CN111899216A - Abnormity detection method for insulator fastener of high-speed rail contact network - Google Patents
Abnormity detection method for insulator fastener of high-speed rail contact network Download PDFInfo
- Publication number
- CN111899216A CN111899216A CN202010549723.2A CN202010549723A CN111899216A CN 111899216 A CN111899216 A CN 111899216A CN 202010549723 A CN202010549723 A CN 202010549723A CN 111899216 A CN111899216 A CN 111899216A
- Authority
- CN
- China
- Prior art keywords
- image
- images
- network
- insulator fastener
- insulator
- 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
- 239000012212 insulator Substances 0.000 title claims abstract description 87
- 238000001514 detection method Methods 0.000 title claims abstract description 42
- 238000005259 measurement Methods 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 27
- 230000002159 abnormal effect Effects 0.000 claims description 24
- 238000012360 testing method Methods 0.000 claims description 21
- 238000000034 method Methods 0.000 claims description 20
- 230000005856 abnormality Effects 0.000 claims description 16
- 239000013598 vector Substances 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 206010063385 Intellectualisation Diseases 0.000 abstract description 2
- 238000012850 discrimination method Methods 0.000 abstract 1
- 238000013527 convolutional neural network Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000003137 locomotive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012216 screening Methods 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an anomaly detection method for insulator fasteners of a high-speed rail contact network, which comprises the steps of collecting contact network images through a 4C system, marking components to form a contact network image data set, training a Faster R-CNN deep network model based on the data set, and identifying and positioning the components in the contact network images by using the model; marking the identified contact net components as 6 types according to the position and the abnormity to form a contact net component detection data set, and then training an image distance measurement depth network model based on the data set; and finally, extracting 128-dimensional characteristics of the contact net component to be detected by using the image distance measurement depth network model, and constructing an abnormity detection and discrimination method for the contact net component for detection. The invention has high detection efficiency and high precision, and provides an effective solution for the intellectualization of the detection of the contact network.
Description
Technical Field
The invention belongs to the technical field of automatic detection of electrified railways, and particularly relates to an abnormity detection method for a high-speed rail contact net insulator fastener.
Background
The high-speed rail contact network system is a key component of an electrified railway, and the working condition of the contact network directly influences the safe operation of a locomotive. The contact net works in the open environment throughout the year, is blown by wind and exposed to the sun, and is easily broken down under the continuous action of the pantograph. Therefore, the railway department places great importance on the anomaly detection of each key component of the contact network.
At present, the anomaly detection of the contact net parts by various railway companies in China is mainly carried out in a mode of on-site inspection and manual contact net image browsing. Because the field inspection work efficiency is low, at present, a 4C detection device is mainly adopted to regularly acquire high-resolution images of the contact network, and contact network components are inspected in a manual browsing mode to find the defects of the contact network components. Important parts concerned by detection personnel comprise abnormal conditions such as looseness, falling and deformation of fasteners of supporting devices such as a contact net insulator fixing part, a double-sleeve connecting part, a sleeve seat, a positioning ring connecting part and a positioner support. And the workload of manually browsing images is large, eyes are easy to be tired, the detection period is long, and the detection period is influenced by factors such as personal emotion and responsibility. How to utilize machine learning and artificial intelligence technique, realize carrying out intelligent detection to the contact net image that high-speed railway 4C system obtained is a technical problem that the railway department is very much regarded.
The early intelligent method for detecting the abnormity of the contact network is mainly based on the traditional computer vision technology, machine learning is carried out through artificial design characteristics to establish a fault detection model, key parts of the contact network are identified and positioned, and then whether the abnormity occurs is judged. However, in actual operation, since the image acquired by the high-speed rail 4C system is affected by factors such as shooting position and weather, each part is diversified in the image, and missing detection and erroneous judgment are easy to occur in the method. In recent years, with the rapid development of deep learning technology, it is a focus of attention to apply the deep learning technology to detect the abnormality of key components of the overhead line system.
Disclosure of Invention
The technical problem to be solved by the invention is to provide an abnormality detection method for insulator fasteners of a high-speed rail contact network, aiming at the defects in the prior art, which is characterized in that the insulator fasteners of the contact network are identified and positioned by using fast R-CNN, and then whether the corresponding insulator fasteners are abnormal or not is judged by measuring the relative distance between the images of the insulator fasteners.
The invention adopts the following technical scheme:
an abnormity detection method for a high-speed rail contact net insulator fastener comprises the following steps:
s1, acquiring contact network image data to form an original image sample set;
s2, uniformly adjusting the images in the overhead line system data set constructed in the step S1 to 1280-1280 pixels, and constructing an overhead line system data set;
s3, training a Faster R-CNN network model for identifying and positioning an insulator fastener of the contact network by using the contact network data set constructed in the step S2, testing the application test set of the Faster R-CNN network model, and storing model parameters;
s4, constructing an insulator fastener data set;
s5, constructing an image distance measurement depth network, and training an image distance measurement network model by using the insulator fastener data set constructed in the step S4;
s6, recognizing and positioning the position of the insulator fastener by using the insulator fastener recognition Faster R-CNN network model trained in the step S3 to obtain an image of the insulator fastener to be detected; and then, carrying out abnormity detection on the image of the insulator fastener to be detected by using the image distance measurement depth network model obtained in the step S5, and judging whether the insulator fastener is abnormal or not.
Specifically, in step S2, the catenary data set is divided into a training set and a test set according to a ratio of 7: 3.
Specifically, in step S3, the Faster R-CNN network model uses ResNet50 as the basic feature extraction network, the learning rate is set to 0.001, the random gradient descent method is used, the momentum is set to 0.9, and the weight attenuation is 0.0001.
Specifically, step S4 specifically includes:
s401, dividing the insulator fastener images marked in the step S2 into 3 types according to the directions, and dividing each type into two types according to the normality and the abnormality, namely dividing the insulator fastener images into 6 types in total and respectively storing the insulator fastener images in 6 folders;
s402, uniformly adjusting the size of the insulator fastener images to 128 × 128 pixel values;
and S403, performing data enhancement through Gaussian noise enhancement, sharpness change and contrast change to obtain 3008 insulator fastener images.
Specifically, step S5 specifically includes:
s501, constructing an image distance measurement depth network comprising an increment-ResNet-V1 network and an L2 normalization and triple loss function of an output vector of the network;
s502, dividing the insulator fastener data set constructed in the step S4 into a training set and a testing set according to the ratio of 7: 3;
s503, training an image distance measurement depth network by using a training set, wherein a marginal value alpha is 0.2, a feature vector dimension n is 128, the learning rate is 0.01, and network parameters are randomly initialized by using an Adagad optimization algorithm;
s504, after iterative training for 50 times, saving image distance measurement depth network model parameters;
and S505, testing the image distance measurement depth network model trained in the step S504 by using a test set.
Further, in step S501, L2 is normalized to a pair vector X ═ X1,x2,...,xnIs divided by the L2 norm to obtain a new vector X2The method specifically comprises the following steps:
further, in step S501, the objective function corresponding to the triple loss is:
wherein the content of the first and second substances,
respectively representing the character representation of Anchor, Positive and Negative; α is a constant greater than zero; + represents [, ]]When the value of [ alpha ], [ alpha]The value of the internal expression is the value of the loss function, [ alpha ], [ alpha]The value of the loss function is zero when the value of the inner expression is less than zero.
Specifically, step S6 specifically includes:
s601, identifying and positioning the insulator fastener in the original image of the contact network to be detected by using the Faster R-CNN target detection network model trained in the step S3 to serve as an image of the insulator fastener to be detected;
s602, randomly selecting 3 images from the 3 normal image sets in the data set constructed in the step S4, forming 10 images together with the insulator fastener image to be detected, and judging whether the insulator fastener image is abnormal or not according to the relative distance D.
Further, in step S602, performing feature normalization on the 10 images through L2 to obtain 128-dimensional feature vectors of each image, and calculating a distance between each two images by using an euclidean distance; taking 3 images with the minimum distance to the image to be detected from the 9 normal images, recording the distance between every two of the 3 images as P1, P2 and P3, and recording the distance between the image to be detected and the 3 images as N1, N2 and N3; and calculating the relative distance D as (N1+ N2+ N3)/(P1+ P2+ P3), if D is larger than 1.2, judging that the image to be detected is abnormal, otherwise, judging that the image to be detected is not abnormal, and if the judgment result is consistent with the label, indicating that the prediction is correct.
Compared with the prior art, the invention has at least the following beneficial effects:
the invention discloses an anomaly detection method of an insulator fastener of a contact network based on deep network learning image distance measurement. The image distance measurement model used for detecting the abnormity of the contact net insulator fastener is the distance measurement model obtained through deep network learning, so the method has the advantages of high abnormity detection speed, high precision and strong robustness of the contact net insulator fastener.
Furthermore, the insulator fastener of the contact net is identified and positioned by training the Faster R-CNN network model, the identification precision can reach 93.2%, and early-stage preparation is made for subsequent abnormal component detection.
Further, the data set learning obtains a network model of image distance measurement, and the distance measurement model can effectively express the distance between the normal image and the distance between the normal image and the abnormal image.
Further, on the basis of the network model of image distance measurement obtained by learning in the steps S3 and S5, the step S6 constructs the method for detecting the abnormality of the insulator fastener of the contact network, the method is simple to calculate, and the detection precision can meet the actual use requirement.
In conclusion, the method for detecting the abnormality of the insulator fastener of the high-speed rail contact network disclosed by the invention is a high-efficiency and high-precision detection method, and an effective solution is provided for the intellectualization of contact network detection.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic illustration of a marking of a contact net insulator fastener according to an embodiment of the invention;
FIG. 3 is a diagram of a fast R-CNN network model structure according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an exemplary image distance measurement network model;
FIG. 5 is a schematic diagram of triple Loss (triple Loss) based training according to an embodiment of the present invention;
fig. 6 is a normal and abnormal state diagram of the contact net insulator fastener in the embodiment of the invention.
Detailed Description
The invention provides an anomaly detection method for insulator fasteners of a high-speed rail contact network, which comprises the steps of establishing an original contact network image data set, identifying and positioning the insulator fasteners of the contact network by using an Faster R-CNN deep network model, and judging whether the corresponding insulator fasteners are abnormal or not by measuring the relative distance between images of the insulator fasteners.
Referring to fig. 1, the method for detecting the abnormality of the insulator fastener of the high-speed rail contact network according to the present invention includes the following steps:
s1, acquiring image data of the overhead line system from the high-speed rail 4C system, shooting high-resolution images of the overhead line system from different angles by a plurality of cameras installed on the roof of the detection vehicle, screening and removing some images which do not meet requirements, and forming an original image sample set, wherein the number of the original image sample set is 600 in the embodiment.
And S2, uniformly adjusting the images in the overhead line system data set constructed in the step S1 to 1280-1280 pixels. As shown in fig. 2, an image labeling tool, Labelimg, is used to label the contact net insulator fastener in a rectangular frame, a contact net data set is constructed, and the contact net data set is divided into a training set and a test set according to a ratio of 7:3, wherein 420 training sets are selected, and 180 test sets are selected.
S3, training a Faster R-CNN network model for identifying and positioning the insulator fasteners of the contact network by using the contact network data set formed in the step S2. The Faster R-CNN network model is shown in fig. 3, and specifically, the Faster R-CNN model uses ResNet50 as a basic feature extraction network, the Learning Rate (Learning Rate) is set to 0.001, the stochastic gradient descent method (SGD) is used, the Momentum (Momentum) is set to 0.9, and the Weight Decay (Weight Decay) is set to 0.0001. The test set is tested by using a FasterR-CNN insulator fastener detection model, the recall rate reaches 98.0 percent, the accuracy rate is 93.2 percent, and model parameters are saved.
S4, constructing an insulator fastener data set
S401, 752 insulator fastener images marked in the step S2 in total, wherein the insulator fastener images are divided into 3 types according to three directions of upper left, upper right and upper right, and then each type is divided into two types according to normal and abnormal, as shown in FIG. 6, namely the insulator fastener images are divided into 6 types and are respectively stored in 6 folders;
s402, uniformly adjusting the size of the insulator fastener images to 128 × 128 pixel values;
and S403, because abnormal data are less, data enhancement is carried out by adding Gaussian noise, sharpness change and contrast change, and 3008 insulator fastener images are obtained in total.
S5 training image distance measurement network model
S501, constructing image distance degreeThe quantity depth network comprises an increment-ResNet-V1 network and an L2 normalization part and a triple Loss function (triple Loss) part of an output vector of the increment-ResNet-V1 network, as shown in FIG. 4, two similar images and one heterogeneous image are input into an increment-Resnet-V1 neural network to extract image characteristics, and an n-128-dimensional vector is obtained through L2 characteristic normalization to be used as a characteristic representation; the term "L2 normalization" means that the vector X is (X)1,x2,...,xn) Is divided by the L2 norm to obtain a new vector X2I.e. by
Utilizing a triple Loss function (triple Loss) to learn separability among the features, wherein the feature distance of the same class is as smaller as possible than the feature distances of different classes; calculating the loss of the triples based on one-dimensional feature representation, wherein the triples are three samples, two samples belong to the same class, and the other samples belong to other classes and are respectively marked as Anchor, Positive and Negative, wherein Anchor and Positive are the same class, and Anchor and Negative are different classes, so that the learning process is one representation, and for the triples as many as possible, the sum of the distance between Anchor and Positive and the marginal value alpha is smaller than the distance between Anchor and Negative, as shown in FIG. 5; the objective function corresponding to the triplet loss here is:
wherein the content of the first and second substances,
respectively representing the character representation of Anchor, Positive and Negative; α is a constant greater than zero; here, the distance is measured by Euclidean distance and + represents [ 2 ]]When the value in (A) is greater than or equal to zero, the value of [ 2 ]]The value of the internal expression is the value of the loss function, [ alpha ], [ alpha]When the value of the internal expression is less than zero, the value of the loss function is zero;
s502, dividing the insulator fastener data set constructed in the step S4 into a training set and a testing set according to the ratio of 7: 3;
s503, training an image distance measurement depth network by using a test set, setting a marginal value alpha to be 0.2, setting a feature vector dimension to be 128, setting a learning rate to be 0.01, randomly initializing network parameters by using an Adagarad optimization algorithm, and setting training iteration times (Epoch) to be 50.
S504, after 50 times of iterative training, saving image distance measurement depth network model parameters;
and S505, testing the image distance measurement depth network model trained in the step S504 by using a test set. Randomly extracting an image in the test set as an image to be detected, recording the label of the image as normal or abnormal (the normal is 1, and the abnormal is 0), randomly selecting 3 images from the 3 normal images respectively, and applying the feature extraction network obtained by training in the step S501 to respectively extract the features of the 10 images, obtaining 128-dimensional feature vectors of the images through L2 feature normalization, and calculating the distance between every two images by using Euclidean distance; taking 3 images with the minimum distance to the image to be detected from the 9 normal images, recording the distance between every two of the 3 images as P1, P2 and P3, and recording the distance between the image to be detected and the 3 images as N1, N2 and N3; calculating the relative distance D (N1+ N2+ N3)/(P1+ P2+ P3), if D is larger than 1.2, judging that the image to be detected is abnormal (marked as 0), otherwise, judging that the image to be detected is not abnormal (marked as 1), and if the judgment result is consistent with the label, judging that the prediction is correct;
and S506, testing to ensure that the accuracy reaches 90.67 percent and the application requirement is met.
And S6, judging whether the insulator fastener is abnormal or not by using the image distance measurement depth network model obtained in the step S5.
S601, identifying and positioning the insulator fastener in the original image of the contact network to be detected by using the Faster R-CNN target detection network model trained in the step S3 to serve as an image of the insulator fastener to be detected;
s602, in the data set constructed in the step S4, 3 images are randomly selected from the 3 normal image sets, 10 images are selected together with the insulator fastener image to be detected, the detection method in the step S505 is applied, the relative distance D is calculated, if D is larger than 1.2, the insulator fastener image is judged to be abnormal, otherwise, the insulator fastener image is judged not to be abnormal.
Experimental results show that the identification accuracy of the insulator fastener in the original image of the contact net is up to more than 93.2% by using the method, and the abnormality detection accuracy of the insulator fastener is up to 90.67%. Therefore, the method can well solve the problem of abnormal detection of the key insulator fastener of the contact network, and is greatly improved compared with the traditional method.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (9)
1. The method for detecting the abnormity of the insulator fastener of the high-speed rail contact network is characterized by comprising the following steps of:
s1, acquiring contact network image data to form an original image sample set;
s2, uniformly adjusting the images in the overhead line system data set constructed in the step S1 to 1280-1280 pixels, and constructing an overhead line system data set;
s3, training a Faster R-CNN network model for identifying and positioning an insulator fastener of the contact network by using the contact network data set constructed in the step S2, testing the application test set of the Faster R-CNN network model, and storing model parameters;
s4, constructing an insulator fastener data set;
s5, constructing an image distance measurement depth network, and training an image distance measurement network model by using the insulator fastener data set constructed in the step S4;
s6, recognizing and positioning the position of the insulator fastener by using the insulator fastener recognition Faster R-CNN network model trained in the step S3 to obtain an image of the insulator fastener to be detected; and then, carrying out abnormity detection on the image of the insulator fastener to be detected by using the image distance measurement depth network model obtained in the step S5, and judging whether the insulator fastener is abnormal or not.
2. The method for detecting the abnormality of the insulator fastener of the high-speed rail catenary according to claim 1, wherein in the step S2, the catenary data set is divided into a training set and a test set according to a ratio of 7: 3.
3. The method for detecting the abnormality of the insulator fastener of the high-speed rail catenary according to claim 1, wherein in step S3, the Faster R-CNN network model uses ResNet50 as a basic feature extraction network, the learning rate is set to 0.001, the random gradient descent method is used, the momentum is set to 0.9, and the weight attenuation is 0.0001.
4. The abnormality detection method for the insulator fastener of the high-speed rail contact network according to claim 1, wherein the step S4 is specifically as follows:
s401, dividing the insulator fastener images marked in the step S2 into 3 types according to the directions, and dividing each type into two types according to the normality and the abnormality, namely dividing the insulator fastener images into 6 types in total and respectively storing the insulator fastener images in 6 folders;
s402, uniformly adjusting the size of the insulator fastener images to 128 × 128 pixel values;
and S403, performing data enhancement through Gaussian noise enhancement, sharpness change and contrast change to obtain 3008 insulator fastener images.
5. The abnormality detection method for the insulator fastener of the high-speed rail contact network according to claim 1, wherein the step S5 is specifically as follows:
s501, constructing an image distance measurement depth network comprising an increment-ResNet-V1 network and an L2 normalization and triple loss function of an output vector of the network;
s502, dividing the insulator fastener data set constructed in the step S4 into a training set and a testing set according to the ratio of 7: 3;
s503, training an image distance measurement depth network by using a training set, wherein a marginal value alpha is 0.2, a feature vector dimension n is 128, the learning rate is 0.01, and network parameters are randomly initialized by using an Adagad optimization algorithm;
s504, after iterative training for 50 times, saving image distance measurement depth network model parameters;
and S505, testing the image distance measurement depth network model trained in the step S504 by using a test set.
6. The method for detecting the abnormality of the insulator fastener of the high-speed rail catenary according to claim 5, wherein in step S501, L2 is normalized to a pair vector X-X1,x2,...,xnIs divided by the L2 norm to obtain a new vector X2The method specifically comprises the following steps:
7. the method for detecting the abnormality of the insulator fastener of the high-speed rail contact network according to claim 5, wherein in the step S501, the objective function corresponding to the triple loss is as follows:
wherein the content of the first and second substances,
respectively representing the character representation of Anchor, Positive and Negative; α is a constant greater than zero; + represents [, ]]When the value of [ alpha ], [ alpha]The value of the internal expression is the value of the loss function, [ alpha ], [ alpha]The value of the loss function is zero when the value of the inner expression is less than zero.
8. The abnormality detection method for the insulator fastener of the high-speed rail contact network according to claim 1, wherein the step S6 is specifically as follows:
s601, identifying and positioning the insulator fastener in the original image of the contact network to be detected by using the Faster R-CNN target detection network model trained in the step S3 to serve as an image of the insulator fastener to be detected;
s602, randomly selecting 3 images from the 3 normal image sets in the data set constructed in the step S4, forming 10 images together with the insulator fastener image to be detected, and judging whether the insulator fastener image is abnormal or not according to the relative distance D.
9. The abnormality detection method for the insulator fastener of the high-speed rail contact network according to claim 8, wherein in step S602, the 128-dimensional feature vectors of the images are obtained by subjecting the 10 images to L2 feature normalization, and the distance between each two images is calculated by using the euclidean distance; taking 3 images with the minimum distance to the image to be detected from the 9 normal images, recording the distance between every two of the 3 images as P1, P2 and P3, and recording the distance between the image to be detected and the 3 images as N1, N2 and N3; and calculating the relative distance D as (N1+ N2+ N3)/(P1+ P2+ P3), if D is larger than 1.2, judging that the image to be detected is abnormal, and if not, judging that the image to be detected is not abnormal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010549723.2A CN111899216A (en) | 2020-06-16 | 2020-06-16 | Abnormity detection method for insulator fastener of high-speed rail contact network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010549723.2A CN111899216A (en) | 2020-06-16 | 2020-06-16 | Abnormity detection method for insulator fastener of high-speed rail contact network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111899216A true CN111899216A (en) | 2020-11-06 |
Family
ID=73207688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010549723.2A Pending CN111899216A (en) | 2020-06-16 | 2020-06-16 | Abnormity detection method for insulator fastener of high-speed rail contact network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111899216A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113160190A (en) * | 2021-04-27 | 2021-07-23 | 西安交通大学 | Method, equipment, device and storage medium for detecting abnormity of fastener of contact network |
| CN113192071A (en) * | 2021-06-29 | 2021-07-30 | 南方电网数字电网研究院有限公司 | Insulator binding wire missing identification method and device and computer equipment |
| CN113487166A (en) * | 2021-06-30 | 2021-10-08 | 桐乡市五疆科技发展有限公司 | Chemical fiber floating filament quality detection method and system based on convolutional neural network |
| CN113487541A (en) * | 2021-06-15 | 2021-10-08 | 三峡大学 | Insulator detection method and device |
| CN116958713A (en) * | 2023-09-20 | 2023-10-27 | 中航西安飞机工业集团股份有限公司 | Quick recognition and statistics method and system for surface fastener of aviation part |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108009591A (en) * | 2017-12-14 | 2018-05-08 | 西南交通大学 | A kind of contact network key component identification method based on deep learning |
| CN109801284A (en) * | 2019-01-25 | 2019-05-24 | 华中科技大学 | A kind of high iron catenary insulator breakdown detection method based on deep learning |
| CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
| US20200082224A1 (en) * | 2018-09-10 | 2020-03-12 | Sri International | Weakly supervised learning for classifying images |
-
2020
- 2020-06-16 CN CN202010549723.2A patent/CN111899216A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108009591A (en) * | 2017-12-14 | 2018-05-08 | 西南交通大学 | A kind of contact network key component identification method based on deep learning |
| US20200082224A1 (en) * | 2018-09-10 | 2020-03-12 | Sri International | Weakly supervised learning for classifying images |
| CN109801284A (en) * | 2019-01-25 | 2019-05-24 | 华中科技大学 | A kind of high iron catenary insulator breakdown detection method based on deep learning |
| CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
Non-Patent Citations (1)
| Title |
|---|
| 孙萌等: "基于TripletLoss损失函数的舌象分类方法研究", 《北京生物医学工程》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113160190A (en) * | 2021-04-27 | 2021-07-23 | 西安交通大学 | Method, equipment, device and storage medium for detecting abnormity of fastener of contact network |
| CN113487541A (en) * | 2021-06-15 | 2021-10-08 | 三峡大学 | Insulator detection method and device |
| CN113192071A (en) * | 2021-06-29 | 2021-07-30 | 南方电网数字电网研究院有限公司 | Insulator binding wire missing identification method and device and computer equipment |
| CN113487166A (en) * | 2021-06-30 | 2021-10-08 | 桐乡市五疆科技发展有限公司 | Chemical fiber floating filament quality detection method and system based on convolutional neural network |
| CN116958713A (en) * | 2023-09-20 | 2023-10-27 | 中航西安飞机工业集团股份有限公司 | Quick recognition and statistics method and system for surface fastener of aviation part |
| CN116958713B (en) * | 2023-09-20 | 2023-12-15 | 中航西安飞机工业集团股份有限公司 | Quick recognition and statistics method and system for surface fastener of aviation part |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111899216A (en) | Abnormity detection method for insulator fastener of high-speed rail contact network | |
| CN111179251B (en) | Defect detection system and method based on twin neural network and by utilizing template comparison | |
| CN109118479B (en) | Capsule network-based insulator defect identification and positioning device and method | |
| CN108074231B (en) | Magnetic sheet surface defect detection method based on convolutional neural network | |
| CN112200178B (en) | Transformer substation insulator infrared image detection method based on artificial intelligence | |
| CN111462076B (en) | Full-slice digital pathological image fuzzy region detection method and system | |
| CN107742130A (en) | High iron catenary based on deep learning supports device fastener failure diagnostic method | |
| Bo et al. | Particle pollution estimation from images using convolutional neural network and weather features | |
| CN107833211B (en) | Infrared image-based zero value insulator automatic detection method and device | |
| CN110598693A (en) | Ship plate identification method based on fast-RCNN | |
| CN111127448B (en) | Method for detecting air spring fault based on isolated forest | |
| CN108664997A (en) | High iron catenary equipotential line defective mode detection method based on cascade Faster R-CNN | |
| CN112381175A (en) | Circuit board identification and analysis method based on image processing | |
| CN114022715A (en) | Method and device for detecting lead sealing defect of high-voltage cable terminal | |
| CN115019294A (en) | Pointer instrument reading identification method and system | |
| CN112084860A (en) | Target object detection method and device and thermal power plant detection method and device | |
| CN108508023A (en) | The defect detecting system of end puller bolt is contacted in a kind of railway contact line | |
| CN110689447A (en) | Real-time detection method for social software user published content based on deep learning | |
| CN112132088B (en) | Inspection point missing inspection identification method | |
| CN112598666B (en) | Cable tunnel anomaly detection method based on convolutional neural network | |
| CN112699824B (en) | Method and device for detecting constant of electric energy meter and storage medium | |
| CN113128555A (en) | Method for detecting abnormality of train brake pad part | |
| Yang et al. | Defect detection scheme of pins for aviation connectors based on image segmentation and improved RESNET-50 | |
| CN110070520B (en) | Pavement crack detection model construction and detection method based on deep neural network | |
| CN112508946B (en) | Cable tunnel anomaly detection method based on antagonistic 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201106 |