CN107437245A - High-speed railway touching net method for diagnosing faults based on depth convolutional neural networks - Google Patents
High-speed railway touching net method for diagnosing faults based on depth convolutional neural networks Download PDFInfo
- Publication number
- CN107437245A CN107437245A CN201710493102.5A CN201710493102A CN107437245A CN 107437245 A CN107437245 A CN 107437245A CN 201710493102 A CN201710493102 A CN 201710493102A CN 107437245 A CN107437245 A CN 107437245A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- mfrac
- equipotential line
- pixel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06T7/0008—Industrial image inspection checking presence/absence
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based 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]
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the high-speed railway touching net method for diagnosing faults based on depth convolutional neural networks, comprise the following steps:Obtain the 2-D gray image of high-speed railway touching net support meanss;By contact net training set pre-training depth convolutional neural networks, and bring into faster RCNN and train, by the equipotential line in the model extraction 2-D gray image that trains and split, obtain equipotential line region picture;The equipotential line region picture of acquisition is handled as follows successively, its brightness and contrast, recurrence Otsu pre-segmentations are adjusted, is split with ICM/MPM, corrodes expansion picture, equipotential line pixel is obtained, largest connected domain is extracted and counts the independent communication domain number N in equipotential pixel region;It is judged as that broken lot failure occur in equipotential line parts if N > m;The present invention can be accurately positioned equipotential line, improve fault diagnosis accuracy rate, meet needs of production.
Description
Technical field
The present invention relates to high-speed railway touching net field of fault detection, and in particular to one kind is based on depth convolutional neural networks
High-speed railway touching net method for diagnosing faults.
Background technology
For high-speed railway as important basic vehicles facility, its fast development proposes higher want to safety problem
Ask;Equipotential line is one of parts of contact net support meanss, and it, which is acted on, is ensured between locator bearing and locator
Equipotential link;On high-speed railway, the positive and negative between locator bearing and locator is respectively mounted equipotential line, it is seen that its
Importance;Application of the non-contact detection technology on railway based on image procossing is concentrated mainly on the survey of contact net geometric parameter
Amount and the detection of bow net defective mode, detection such as to locator gradient, lead the measurement higher than stagger, Contact-line Wind deviator is examined
Survey, pantograph pan crack detection etc.;Detected for the component failure of contact net support meanss, use is traditional characteristic
Extracting method is to contact net position components;Due to equipotential line, to belong to non-rigid, form more, utilizes existing HOG features
Or SIGT features, gratifying effect can not be obtained using traditional template matches mode.
The content of the invention
The present invention provides a kind of high-speed railway touching net based on depth convolutional neural networks having compared with high measurement accuracy
The broken lot method for diagnosing faults of equipotential line.
The technical solution adopted by the present invention is:High-speed railway touching net fault diagnosis side based on depth convolutional neural networks
Method, comprise the following steps:
Obtain the 2-D gray image of high-speed railway touching net support meanss;
By contact net training set pre-training depth convolutional neural networks, and bring target detection framework faster RCNN into
Middle training, by the equipotential line in the model extraction 2-D gray image that trains and split, obtain equipotential line area
Domain picture;
The equipotential line region picture of acquisition is handled as follows successively, adjusts its brightness and contrast, recurrence maximum
The pre-segmentation of Ostu method Otsu methods, with condition iterative model-maximal margin posterior probability algorithm (Iteration
Condition model/maximization of the posterior marginal, ICM/MPM) algorithm segmentation, corrosion
Picture is expanded, equipotential line pixel is obtained, extracts largest connected domain and count the independent communication in equipotential pixel region
Domain number N;
It is judged as that broken lot failure occur in equipotential line parts if N > m.
Further, obtained by the size of the standard deviation for the pixel value for normalizing the equipotential line after formula if N≤m
Go out the possibility of incipient fault;
Method for normalizing is as follows:
Standard deviation sigma computational methods are as follows:
In formula:wiFor i-th of position pixel value of equipotential line;wminFor the minimum value in equipotential pixel value;wmaxFor
Maximum in equipotential pixel value;It is equipotential line after normalization as average;viPut for equipotential line i-th bit after normalization
Pixel value;N is equipotential line pixel number, and σ is poor for equipotential pixel criterion after normalization.
Further, the method for the maximum variance between clusters Otsu method pre-segmentations comprises the following steps:
Obtain equipotential line region picture;
Calculate the grey level histogram in picture;
Calculate the probability that each pixel value occurs;
Travel through each pixel and calculate inter-class variance;
Corresponding pixel value when obtaining inter-class variance maximum.
Further, the ICM/MPM algorithms comprise the following steps:
According to Bayesian formula:
In formula:θ is model parameter matrix, and y and x are respectively the sample data of observation field and label field;P (x | y, θ) it is to see
Survey conditional probability of the field to label field;P (y) is the prior probability of observation field, is a constant;According to MPM criterions, by image point
The problem of cutting is converted into optimization problem;
U (x)=∑c∈CVc(xc) (5)
In formula:For the target equation of optimization, Z is normaliztion constant, and C closes for all gesture agglomerations, and U (x) closes for gesture agglomeration
Sum of interior all gesture group potential energy, T is constant, generally takes 1, and its control P (x) shape, T is bigger, and P (x) shape more levels off to flat
It is slow.Vc(xc) for gesture roll into a ball potential energy, S be picture S location, ysFor the pixel value of S location observation pictures,To possess label xs
All pixels point average,To possess label xsAll pixels point variance;
It is iterated using ICM algorithms, detailed process is as follows:
In formula:In formula:μk(p) the pixel average of kth class, σ are belonged to during iteration secondary for observation field pthk(p) it is observation field
Belong to the pixel variance of kth class, N during pth time iterationk(p) of kth class pixel is belonged to during iteration secondary for observation field pth
Number, k are the classification number of image segmentation, and p is pth time iteration.
Further, the depth convolutional neural networks include six convolutional layers, two pond layers and two full articulamentums,
A down-sampling pond layer is connect behind the first two convolutional layer respectively, second pond layer is followed by 4 convolutional layers, this four convolutional layers
It is sequentially connected, 2 full articulamentums, the vector of last full articulamentum output 1000 × 1 is followed by the 6th convolutional layer.
The beneficial effects of the invention are as follows:
(1) present invention learns to extract feature automatically using deep neural network, can more accurately position equipotential line;
(2) present invention carries out the segmentation result for using ICM/MPM to optimize Otsu after pre-segmentation, picture using Otsu to picture
Segmentation result is more accurate, improves the accuracy rate of fault diagnosis;
(3) present invention can provide the possibility of incipient fault when cannot recognize that failure, meet the needs of actual production.
Brief description of the drawings
Fig. 1 is schematic flow sheet of the present invention.
Fig. 2 is faster RCNN Organization Charts in the present invention.
Fig. 3 is CATENARNET convolutional neural networks Organization Charts in the present invention.
Fig. 4 is ICM/MPM algorithm flow charts in the present invention.
Fig. 5 is Otsu method flow diagrams of the present invention.
Fig. 6 is medium potential line segmentation effect figure of the present invention.
Embodiment
The present invention will be further described with specific embodiment below in conjunction with the accompanying drawings.
As Figure 1-5, a kind of high-speed railway touching net method for diagnosing faults based on depth convolutional neural networks, including
Following steps:
Obtain the 2-D gray image of high-speed railway touching net support meanss;
It is a kind of pin by contact net training set pre-training depth convolutional neural networks CATENARYNET, CATENARYNET
To the convolutional Neural neutral net framework of contact net equipotential contour piece design, the architecture design is by six convolutional layers, two ponds
Change layer and two full articulamentums form, connect a down-sampling pond layer behind the first two convolutional layer respectively, after second pond layer
4 convolutional layers are connect, this four convolutional layers are sequentially connected, and 2 full articulamentums, last full articulamentum are followed by the 6th convolutional layer
The vector of output 1000 × 1;CATENARYNET is brought into faster RCNN and trained, passes through the model extraction two dimension trained
Equipotential line in gray level image is simultaneously split, and obtains equipotential line region picture;
The equipotential line region picture of acquisition is handled as follows successively, adjusts its brightness and contrast, recurrence Otsu
Method pre-segmentation, parameter is initialized, split with ICM/MPM algorithms, corrosion expansion picture, obtains equipotential line pixel
Point, extract largest connected domain and count the independent communication domain number N in equipotential pixel region;
It is judged as that broken lot failure occur in equipotential line parts if N > m.
Further, obtained by the size of the standard deviation for the pixel value for normalizing the equipotential line after formula if N≤m
Go out the possibility of incipient fault;Due to whether either with or without broken lot neither one determine standard;This method thinks only to scatter
Failure is identified as, remaining tendency that has a case that to scatter is considered to have the possibility of potential broken lot failure;This possibility
Property is provided by the standard deviation of the pixel value of normalized equipotential line;When equipotential line scatters, this method is by detecting it
The number in the independent communication domain in region judges whether broken lot;When broken lot occurs due to equipotential line, the ground of everywhere broken lot
Side can become an independent communication domain after treatment, and due in practical application, equipotential line will not be only when occurring broken lot
There is generation broken lot at one;So sentencing as equipotential line broken lot when the number m in independent communication domain typically being more than into 3 in this method
According to;
Method for normalizing is as follows:
Standard deviation sigma computational methods are as follows:
In formula:wiFor i-th of position pixel value of equipotential line;wminFor the minimum value in equipotential pixel value;wmaxFor
Maximum in equipotential pixel value;It is equipotential line after normalization as average;viFor the equipotential line i-th bit after normalization
The pixel value put;N is equipotential line pixel number, and σ is pixel standard deviation after normalization.
Further, the method for the Otsu methods pre-segmentation comprises the following steps:
Obtain equipotential line region picture;
Calculate the grey level histogram in picture;
Calculate the probability that each pixel value occurs;
Travel through each pixel and calculate inter-class variance;
Corresponding pixel value when obtaining inter-class variance maximum.
Further, the ICM/MPM algorithms comprise the following steps:ICM/MPM algorithms include two parts, and Part I makes
Equipotential line segmentation problem is converted into optimization problem with MPM criterions;Part II is to model to be optimized using ICM algorithms
Solve;
According to Bayesian formula:
In formula:θ is model parameter matrix, and y and x are respectively the sample data of observation field and label field;P (x | y, θ) it is to see
Survey conditional probability of the field to label field;P (y) is the prior probability of observation field, is a constant;Image is divided according to MPM criterions
The problem of cutting is converted into optimization problem;
U (x)=∑c∈CVc(xc) (5)
In formula:For target equation to be optimized, Z is normaliztion constant, and C closes for all gesture agglomerations, and U (x) is gesture agglomeration
Sum of all gesture group potential energy in conjunction, T are that constant generally takes 1, and control P (x) shape, T is bigger, and P (x) shape more levels off to flat
It is slow, Vc(xc) for gesture roll into a ball potential energy, S be picture S location, ysFor the pixel value of S location observation pictures,To possess label xs
All pixels point average,To possess label xsAll pixels point variance;
It is iterated using ICM algorithms, detailed process is as follows:
In formula:μk(p) the pixel average of kth class, σ are belonged to during iteration secondary for observation field pthk(p) it is observation field pth time
Belong to the pixel variance of kth class, N during iterationk(p) to belong to the number of kth class pixel during observation field pth time iteration, k is
The classification number of image segmentation, p are pth time iteration.
Further, the depth convolutional neural networks include six convolutional layers, two pond layers and two full articulamentums,
A down-sampling pond layer is connect behind the first two convolutional layer respectively, second pond layer is followed by 4 convolutional layers, this four convolutional layers
It is sequentially connected, 2 full articulamentums, the vector of last full articulamentum output 1000 × 1 is followed by the 6th convolutional layer;Each volume
Lamination uses the linear activation primitives of Relu.
In use, specific works step is as follows:
(1) by the overhead contact line support meanss condition detecting system on inspection car, contact net is supported and filled
The positive and overall and local two dimensional gray picture for being continuously shot, gathering contact net support meanss in real time put;From obtaining
The picture box obtained selects the picture of contact net equipotential line component area, and training dataset and survey are made according to VOC2007 standards
Try data set;At the same time, common elements of contacting net is cut out and is used for pre-training to make contact net training set
CATENARYNET;
(2) in order to improve the precision of positioning, the present invention proposes a kind of neural network structure of new equipotential line identification
CATENARYNET, it is more suitable for the feature extraction of contact net picture;CATENARYNET is by six convolutional layers and two full articulamentums
Composition;
(3) the pre-training CATENARYNET on contact net data set, and be brought into faster RCNN and train, pass through instruction
Position and segmentation of the model extraction contact net equipotential line practised in two dimensional gray picture;To the equipotential line area being partitioned into
Domain picture adjustment brightness and contrast, based on Ostu algorithm segmentation results, is tentatively divided picture using ICM/MPM algorithms
Cut, by the Morphological scale-space (corrosion expansive working is used in the present invention) of digital picture, be partitioned into accurate equipotential line picture
Vegetarian refreshments simultaneously counts the independent communication domain number in equipotential line pixel region;
ICM/MPM algorithms are as follows:
In formula:θ is model parameter matrix, and y and x are respectively the sample data of observation field and label field;P (x | y, θ) it is to see
Survey conditional probability of the field to label field;P (y) is the prior probability of observation field;
U (x)=∑c∈CVc(xc) (5)
In formula:For the target equation of optimization, Z is normaliztion constant, and C closes for all gesture agglomerations, and U (x) closes for gesture agglomeration
The sum of interior all gesture groups potential energy, T control P (x) shapes, T is bigger, and P (x) shape more levels off to gently.Vc(xc) it is that gesture rolls into a ball gesture
Can, S be picture S location, ysFor the pixel value of S location observation pictures,To possess label xsAll pixels point it is equal
Value,To possess label xsAll pixels point variance.
It is iterated using ICM algorithms, detailed process is as follows:
In formula:μk(p) the pixel average of kth class, σ are belonged to during iteration secondary for observation field pthk(p) it is observation field pth time
Belong to the pixel variance of kth class, N during iterationk(p) to belong to the number of kth class pixel during observation field pth time iteration, k is
The classification number of image segmentation, p are pth time iteration.
(4) according to independent communication in the statistics rule of equipotential line pixel grey scale pixel value and equipotential line region
Domain number, provide the incipient fault possibility and malfunction of equipotential line parts broken lot.
Embodiment
Using including high definition industrial camera, the large-scale array of source of integration, triggering control function module and high performance service
The overhead contact line support meanss state-detection supervising device of the equipment such as device composition;High definition industrial camera and the large-scale light source of integration
Array is arranged on roof, when inspection car is travelled on the line with certain speed, the equipment interconnection net-fault support meanss
Positive and negative and overall and part are shot, and simultaneously by the corresponding preservation of the positional information of picture;This method mainly includes three
The individual stage:First stage is equipotential line positioning stage;Second stage is that equipotential line splits the stage;Three phases are
The Fault Identification stage;Input picture is obtained from overhead contact line support meanss state-detection monitoring device;Using what is trained
Faster RCNN models position to input picture and come out the picture segmentation of the rectangle frame of positioning;Picture after segmentation is adopted
The region of the pixel and equipotential line of equipotential line in picture is got with the segmentation of ICM/MPM algorithms;By detecting above-mentioned area
The standard deviation of the equipotential pixel after independent communication domain number and normalization in domain carries out breakdown judge and calculates incipient fault
Possibility;Table 1 is CAENARYNET detail parameters allocation lists:
The CAENARYNET detail parameters of table 1 are set
Because it is that single pass size is 6600 × 4400 pixels that contact net, which shoots the picture come, depth convolution is refreshing
The parameter of the convolutional layer of first layer through network structure is arranged to 1 × 660 × 440;In order to shorten the time of pre-training and not shadow
Nicety of grading is rung, the sample pre-training of 1/10th resolution ratio is used in training;In test still using 6600 ×
The picture of 4400 pixels, it can so improve measuring accuracy.
Fig. 2 is faster RCNN Organization Charts, in hands-on, by the convolution trained of feature extraction network in Fig. 3
Trained in the feature extraction network that layer parameter is brought into Fig. 2;Faster RCNN are mainly made up of two parts, and Part I is
Risk factor RPN, it act as body region candidate frame;Part II is fast RCNN;Fast RCNN are according to region candidate frame
Suggestion and a ROI feature vector is generated by ROI ponds layer according to the feature of feature extraction network and is respectively fed to
Softmax layers and Bbox return device.
By positioning equipotential line, the adjustment of contrast and brightness is carried out to the picture after cutting, then using recurrence
Otsu methods carry out pre-segmentation to the picture after adjustment, and segmentation result is corrected using ICM/MPM algorithms afterwards, then right
Result after correction is corroded and is expanded and extract final segmentation result of the largest connected domain as equipotential line, ICM/MPM
The core concept of algorithm is that Otsu misclassification result is corrected, so as to improve the accuracy of pixel classification.
Black hole in Fig. 6 in (j) and (k) is independent communication domain;Carry out needing to detect these during broken lot fault detect
The number in black hole;Using fault detection method proposed by the present invention, fault diagnosis is carried out to the image in Fig. 6 respectively and obtains table 2.
As can be seen from Table 2 using the method for the present invention, (a) (i) (j) (k) (l) (m) can be accurately judged to broken lot failure occurs,
(b) (c) (d) (e) (f) (g) (h) (n) (o) does not break down, from the point of view of the size of incipient fault possibility, (g) most has
Broken lot may occur, (b) has the minimum possibility that broken lot failure occurs, and this result meets the artificial observed result to picture.
The fault diagnosis result of table 2
The present invention is related to feature instead of traditional-handwork using depth convolutional neural networks and learns to extract feature come automatic, simultaneously
Positioned using neural net regression generating region candidate frame and replace traditional sliding window;Equipotential can more accurately be positioned
Line, while shorten positioning time;ICM/MPM optimization Otsu segmentation results are used after the segmentation result based on Otsu;Segmentation
As a result it is more accurate, the error rate of segmentation result is effectively reduced, greatly improves the accuracy rate of the fault diagnosis of system;Utilize
Unique characteristics identification failure when equipotential line breaks down, the possibility of incipient fault is provided when failure can not be identified, this
The design of sample more hommization, can shift to an earlier date the generation of trouble saving, taken into full account the hommization of system in practical application and needed
Will;Potential likelihood of failure is provided when equipotential line is not failure, more conforms to the needs of actual production;Depth is rolled up
Product neutral net CATENARYNET obtains preferable compromise on deep layer network and shallow-layer network, and parameter setting is more suitable for connecing
Net-fault picture;Shorten the training time has heightened classification accuracy again, is advantageous to the raising of positioning precision;The inventive method can
The feature of automatic study equipotential line, without using the feature of engineer, while shortens the positioning time of equipotential line, carries
High positioning precision, improves the accuracy rate of fault diagnosis and can provide the possibility of incipient fault, be more suitable for it is actual should
With.
Claims (5)
- A kind of 1. high-speed railway touching net method for diagnosing faults based on depth convolutional neural networks, it is characterised in that including with Lower step:Obtain the 2-D gray image of high-speed railway touching net support meanss;By contact net training set pre-training depth convolutional neural networks, and bring into target detection framework faster RCNN and instruct Practice, by the equipotential line in the model extraction 2-D gray image that trains and split, obtain equipotential line administrative division map Piece;The equipotential line region picture of acquisition is handled as follows successively, adjusted between its brightness and contrast, recurrence maximum kind Variance method Otsu pre-segmentations, with the segmentation of condition iterative model-maximal margin posterior probability algorithm ICM/MPM, corrosion expansion picture, Equipotential line pixel is obtained, largest connected domain is extracted and counts the independent communication domain number N in equipotential pixel region;It is judged as that broken lot failure occur in equipotential line parts if N > m.
- A kind of 2. high-speed railway touching net fault diagnosis side based on depth convolutional neural networks according to claim 1 Method, it is characterised in that drawn by the size of the standard deviation for the pixel value for normalizing the equipotential line after formula if N≤m The possibility of incipient fault;Method for normalizing is as follows:<mrow> <msub> <mi>v</mi> <mi>i</mi> </msub> <mo>=</mo> <mfrac> <mrow> <msub> <mi>w</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>w</mi> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> </msub> </mrow> <mrow> <msub> <mi>w</mi> <mi>max</mi> </msub> <mo>-</mo> <msub> <mi>w</mi> <mrow> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> </msub> </mrow> </mfrac> </mrow><mrow> <mover> <mi>v</mi> <mo>&OverBar;</mo> </mover> <mo>=</mo> <mfrac> <mn>1</mn> <mi>N</mi> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msub> <mi>v</mi> <mi>i</mi> </msub> </mrow>Standard deviation sigma computational methods are as follows:<mrow> <mi>&sigma;</mi> <mo>=</mo> <msqrt> <mrow> <mfrac> <mn>1</mn> <mi>N</mi> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>v</mi> <mi>i</mi> </msub> <mo>-</mo> <mover> <mi>v</mi> <mo>&OverBar;</mo> </mover> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> </msqrt> </mrow>In formula:wiFor i-th of position pixel value of equipotential line;wminFor the minimum value in equipotential pixel value;wmaxFor equipotential Maximum in pixel value;It is equipotential line after normalization as average;viFor the pixel that equipotential line i-th bit is put after normalization Value;N is equipotential line pixel number, and σ is poor for equipotential pixel criterion after normalization.
- A kind of 3. high-speed railway touching net fault diagnosis side based on depth convolutional neural networks according to claim 1 Method, it is characterised in that the method for the maximum variance between clusters Otsu pre-segmentations comprises the following steps:Obtain equipotential line region picture;Calculate the grey level histogram in picture;Calculate the probability that each pixel value occurs;Travel through each pixel and calculate inter-class variance;Corresponding pixel value when obtaining inter-class variance maximum.
- A kind of 4. high-speed railway touching net fault diagnosis side based on depth convolutional neural networks according to claim 1 Method, it is characterised in that the ICM/MPM algorithms comprise the following steps:<mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>|</mo> <mi>y</mi> <mo>,</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>|</mo> <mi>x</mi> <mo>,</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> </mrow> <mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>In formula:θ is model parameter matrix, and y and x are respectively the sample data of observation field and label field;P (x | y, θ) it is observation field To the conditional probability of label field;P (y) is the prior probability of observation field;<mrow> <mover> <mi>X</mi> <mo>~</mo> </mover> <mo>=</mo> <mi>arg</mi> <mi> </mi> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <mo>-</mo> <mi>l</mi> <mi>n</mi> <mo>(</mo> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mi>P</mi> <mrow> <mo>(</mo> <mrow> <mi>y</mi> <mo>|</mo> <mi>x</mi> <mo>,</mo> <mi>&theta;</mi> </mrow> <mo>)</mo> </mrow> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> 1<mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>Z</mi> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mn>1</mn> <mi>T</mi> </mfrac> <mi>U</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow><mrow> <mi>Z</mi> <mo>=</mo> <msub> <mi>&Sigma;</mi> <mi>x</mi> </msub> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mfrac> <mrow> <mi>U</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow> <mi>T</mi> </mfrac> </mrow> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>U (x)=∑c∈CVc(xc) (5)<mrow> <msub> <mi>V</mi> <mi>c</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>c</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>V</mi> <mrow> <mi>s</mi> <mo>,</mo> <mi>r</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>s</mi> </msub> <mo>,</mo> <msub> <mi>x</mi> <mi>r</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <mrow> <msub> <mi>x</mi> <mi>s</mi> </msub> <mo>=</mo> <msub> <mi>x</mi> <mi>r</mi> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mn>1</mn> </mtd> <mtd> <mrow> <msub> <mi>x</mi> <mi>s</mi> </msub> <mo>&NotEqual;</mo> <msub> <mi>x</mi> <mi>r</mi> </msub> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> </mrow><mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>|</mo> <mi>x</mi> <mo>,</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&Pi;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mfrac> <mn>1</mn> <msqrt> <mrow> <mn>2</mn> <msub> <mi>&pi;&sigma;</mi> <msub> <mi>x</mi> <mi>s</mi> </msub> </msub> </mrow> </msqrt> </mfrac> <mi>exp</mi> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <msup> <mrow> <mo>(</mo> <msub> <mi>y</mi> <mi>s</mi> </msub> <mo>-</mo> <msub> <mi>&mu;</mi> <msub> <mi>x</mi> <mi>s</mi> </msub> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mrow> <mn>2</mn> <msubsup> <mi>&sigma;</mi> <msub> <mi>x</mi> <mi>s</mi> </msub> <mn>2</mn> </msubsup> </mrow> </mfrac> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> </mrow>In formula:For the target equation of optimization, Z is normaliztion constant, and C closes for all gesture agglomerations, and U (x) is to own in gesture agglomeration closes The sum of gesture group's potential energy, T are constant, Vc(xc) for gesture roll into a ball potential energy, S be picture S location, ysFor the pixel of S location observation pictures Value,To possess label xsAll pixels point average,To possess label xsAll pixels point variance;It is iterated using ICM algorithms, detailed process is as follows:<mrow> <msub> <mi>&mu;</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <msub> <mi>N</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msub> <mi>y</mi> <mi>s</mi> </msub> <mi>p</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>|</mo> <mi>y</mi> <mo>,</mo> <mi>&theta;</mi> <mo>(</mo> <mrow> <mi>p</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>8</mn> <mo>)</mo> </mrow> </mrow><mrow> <msub> <mi>&sigma;</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <msub> <mi>N</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msup> <mrow> <mo>(</mo> <msub> <mi>y</mi> <mi>s</mi> </msub> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>k</mi> </msub> <mo>(</mo> <mi>p</mi> <mo>)</mo> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mi>p</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>|</mo> <mi>y</mi> <mo>,</mo> <mi>&theta;</mi> <mo>(</mo> <mrow> <mi>p</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>9</mn> <mo>)</mo> </mrow> </mrow><mrow> <msub> <mi>N</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mi>p</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>|</mo> <mi>y</mi> <mo>,</mo> <mi>&theta;</mi> <mo>(</mo> <mrow> <mi>p</mi> <mo>-</mo> <mn>1</mn> </mrow> <mo>)</mo> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>10</mn> <mo>)</mo> </mrow> </mrow>In formula:μk(p) the pixel average of kth class, σ are belonged to during iteration secondary for observation field pthk(p) it is observation field pth time iteration When belong to the pixel variance of kth class, Nk(p) to belong to the number of kth class pixel during observation field pth time iteration, k is image The classification number of segmentation, p are pth time iteration.
- A kind of 5. high-speed railway touching net fault diagnosis side based on depth convolutional neural networks according to claim 1 Method, it is characterised in that the depth convolutional neural networks include six convolutional layers, two pond layers and two full articulamentums, preceding A down-sampling pond layer is connect behind two convolutional layers respectively, second pond layer is followed by 4 convolutional layers, this four convolutional layers according to It is secondary to be connected, it is followed by 2 full articulamentums, the vector of last full articulamentum output 1000 × 1 in the 6th convolutional layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710493102.5A CN107437245B (en) | 2017-06-26 | 2017-06-26 | High-speed railway contact net fault diagnosis method based on deep convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710493102.5A CN107437245B (en) | 2017-06-26 | 2017-06-26 | High-speed railway contact net fault diagnosis method based on deep convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107437245A true CN107437245A (en) | 2017-12-05 |
CN107437245B CN107437245B (en) | 2020-10-20 |
Family
ID=60458622
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710493102.5A Expired - Fee Related CN107437245B (en) | 2017-06-26 | 2017-06-26 | High-speed railway contact net fault diagnosis method based on deep convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107437245B (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108387814A (en) * | 2018-01-12 | 2018-08-10 | 广州市扬新技术研究有限责任公司 | The urban rail transit contact network fault locator and method of the bilateral switching power supply of size |
CN108573277A (en) * | 2018-03-12 | 2018-09-25 | 北京交通大学 | A kind of pantograph carbon slide surface disease automatic recognition system and method |
CN108647045A (en) * | 2018-03-20 | 2018-10-12 | 科大讯飞股份有限公司 | Method and device for realizing activation function, storage medium and electronic equipment |
CN108664997A (en) * | 2018-04-20 | 2018-10-16 | 西南交通大学 | High iron catenary equipotential line defective mode detection method based on cascade Faster R-CNN |
CN108828405A (en) * | 2018-06-06 | 2018-11-16 | 西南交通大学 | A kind of electric transmission line fault detection method based on random matrix |
CN109241982A (en) * | 2018-09-06 | 2019-01-18 | 广西师范大学 | Object detection method based on depth layer convolutional neural networks |
CN109360188A (en) * | 2018-09-13 | 2019-02-19 | 西南交通大学 | A kind of detection method of high iron catenary equipotential line defective mode |
CN109583573A (en) * | 2018-12-13 | 2019-04-05 | 银河水滴科技(北京)有限公司 | A kind of part missing detection method and device of rail clip |
CN109636789A (en) * | 2018-12-13 | 2019-04-16 | 银河水滴科技(北京)有限公司 | The part of rail clip based on image segmentation lacks detection method and device |
CN109753962A (en) * | 2019-01-13 | 2019-05-14 | 南京邮电大学盐城大数据研究院有限公司 | Text filed processing method in natural scene image based on hybrid network |
CN110008971A (en) * | 2018-08-23 | 2019-07-12 | 腾讯科技(深圳)有限公司 | Image processing method, device, computer readable storage medium and computer equipment |
CN110084852A (en) * | 2019-04-22 | 2019-08-02 | 西南交通大学 | A kind of accurate positioning method of high iron catenary support device sleeve puller bolt |
CN110135273A (en) * | 2019-04-19 | 2019-08-16 | 中铁第一勘察设计院集团有限公司 | Contact net video image cloud intellectual monitoring and fault recognition method |
CN110210427A (en) * | 2019-06-06 | 2019-09-06 | 中国民航科学技术研究院 | A kind of shelter bridge working condition detection system and method based on image processing techniques |
CN110428400A (en) * | 2019-07-09 | 2019-11-08 | 北京百分点信息科技有限公司 | A kind of localization method of equipment fault, device, equipment and storage medium |
CN110503048A (en) * | 2019-08-26 | 2019-11-26 | 中铁电气化局集团有限公司 | The identifying system and method for rigid contact net suspension arrangement |
CN111311591A (en) * | 2020-03-07 | 2020-06-19 | 西南交通大学 | Method for detecting lifting amount of high-speed railway contact net |
CN111784656A (en) * | 2020-06-28 | 2020-10-16 | 京东数字科技控股有限公司 | Railway contact network fault detection method and device, electronic equipment and storage medium |
CN112508911A (en) * | 2020-12-03 | 2021-03-16 | 合肥科大智能机器人技术有限公司 | Rail joint touch net suspension support component crack detection system based on inspection robot and detection method thereof |
CN113658112A (en) * | 2021-07-27 | 2021-11-16 | 北京交通大学 | Bow net anomaly detection method based on template matching and neural network algorithm |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101247030A (en) * | 2007-08-01 | 2008-08-20 | 北京深浪电子技术有限公司 | Overhead network obstacle detouring inspection robot and its obstacle detouring control method |
CN101577003A (en) * | 2009-06-05 | 2009-11-11 | 北京航空航天大学 | Image segmenting method based on improvement of intersecting visual cortical model |
CN101699511A (en) * | 2009-10-30 | 2010-04-28 | 深圳创维数字技术股份有限公司 | Color image segmentation method and system |
CN101847259A (en) * | 2010-01-21 | 2010-09-29 | 西北工业大学 | Infrared object segmentation method based on weighted information entropy and markov random field |
CN102968637A (en) * | 2012-12-20 | 2013-03-13 | 山东科技大学 | Complicated background image and character division method |
CN104036491A (en) * | 2014-05-14 | 2014-09-10 | 西安电子科技大学 | SAR image segmentation method based on area division and self-adaptive polynomial implicit model |
CN105741291A (en) * | 2016-01-30 | 2016-07-06 | 西南交通大学 | Method for detecting faults of equipotential lines of high-speed railway overhead line system suspension devices |
CN105957073A (en) * | 2015-04-29 | 2016-09-21 | 国网河南省电力公司电力科学研究院 | Fault detection method for scattered strands in power transmission line |
CN106683099A (en) * | 2016-11-17 | 2017-05-17 | 南京邮电大学 | Product surface defect detection method |
-
2017
- 2017-06-26 CN CN201710493102.5A patent/CN107437245B/en not_active Expired - Fee Related
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101247030A (en) * | 2007-08-01 | 2008-08-20 | 北京深浪电子技术有限公司 | Overhead network obstacle detouring inspection robot and its obstacle detouring control method |
CN101577003A (en) * | 2009-06-05 | 2009-11-11 | 北京航空航天大学 | Image segmenting method based on improvement of intersecting visual cortical model |
CN101699511A (en) * | 2009-10-30 | 2010-04-28 | 深圳创维数字技术股份有限公司 | Color image segmentation method and system |
CN101847259A (en) * | 2010-01-21 | 2010-09-29 | 西北工业大学 | Infrared object segmentation method based on weighted information entropy and markov random field |
CN102968637A (en) * | 2012-12-20 | 2013-03-13 | 山东科技大学 | Complicated background image and character division method |
CN104036491A (en) * | 2014-05-14 | 2014-09-10 | 西安电子科技大学 | SAR image segmentation method based on area division and self-adaptive polynomial implicit model |
CN105957073A (en) * | 2015-04-29 | 2016-09-21 | 国网河南省电力公司电力科学研究院 | Fault detection method for scattered strands in power transmission line |
CN105741291A (en) * | 2016-01-30 | 2016-07-06 | 西南交通大学 | Method for detecting faults of equipotential lines of high-speed railway overhead line system suspension devices |
CN106683099A (en) * | 2016-11-17 | 2017-05-17 | 南京邮电大学 | Product surface defect detection method |
Non-Patent Citations (4)
Title |
---|
刘荣涛: "基于计算机视觉的刀具后刀面磨损检测技术", 《中国优秀硕士学位论文全文数据库》 * |
王忠强: "输电线路远程智能巡线***的设计与实现", 《万方数据》 * |
陈东杰等: "基于深度学习的高铁接触网***检测与识别", 《中国科学技术大学学报》 * |
黄帅: "基于Markov随机场和K均值聚类的小麦叶部病害图像分割", 《万方数据》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108387814A (en) * | 2018-01-12 | 2018-08-10 | 广州市扬新技术研究有限责任公司 | The urban rail transit contact network fault locator and method of the bilateral switching power supply of size |
CN108573277A (en) * | 2018-03-12 | 2018-09-25 | 北京交通大学 | A kind of pantograph carbon slide surface disease automatic recognition system and method |
CN108647045A (en) * | 2018-03-20 | 2018-10-12 | 科大讯飞股份有限公司 | Method and device for realizing activation function, storage medium and electronic equipment |
CN108647045B (en) * | 2018-03-20 | 2021-10-01 | 科大讯飞股份有限公司 | Method and device for realizing activation function, storage medium and electronic equipment |
CN108664997A (en) * | 2018-04-20 | 2018-10-16 | 西南交通大学 | High iron catenary equipotential line defective mode detection method based on cascade Faster R-CNN |
CN108828405A (en) * | 2018-06-06 | 2018-11-16 | 西南交通大学 | A kind of electric transmission line fault detection method based on random matrix |
CN108828405B (en) * | 2018-06-06 | 2020-05-12 | 西南交通大学 | Power transmission line fault detection method based on random matrix |
CN110008971A (en) * | 2018-08-23 | 2019-07-12 | 腾讯科技(深圳)有限公司 | Image processing method, device, computer readable storage medium and computer equipment |
US11604949B2 (en) | 2018-08-23 | 2023-03-14 | Tencent Technology (Shenzhen) Company Limited | Image processing method and apparatus, computer-readable storage medium, and computer device |
CN109241982A (en) * | 2018-09-06 | 2019-01-18 | 广西师范大学 | Object detection method based on depth layer convolutional neural networks |
CN109360188A (en) * | 2018-09-13 | 2019-02-19 | 西南交通大学 | A kind of detection method of high iron catenary equipotential line defective mode |
CN109636789A (en) * | 2018-12-13 | 2019-04-16 | 银河水滴科技(北京)有限公司 | The part of rail clip based on image segmentation lacks detection method and device |
CN109583573A (en) * | 2018-12-13 | 2019-04-05 | 银河水滴科技(北京)有限公司 | A kind of part missing detection method and device of rail clip |
CN109636789B (en) * | 2018-12-13 | 2021-07-30 | 银河水滴科技(北京)有限公司 | Image segmentation-based part missing detection method and device for track fastener |
CN109753962A (en) * | 2019-01-13 | 2019-05-14 | 南京邮电大学盐城大数据研究院有限公司 | Text filed processing method in natural scene image based on hybrid network |
CN110135273A (en) * | 2019-04-19 | 2019-08-16 | 中铁第一勘察设计院集团有限公司 | Contact net video image cloud intellectual monitoring and fault recognition method |
CN110084852A (en) * | 2019-04-22 | 2019-08-02 | 西南交通大学 | A kind of accurate positioning method of high iron catenary support device sleeve puller bolt |
CN110210427A (en) * | 2019-06-06 | 2019-09-06 | 中国民航科学技术研究院 | A kind of shelter bridge working condition detection system and method based on image processing techniques |
CN110428400A (en) * | 2019-07-09 | 2019-11-08 | 北京百分点信息科技有限公司 | A kind of localization method of equipment fault, device, equipment and storage medium |
CN110503048A (en) * | 2019-08-26 | 2019-11-26 | 中铁电气化局集团有限公司 | The identifying system and method for rigid contact net suspension arrangement |
CN111311591A (en) * | 2020-03-07 | 2020-06-19 | 西南交通大学 | Method for detecting lifting amount of high-speed railway contact net |
CN111784656A (en) * | 2020-06-28 | 2020-10-16 | 京东数字科技控股有限公司 | Railway contact network fault detection method and device, electronic equipment and storage medium |
CN112508911A (en) * | 2020-12-03 | 2021-03-16 | 合肥科大智能机器人技术有限公司 | Rail joint touch net suspension support component crack detection system based on inspection robot and detection method thereof |
CN113658112A (en) * | 2021-07-27 | 2021-11-16 | 北京交通大学 | Bow net anomaly detection method based on template matching and neural network algorithm |
CN113658112B (en) * | 2021-07-27 | 2024-03-01 | 北京交通大学 | Bow net anomaly detection method based on template matching and neural network algorithm |
Also Published As
Publication number | Publication date |
---|---|
CN107437245B (en) | 2020-10-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107437245A (en) | High-speed railway touching net method for diagnosing faults based on depth convolutional neural networks | |
CN108830188B (en) | Vehicle detection method based on deep learning | |
CN111079747B (en) | Railway wagon bogie side frame fracture fault image identification method | |
WO2023155069A1 (en) | Deep-learning-based surface defect detection method for mobile phone battery | |
CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
CN107316064B (en) | Asphalt pavement crack classification and identification method based on convolutional neural network | |
CN108492291B (en) | CNN segmentation-based solar photovoltaic silicon wafer defect detection system and method | |
CN113469953B (en) | Transmission line insulator defect detection method based on improved YOLOv4 algorithm | |
CN102902974B (en) | Image based method for identifying railway overhead-contact system bolt support identifying information | |
CN111091544B (en) | Method for detecting breakage fault of side integrated framework of railway wagon bogie | |
CN105044122A (en) | Copper part surface defect visual inspection system and inspection method based on semi-supervised learning model | |
CN102854191A (en) | Real-time visual detection and identification method for high speed rail surface defect | |
CN110490842B (en) | Strip steel surface defect detection method based on deep learning | |
CN109214308A (en) | A kind of traffic abnormity image identification method based on focal loss function | |
CN106198551A (en) | Method and device for detecting defects of power transmission line | |
CN104598885A (en) | Method for detecting and locating text sign in street view image | |
CN112419289B (en) | Intelligent detection method for urban subway rail fastener defects | |
CN112258490A (en) | Low-emissivity coating intelligent damage detection method based on optical and infrared image fusion | |
CN113643268A (en) | Industrial product defect quality inspection method and device based on deep learning and storage medium | |
CN112258459A (en) | Lithium battery defect detection method based on integrated deep learning | |
CN113469950A (en) | Method for diagnosing abnormal heating defect of composite insulator based on deep learning | |
CN115526864A (en) | Steel rail surface defect detection method based on improved characteristic pyramid network and metric learning | |
CN112949510A (en) | Human detection method based on fast R-CNN thermal infrared image | |
CN115424128A (en) | Fault image detection method and system for lower link of freight car bogie | |
CN114219763A (en) | Infrared picture detection method for abnormal heating point of power distribution equipment based on fast RCNN algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201020 Termination date: 20210626 |
|
CF01 | Termination of patent right due to non-payment of annual fee |