CN115311261A - Method and system for detecting abnormality of cotter pin of suspension device of high-speed railway contact network - Google Patents
Method and system for detecting abnormality of cotter pin of suspension device of high-speed railway contact network Download PDFInfo
- Publication number
- CN115311261A CN115311261A CN202211219348.0A CN202211219348A CN115311261A CN 115311261 A CN115311261 A CN 115311261A CN 202211219348 A CN202211219348 A CN 202211219348A CN 115311261 A CN115311261 A CN 115311261A
- Authority
- CN
- China
- Prior art keywords
- image
- cotter pin
- cotter
- suspension device
- reconstruction
- 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
- 239000000725 suspension Substances 0.000 title claims abstract description 97
- 238000000034 method Methods 0.000 title claims abstract description 48
- 230000005856 abnormality Effects 0.000 title claims abstract description 24
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 83
- 238000001514 detection method Methods 0.000 claims abstract description 44
- 230000002159 abnormal effect Effects 0.000 claims abstract description 23
- 238000005520 cutting process Methods 0.000 claims abstract description 18
- 238000004590 computer program Methods 0.000 claims description 20
- 239000000126 substance Substances 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 description 33
- 238000004364 calculation method Methods 0.000 description 19
- 230000006870 function Effects 0.000 description 17
- 238000012549 training Methods 0.000 description 16
- 238000002372 labelling Methods 0.000 description 15
- 230000008569 process Effects 0.000 description 14
- 238000000605 extraction Methods 0.000 description 9
- 230000004927 fusion Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000007689 inspection Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 244000141353 Prunus domestica Species 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 208000003464 asthenopia Diseases 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000013138 pruning Methods 0.000 description 1
- 238000006467 substitution reaction 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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]
-
- 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
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Image Processing (AREA)
Abstract
The application provides a method and a system for detecting abnormality of a cotter pin of a suspension device of a high-speed railway contact network. The method comprises the following steps: acquiring an image of a contact net suspension device to be detected; extracting a connection area in an image of the overhead line system suspension device to obtain an image of the connection area; positioning and cutting the cotter pin component in the connection area image to obtain a cotter pin image; carrying out image reconstruction on the cotter image based on an image reconstruction model to obtain a cotter reconstruction image, wherein the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm; and calculating a reconstruction error according to the cotter pin image and the cotter pin reconstruction image, and carrying out abnormity detection on the contact net suspension device image to be detected through the reconstruction error. The application can improve the accuracy of the abnormal detection of the split pin of the contact net suspension device.
Description
Technical Field
The application relates to the technical field of fault diagnosis of mechanical parts, in particular to a method and a system for detecting the abnormality of a cotter pin of a suspension device of a high-speed railway contact network.
Background
The overhead line system suspension device is one of the most important devices of a high-speed railway system, and has very important influence on the safe and stable operation of the railway. However, the contact force between the pantograph and the overhead line system is very large, and the vibration and impact transmitted by the interaction between the vehicle and the rail and the pantograph and overhead line system can gradually damage parts of the overhead line system. At present, the detection means mainly comprises reading a large amount of image data manually and offline. However, with the large-scale construction of high-speed electrified railways, different cameras mounted on inspection vehicles are usually photographed at night, the number of images obtained by single inspection is huge, the angles of the obtained images are various, abnormal targets are small, the number of abnormal targets is small, the efficiency of abnormality detection based on manual work is low, and the conditions of missed inspection and false inspection are easy to occur due to visual fatigue of inspectors.
With the development of digital image processing technology in recent years, some researchers propose an image detection means based on traditional computer vision, and although the appearance abnormality can be detected to a certain extent, the traditional algorithm is limited by sample difference and surrounding environment, and cannot meet ideal requirements on precision and speed.
Disclosure of Invention
The application provides a method and a system for detecting the abnormality of a cotter pin of a suspension device of a high-speed railway contact network, which aim to solve the problem of low accuracy of detecting the abnormality of the cotter pin in the suspension device of the contact network in the prior art.
In a first aspect, the application provides a method for detecting the abnormality of a cotter pin of a suspension device of a high-speed railway contact network, which comprises the following steps:
acquiring an image of a catenary suspension device to be detected;
extracting a connection area in the contact net suspension device image to obtain a connection area image;
positioning and cutting the cotter pin component in the connecting area image to obtain a cotter pin image;
carrying out image reconstruction on the cotter image based on an image reconstruction model to obtain a cotter reconstruction image, wherein the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm;
and calculating a reconstruction error according to the cotter pin image and the cotter pin reconstruction image, and carrying out abnormity detection on the contact net suspension device image to be detected through the reconstruction error.
In a second aspect, the application provides an unusual detection device of high-speed railway contact net linkage device cotter pin, includes:
the acquisition module is used for acquiring an image of the contact net suspension device to be detected;
the connecting area determining module is used for extracting a connecting area in the contact net suspension device image to obtain a connecting area image;
the cotter determining module is used for positioning and cutting the cotter area in the connecting area image to obtain a cotter image, and the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm;
the image reconstruction module is used for carrying out image reconstruction on the cotter pin image based on an image reconstruction model to obtain a cotter pin reconstruction image;
and the detection module is used for calculating a reconstruction error according to the cotter image and the cotter reconstruction image and carrying out abnormity detection on the contact net suspension device image to be detected through the reconstruction error.
In a third aspect, the present application provides a terminal, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to the first aspect or any one of the possible implementation manners of the first aspect when executing the computer program.
In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the steps of the method as described above in the first aspect or any one of the possible implementations of the first aspect.
The invention provides a method and a system for detecting the abnormal condition of a cotter pin of a suspension device of a high-speed railway contact network.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a flowchart of an implementation of a method for detecting an abnormality of a cotter pin of a suspension device of a high-speed railway overhead line system according to an embodiment of the present application;
FIG. 2 is a flowchart of an implementation of data enhancement incorporating a GridMask according to an embodiment of the present application;
fig. 3 is a network structure diagram of improvement of the YOLOv5 model provided in the embodiment of the present application;
FIG. 4 is a basic flowchart of the YOLOv5 algorithm training provided by the embodiments of the present application;
fig. 5 is a schematic structural diagram of an AutoEncoder + StyleGAN model provided in an embodiment of the present application;
fig. 6 is a schematic structural diagram of a ResNet50 network provided in an embodiment of the present application;
fig. 7 is a schematic structural diagram of an FPN network provided in the embodiment of the present application;
fig. 8 is a schematic structural diagram of a decoder network provided in an embodiment of the present application;
FIG. 9 is a basic flow chart of the reconstruction error calculation classification provided by the embodiments of the present application;
fig. 10 is a schematic view of a connection area in an image of a catenary suspension device provided in an embodiment of the present application;
FIG. 11 is a schematic view of a cotter pin member in an image of a connection area provided by an embodiment of the application;
FIG. 12 is a graph of image contrast before and after connected region extraction provided by embodiments of the present application;
FIG. 13 is an image comparison graph of a cotter pin section provided in an embodiment of the present application before and after extraction;
FIG. 14 is an image comparison graph of a cotter pin component provided in an embodiment of the present application before and after reconstruction;
FIG. 15 is a diagram showing an abnormality detection result of a cotter pin member according to an embodiment of the present application;
fig. 16 is a schematic structural diagram of an abnormal detection device for a cotter pin of a suspension device of a high-speed railway contact network provided in an embodiment of the present application;
fig. 17 is a schematic diagram of a terminal provided in an embodiment of the present application.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
To make the objects, technical solutions and advantages of the present application more clear, the following description is made by way of specific embodiments with reference to the accompanying drawings.
Fig. 1 is a flowchart of an implementation of a method for detecting an abnormality of a cotter pin of a suspension device of a high-speed railway contact network according to an embodiment of the present application, which is detailed as follows:
in S101, an image of the catenary suspension device to be detected is acquired.
At present, with the large-scale construction of high-speed electrified railways, images of a contact network suspension device are acquired by shooting the contact network suspension device of the high-speed railway at night according to different cameras mounted on a detection vehicle, and then the shot images are packed and recorded by related personnel, wherein the images at the moment are RGB images, and the resolution of the images is kept consistent.
In S102, a connection region in the catenary suspension apparatus image is extracted to obtain a connection region image.
Since the images of the catenary suspension devices acquired in S101 have various angles, small abnormal targets and small abnormal numbers, the connection areas in the images of the catenary suspension devices need to be located.
In a possible implementation manner, S101 may specifically include:
labeling an image of the contact net suspension device to be detected based on a connection region labeling model, wherein the labeling comprises an image name, a connection region category and a connection region coordinate;
performing data enhancement on the marked image of the catenary suspension device to obtain a first data enhanced image;
and positioning the connection area in the first data enhanced image to obtain a connection area image.
The training process of the connection region labeling model comprises the following steps: firstly, manually labeling a connecting region in an image of a contact net suspension device by using a related image labeling tool, wherein the labeling comprises an image name, a connecting region type and a connecting region coordinate, then dividing the labeled image of the contact net suspension device into a training set, a verification set and a test set according to a proper proportion, training and learning a labeling target, and finally obtaining a connecting region labeling model.
Based on the connection region labeling model, inputting the to-be-detected contact net suspension device image obtained in the step S101 into the connection region labeling model, and labeling the connection region, wherein the obtained labeled contact net suspension device image comprises an image name, a connection region category and a connection region coordinate.
In a possible implementation manner, performing data enhancement on the labeled image of the suspension device of the overhead line system to obtain a first data enhanced image may include:
and performing data enhancement on the marked image of the contact net suspension device by adopting a GridMask data enhancement algorithm to obtain a first data enhanced image.
The GridMask data enhancement algorithm is an algorithm for generating a mask with the same resolution as the original image and multiplying the mask and the original image to obtain an image.
The GridMask data enhancement algorithm is added into the YOLOv5 algorithm as a new data enhancement processing mode and is fused with other data enhancement modes.
The specific implementation process is as follows: referring to fig. 2, in a program, an image data set for training is read first, fusion of data enhancement modes such as image scaling, image mirroring, mosaic enhancement and the like originally provided by a YOLOv5 algorithm is used, and meanwhile, a label of an image is also synchronously processed along with the image; then, a GridMask data enhancement algorithm is cited in the YOLOv5 algorithm, and the specific operation is that the image data after data enhancement takes the size asThe probability of the active site is enhanced by GridMask data and is analyzed through experimentsThe influence of the value on the expression effect of the model on the evaluation index,taking the value of 0.5, namely performing GridMask enhancement on the image after data enhancement with the probability of 50%; the label is not required to be processed when the GridMask enhancement is carried out, because the GridMask enhancement mode only forms a rectangular array with a certain size and arrangement mode on the image to cover the original image and does not influence the specific coordinate of the target in the image.
In the embodiment of the invention, various data enhancement modes are automatically completed through programs without manual operation.
After the marked image of the catenary suspension device is obtained, when image data and marked label data are read by a YOLOv5 algorithm, the image and the label are processed synchronously, a GridMask data enhancement algorithm is added to the YOLOv5 algorithm and fused with other data enhancement algorithms in the YOLOv5 algorithm, wherein the enhancement algorithms in the YOLOv5 algorithm comprise a plurality of data enhancement modes such as image rotation, image scaling and mosaic enhancement, and therefore a first data enhancement image is obtained.
In a possible implementation manner, the locating a connection region in the first data enhanced image to obtain a connection region image may include:
based on the first target positioning model, performing target positioning on the connection area in the first-time data enhanced image, and cutting the connection area subjected to the target positioning to obtain a connection area image;
the first target positioning model is constructed based on a YOLOv5 algorithm.
The YOLOv5 algorithm uses a deep neural network to detect and classify the position of an object, has the main characteristics of high speed and high accuracy, and combines two stages of a candidate area and object identification into a whole by adopting a method of directly predicting a boundary box of a target object. The image size of the YOLOv5 input is uniformThe predicted three feature layer sizes are respectively,,。The connection areas in the images of the catenary suspension device in the embodiment of the invention occupy small specific gravity in the whole image, so that the Yolov5 algorithm in the embodiment of the invention is improved on the conventional structure, multi-scale detection is added, the characteristics of the image are extracted by a deeper network structure, and the characteristics of the image are removedAnd the anchors are reduced from 9 to 3 at the same time, so that the improved YOLOv5 network is more suitable for the identification and detection of the connection areas of smaller objects, and the identified connection areas are extracted and cut out so as to facilitate the detection and cutting of cotter pin parts in the connection area images.
The three feature map layers are as follows:
removingThe layer YOLOv5 algorithm is more suitable for detecting a connection region occupying a smaller proportion of an image, and an original image is subjected to data enhancement and then input into a YOLOv5 network to be trained and detected to obtain a connection region image. The improved YOLOv5 model framework is subjected to algorithm pruning, and a network structure diagram of a specific model improvement refers to FIG. 3:
the size of the input image of the network structure isFinally, the feature map scale size used for prediction is. The improved YOLOv5 model comprises four main substructures of CBS, resUnit, C3-N and SPPF, and each substructure is specifically introduced as follows:
and (3) CBS: the structure consists of convolution operations (Conv), batch Normalization (Batch Normalization) and activation function SiLu operations.
ResUnit: the input of the structure is an output characteristic diagram of a CBS substructure, and the output characteristic diagram is added with an input characteristic diagram of a ResUnit structure after 2 substructures CBS operations.
C3-N: the input of the structure is an output characteristic diagram of a CBS substructure, firstly, the input characteristic diagram is respectively calculated by two branches, the first branch is CBS structure operation, the second branch is CBS structure operation, then N ResUnit structure operations are carried out, and the output characteristic diagrams of the two branches are spliced and then CBS structure operation is carried out.
SPPF: the input of the structure is an output feature map of a CBS substructure, firstly, CBS structure operation is carried out on the input feature map, the obtained feature map is subjected to maximum pooling operation (Max Pool) for three times, and the feature map after the CBS structure operation and the feature map after each maximum pooling operation (four feature maps in total) are spliced and then CBS operation is carried out.
In the embodiment of the present invention, the improved YOLOv5 network is constructed according to the above four substructures in the manner shown in fig. 3, and the size of the final output feature graph is as follows。
And performing target positioning and cutting on a first data enhanced image obtained by data enhancement by adopting a first target positioning model constructed based on an improved YOLOv5 network to obtain a connection region image. The connecting area in the contact net suspension device image is positioned and cut, so that redundant useless areas are removed, the opening pin component of the subsequent connecting area can be conveniently identified, the schematic diagram of the connecting area image refers to fig. 10, and the image comparison diagram before and after the connecting area image is extracted refers to fig. 12.
The embodiment of the invention applies the improved YOLOv5 algorithm to the detection of the connecting area of the contact net suspension device, and the branches of the algorithm are trimmed aiming at the proportion of the connecting area in the image, so that the automatic identification and extraction of the connecting area are realized; the improved YOLOv5 algorithm is used for identifying the connection area, and a one-stage target detection algorithm is adopted, so that the detection efficiency is improved, and the identification precision is also ensured.
In S103, the cotter members in the connection region image are positioned and cut out, and a cotter image is obtained.
Because the proportion of pixels occupied by the connecting area in the image of the catenary suspension device is small, the proportion of pixels occupied by the cotter component in the image of the catenary suspension device is small, the cotter component is directly marked on the image of the catenary suspension device, and the positioning of the cotter component on the image of the catenary suspension device by using the improved YOLOv5 model cannot be realized, the positioning and the cutting of the cotter component are performed on the image of the connecting area after the image of the connecting area is obtained on the basis of S102, so that the cotter image is obtained, the distribution training and the extraction are completed, the whole process needs to be completed under the improved YOLOv5 algorithm, and the basic flow chart of the YOLOv5 algorithm training refers to FIG. 4.
In a possible implementation manner, S103 may specifically include:
marking the cotter pin component in the connecting area image based on the cotter pin marking model;
performing data enhancement on the marked image of the connection area to obtain a second data enhanced image;
based on the second target positioning model, performing target positioning on the cotter component in the second data enhanced image, and cutting the cotter component subjected to target positioning to obtain a cotter image;
and the second target positioning model is constructed based on a YOLOv5 algorithm.
The training process of the cotter marking model is consistent with that of the connecting area marking model, and the difference is that the cotter components in the connecting area need to be marked manually, and the marking comprises an image name, a cotter category and cotter coordinates.
Under the condition that the pixel accuracy of the images of the connecting regions is guaranteed, the cotter components in the images of the connecting regions are marked on the basis of the cotter marking model, the image data of the connecting regions and part of original image data of the overhead contact system suspension device are mixed to form a new image set as training input data, a GridMask data enhancement algorithm is adopted to be added into a YOLOv5 algorithm to be fused with other data enhancement modes, and the incremental training of mixed data is carried out to obtain a second data enhancement image. The data enhancement process is the same as the data enhancement process in S102, and will not be described in detail.
And after the second data enhanced image is obtained, constructing an obtained second target positioning model based on a YOLOv5 algorithm, carrying out target positioning on the connection region image, and cutting the cotter pin image subjected to target positioning to obtain the cotter pin image. The schematic diagram of the cotter image is shown in fig. 11, and the image comparison diagram before and after cotter image extraction is shown in fig. 13.
Based on S102 and S103, it can be known that the training strategy of the YOLOv5 algorithm provided in the embodiment of the present invention is consistent, and the specific expression is as follows: firstly, completing model training of coarse extraction of a connecting area of a contact net suspension device, obtaining a cut connecting area image, and after manually marking the cut connecting area image, mixing the connecting area image with an original contact net suspension device image to form a new data set; and secondly, performing incremental training on the mixed data on the basis of the model which is trained in the previous step, and finally obtaining the well-trained improved YOLOv5 algorithm. At the moment, crude extraction of the connecting area of the overhead line system suspension device can be completed only through a final algorithm, and the cotter pin image is accurately positioned on the crude extracted connecting area image.
The evaluation index for the improved YOLOv5 algorithm can be evaluated according to the following criteria:
by using an improved target detection algorithm of a YOLOv5 model, a connecting area and an opening pin part of a contact net suspension device are realizedObject recognition of the object, recording the overlapping rate of the results asSetting the threshold value to be 0.5 whenIf the target is detected, the type of the prediction box is obtained as a real exampleExample of falseTrue inverse exampleFalse negative exampleCalculating the improved performance evaluation index of the YOLOv5 model,(the accuracy rate),(recall ratio) of the number of times,(average value of precision),(average precision mean).
and meanwhile, the recognized cotter pin image is restored to a connection area image, and the connection area image is restored to the initial contact net suspension device image data, so that the final abnormal detection result is used as an auxiliary detection result to be submitted to relevant detection personnel for final judgment.
For example, the embodiment of the invention adopts the image data of the catenary suspension device of the Jinghush railway gallery section for verification. The number of images of the contact net suspension device of the Jing and Shanghai railway gallery workshop section is 175, and the image resolution isA pixel. According toIs divided into a training set, a validation set and a test set. Positioning, cutting and extracting the connecting area of the contact net suspension device by an improved YOLOv5 algorithm, filling 0 element value into the original image and adjusting the original image to beThe size of the image data is increased, so that the occupation of a model on a memory caused by an overlarge data image is reduced while the subsequent detection precision is improved. The images of the contact net suspension device of the connection area image mixing part obtained by cutting are 200 in total, the image resolution ratio is unequal according toThe proportions of (a) are divided into a training set, a validation set and a test set. The blended image data set is further trained based on the trained modified YOLOv5 network. The results show that: the detection system for detecting the cotter pin at the connection position of the suspension device of the high-speed railway contact net based on the improved YOLOv5 algorithm can quickly and accurately identify the connection area and the cotter pin component. The target detection model has stronger robustness and generalization capability on the identification of the connection region and the cotter pin component.
The embodiment of the invention applies the improved YOLOv5 algorithm to the detection of the cotter pin component in the connection area, and prunes the branches of the algorithm according to the proportion of the cotter pin component in the image, thereby realizing the automatic identification and extraction of the cotter pin component; the improved YOLOv5 algorithm is used for identifying the cotter pin part, a one-stage target detection algorithm is adopted, the detection efficiency is improved, the identification precision is guaranteed, two times of marking and training of images (images of a contact net suspension device and images of a connection area) in different types are carried out, and automatic identification and extraction of targets on the images in different scales and different types can be realized only through the improved YOLOv5 algorithm.
In S104, image reconstruction is carried out on the cotter image based on the image reconstruction model to obtain a cotter reconstruction image, and the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm.
Among them, an Automatic Encoder (AE) is a type of Artificial Neural Networks (ANNs) used in semi-supervised learning and unsupervised learning, and functions to perform representation learning (representation learning) on input information by using the input information as a learning target. The AutoEncoder algorithm can realize the reconstruction task of the image, but the cotter image has the characteristics of lower resolution, smaller target and the like, and the simple AutoEncoder algorithm has great difficulty in reconstructing the cotter image and poor reconstruction effect.
The StyleGAN is a human face image generation algorithm, is a countercheck production network, has the advantages of learning various style characteristics of images and still generating high-quality pseudo images when facing low-resolution images.
Therefore, an image reconstruction model is constructed by combining the AutoEncoder algorithm and the StyleGAN algorithm and is used for the task of high-quality reconstruction of cotter images, the structural schematic diagram of the AutoEncoder + StyleGAN model refers to FIG. 5, and the image comparison diagrams before and after the cotter pin member is reconstructed and obtained based on the image reconstruction model refer to FIG. 14.
Wherein, the AutoEncoder mainly includes two parts, which are an Encoder (Encoder) and a Decoder (Decoder) in fig. 6, respectively, wherein the Encoder structure adopts a ResNet50 network structure, and performs multi-scale fusion of feature maps by using an FPN network (pyramid network) fusion mode, the ResNet50 network structure diagram refers to fig. 6, the FPN network structure diagram refers to fig. 7, the Decoder structure is derived from a generator in a StyleGAN model, a specific network structure diagram refers to fig. 8, and is used for encoding features extracted by the Encoder (refer to fig. 5,representing the extracted feature codes) to generate a reconstructed image. Furthermore, the data dimensions of feature encodings are consistent with those in StyleGAN, i.e.,Has the dimension of. The role of the StyleBlock structure in FIG. 5 is: sampling the characteristic graphs of different scales extracted by the encoder to finally obtainThe data dimension of (c).
Feature maps under different scales are obtained from the cotter pin image obtained in S103 through a ResNet50 network of the encoder, the feature maps of three scales of the encoder structure are respectively from the output features of layers 6, 20 and 23 of the ResNet50, the up-sampling mode adopted by feature fusion of different scales is bilinear interpolation, and the feature map to be fused is directly added with the up-sampling feature map, which is referred to fig. 7. The specific calculation process is as follows:
wherein, the first and the second end of the pipe are connected with each other,respectively representing a feature map to be upsampled, a feature map to be channel number changed, wherein,the number of channels and the size of the feature map are respectivelyAnd,the number of channels and the size of the feature map are respectivelyAnd,respectively showing the characteristic diagram after the up-sampling and the characteristic diagram after the channel number is changed, wherein,the number of channels and the size of the feature map are respectivelyAnd,the number of channels and the size of the feature map are respectivelyAnd,representing two feature maps at different scalesAnd (4) fusing the feature maps.
In an encoder, a model can obtain three fused feature maps with different scales, and the three fused feature maps are respectively used as original images from top to bottom compared with the original images、、. Coding and mapping the obtained feature maps under three scales through a StyleBlock structure, wherein each StyleBlock structure maps the feature map into oneThe dimension tensor is the 10 total feature maps of the upper layer, the middle layer and the lower layer of the encoder structure of FIG. 5, which are mapped by three, four and three StyleBlock structuresTensor of dimensionTo. The resulting 10 feature encodings will be used in the decoder for the generation of the reconstructed image, and the schematic structure of the decoder network is shown in fig. 8.
The decoder initializes a tensor of dimensionsThe initialized tensor is regarded as an eigen map, and the eigen map gradually enlarges the scale size of the eigen map through operations such as convolution, encoding of 10 eigen maps extracted by an encoder and the like, and finally a reconstructed image is generated. Each substructure contains operations including Equal line, modulation, and DeModulation. It is composed ofIn the method, equal Linear is a full-link layer structure, wherein Fused Leaky Relu is a dimension-adjusted Leaky Relu activation function, and the calculation formula is as follows:
wherein the content of the first and second substances,is a scale adjustment factor, which takes the value in the model as,Is a very small offset of the bias voltage,the value in the model is 0.2,is the input tensor of Equal Linear.
Modulation is a simple calculation structure, and the calculation formula is as follows:
wherein the content of the first and second substances,is corresponding toAnThe scaling of (a) to (b) is,for the computing structureAn initial weight is required that is a function of,is one dimension ofThe feature code is one of 10 feature codes extracted by the encoder.
The Demodulation is also a simple calculation structure, and the calculation formula is as follows:
wherein the content of the first and second substances,is a very small positive number to ensure that the value is stable.
The StyleBlock1, styleBlock2, toRGB sub-structures are also involved in the decoder process. The function of the StyleBlock1 substructure is to fuse a feature code obtained by the encoder with an output feature map of a previous substructure, the input of the structure is the output feature map of the previous substructure and the feature code to be fused, and the structure does not change the dimension of the input feature map. Firstly, the characteristic code is subjected to Equal Linear, modulation and DeModulation operations to obtain a weight, and the weight is used as the weight of convolution to perform on an input characteristic diagramAnd performing convolution operation with the convolution kernel size, and activating by adding noise and Fused Leaky Relu to finally obtain an output characteristic diagram.
The StyleBlock2 substructure functions in accordance with StyleBlock 1. Unlike this structure, which changes the dimension of the input feature map, the input feature map is expanded by two times, and the difference with StyleBlock1 in the operation process is that: in the process ofBefore convolution operation of the convolution kernel size, the characteristic diagram is up-sampled, and the characteristic diagram scale is enlarged.
The function of the tomgb sub-structure is to adjust the number of channels of the output feature map of the previous sub-structure to 3, facilitating the output of the final reconstructed image, the input of the structure being the output feature map of the previous sub-structure and the feature code with fusion. Firstly, the weight obtained after the characteristic code is subjected to Equal line and Modulation operationPerforming convolution on the input feature map as a weightAnd (3) performing convolution operation by the size of the convolution kernel, changing the number of channels of the feature map from 32 to 3, and outputting the feature maps of 3 channels.
In the embodiment of the invention, besides designing the image reconstruction model based on the AutoEncoder algorithm and the StyleGAN algorithm, the loss function based on the AutoEncoder algorithm and the StyleGAN algorithm is also provided and recordedA cotter pin image representing the input AutoEncoder algorithm and StyleGAN algorithm,separately representing the encoder and decoder of the algorithm, the cotter reconstructed image may be represented asThe loss function includes two terms, and the calculation formula is as follows:
wherein the content of the first and second substances,and the extractor is used for extracting the comprehensive numerical value of the cotter pin image and the reconstructed image on the characteristic diagram.
The calculation formula of the total loss function of the AutoEncoder algorithm and the StyleGAN algorithm according to the loss function is as follows:
wherein the content of the first and second substances,andis a constant that defines the loss weight.
The embodiment of the invention adopts the AutoEncoder algorithm and the StyleGAN algorithm to be applied to reconstruction of the cotter image, integrates the advantage that the StyleGAN algorithm can learn the style characteristics of the image compared with the common AutoEncoder algorithm, can reconstruct an image with higher quality, and is favorable for calculation of reconstruction errors and classification of the image for the reconstructed image with high quality.
In S105, a reconstruction error is calculated according to the cotter image and the cotter reconstructed image, and an anomaly detection is performed on the catenary suspension device image to be detected according to the reconstruction error.
And calculating a reconstruction error based on the cotter pin image obtained in the S103 and the S104 and the cotter pin reconstruction image corresponding to the cotter pin image, and judging whether the catenary suspension device image to be detected is an abnormal image according to the reconstruction error.
In a possible implementation manner, S105 may specifically include:
calculating the average absolute error of the cotter image and the cotter reconstructed image through a first formula, calculating the structural similarity of the cotter image and the cotter reconstructed image through a second formula, and performing weighted summation on the average absolute error and the structural similarity to obtain a reconstructed error;
judging whether the reconstruction error is larger than a classification threshold value;
if the reconstruction error is larger than the classification threshold, determining the cotter pin image as an abnormal image;
wherein the first formula is:
wherein the content of the first and second substances,in order to average the absolute error of the signal,as an image of the cotter pinThe value of each of the pixels is calculated,reconstruction of images for cotter pinsThe value of each of the pixels is calculated,the total number of pixel values of the reconstructed images for the cotter image and the cotter pin;
the second formula is:
wherein, the first and the second end of the pipe are connected with each other,in order to achieve the similarity of the structure,is the average of the pixel values of the cotter pin image,the mean of the pixel values of the reconstructed image for the cotter pin,is the variance of the pixel values of the cotter image,the variance of the pixel values of the image is reconstructed for the cotter pin,the covariance of the pixel values for the cotter pin image and the cotter pin reconstructed image,for the purpose of a first given constant, the value of,is a second given constant, wherein,。。
and acquiring the accurate positioning of the cotter pin component through an improved YOLOv5 network, and selecting the cut image to carry out reconstruction error calculation classification. Aiming at the end-to-end reconstruction task of the cotter pin image, the calculation index of the reconstruction error is designed to be the average absolute error between the original image and the reconstructed imageStructural similarity of original image and reconstructed imageA classification threshold is set according to the weighted numerical value of the index, and when the reconstruction error is larger than the classification threshold, the corresponding cotter pin image is considered as an abnormal image, and a basic flow chart of error calculation classification is reconstructed with reference to fig. 9.
And weighting and summing the average absolute error and the structural similarity to obtain a reconstruction error, wherein a calculation formula of the reconstruction error is as follows:
wherein the content of the first and second substances,is the mean absolute errorThe weight constant of (a) is set,is the degree of structural similarityIs calculated.
Normal and abnormal cotter pin images are classified according to the calculated and reconstructed error values, and a partial cotter pin component abnormality detection result map is shown in fig. 15.
The embodiment of the invention belongs to a data driving method, features do not need to be extracted manually, and the abnormity identification of the cotter pin component at the connecting part of the contact net suspension device can be realized through the cotter pin abnormity detection system at the connecting part of the contact net suspension device of the high-speed railway, so that the professional requirements are reduced, and the engineering applicability is increased.
The invention provides a method for detecting the abnormal condition of a cotter pin of a suspension device of a high-speed railway overhead line system, which is compared with the prior art, the method comprises the steps of firstly carrying out coarse positioning on an image of the suspension device of the overhead line system to be detected to obtain an image of a connection area, then carrying out accurate positioning on the image of the connection area to obtain an image of the cotter pin, and judging whether the image of the cotter pin is an abnormal image or not by carrying out reconstruction error calculation on the image of the cotter pin and a corresponding reconstructed image of the cotter pin, so that the accuracy of the abnormal condition detection of the cotter pin is improved, and the abnormal condition detection speed of the image of the cotter pin is also improved.
It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.
The following are apparatus embodiments of the present application, and for details not described in detail therein, reference may be made to the corresponding method embodiments described above.
Fig. 16 shows a schematic structural diagram of a device for detecting an abnormality of a cotter pin of a suspension device of a high-speed railway contact network provided in an embodiment of the present application, and for convenience of description, only the parts related to the embodiment of the present application are shown, and details are as follows:
as shown in fig. 16, the high-speed railway contact line suspension device cotter pin abnormality detection device 16 includes:
the acquiring module 1601 is used for acquiring an image of the catenary suspension device to be detected;
a connection area determining module 1602, configured to extract a connection area in the image of the catenary suspension device to obtain a connection area image;
a cotter determining module 1603, configured to position and cut the cotter component in the connection area image to obtain a cotter image;
the image reconstruction module 1604 is configured to perform image reconstruction on the cotter image based on an image reconstruction model to obtain a cotter reconstructed image, where the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm;
the detection module 1605 is configured to calculate a reconstruction error according to the cotter pin image and the cotter pin reconstructed image, and perform anomaly detection on the image of the catenary suspension device to be detected according to the reconstruction error.
The invention provides a system for detecting the abnormal cotter pin of a suspension device of a high-speed railway overhead line system, which is compared with the prior art, the system comprises a coarse positioning device for roughly positioning an image of the suspension device of the overhead line system to be detected to obtain an image of a connection area, and then, the image of the connection area is accurately positioned to obtain an image of the cotter pin, and whether the image of the cotter pin is an abnormal image or not is judged by carrying out reconstruction error calculation on the image of the cotter pin and a corresponding reconstructed image of the cotter pin, so that the accuracy of the abnormal cotter pin detection is improved, and the abnormal cotter pin image detection speed is also improved.
In a possible implementation manner, the apparatus may further include a preprocessing module, and the connection region determining module is specifically configured to:
labeling an image of the contact net suspension device to be detected based on a connection region labeling model, wherein the labeling comprises an image name, a connection region category and a connection region coordinate;
performing data enhancement on the marked image of the overhead line system suspension device to obtain a first data enhanced image;
and positioning the connection area in the first data enhanced image to obtain a connection area image.
In one possible implementation manner, the connection region determining module may be further configured to:
and performing data enhancement on the marked image of the contact net suspension device by adopting a GridMask data enhancement algorithm to obtain a first data enhanced image.
In one possible implementation, the connection region determining module may be further configured to:
based on the first target positioning model, performing target positioning on the connection area in the first-time data enhanced image, and cutting the connection area subjected to the target positioning to obtain a connection area image;
the first target positioning model is constructed based on a YOLOv5 algorithm.
In one possible implementation, the cotter determining module may be specifically configured to:
marking the cotter pin component in the connecting area image based on the cotter pin marking model;
performing data enhancement on the marked image of the connecting area to obtain a second data enhanced image;
based on the second target positioning model, performing target positioning on the cotter component in the second data enhanced image, and cutting the cotter component subjected to target positioning to obtain a cotter image;
and the second target positioning model is constructed based on a YOLOv5 algorithm.
In a possible implementation manner, the detection module may specifically be configured to:
calculating the average absolute error of the cotter image and the cotter reconstructed image through a first formula, calculating the structural similarity of the cotter image and the cotter reconstructed image through a second formula, and performing weighted summation on the average absolute error and the structural similarity to obtain a reconstructed error;
judging whether the reconstruction error is larger than a classification threshold value;
if the reconstruction error is larger than the classification threshold value, determining the cotter pin image as an abnormal image;
wherein the first formula is:
wherein the content of the first and second substances,in order to average the absolute error of the signal,as an image of the cotter pinThe value of each of the pixels is calculated,reconstruction of images for cotter pinsThe value of each of the pixels is calculated,the total number of pixel values for the aperture image and the cotter reconstructed image;
the second formula is:
wherein, the first and the second end of the pipe are connected with each other,in order to achieve the similarity of the structure,is the average of the pixel values of the cotter image,the mean of the pixel values of the reconstructed image for the cotter pin,is the variance of the pixel values of the cotter pin image,the variance of the pixel values of the image is reconstructed for the cotter pin,the covariance of the pixel values for the cotter pin image and the cotter pin reconstructed image,for the purpose of a first given constant, the value of,is a second given constant, wherein,。
fig. 17 is a schematic diagram of a terminal provided in an embodiment of the present application. As shown in fig. 17, the terminal 17 of this embodiment includes: a processor 1701, a memory 1702, and a computer program 1703 stored in the memory 1702 and executable on the processor 1701. The processor 1701 implements the steps in each of the above embodiments of the method for detecting an abnormality of a cotter pin of a suspension device of a catenary of a high-speed railway, such as S01 to S105 shown in fig. 1, when executing the computer program 1703. Alternatively, the processor 1701 implements the functions of the modules/units in the above-described device embodiments, such as the functions of the modules 1601 to 1605 shown in fig. 16, when executing the computer program 1703.
Illustratively, the computer program 1703 may be divided into one or more modules/units, which are stored in the memory 1702 and executed by the processor 1701 to accomplish the present application. The one or more modules/units may be a series of computer program instruction segments capable of performing certain functions, which are used to describe the execution of the computer program 1703 in the terminal 17. For example, the computer program 1703 may be partitioned into modules 1604 through 1605 shown in FIG. 16.
The terminal 17 may be a desktop computer, a notebook, a palm computer, a cloud server, or other computing devices. The terminal 17 may include, but is not limited to, a processor 1701 and a memory 1702. Those skilled in the art will appreciate that fig. 17 is only an example of a terminal 17 and does not constitute a limitation of terminal 17 and may include more or less components than those shown, or some components in combination, or different components, for example, the terminal may also include input output devices, network access devices, buses, etc.
The Processor 1701 may be a Central Processing Unit (CPU), or other general purpose Processor, digital Signal Processor (DSP), application Specific Integrated Circuit (ASIC), field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory 1702 may be an internal storage unit of the terminal 17, such as a hard disk or a memory of the terminal 17. The memory 1702 may also be an external storage device of the terminal 17, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are equipped on the terminal 17. Further, the memory 1702 may also include both an internal storage unit of the terminal 17 and an external storage device. The memory 1702 is used for storing the computer programs and other programs and data required by the terminal. The memory 1702 may also be used to temporarily store data that has been output or is to be output.
It should be clear to those skilled in the art that, for convenience and simplicity of description, the foregoing division of the functional units and modules is only used for illustration, and in practical applications, the above function distribution may be performed by different functional units and modules as needed, that is, the internal structure of the apparatus may be divided into different functional units or modules to perform all or part of the above described functions. Each functional unit and module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated unit may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional units and modules are only used for distinguishing one functional unit from another, and are not used for limiting the protection scope of the present application. For the specific working processes of the units and modules in the system, reference may be made to the corresponding processes in the foregoing method embodiments, which are not described herein again.
In the above embodiments, the description of each embodiment has its own emphasis, and reference may be made to the related description of other embodiments for parts that are not described or recited in any embodiment.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus/terminal and method may be implemented in other ways. For example, the above-described apparatus/terminal embodiments are merely illustrative, and for example, the division of the modules or units is only one logical division, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated module/unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, all or part of the flow in the method of the embodiments described above can be realized by the present application, and the computer program can be stored in a computer readable storage medium, and when the computer program is executed by a processor, the steps of the embodiments of the method for detecting an abnormality of a cotter pin of a suspension device of a catenary of a high speed railway can be realized. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, read-Only Memory (ROM), random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, and the like. It should be noted that the computer readable medium may contain other components which may be suitably increased or decreased as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable media which may not include electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the embodiments of the present application, and they should be construed as being included in the present application.
Claims (10)
1. The method for detecting the abnormity of the cotter pin of the suspension device of the high-speed railway contact network is characterized by comprising the following steps of:
acquiring an image of a contact net suspension device to be detected;
extracting a connection area in the contact net suspension device image to obtain a connection area image;
positioning and cutting the cotter pin component in the connecting area image to obtain a cotter pin image;
performing image reconstruction on the cotter image based on an image reconstruction model to obtain a cotter reconstructed image, wherein the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm;
and calculating a reconstruction error according to the cotter image and the cotter reconstruction image, and carrying out anomaly detection on the contact net suspension device image to be detected through the reconstruction error.
2. The method for detecting the abnormality of the cotter pin of the suspension device of the overhead line system of the high-speed railway according to claim 1, wherein the step of extracting the connecting area in the image of the suspension device of the overhead line system to obtain the image of the connecting area comprises the following steps:
marking the contact net suspension device image to be detected based on a connection area marking model, wherein the marking comprises an image name, a connection area type and a connection area coordinate;
performing data enhancement on the marked image of the overhead line system suspension device to obtain a first data enhanced image;
and positioning a connection region in the first data enhanced image to obtain the connection region image.
3. The method for detecting the abnormality of the cotter pin of the suspension device of the overhead line system of the high-speed railway according to claim 2, wherein the step of performing data enhancement on the marked image of the suspension device of the overhead line system to obtain a first data enhanced image comprises the following steps:
and performing data enhancement on the marked image of the contact net suspension device by adopting a GridMask data enhancement algorithm to obtain a first data enhanced image.
4. The method for detecting the abnormality of the cotter pin of the suspension device of the high-speed railway contact line system according to claim 2, wherein the positioning the connecting area in the first data enhanced image to obtain the connecting area image comprises the following steps:
based on a first target positioning model, performing target positioning on a connection region in the first-time data enhanced image, and cutting the connection region subjected to target positioning to obtain a connection region image;
the first target positioning model is constructed based on a YOLOv5 algorithm.
5. The method for detecting the cotter pin abnormality of the suspension device of the high-speed railway catenary according to claim 1, wherein the positioning and cutting the cotter pin component in the connection area image to obtain the cotter pin image comprises the following steps:
marking the cotter pin component in the connecting area image based on the cotter pin marking model;
performing data enhancement on the marked image of the connecting area to obtain a second data enhanced image;
based on a second target positioning model, performing target positioning on the cotter component in the second data enhanced image, and cutting the cotter component subjected to target positioning to obtain a cotter image;
and the second target positioning model is constructed based on a YOLOv5 algorithm.
6. The method for detecting the abnormity of the cotter pin of the suspension device of the high-speed railway contact network according to claim 1, wherein the method for calculating the reconstruction error according to the cotter pin image and the cotter pin reconstruction image and carrying out abnormity detection on the image of the suspension device of the contact network to be detected through the reconstruction error comprises the following steps:
calculating the average absolute error of the cotter image and the cotter reconstructed image through a first formula, calculating the structural similarity of the cotter image and the cotter reconstructed image through a second formula, and performing weighted summation on the average absolute error and the structural similarity to obtain a reconstructed error;
judging whether the reconstruction error is larger than a classification threshold value or not;
if the reconstruction error is larger than the classification threshold, determining that the cotter pin image is an abnormal image;
wherein the first formula is:
wherein, the first and the second end of the pipe are connected with each other,in order to be the average absolute error,is the first of the cotter pin imageThe value of each of the pixels is calculated,reconstructing an image for the cotter pinThe value of each of the pixels is calculated,a total number of pixel values for the aperture image and the cotter pin reconstructed image;
the second formula is:
wherein the content of the first and second substances,in order to be a measure of the structural similarity,is the mean of the pixel values of the cotter image,reconstructing a mean value of pixel values of an image for the cotter pin,as pixel values of the cotter pin imageThe variance of (a) is determined,reconstructing a variance of pixel values of an image for the cotter pin,covariance of pixel values for the cotter pin image and the cotter pin reconstructed image,for the purpose of a first given constant, the value of,is a second given constant, wherein,。
7. the utility model provides an unusual detecting system of high-speed railway contact net linkage suspension device cotter pin which characterized in that includes:
the acquisition module is used for acquiring an image of the contact net suspension device to be detected;
the connecting area determining module is used for extracting a connecting area in the contact net suspension device image to obtain a connecting area image;
the cotter determining module is used for positioning and cutting the cotter area in the connection area image to obtain a cotter image;
the image reconstruction module is used for carrying out image reconstruction on the cotter pin image based on an image reconstruction model to obtain a cotter pin reconstruction image, and the image reconstruction model is constructed based on an AutoEncoder algorithm and a StyleGAN algorithm;
and the detection module is used for calculating a reconstruction error according to the cotter image and the cotter reconstruction image and carrying out abnormity detection on the contact net suspension device image to be detected through the reconstruction error.
8. The system for detecting the abnormality of the cotter pin of the suspension device of the high-speed railway contact network according to claim 7, wherein the detection module is used for:
calculating the average absolute error of the cotter pin image and the cotter pin reconstructed image through a first formula, calculating the structural similarity of the cotter pin image and the cotter pin reconstructed image through a second formula, and performing weighted summation on the average absolute error and the structural similarity to obtain a reconstructed error;
judging whether the reconstruction error is larger than a classification threshold value;
if the reconstruction error is larger than the classification threshold, determining that the cotter pin image is an abnormal image;
wherein the first formula is:
wherein, the first and the second end of the pipe are connected with each other,in order to be the average absolute error,is the first of the cotter pin imageThe value of each of the pixels is calculated,reconstructing an image for the cotter pinThe value of each of the pixels is calculated,is the opening imageAnd the total number of pixel values of the cotter pin reconstructed image;
the second formula is:
wherein, the first and the second end of the pipe are connected with each other,in order to be the degree of similarity of the structure,is the mean of the pixel values of the cotter pin image,reconstructing a mean value of pixel values of an image for the cotter pin,is the variance of the pixel values of the cotter image,reconstructing a variance of pixel values of an image for the cotter pin,covariance of pixel values for the cotter pin image and the cotter pin reconstructed image,for the purpose of a first given constant, the value of,is a second given constant, wherein,。
9. a terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor when executing the computer program implements the steps of the method for detecting an abnormality of a cotter pin of a suspension device of a high speed railway catenary as claimed in any one of claims 1 to 6.
10. A computer-readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the method for detecting an abnormality of a cotter pin of a suspension device of a high-speed railway catenary according to any of claims 1 to 6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211219348.0A CN115311261A (en) | 2022-10-08 | 2022-10-08 | Method and system for detecting abnormality of cotter pin of suspension device of high-speed railway contact network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211219348.0A CN115311261A (en) | 2022-10-08 | 2022-10-08 | Method and system for detecting abnormality of cotter pin of suspension device of high-speed railway contact network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115311261A true CN115311261A (en) | 2022-11-08 |
Family
ID=83867819
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211219348.0A Pending CN115311261A (en) | 2022-10-08 | 2022-10-08 | Method and system for detecting abnormality of cotter pin of suspension device of high-speed railway contact network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115311261A (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107154033A (en) * | 2016-03-03 | 2017-09-12 | 成都交大光芒科技股份有限公司 | A kind of high ferro contact net rotation ears vertical openings pin missing detection method and system |
CN111563896A (en) * | 2020-07-20 | 2020-08-21 | 成都中轨轨道设备有限公司 | Image processing method for catenary anomaly detection |
CN111783658A (en) * | 2020-07-01 | 2020-10-16 | 河北工业大学 | Two-stage expression animation generation method based on double generation countermeasure network |
CN113553979A (en) * | 2021-07-30 | 2021-10-26 | 国电汉川发电有限公司 | Safety clothing detection method and system based on improved YOLO V5 |
CN114494479A (en) * | 2020-10-23 | 2022-05-13 | 美国西门子医疗***股份有限公司 | System and method for simultaneous attenuation correction, scatter correction, and denoising of low dose PET images using neural networks |
CN115131747A (en) * | 2022-08-25 | 2022-09-30 | 合肥中科类脑智能技术有限公司 | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system |
-
2022
- 2022-10-08 CN CN202211219348.0A patent/CN115311261A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107154033A (en) * | 2016-03-03 | 2017-09-12 | 成都交大光芒科技股份有限公司 | A kind of high ferro contact net rotation ears vertical openings pin missing detection method and system |
CN111783658A (en) * | 2020-07-01 | 2020-10-16 | 河北工业大学 | Two-stage expression animation generation method based on double generation countermeasure network |
CN111563896A (en) * | 2020-07-20 | 2020-08-21 | 成都中轨轨道设备有限公司 | Image processing method for catenary anomaly detection |
CN114494479A (en) * | 2020-10-23 | 2022-05-13 | 美国西门子医疗***股份有限公司 | System and method for simultaneous attenuation correction, scatter correction, and denoising of low dose PET images using neural networks |
CN113553979A (en) * | 2021-07-30 | 2021-10-26 | 国电汉川发电有限公司 | Safety clothing detection method and system based on improved YOLO V5 |
CN115131747A (en) * | 2022-08-25 | 2022-09-30 | 合肥中科类脑智能技术有限公司 | Knowledge distillation-based power transmission channel engineering vehicle target detection method and system |
Non-Patent Citations (1)
Title |
---|
蔚焘等: "基于记忆增强的对抗自编码器异常检测算法", 《南京邮电大学学报》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111160311B (en) | Yellow river ice semantic segmentation method based on multi-attention machine system double-flow fusion network | |
CN110738697B (en) | Monocular depth estimation method based on deep learning | |
CN108537742B (en) | Remote sensing image panchromatic sharpening method based on generation countermeasure network | |
CN107564025B (en) | Electric power equipment infrared image semantic segmentation method based on deep neural network | |
CN110728658A (en) | High-resolution remote sensing image weak target detection method based on deep learning | |
CN113887459B (en) | Open-pit mining area stope change area detection method based on improved Unet + | |
CN112017192B (en) | Glandular cell image segmentation method and glandular cell image segmentation system based on improved U-Net network | |
CN112862830B (en) | Multi-mode image segmentation method, system, terminal and readable storage medium | |
CN111044570A (en) | Defect identification and early warning method and device for power equipment and computer equipment | |
CN109977834B (en) | Method and device for segmenting human hand and interactive object from depth image | |
CN114694039A (en) | Remote sensing hyperspectral and laser radar image fusion classification method and device | |
CN110738663A (en) | Double-domain adaptive module pyramid network and unsupervised domain adaptive image segmentation method | |
CN115439654B (en) | Method and system for finely dividing weakly supervised farmland plots under dynamic constraint | |
CN113887472A (en) | Remote sensing image cloud detection method based on cascade color and texture feature attention | |
CN116665153A (en) | Road scene segmentation method based on improved deep bv3+ network model | |
CN114612315A (en) | High-resolution image missing region reconstruction method based on multi-task learning | |
CN116664839B (en) | Weak supervision semi-supervision defect perception segmentation method and device and electronic equipment | |
CN116129280B (en) | Method for detecting snow in remote sensing image | |
CN113393385A (en) | Unsupervised rain removal method, system, device and medium based on multi-scale fusion | |
CN112990213A (en) | Digital multimeter character recognition system and method based on deep learning | |
CN115456957B (en) | Method for detecting change of remote sensing image by full-scale feature aggregation | |
CN115311261A (en) | Method and system for detecting abnormality of cotter pin of suspension device of high-speed railway contact network | |
CN111047525A (en) | Method for translating SAR remote sensing image into optical remote sensing image | |
CN115909077A (en) | Hyperspectral image change detection method based on unsupervised spectrum unmixing neural network | |
CN115131244A (en) | Single image rain removing method and system based on counterstudy |
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 |
Application publication date: 20221108 |
|
RJ01 | Rejection of invention patent application after publication |