softmax as a classification function in the last layer of the network, inputting a training set into the U-Net model for training, setting the epoch to be 30 during the first training, performing 50 iterations in each epoch, and performing model parameter training for 1500 times in total; performing second model training after adjusting the parameters, verifying the training parameters by using the verification set every time the training is finished, and if the training parameters with the minimum loss on the verification set meet the MIoU function, taking the training parameters with the minimum loss on the verification set as final parameters of the U-Net model, namely finishing the training of the U-Net model; the MIoU function is as follows:
in the formula, i represents a true value, j represents a predicted value, K +1 is the number of categories (including empty categories), and pijRepresenting the number of pixels, p, that would have been in class i but predicted to be in class jiiRepresenting the true number of pixels, p, of the class ijiRepresents the number of pixels that would have been in category j but predicted to be in category i;
since the problem is a multi-classification problem, the loss function adopted for calculating the loss in the U-Net model is classification cross entropy, and whether the parameters of the U-Net model reach the optimum or not is measured by using the classification cross entropy, and the specific loss function adopted is as follows:
in the formula, i is a verification set sample, namely a single-channel labeled graph in the verification set; c-sequence of classes from 1 to M, M-Total number of classes in the verification set, pic-predicted probability of sample i belonging to class c, yicIndicating a variable, if the sample i belongs to the class c, yicIs 1, otherwise is 0; n-total number of validation set samples.
In this embodiment, the batch size is set to be 32, and the model corresponding to the epoch with the minimum loss value on the verification set is epoch-32, that is, the epoch-32 is the final improved U-Net model after training.
And 3.4, improving the pixel precision of the test set by adopting the trained U-Net model, and obtaining a mask image with the optimal pixel precision.
The calculation formula of the optimal pixel precision PK is as follows:
fig. 7 is a mask image of an original insulator image one for model testing, fig. 8 is a mask image of an original insulator image two for model testing, fig. 9 is a mask image of an original insulator image three for model testing, and fig. 10 is a mask image of an original insulator image four for model testing.
Step 4, with reference to fig. 11, constructing a convolutional neural network CNN model, training the convolutional neural network CNN model by using a single-channel label graph, and obtaining the trained CNN model, wherein the specific process is as follows:
step 4.1, constructing a convolutional neural network CNN model
Step 4.1.1, establishing a frame structure of a convolutional neural network CNN, wherein a basic network is a VGG16 network in ImageNet competition, a final output layer is modified into a two-class softmax layer, and the whole network structure comprises 3 convolutional layers, 3 complete connection layers and 3 pooling layers;
3 complete connection layers, wherein the activation function of the first two complete connection layers is ReLU, the activation function of the last layer is sigmoid, and the number of the neurons of the 3 complete connection layers is 128, 64 and 1 respectively; the convolution kernel size of all 3 convolution layers is 3x3, the step length is 1x1, and the activation functions are all ReLU; each convolutional layer is followed by a pooling layer, the size of a pooling window of the pooling layer is 2x2, the step length is 2x2, the number of feature maps after pooling is kept unchanged, and the size is reduced to half of the original size.
Step 4.1.2, performing feature extraction operation on the convolutional layer and the pooling layer part, and performing classification operation on the complete connection layer part;
4.1.3, in order to avoid overfitting of the model, performing Dropout regularization after the pooling operation, wherein the Dropout ratio is selected to be 0.2, vectorizing the characteristic diagram by the Flatten operation after the last convolution layer, and then following three complete connection layers to complete the construction of the CNN model;
and 4.2, training the CNN model by adopting an Adam optimizer, setting the learning rate to be 0.0001, inputting the training set into the CNN model for model training, verifying by using a binary cross entropy loss function every time the CNN model is trained, stopping training when the binary cross entropy loss function value does not decrease any more, and storing the current weight value as a trained CNN model parameter to finish the training of the CNN model.
Different from the improved U-Net model, the CNN model is a binary problem, so the loss function adopted for calculating the loss is a binary cross entropy, and the binary cross entropy loss function is as follows:
in the formula, yiLabel of sample i, if sample i is positive, yiIs 1, if sample i is negative, then yiIs 0; p is a radical ofi-probability that sample i is predicted as positive class.
In the model training, epoch is set to 20, batch _ size is set to 10, and the total number of model parameter training is 560.
Step 5, inputting the 4 mask images obtained in the step 3 into a trained CNN model, wherein the CNN model can output any numerical value between 0 and 1, and if the numerical value output by the CNN model is greater than 0.5, the insulator image is considered to be complete, namely, the insulator is not subjected to spontaneous explosion; if the numerical value output by the CNN model is less than or equal to 0.5, the insulator image is considered to be missing, namely the insulator is subjected to spontaneous explosion, and the result is shown in the following table:
fig. 2 is a first original insulator image for model testing, fig. 3 is a second original insulator image for model testing, fig. 4 is a third original insulator image for model testing, and fig. 5 is a fourth original insulator image for model testing. Fig. 7 is a mask image of an original insulator image one for model testing, fig. 8 is a mask image of an original insulator image two for model testing, fig. 9 is a mask image of an original insulator image three for model testing, and fig. 10 is a mask image of an original insulator image four for model testing.
TABLE 1 CNN model test results
As can be seen from table 1, the trained CNN model better learns the characteristics of insulator integrity and loss, the test result on the test set is nearly perfect, and the test accuracy reaches 100%.
Claims (10)
1. An insulator spontaneous explosion detection method based on a deep learning model is characterized by comprising the steps of collecting an insulator image, converting the insulator image into a single-channel label graph, constructing a U-Net model and a CNN model, training the U-Net model and the CNN model by using part of the single-channel label graph, improving the pixel precision of the other part of the single-channel label graph through the trained U-Net model to obtain a mask image of an optimal pixel, inputting the mask image into the trained CNN model, and if the output value of the CNN model is greater than 0.5, determining that the insulator has no spontaneous explosion; otherwise, the insulator is considered to have spontaneous explosion.
2. The insulator spontaneous explosion detection method based on the deep learning model is characterized by comprising the following steps of:
step 1, collecting a plurality of insulator images of the same insulator, and performing unified adjustment on pixels of the collected insulator images;
step 2, an image marking tool is adopted to mark and format the insulator image to obtain a single-channel marking graph;
step 3, constructing a U-Net model to enable pixels of the images to be consistent before and after the characteristic diagram is subjected to convolution operation; training the U-Net model by adopting the part of the single-channel annotation graphs obtained in the step 2, taking the training parameter with the minimum loss as the final parameter of the U-Net model, and then improving the pixel precision of the rest part of the single-channel annotation graphs by the trained U-Net model to obtain the mask image of the optimal pixel;
step 4, constructing a convolutional neural network CNN model, training the convolutional neural network CNN model by adopting a single-channel label graph of a training U-Net model, and taking a training parameter with minimum loss as a final parameter of the CNN model to obtain the trained CNN model;
step 5, inputting the mask image obtained in the step 3 into the trained CNN model, and if the output value of the CNN model is greater than 0.5, determining that the insulator image is complete, namely the insulator does not have spontaneous explosion; otherwise, the insulator image is considered to have a defect, namely the insulator has spontaneous explosion.
3. The insulator spontaneous explosion detection method based on the deep learning model as claimed in claim 2, wherein in the step 1, an unmanned aerial vehicle, a steel wire rolling robot or a climbing robot is used for collecting insulator images.
4. The deep learning model-based insulator spontaneous explosion detection method according to claim 2, wherein the specific operation process of the step 2 is as follows:
step 2.1, marking the insulator image by using an image marking tool, marking the insulator as disc, and marking the connecting piece as ca;
step 2.2, storing the marked image into a Json format to obtain a three-channel marked image;
and 2.3, carrying out image format conversion on the three-channel labeling graph by using MATLAB to obtain a single-channel labeling graph.
5. The insulator spontaneous explosion detection method based on the deep learning model according to claim 2 or 4, wherein the step 3 specifically comprises the following steps:
step 3.1, constructing a U-Net model, and performing zero filling on the feature map to improve the model before convolution operation of each layer so as to enable pixels of the image to be consistent before and after the convolution operation of the feature map;
step 3.2, dividing the single-channel label graph obtained in the step 2 into three parts, namely a training set, a verification set and a test set;
step 3.3, training the U-Net model obtained in the step 3.1 by adopting an optimizer, inputting a training set into the U-Net model for training, verifying by using a verification set every time the training is finished, and if the training parameter with the minimum loss on the verification set meets an MIoU function, finishing the training of the U-Net model by taking the training parameter with the minimum loss on the verification set as a final parameter of the U-Net model, wherein the MIoU function is as follows:
in the formula, i represents a true value, j represents a predicted value, K +1 is the number of categories including a null category, and pijRepresenting the number of pixels, p, that would have been in class i but predicted to be in class jiiRepresenting the true number of pixels, p, of the class ijiRepresents the number of pixels that would have been in category j but predicted to be in category i;
and 3.4, improving the pixel precision of the test set by adopting the trained U-Net model, and obtaining a mask image with the optimal pixel precision.
6. The method for detecting the insulator spontaneous explosion based on the deep learning model according to claim 5, wherein in the step 3.3, when the verification is performed by using the verification set, the cross entropy evaluation training result is used, and the loss function adopted for calculating the loss is as follows:
in the formula, i is a verification set sample, namely a single-channel labeled graph in the verification set; c-the sequence of classes from 1 to M, M-the total number of classes in the validation set, pic-the predicted probability that sample i belongs to class c, yic-an indicator variable, if sample i belongs to class c, yic is 1, otherwise 0; n-total number of validation set samples.
7. The method for detecting the self-explosion of the insulator based on the deep learning model according to claim 6, wherein in the step 3.4, a calculation formula of the optimal pixel precision is as follows:
8. the deep learning model-based insulator spontaneous explosion detection method according to claim 7, wherein the specific process of the step 4 is as follows:
step 4.1, constructing a Convolutional Neural Network (CNN) model;
step 4.1.1, establishing a frame structure of a Convolutional Neural Network (CNN), wherein the whole network structure comprises 3 convolutional layers, 3 complete connection layers and 3 pooling layers;
step 4.1.2, performing feature extraction operation on the convolutional layer and the pooling layer part, and performing classification operation on the complete connection layer part;
4.1.3, after the last layer of convolution layer, vectorizing the characteristic diagram by a Flatten operation, and then following the three complete connection layers to obtain a CNN model;
and 4.2, training the CNN model by adopting an optimizer, inputting the training set into the CNN model for model training, verifying by using a binary cross entropy loss function every time the CNN model is trained, stopping training when the binary cross entropy loss function value does not decrease any more, and storing the current weight value as a trained CNN model parameter to finish the training of the CNN model.
9. The insulator spontaneous explosion detection method based on the deep learning model according to claim 8, wherein in the step 4.1, the number of the neurons of 3 complete connection layers is 128, 64, 1; the convolution kernel sizes of the 3 convolutional layers are all 3x3, and the step sizes are all 1x 1; each convolutional layer is followed by a pooling layer, the size of a pooling window of the pooling layer is 2x2, the step length is 2x2, the number of feature maps after pooling is kept unchanged, and the size is reduced to half of the original size.
10. The insulator spontaneous explosion detection method based on the deep learning model according to claim 8, wherein in the step 4.2, the binary cross entropy loss function is as follows:
in the formula, yiLabel of sample i, if sample i is positive, yiIs 1, if sample i is negative, then yiIs 0; p is a radical ofi-probability that sample i is predicted as positive class.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011485662.4A CN112634216B (en) | 2020-12-16 | 2020-12-16 | Insulator self-explosion detection method based on deep learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011485662.4A CN112634216B (en) | 2020-12-16 | 2020-12-16 | Insulator self-explosion detection method based on deep learning model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112634216A true CN112634216A (en) | 2021-04-09 |
| CN112634216B CN112634216B (en) | 2024-02-09 |
Family
ID=75313680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011485662.4A Active CN112634216B (en) | 2020-12-16 | 2020-12-16 | Insulator self-explosion detection method based on deep learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112634216B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780438A (en) * | 2016-11-11 | 2017-05-31 | 广东电网有限责任公司清远供电局 | Defects of insulator detection method and system based on image procossing |
| CN109934222A (en) * | 2019-03-01 | 2019-06-25 | 长沙理工大学 | A kind of insulator chain self-destruction recognition methods based on transfer learning |
| WO2020062088A1 (en) * | 2018-09-28 | 2020-04-02 | 安徽继远软件有限公司 | Image identification method and device, storage medium, and processor |
| WO2020172838A1 (en) * | 2019-02-26 | 2020-09-03 | 长沙理工大学 | Image classification method for improvement of auxiliary classifier gan |
-
2020
- 2020-12-16 CN CN202011485662.4A patent/CN112634216B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780438A (en) * | 2016-11-11 | 2017-05-31 | 广东电网有限责任公司清远供电局 | Defects of insulator detection method and system based on image procossing |
| WO2020062088A1 (en) * | 2018-09-28 | 2020-04-02 | 安徽继远软件有限公司 | Image identification method and device, storage medium, and processor |
| WO2020172838A1 (en) * | 2019-02-26 | 2020-09-03 | 长沙理工大学 | Image classification method for improvement of auxiliary classifier gan |
| CN109934222A (en) * | 2019-03-01 | 2019-06-25 | 长沙理工大学 | A kind of insulator chain self-destruction recognition methods based on transfer learning |
Non-Patent Citations (3)
| Title |
|---|
| 朱有产;王雯瑶;: "基于改进Mask R-CNN的绝缘子目标识别方法", 微电子学与计算机, no. 02 * |
| 王义军;曹培培;王雪松;闫星宇;: "基于深度学习的绝缘子自爆检测方法研究", 东北电力大学学报, no. 03 * |
| 陈俊杰;叶东华;产焰萍;陈凌睿;: "基于Faster R-CNN模型的绝缘子故障检测", 电工电气, no. 04 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112634216B (en) | 2024-02-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110827251B (en) | Power transmission line locking pin defect detection method based on aerial image | |
| CN111444939B (en) | Small-scale equipment component detection method based on weak supervision cooperative learning in open scene of power field | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| CN109376606B (en) | Power inspection image tower foundation fault detection method | |
| CN111754498A (en) | Conveyor belt carrier roller detection method based on YOLOv3 | |
| CN111814850A (en) | Defect detection model training method, defect detection method and related device | |
| CN113436169A (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
| CN111008641B (en) | Power transmission line tower external force damage detection method based on convolutional neural network | |
| CN116612098B (en) | Insulator RTV spraying quality evaluation method and device based on image processing | |
| CN115861190A (en) | Comparison learning-based unsupervised defect detection method for photovoltaic module | |
| CN112419268A (en) | Method, device, equipment and medium for detecting image defects of power transmission line | |
| CN116128839A (en) | Wafer defect identification method, device, electronic equipment and storage medium | |
| CN112598666B (en) | Cable tunnel anomaly detection method based on convolutional neural network | |
| CN112132088B (en) | Inspection point missing inspection identification method | |
| CN116485802B (en) | Insulator flashover defect detection method, device, equipment and storage medium | |
| CN111091533B (en) | Battery piece EL defect detection method based on improved SSD algorithm | |
| CN112634216B (en) | Insulator self-explosion detection method based on deep learning model | |
| CN114596273B (en) | Intelligent detection method for multiple defects of ceramic substrate by using YOLOV4 network | |
| CN116385950A (en) | Electric power line hidden danger target detection method under small sample condition | |
| CN113344915B (en) | Method and system for detecting defects of key parts of power transmission line | |
| CN112508946B (en) | Cable tunnel anomaly detection method based on antagonistic neural network | |
| CN114418968A (en) | Power transmission line small target defect detection method based on deep learning | |
| CN114861771A (en) | Industrial CT image defect classification method based on feature extraction and deep learning | |
| CN113284103A (en) | Substation equipment defect online detection system based on space transformation fast R-CNN model | |
| CN117781661B (en) | Silicon wafer drying improvement method and device based on D-LKA network model |
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 |