CN116309407A - Method for detecting abnormal state of railway contact net bolt - Google Patents
Method for detecting abnormal state of railway contact net bolt Download PDFInfo
- Publication number
- CN116309407A CN116309407A CN202310209896.3A CN202310209896A CN116309407A CN 116309407 A CN116309407 A CN 116309407A CN 202310209896 A CN202310209896 A CN 202310209896A CN 116309407 A CN116309407 A CN 116309407A
- Authority
- CN
- China
- Prior art keywords
- bolt
- image
- image block
- bolts
- target
- 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
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 22
- 238000000034 method Methods 0.000 title claims description 29
- 238000001514 detection method Methods 0.000 claims abstract description 38
- 238000012544 monitoring process Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000004364 calculation method Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 10
- 238000013459 approach Methods 0.000 claims description 6
- 230000004044 response Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 4
- 230000010287 polarization Effects 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 230000002401 inhibitory effect Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 230000036961 partial effect Effects 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 13
- 238000004458 analytical method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000001228 spectrum Methods 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000003137 locomotive effect Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002688 persistence Effects 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000010959 steel Substances 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4007—Interpolation-based scaling, e.g. bilinear interpolation
-
- 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
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/446—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering using Haar-like filters, e.g. using integral image techniques
-
- 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
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- 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
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention discloses a detection method for abnormal states of bolts of a railway overhead line system, which comprises the following steps: s1, acquiring a monitoring image of the state of a contact net bolt, and performing cutting pretreatment to obtain an image block; s2, detecting the image block obtained by preprocessing to obtain a bolt target; s3, classifying the bolt targets in the image block output in the step S2 through a classification network to obtain a bolt target state; the invention can rapidly and accurately detect the abnormal state of the bolt, improves the average detection rate of the bolt positioning and the average identification precision of the bolt defect of the box, and reduces the labor intensity.
Description
Technical Field
The invention relates to the technical field of railway detection, in particular to a method for detecting abnormal states of bolts of a railway contact net.
Background
The high-speed railway overhead contact system is a power transmission line erected above a relatively fixed position along a railway line, an electric traction locomotive takes electricity from the overhead contact system through a pantograph, the electric traction locomotive consists of a wire, a steel rope traction device, a fixed point device and a strut infrastructure, is an important component of the high-speed railway, the overhead contact system is influenced by various factors such as working time and working environment, has higher standard requirements, and has a large number of railway overhead contact system image anomaly detection tasks in daily railway overhead contact system maintenance, wherein due to the fact that bolts are typical small-size components, the number and the types are numerous, the probability of occurrence of state anomalies is high, and actually acquired anomaly samples are relatively less, so that the difficulty of bolt state anomaly detection task is highest.
In the prior art, for example, the Chinese patent number is: the method comprises the steps of CN114743119A, firstly, utilizing an unmanned aerial vehicle to patrol and collect sample pictures, utilizing a generated countermeasure network to generate a defect sample, carrying out data enhancement on the defect sample, and then sending the defect sample into a constructed target detection network to position a target. Deleting the shot biased target according to the positioning result, cutting the target frame, sending the shot biased target into a constructed semantic segmentation network, carrying out pixel level segmentation on the exposed bolt and nut parts of the target, and carrying out threshold judgment on the segmentation result to finish detection. The invention can well make up the shooting defect of the 4C device, can realize the automatic identification and positioning of the upper nut of the dropper of the high-speed railway contact net and the detection of the defect of the nut, and can efficiently and safely ensure the safe state of the dropper on the high-speed railway contact net.
The existing overhead line system bolt state abnormality detection task mainly depends on professional personnel to carry out image screening, the detection efficiency of a manual analysis method is low, the detection persistence is poor, the judgment is full of subjectivity and the like, the analysis task of the bolt abnormal state is seriously affected, the existing railway image abnormality detection method is mainly based on a supervision type learning method, and the existing overhead line system bolt state abnormality detection method can realize detection of the bad state of the overhead line system bolt, but different characteristic extraction methods are designed aiming at different bolts, and the problems of high precision improvement difficulty and low operation efficiency exist.
Disclosure of Invention
The invention overcomes the defects existing in the prior art, and solves the technical problems that: the method for detecting the abnormal state of the bolt of the railway overhead line system is provided to realize automatic detection of the abnormal state of the bolt of the railway overhead line system and improve the safety of railway operation.
In order to solve the technical problems, the invention adopts the following technical scheme: a detection method of abnormal states of bolts of a railway contact net comprises the following steps:
s1, acquiring a monitoring image of the state of a contact net bolt, and performing cutting pretreatment to obtain an image block;
s2, detecting the image block obtained by preprocessing to obtain a bolt target;
the method specifically comprises the following steps:
s201, extracting features of an image block by adopting ResNet as a backbone network for feature extraction; the number of the characteristic diagrams of the 1 st to 4 th stages of the network is respectively set to 16, 32, 48 and 72;
s202, firstly pooling the feature map into a pyramid level feature map with 2 pixels by 2 pixels, and recovering the size of the feature map by using a bilinear interpolation method;
s203, fusing the features of different pyramid levels through a global attention module;
s204, predicting the probability of each image block existing a target bolt through a classifier, and outputting the image blocks with the probability larger than a threshold value; acquiring probability distribution of each candidate region in the image block on K+1 categories and frame regression offset of the K categories through a detector;
s3, classifying the bolt targets in the image block output in the step S2 through a classification network to obtain a bolt target state; the method specifically comprises the following steps:
s301, performing enhancement processing on the target contour and texture by using the polarization fusion image;
s302, primarily positioning a target bolt area;
s303, extracting local binary feature quantities of the bolts, converting the local binary feature quantities into cosine vectors, performing similarity calculation, and judging fault grades of the bolts according to similarity calculation results.
The step S2 is realized by adopting a catenary bolt detection network, wherein the catenary bolt detection network comprises a characteristic extraction network, a global attention module, and a classifier and a detector which are mutually enhanced.
In the overhead line system bolt detection network, a classifier is used for outputting the probability of each image block that a target bolt exists, a detector is used for acquiring probability distribution of each candidate region in the image block on K+1 categories and frame regression offset of the K categories, and a loss function of the detector is as follows:
L(m,b,p,u,t u ,v)＝L cls (m,b)+u[b＝2]×L cls (t u ,v)；
wherein L is cls (m, b) is a softmax penalty, u [ b=2]×L cls (t u V) represents the frame regression loss for each candidate region in the detector as the region labeled 2 that matches it; during training, the loss of the detector is counter-propagated only when the target bolt is detected in the corresponding image block.
The specific steps of the step S302 are as follows:
solving an integral image of the anchor ear region, and introducing a 2-order black plug matrix as a filter to calculate a 2-order partial derivative by convolution of the filter and the kernel;
changing the size of a filter box and detecting textures through image integration, solving different response values of a matrix and greatly inhibiting;
and performing gradient operation on the image by utilizing a harr wavelet, distributing a main direction after weighting response values, dividing the region into blocks along the main direction, calculating four feature vectors, and performing feature point matching on the image to be detected according to texture differences.
And when the characteristic points are matched, judging the direction of the characteristic points according to the black plug matrix trace, matching 2 images by adopting Euclidean distance, determining the size of a characteristic region according to the angular points of the template, sequentially reading and calibrating 4 vertex angle coordinates of the region from left to right, and actively drawing a bolt positioning frame.
In step S303, the similarity calculation method includes mapping the image to a two-dimensional space after performing feature transformation, using cosine values between vectors as similarity values, wherein the more similar the cosine values approach 1, the more different the image structure approaches 0, and each pixel feature value corresponds to an element, and judging the fault level of the bolt defect through the difference between the cosine values obtained by calculation.
In the step S1, the overhead line system bolt status monitoring image is cut into 512×512 pixel image blocks, and the overlapping rate between the image blocks is 35%.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention provides a detection method of abnormal states of bolts of a railway overhead line system, which comprises the steps of designing an overhead line system bolt detection network, and realizing high-efficiency detection of a target bolt, wherein the overhead line system bolt detection network consists of a lightweight characteristic extraction network, a global attention module, a mutually enhanced classifier and a detector;
2. the detection method of the invention adopts the 2-order cascade convolutional neural network, can rapidly and accurately detect the abnormal state of the bolt, has the average detection rate of bolt positioning up to 98.2%, has the average identification precision of bolt defects up to 95.8%, improves the analysis efficiency of image data by about 21.5% compared with a single detection network, reduces the manual labor intensity, can continuously detect the bolt of the railway overhead line system, and avoids seriously affecting the analysis task of the abnormal state of the bolt.
Drawings
Fig. 1 is a schematic flow chart of a method for detecting an abnormal state of a bolt of a railway overhead line system according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the embodiment of the invention provides a method for detecting an abnormal state of a bolt of a railway overhead line system, which comprises the following steps:
s1, acquiring a monitoring image of the state of a contact net bolt, and performing cutting pretreatment to obtain an image block.
Installing imaging devices at corresponding positions of the railway overhead contact system, and shooting from a plurality of angles globally and locally through the imaging devices to obtain overhead contact system bolt state monitoring images; before the overhead line system bolt state monitoring image is input into the detection network, small batches of image blocks are needed to be processed, the overhead line system bolt state monitoring image with the pixels (6600 multiplied by 4400) is automatically cut into the image blocks with the pixels (512 multiplied by 512), and the overlapping rate between the image blocks is 35%.
S2, detecting the image block obtained by preprocessing to obtain a bolt target;
the method specifically comprises the following steps:
s201, extracting features of an image block by adopting ResNet as a backbone network for feature extraction; the number of the 1 st to 4 th stage characteristic diagrams of the network is respectively set to 16, 32, 48 and 72. The parameters of the network are much smaller than res net-18.
In the embodiment, the number of the feature graphs of each stage is obtained based on the feature spectrum of the covariance matrix of the feature graph of each stage in the initial convergence network by fine adjustment in the training process; the specific adjustment method comprises the following steps:
calculating the characteristic values of characteristic spectrums sigma of each image block in each stage of the ResNet network, and taking the number of the characteristic values larger than a preset threshold value as the number of the characteristic graphs in the stage; the characteristic spectrum sigma of any characteristic diagram is obtained by normalizing the covariance matrix of the characteristic diagram through the spatial resolution, and the calculation formula of the characteristic spectrum sigma is as follows:
wherein n is the number of image blocks; h i And W is i The width and the height of the feature map space are respectively; f (F) i,x,y Feature vectors for (x, y) pixels in space coordinates in the feature map for the ith image block;
in the embodiment, the final feature map number is obtained by fine adjustment in the training process, the total downsampling step length of a lightweight feature extraction network is reduced, and the fine features of the overhead line bolts are finally extracted;
s202, firstly pooling the feature map into a pyramid level feature map of 2 x 2 pixels, and then restoring the size of the feature map by using a bilinear interpolation method.
S203, fusing the features of different pyramid levels through the global attention module, so that the detector can obtain more features containing context information.
S204, predicting the probability of each image block existing a target bolt through a classifier, and outputting the image blocks with the probability larger than a threshold value; and acquiring probability distribution of each candidate region in the image block on K+1 categories and frame regression offsets of the K categories through a detector.
The whole overhead line system bolt detection network has three outputs, one is the output of a classifier, represents the probability m of whether a target bolt exists in a current image block, and the other two are the outputs of a detector; the output of the detector comprises a probability distribution p= (p) over k+1 categories for each candidate region in the image block 0 ,...,p k ) And K classes of frame regression offsets
Where k is a category index number, t k Is a scale invariant transformation and a spatial shift of width and height (w, h) relative to the marked frame center point (x, y).
The step S2 is realized by adopting a catenary bolt detection network, wherein the catenary bolt detection network comprises a characteristic extraction network, a global attention module, and a classifier and a detector which are mutually enhanced. The classifier is used for outputting the probability of the existence of the target bolt of each image block, and the detector is used for acquiring probability distribution of each candidate region in the image block on K+1 categories and frame regression offsets of the K categories.
When the overhead line system bolt detection network is trained, each image block participating in training is marked as a binary label b according to whether a target bolt is contained or not, each candidate area in the detector is marked as a label frame type label u matched with the candidate area, and the target of frame regression is v.
The loss function of the detector is set as a multitasking loss function, which is defined as follows:
L(m,b,p,u,t u ,v)＝L cls (m,b)+u[b＝2]×L cls (t u ,v)； (2)
wherein L is cls (m, b) is a softmax penalty, u [ b=2]×L cls (t u V) represents the frame regression loss for each candidate region in the detector as the region labeled 2 that matches it; during training, the loss of the detector is counter-propagated only when the target bolt is detected in the corresponding image block.
S3, classifying the bolt targets in the image block output in the step S2 through a classification network to obtain a bolt target state; the classification network adopted in the embodiment is a lightweight class network of running states, which can finish the fine classification of the running states of the overhead line bolts and realize defect identification.
The step S3 specifically comprises the following steps:
s301, performing enhancement processing on the target contour and texture by using the polarization fusion image;
s302, primarily positioning a target bolt area;
s303, extracting local binary feature quantities of the bolts, converting the local binary feature quantities into cosine vectors, performing similarity calculation, and judging fault grades of the bolts according to similarity calculation results.
In the step S302, the initial positioning of the target bolt area is performed by adopting acceleration robust feature detection, which specifically comprises the following steps:
solving an integral image of the anchor ear region, and introducing a 2-order black plug matrix as a filter to calculate a 2-order partial derivative by convolution of the filter and the kernel;
changing the size of a filter box and detecting textures through image integration, solving different response values of a matrix and greatly inhibiting;
and performing gradient operation on the image by utilizing a harr wavelet, distributing a main direction after weighting response values, dividing the region into blocks along the main direction, calculating four feature vectors, and performing feature point matching on the image to be detected according to texture differences.
When the characteristic points are matched, the direction of the characteristic points is judged according to the black plug matrix trace, the Euclidean distance is adopted to pair 2 images, the size of a characteristic area is determined according to the angular points of the template, 4 vertex angle coordinates of the area are sequentially read and calibrated from left to right, and a bolt positioning frame is actively drawn.
In step S303, the method of similarity calculation includes mapping the image to a two-dimensional space after performing feature transformation, using cosine values between vectors of the image and a standard image as similarity values, wherein the cosine values approach 1, the image structures are more similar, the difference is generated when the cosine values approach 0, each pixel feature value corresponds to an element, and judging the fault level of the bolt defect through the difference between the cosine values obtained by calculation.
Wherein, each nut rotates slowly until falling off, and a plurality of groups of cosine values cos theta 1, cos theta 2, … and cos theta m are obtained, and when loosening, the nuts are used as a judging basis according to a plurality of groups of value range ranges; when falling off, measuring cosine values cos theta 1, cos theta 2, … and cos theta n in various states, and taking the minimum value as a judgment cut-off value; and (3) preventing the factors such as position movement, illumination change, shooting shake, background change and the like from affecting the measurement result, respectively carrying out experiments on 35 samples of the falling-off samples of the anchor ear nut and 100 samples with loosening phenomenon in the experiments, counting cosine values and classifying.
The invention designs a contact net bolt detection network, which consists of a lightweight characteristic extraction network, a global attention module, a classifier and a detector which are mutually enhanced, so that efficient detection of a target bolt is realized, then a polarization fusion image is utilized to enhance the target contour and texture, acceleration robust characteristic detection is adopted to initially position a bolt area, local binary characteristic quantity of the bolt is extracted, similarity identification measurement is carried out on the defect state of the bolt after the local binary characteristic quantity of the bolt is converted, fine classification of the bolt state is completed, and defect identification is realized; the 2-level convolutional neural network method can rapidly and accurately detect abnormal states of bolts, improves analysis efficiency of image data and reduces labor intensity.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (8)
1. The method for detecting the abnormal state of the bolt of the railway contact net is characterized by comprising the following steps of:
s1, acquiring a monitoring image of the state of a contact net bolt, and performing cutting pretreatment to obtain an image block;
s2, detecting the image block obtained by preprocessing to obtain a bolt target; the method specifically comprises the following steps:
s201, extracting features of an image block by adopting ResNet as a backbone network for feature extraction;
s202, firstly pooling the feature map into a pyramid level feature map with 2 pixels by 2 pixels, and recovering the size of the feature map by using a bilinear interpolation method;
s203, fusing the features of different pyramid levels through a global attention module;
s204, predicting the probability of each image block existing a target bolt through a classifier, and outputting the image blocks with the probability larger than a threshold value; acquiring probability distribution of each candidate region in the image block on K+1 categories and frame regression offset of the K categories through a detector;
s3, classifying the bolt targets in the image block output in the step S2 through a classification network to obtain a bolt target state; the method specifically comprises the following steps:
s301, performing enhancement processing on the target contour and texture by using the polarization fusion image;
s302, primarily positioning a target bolt area;
s303, extracting local binary feature quantities of the bolts, converting the local binary feature quantities into cosine vectors, performing similarity calculation, and judging fault grades of the bolts according to similarity calculation results.
2. The method for detecting abnormal states of bolts of a railway overhead line system according to claim 1, wherein the step S2 is implemented by adopting an overhead line system bolt detection network, and the overhead line system bolt detection network comprises a feature extraction network, a global attention module, and a classifier and a detector which are mutually enhanced.
3. The method for detecting abnormal states of bolts of a railway overhead line system according to claim 2, wherein in the overhead line system bolt detection network, a classifier is used for outputting probability of existence of target bolts of each image block, the detector is used for obtaining probability distribution of each candidate region in the image block on k+1 categories and frame regression offset of the K categories, and a loss function of the detector is as follows:
L(m,b,p,u,t u ,v)＝L cls (m,b)+u[b＝2]×L cls (t u ,v)；
wherein L is cls (m, b) is a softmax penalty, u [ b=2]×L cls (t u V) represents the frame regression loss for each candidate region in the detector as the region labeled 2 that matches it; during training, the loss of the detector is counter-propagated only when the target bolt is detected in the corresponding image block.
4. The method for detecting abnormal states of bolts of a railway overhead line system according to claim 1, wherein the specific steps of step S302 are as follows:
solving an integral image of the anchor ear region, and introducing a 2-order black plug matrix as a filter to calculate a 2-order partial derivative by convolution of the filter and the kernel;
changing the size of a filter box and detecting textures through image integration, solving different response values of a matrix and greatly inhibiting;
and performing gradient operation on the image by utilizing a harr wavelet, distributing a main direction after weighting response values, dividing the region into blocks along the main direction, calculating four feature vectors, and performing feature point matching on the image to be detected according to texture differences.
5. The method for detecting the abnormal state of the bolt of the railway catenary according to claim 4, wherein when characteristic points are matched, the direction of the characteristic points is judged according to black plug matrix tracks, 2 images are matched by adopting Euclidean distance, the size of a characteristic area is determined according to corner points of a template, 4 vertex angle coordinates of the area are read and calibrated in sequence from left to right, and a bolt positioning frame is actively drawn.
6. The method for detecting abnormal states of bolts in a railway catenary according to claim 1, wherein in the step S303, the similarity calculation method is that after the image is subjected to feature transformation, the values of the images are mapped into a two-dimensional space, cosine values among vectors are used as similarity values, when the cosine values approach 1, the image structures are more similar, when the cosine values approach 0, the difference is generated, each pixel feature value corresponds to one element, and the fault level of the bolts is judged by calculating the difference among the cosine values.
7. The method according to claim 1, wherein in the step S1, the overhead line system bolt status monitoring image is cut into 512 x 512 pixel image blocks, and the overlapping rate between the image blocks is 35%.
8. The method for detecting abnormal states of bolts of a railway catenary according to claim 1, wherein in the step S201, the number of the 1 st to 4 th phase feature maps of the res net network adopted is set to 16, 32, 48 and 72 respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310209896.3A CN116309407A (en) | 2023-03-07 | 2023-03-07 | Method for detecting abnormal state of railway contact net bolt |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310209896.3A CN116309407A (en) | 2023-03-07 | 2023-03-07 | Method for detecting abnormal state of railway contact net bolt |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116309407A true CN116309407A (en) | 2023-06-23 |
Family
ID=86800815
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310209896.3A Pending CN116309407A (en) | 2023-03-07 | 2023-03-07 | Method for detecting abnormal state of railway contact net bolt |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116309407A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117105038A (en) * | 2023-10-17 | 2023-11-24 | 山西戴德测控技术股份有限公司 | Elevator operation monitoring method, device, equipment and storage medium |
-
2023
- 2023-03-07 CN CN202310209896.3A patent/CN116309407A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117105038A (en) * | 2023-10-17 | 2023-11-24 | 山西戴德测控技术股份有限公司 | Elevator operation monitoring method, device, equipment and storage medium |
| CN117105038B (en) * | 2023-10-17 | 2024-01-05 | 山西戴德测控技术股份有限公司 | Elevator operation monitoring method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111428748B (en) | HOG feature and SVM-based infrared image insulator identification detection method | |
| CN109344753A (en) | A kind of tiny fitting recognition methods of Aerial Images transmission line of electricity based on deep learning | |
| CN109325935B (en) | Power transmission line detection method based on unmanned aerial vehicle image | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| CN111814686A (en) | Vision-based power transmission line identification and foreign matter invasion online detection method | |
| CN111008961B (en) | Transmission line equipment defect detection method and system, equipment and medium thereof | |
| CN114743119B (en) | High-speed rail contact net hanger nut defect detection method based on unmanned aerial vehicle | |
| Zhang et al. | Bridge damage detection using a single-stage detector and field inspection images | |
| CN108694349B (en) | Pantograph image extraction method and device based on linear array camera | |
| CN112329623A (en) | Early warning method for visibility detection and visibility safety grade division in foggy days | |
| CN111080691A (en) | Infrared hot spot detection method and device for photovoltaic module | |
| CN111768417B (en) | Railway wagon overrun detection method based on monocular vision 3D reconstruction technology | |
| CN113111727A (en) | Method for detecting rotating target in remote sensing scene based on feature alignment | |
| Tan | Automatic insulator detection for power line using aerial images powered by convolutional neural networks | |
| CN116309407A (en) | Method for detecting abnormal state of railway contact net bolt | |
| CN112950576A (en) | Power transmission line defect intelligent identification method and system based on deep learning | |
| CN116681979A (en) | Power equipment target detection method under complex environment | |
| CN112288682A (en) | Electric power equipment defect positioning method based on image registration | |
| CN112884795A (en) | Power transmission line inspection foreground and background segmentation method based on multi-feature significance fusion | |
| CN113962973A (en) | Power transmission line unmanned aerial vehicle intelligent inspection system and method based on satellite technology | |
| CN111931577A (en) | Intelligent inspection method for specific foreign matters of power grid line | |
| CN116994161A (en) | Insulator defect detection method based on improved YOLOv5 | |
| Zhang et al. | Defect detection of anti-vibration hammer based on improved faster R-CNN | |
| CN112418241A (en) | Power equipment identification method based on infrared imaging technology | |
| CN111626175B (en) | Shaft type identification method based on deep convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |