CN110288571B - High-speed rail contact net insulator abnormity detection method based on image processing - Google Patents
High-speed rail contact net insulator abnormity detection method based on image processing Download PDFInfo
- Publication number
- CN110288571B CN110288571B CN201910489410.XA CN201910489410A CN110288571B CN 110288571 B CN110288571 B CN 110288571B CN 201910489410 A CN201910489410 A CN 201910489410A CN 110288571 B CN110288571 B CN 110288571B
- Authority
- CN
- China
- Prior art keywords
- point
- image
- insulator
- points
- curve
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
- G01R31/1245—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of line insulators or spacers, e.g. ceramic overhead line cap insulators; of insulators in HV bushings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
Abstract
The invention discloses a high-speed rail contact net insulator abnormity detection method based on image processing. The method mainly comprises the following steps: preprocessing an image to be processed; detecting the edge of the insulator; and determining the fault by detecting the shape characteristics of the insulator region and the edge. The method can effectively detect the fault of the contact network insulator, has small calculation amount of the algorithm, and ensures the accuracy of fault detection because the corresponding image processing algorithm is added aiming at the interference existing in the example in the image preprocessing process. The omission factor is below 1 percent, and the fault detection accuracy rate is above 97 percent, which all meet the actual engineering requirements. The invention provides a feasible solution for image processing-based fault detection with insufficient negative samples.
Description
Technical Field
The invention belongs to the technical field of image processing and analysis, and particularly relates to a method for detecting abnormality of an insulator of a contact network.
Background
The contact net is an important part in the construction of the electrified transmission line and is erected on a railway line through a line-along strut device. The electric locomotive mainly obtains the electric energy required by operation through the transmission of a contact network, so that the good working state of the contact network is guaranteed at all times. In contact net systems, insulators are one of the important parts of the suspension device, in addition to the mechanical support, on the one hand to allow a sufficient distance between the contact net and the electrical conductors and on the other hand to ensure insulation between the electrical conductors and the ground. Because the working environment of the insulator needs to be exposed in the atmospheric environment for a long time and also needs to be subjected to a strong electric field and strong mechanical stress for a long time, the failure probability is high, and the insulator is inevitably damaged in different degrees. The insulator porcelain body is broken to reduce the insulation strength, and if the insulator porcelain body is not discovered for a long time and cannot be replaced, the insulator porcelain body can be broken to cause other unpredictable faults in a circuit. At present, the traditional manual detection method has low efficiency, high working strength and high danger coefficient, and the electric field method cannot detect external insulation defects which do not influence the electric field, and the detection methods do not have certain practicability. The intelligent routing inspection method has the advantages that the safe operation of the electric locomotive is guaranteed, meanwhile, the detection efficiency is improved, and the intelligent routing inspection is realized, so that the method is particularly important for increasing railway mileage, and therefore, the research on the intelligent routing inspection technology has important significance in future railway development.
In recent years, with the development of digital image processing techniques and machine learning techniques, computer vision techniques have been widely applied to tasks of various object detection and industrial fault detection. The invention provides a contact network insulator fault detection method based on a computer vision technology, which is characterized in that target detection and classification are realized on insulators in ultrahigh-definition videos and pictures in a contact network area shot by an industrial camera through a target detection algorithm based on deep learning, geometric features of edges and textural features of the area are extracted from the detected insulators through the computer vision technology, and finally fault diagnosis is realized through a classification algorithm and an edge tracking detection algorithm learned by a machine. The method is beneficial to realizing non-contact and online monitoring of the insulator defects, and provides a new idea for fault diagnosis of the contact net suspension device. Has important scientific significance and practical application value.
Disclosure of Invention
The invention aims to provide a high-speed rail contact net insulator abnormity detection method based on image processing, which can efficiently identify whether a contact net insulator has a fault. The method comprises the following steps:
step (1): the method comprises the steps of shooting a contact net area through a contact net suspension state detection device to obtain images of a contact net supporting device, screening out the insulator-containing images as research samples, and making the images into data with labels.
Step (2): and manufacturing insulator detection training samples according to the images, and putting the samples into a deep convolution network for training to obtain an insulator target detection model.
And (3): and detecting a test image to be processed through the target detection model to obtain an insulator region and intercepting the region.
And (4): carrying out filtering pretreatment on the intercepted insulator region;
considering the noise condition of the image, firstly, the image is median filtered, which is a nonlinear signal processing technology based on the ordering statistical theory and capable of effectively suppressing the noise, and the basic principle is to replace the value of one point in the digital image or digital sequence with the median of each point value in a neighborhood of the point.
The method comprises the following steps of correcting the direction of an insulator image in advance to obtain the size of the insulator, wherein the specific steps are as follows: performing Otsu threshold segmentation on the insulators, wherein the Otsu method is used for segmenting the image into a target and a background by maximizing the possibility of selecting the threshold; respectively calculating the projection widths of the binary image in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, searching by using a dichotomy to obtain the minimum projection width, and obtaining a correction angle, wherein the algorithm flow is as follows:
1) setting three variables a, mid and b to point to the left end point, the middle and the right end point of the angle value respectively, and respectively representing the projection widths in the angle direction as f (a), f (mid) and f (b);
2) calculating the projection widths of the binary image in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, selecting an angle with the minimum projection width as mid, wherein (mid-45 degrees) and (mid +45 degrees) are respectively a and b, the angle value is 0-180 degrees, and if the angle value is not in the range in the calculation process, the angle value can be adjusted in the range by adding or subtracting 180 degrees;
3) (a) f (b) l < d and f ((a + b)/2) > Max (f (a), f (b)) are judged, d is less than 10-3If at least one of the two formulas is true, stopping the flow, and taking mid as a correction angle; if the two formulas are not established, the operation is continued;
4) comparing the sizes of f (a) and f (b), if f (a) < f (b), the correction angle needs to be searched in the left half section, updating the value of b to mid, if f (a) > f (b), the correction angle needs to be searched in the right half section, and updating the value of a to mid; and returning to the previous step until the flow is finished.
Because the middle area of the insulator is greatly influenced by illumination, a method for eliminating compensation of a reflection area based on a gray value of a trunk area is adopted to eliminate the influence of illumination on edge extraction, and the method comprises the following specific steps: and performing threshold segmentation on the insulator, marking a white area in the binary image, solving a gray average value X of pixel points of the white area at the position in the original image, and performing smooth interpolation on the pixel points of the gray value in the interval (alpha X, 1) in the original image.
And (5): and performing feature extraction on the preprocessed regions, including extraction of regions on two sides of the insulator and edge detection extraction.
The slope and direction of a surface is defined, mathematically by a gradient,
where I (x) represents the original image, J (x) represents the local gradient direction, which points in the direction of the steeply rising luminance function;
considering that the derivation of the image emphasizes the high frequency part and amplifies the noise, the image is smoothed by a low-pass filter before calculating the gradient, a gaussian function is selected as a circularly symmetric filter, and the smoothed gradient graph is written as:
where the general form of the two-dimensional gaussian function is:
in the formula, the parameter σ reflects the bandwidth of the gaussian function;
then connecting the boundary primitives into a chain shape to extract features, wherein the specific processing steps are as follows:
1) 8-connection normalization processing is carried out on the curve obtained by edge detection, namely any pixel point in the line can only have two points on the line at most and is adjacent to the point;
2) eliminating the branch point;
3) if the distance between the end points of any two curves is less than 3 pixel points and the slope of the end point is close to that of the end point connecting line, the two curves are connected.
In order to extract the two side edges of the complete insulator and avoid the influence of irregular textures of the edges on the edge extraction, a region segmentation algorithm is mainly adopted as a region growing algorithm taking background points as starting points, the region growing is a process of aggregating pixels or sub-regions into a larger region according to a predefined criterion, the basic idea is to start from a group of growing points (the growing points can be single pixels or a certain small region), combine adjacent pixels or regions with similar properties to the growing points with the growing points to form new growing points, repeat the process until the growing points cannot grow, and then perform feature extraction.
The contour shape features to be extracted are:
a) length of boundary
The boundary length L is the basic attribute of the boundary, the vertical and horizontal stride is the unit length, and the length of the diagonal stride under 8 connectivity is
b) Curvature of boundary
Traversing a plane curve, assuming that A is a point in the neighborhood of point B on the curve, delta is an intersection angle formed by positive tangents of the two points, AB represents the distance between the point A and the point B, and AB defines a continuous boundary curvature k as:
in a binary image, use is made of a point A on the curveiAnd its predecessor point Ai-bpAnd successor point Ai+bsThe position relationship of (A) is approximated to two points on the curve at a certain chessboard distance from the point, pass through Ai-bp,Ai,Ai+bs
The boundary curvature is calculated by the precursor digital straight line segmentation determined by the three points, and the algorithm is as follows:
1) computing
lp=||Ai-bp,Ai||,ls=||Ai,Ai+bs||,
δp=|Θp-Θ|,δs=|Θs-Θ|
Wherein lp,lsAre respectively point AiDistance from its predecessor and successor points, Θp,ΘsRespectively a precursor bevel angle and a subsequent bevel angle, deltap,δsRespectively a front driving deflection angle and a subsequent deflection angle;
xi+bs,yi+bsrespectively as successor point Ai+bsThe abscissa and the ordinate of the graph (a),
xi-bp,yi-bpare respectively a front driving point Ai-bpThe abscissa and the ordinate of the graph (a),
xi,yiare respectively point AiThe abscissa and the ordinate.
2) In AiBoundary curvature k (A) of pointi) Is composed of
Each pixel point is in a discrete space, and the inclination angle of the tangent line is approximate to the inclination angle of two points on the curve which are at a certain chessboard distance away from the point.
c) Bending energy
The bending energy BE of the boundary is the energy required to bend a beam into the desired shape,
calculated as boundary curvature k (A)i) The sum of squares divided by the number of boundary pixels N:
the shape characteristics of the region to be extracted mainly include:
a) area of
The area of the region refers to the number of pixels contained within the closed region.
b) Projection (projector)
Defining horizontal and vertical region projections g, respectivelyh(u) and gv(j) Is composed of
I (u, j) is the image function.
c) Eccentricity of a rotor
Eccentricity e is characterized by the ratio of the length of the longest chord Q of the region to the longest chord P perpendicular to Q.
d) Center moment
The moment of the region is expressed by considering a normalized gray scale image function as the probability density of a two-dimensional random variable and the central moment is expressed as
Where p, q denote the order of the moment, xc,ycIs the coordinates of the center of gravity of the region.
And (6): and performing feature detection and matching to finish the fault detection of the target.
Fault detection can be divided into two parts:
1) and (4) detecting the abnormity of the edges of the two sides, extracting the shape characteristic area, the projection, the eccentricity and the central moment of the edge areas of the two sides of the insulator according to the step (5), and identifying the abnormity by using a support vector machine.
2) The middle edge anomaly detection, the normal insulator middle edge is smooth, namely the tangent slope of the point on the curve should be monotonously changed, because the curve in the binary image is not a strict curve but is composed of 8 connected pixel points, the change of the inclination angle of the tangent on the pixel point is caused to oscillate, therefore, the inclination angle of each pixel point tangent on the insulator middle edge curve is drawn on a rectangular coordinate system, and the sequence image is subjected to smoothing treatment, namely: the gray value of each pixel point except the end point on the curve is iterated to be the average gray of the point and two adjacent points, the iteration frequency is the length of the curve, the monotonicity of the smooth image is observed to be used as a judgment basis for whether the edge is abnormal or not, and the specific steps of the algorithm are as follows:
traversing a plane curve I, wherein the length of the curve, namely the number of pixel points is n, the point on the curve is p1,p2…pi…pnTo represent;
for any point p on curve Ii2 < i < n-1, the coordinates of the point being (x)pi,ypi) The slope of the tangent at this point is defined as:
drawing the inclination angle of the tangent line of the point on the curve on a rectangular coordinate system, wherein the abscissa is a point sequence, and the ordinate is the inclination angle of the tangent line of the point, and the inclination angle of the tangent line of the point is obtained byThe value of (a) is subjected to m smoothing iterations:j is the current iteration number, m is INT (beta n), INT is an integer function, and the value range of beta is a constant of 0-1, which indicates the iteration number and the number of pixel pointsThe numbers are positively correlated;
and drawing the updated inclination angle on a rectangular coordinate system, and if the obtained image is a non-monotone changing curve, indicating the edge fault.
The value of alpha in the step 4) is 1.5.
The value of beta is 0.8.
The invention has small calculation amount, and the corresponding image processing algorithm is added to the existing interference in the image preprocessing process, thereby ensuring the accuracy of fault detection. The omission factor reaches below 1 percent, and the fault detection accuracy rate is above 97 percent, which all meet the actual engineering requirements. The invention provides a feasible solution for image processing-based fault detection with insufficient negative samples.
Drawings
Fig. 1 is a schematic view of a fault detection process of an insulator according to the present invention;
FIG. 2 is a schematic view of an example of an insulator of an object of interest provided by the present invention;
FIG. 3 is a schematic view of a process for performing directional calibration on an insulator by using a binary search method according to the present invention;
FIG. 4 is a schematic diagram of an embodiment of the compensation for the light reflection area elimination of the insulator according to the present invention, where (a) is an image before processing and (b) is an image after processing;
FIG. 5 is a schematic diagram of an embodiment of edge extraction after direction correction provided by the present invention, (a) is an original image, (b) is an image after direction correction, (c) is an image after edge operator processing, and (d) is an image after edge extraction;
fig. 6 is a schematic diagram of an embodiment of image segmentation and edge extraction by a region growing method according to the present invention, where (a) is an original image, (b) is an image processed by a Canny operator, (c) is an image segmented by the region growing method, and (d) is an edge image extracted according to (c).
FIG. 7 is a schematic diagram of a tangential tilt angle sequence of curved pixel points at the edge of an insulator according to the present invention;
fig. 8 is a schematic diagram of a tangential tilt angle sequence of the smooth and iterative processed pixel points of the edge curve of the insulator provided by the present invention.
Detailed Description
In order to make the technical scheme and implementation steps of the invention more clear, the invention is further described in detail in the following with reference to the specific embodiments and the attached drawings.
Referring to fig. 1, in this embodiment, the method for detecting the target of the key component of the overhead line system includes the following steps:
firstly, obtaining a sample image
The method comprises the steps of acquiring images of the touch screen supporting device acquired by a high-definition camera in the running process of a train, screening out the images with insulators as research samples, and making the images into data with labels. The size of the collected sample image was 4000 x 6000.
Secondly, detecting and classifying insulator targets
To detect the fault of the insulator, the target detection of the insulator is firstly realized, and the method with the best application effect at present is a target detection network based on deep learning. And putting the label data in the step one into a deep learning network for training, and carrying out target detection and classification on the insulators by adopting a one stage method.
Thirdly, preprocessing the picture
The adopted pretreatment modes mainly comprise linear filtering, insulator direction correction and elimination compensation of a reflective area. The insulator image obtained in step two is shown in fig. 2. Considering the situation that the noise exists in the image, firstly, the image is subjected to median filtering to remove salt and pepper noise, wherein the salt and pepper noise is generated by decoding errors in an image system and the like, and white points and black points appear in the image. The median filtering is generally implemented by using a template method, and the pixels in the template are sorted according to the gray value of the pixel points to generate a monotonously rising (or falling) two-dimensional data sequence, and the two-dimensional data sequence is output by using the following formula:
g(x,y)=med{f(x-m,y-n),(m,n∈N)}
where f (x, y) represents the original image, g (x, y) represents the processed image, W is a two-dimensional template, and m and n are the rows and columns of W, respectively.
Median filtering typically uses a two-dimensional template, with the filter window typically being 3 x 3, 5 x 5, 7 x 7 regions, with 3 x 3 rectangular regions being used in the present invention. The method is realized by taking odd number of data out of a certain sampling window in the image for sorting. And replacing the data to be processed by the sorted median value.
Considering the noise condition of the image, firstly, median filtering is carried out on the image, the median filtering is a nonlinear signal processing technology which is based on the ordering statistical theory and can effectively inhibit the noise, and the basic principle is that the value of one point in the digital image or the digital sequence is replaced by the median of all point values in a neighborhood of the point. One common function of the middle finger is to remove salt and pepper noise.
The method comprises the following steps of correcting the direction of an insulator image in advance to obtain the size of the insulator, wherein the specific steps are as follows: performing Otsu threshold segmentation on the insulators, wherein the Otsu method is used for segmenting the image into a target and a background by maximizing the possibility of selecting the threshold; respectively calculating the projection widths of the binary image in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, searching by using a dichotomy to obtain the minimum projection width, and obtaining a correction angle, wherein the algorithm flow is as follows:
1) setting three variables a, mid and b to point to the left end point, the middle and the right end point of the angle value respectively, and respectively representing the projection widths in the angle direction as f (a), f (mid) and f (b);
2) calculating the projection widths of the binary image in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, selecting an angle with the minimum projection width as mid, wherein (mid-45 degrees) and (mid +45 degrees) are respectively a and b, the angle value is 0-180 degrees, and if the angle value is not in the range in the calculation process, the angle value can be adjusted in the range by adding or subtracting 180 degrees;
3) (a) f (b) l < d and f ((a + b)/2) > Max (f (a), f (b)) are judged, d is less than 10-3If at least one of the two formulas is true, stopping the flow, and taking mid as a correction angle; if the two formulas are not established, the operation is continued;
4) comparing the sizes of f (a) and f (b), if f (a) < f (b), the correction angle needs to be searched in the left half section, updating the value of b to mid, if f (a) > f (b), the correction angle needs to be searched in the right half section, and updating the value of a to mid; returning to the previous step until the flow is finished; the flow chart is shown in fig. 3.
Because the middle area of the insulator is greatly influenced by illumination, a method for eliminating and compensating the reflection area based on the gray value of the trunk area is adopted to eliminate the influence of illumination on edge extraction. The method comprises the following specific steps: the insulator is subjected to threshold segmentation (the selection of the threshold is determined by the gray value distribution of the picture), a white area in a binary image is marked, a gray average value X is obtained for pixel points of the white area in the position of an original image, and then smooth interpolation is carried out for pixel points of the original image, wherein the gray value is in a range (alpha X, 1) (alpha is a constant coefficient and is obtained through experiments, and the effect of taking 1.5 for the research object of the invention is better). The effect is schematically shown in fig. 4.
Four, edge extraction and region segmentation algorithm
Mathematically defining the slope and direction of a surface is accomplished by its gradient.
Where I (x) represents the original image, J (x) represents the local gradient direction, which points in the direction of the steeply rising luminance function;
however, taking the derivative of the image emphasizes the high frequency part and thus amplifies the noise, because the ratio of noise to signal is larger in the high frequency part, so that smoothing the image with a low pass filter is considered before calculating the gradient. The gaussian function is the only separable circularly symmetric filter. Since the differentiation is a linear operation, it is interchangeable with other linear filtering operations. Therefore, a smoothed gradient graph can be written
Where the general form of the two-dimensional gaussian function is:
in the formula, the parameter σ reflects the bandwidth of the gaussian function.
On the basis, in order to better separate the edges of the independent insulators and prevent the edges from being disconnected due to rotation and salt and pepper noise, the invention connects the boundaries into a chain shape according to curve characteristics so as to extract the characteristics, and the specific processing steps are as follows:
1) 8-connection normalization processing is carried out on the curve obtained by edge detection, namely any pixel point in the line can only have two points on the line at most and is adjacent to the point;
2) eliminating the branch point;
3) if the distance between the end points of any two curves is less than 3 pixel points and the slope of the end point is close to that of the end point connecting line, the two curves are connected. The effect diagram is shown in fig. 5.
In order to extract the two side edges of the complete insulator and avoid the influence of irregular texture of the edges on the edge extraction, the region segmentation algorithm adopted by the invention is mainly a region growing algorithm taking background points as starting points. Region growing is the process of grouping pixels or sub-regions into larger regions according to a predefined criterion. The basic idea is to start with a group of growing points (the growing points can be single pixels or some small area), merge the adjacent pixel points or areas with similar properties to the growing points with the growing points to form new growing points, and repeat the process until the growing points cannot grow. The effect graph is shown in fig. 6.
Fifth, feature extraction
And then, extracting the features, wherein the contour shape features to be extracted comprise:
a) length of boundary
The boundary length L is the basic attribute of the boundary, the vertical and horizontal stride is the unit length, and the length of the diagonal stride under 8 connectivity is
b) Curvature of boundary
Traversing a plane curve, assuming that A is a point in the neighborhood of point B on the curve, delta is an intersection angle formed by positive tangents of the two points, AB represents the distance between the point A and the point B, and AB defines a continuous boundary curvature k as:
in a binary image, use is made of a point A on the curveiAnd its predecessor point Ai-bpAnd successor point Ai+bsThe position relationship of (A) is approximated to two points on the curve at a certain chessboard distance from the point, pass through Ai-bp,Ai,Ai+bsThe precursor digital straight line segmentation determined by the three points calculates the boundary curvature. The algorithm can be briefly summarized as follows:
1) computing
lp=||Ai-bp,Ai||,ls=||Ai,Ai+bs||,
δp=|Θp-Θ|,δs=|Θs-Θ|
Wherein lp,lsRespectively the distance theta from its predecessor and successor pointsp,ΘsRespectively a precursor bevel angle and a subsequent bevel angle, deltap,δsRespectively a forward-driving deflection angle and a subsequent deflection angle.
xi+bs,yi+bsRespectively as successor point Ai+bsThe abscissa and the ordinate of the graph (a),
xi-bp,yi-bpare respectively a front driving point Ai-bpThe abscissa and the ordinate of the graph (a),
xi,yiare respectively point AiThe abscissa and the ordinate.
2) In AiCurvature k (A) of pointi) Is composed of
In digital image processing, each pixel point is in a discrete space, and the inclination angle of the tangent line can be approximate to the inclination angle of two points on the curve which are at a certain chessboard distance away from the point.
c) Bending energy
The Bending Energy (BE) of the boundary is the energy required to bend a beam into the desired shape,
can be calculated as boundary curvature k (A)i) The sum of the squares of (a) is divided by the number of border pixels N.
The shape characteristics of the region to be extracted mainly include:
a) area of
The area of the region refers to the number of pixels contained in the closed region;
b) projection (projector)
Defining horizontal and vertical region projections g, respectivelyh(u) and gv(j) Is composed of
I (u, j) is an image function;
c) eccentricity of a rotor
Eccentricity e is characterized by the ratio of the length of the longest chord Q of the region to the longest chord P perpendicular to Q.
d) Center moment
The moments of the regions represent the probability density of interpreting a normalized gray-scale image function as a two-dimensional random variable. The central moment is expressed as
Where p, q denote the order of the moment, xc,ycIs the coordinates of the center of gravity of the region.
Sixth, fault detection
The fault can be divided into two parts, 1) two side edge anomaly detection. Because the curvature change of the two side edges and the influence of noise interference are large, and whether the two side edges are abnormal or not is inconvenient to judge according to the profile characteristics, the method firstly extracts the shape characteristic area, projection, eccentricity and central moment of the two side edge areas of the insulator, and then uses a support vector machine to realize the identification of the abnormality. 2) And detecting the middle edge abnormity. The normal middle edge of the insulator should be smooth, that is, the slope of the tangent line of a point on the curve should be monotonously changed, but because the curve in the binary image is not a strict curve but is composed of 8 connected pixel points, the change of the inclination angle of the tangent line on the pixel point is also oscillatory, based on this, the invention draws the inclination angle of the tangent line of each pixel point on the middle edge curve of the insulator on a rectangular coordinate system, and performs smoothing processing on the sequence image (the gray value of each pixel point except the end point on the curve is iterated to be the average gray value of the point and two points adjacent to the point, and the iteration times are the length of the curve). And observing the monotonicity of the smooth image, and taking the monotonicity as a judgment basis for judging whether the edge is abnormal or not.
The specific steps of the algorithm are as follows:
1) traversing a plane curve I, the length (i.e. pixel point) of the curve is n, then the point on the curve can be p1,p2…pi…pnTo represent;
2) for curveAny point p on Ii(2 < i < n-1), the coordinates of the point areThe slope of the tangent to this point can be roughly defined as:
4) drawing the inclination angle of the tangent line of the point on the curve on a rectangular coordinate system, wherein the abscissa is a point sequence, and the ordinate is the inclination angle of the tangent line of the point, taking fig. 2 as an example, and the processing result is shown in fig. 7;
j is the current iteration frequency, m is INT (beta n), INT is an integer function, and the value range of beta is a constant of 0-1, which indicates that the iteration frequency and the number of pixel points are in positive correlation;
6) and drawing the updated inclination angle on a rectangular coordinate system, and if the obtained image is a curve which changes in a non-monotone way, indicating the edge fault, as shown in fig. 8.
Claims (3)
1. A high-speed rail contact net insulator abnormity detection method based on image processing is characterized by comprising the following steps:
step (1): shooting a contact net area through a contact net suspension state detection device to obtain an image of a contact net supporting device, screening out the insulator-containing images as research samples, and making the images into data with labels;
step (2): making insulator detection training samples according to the images, and putting the samples into a deep convolution network for training to obtain an insulator target detection model;
and (3): detecting a test image to be processed through a target detection model to obtain an insulator region and intercepting the region;
and (4): carrying out filtering pretreatment on the intercepted insulator region;
considering the condition that the noise exists in the researched image, firstly, median filtering is carried out on the image, and the median filtering is a nonlinear signal processing method which is based on a sequencing statistical theory and can effectively inhibit the noise;
the method comprises the following steps of correcting the direction of an insulator image in advance to obtain the size of the insulator, wherein the specific steps are as follows: performing Otsu threshold segmentation on the insulators, wherein the Otsu method is used for segmenting the image into a target and a background by maximizing the possibility of selecting the threshold; respectively calculating the projection widths of the binary image in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, searching by using a dichotomy to obtain the minimum projection width, and obtaining a correction angle, wherein the algorithm flow is as follows:
1) setting three variables a, mid and b to point to the left end point, the middle and the right end point of the angle value respectively, and respectively representing the projection widths in the angle direction as f (a), f (mid) and f (b);
2) calculating the projection widths of the binary image in the directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, selecting an angle with the minimum projection width as mid, wherein (mid-45 degrees) and (mid +45 degrees) are respectively a and b, the angle value is 0-180 degrees, and if the angle value is not in the range in the calculation process, adjusting the angle value to the range by adding or subtracting 180 degrees;
3) determining | f (a) -f (b) & gtY<d and f ((a + b)/2)>Max (f (a), f (b)), d is less than 10-3If at least one of the two formulas is true, stopping the flow, and taking mid as a correction angle; if the two formulas are not established, the operation is continued;
4) comparing the sizes of f (a) and f (b), if f (a) and f (b) indicate that the correction angle needs to be searched in the left half section, updating the value of b to mid, and if f (a) and f (b) indicate that the correction angle needs to be searched in the right half section, updating the value of a to mid; returning to the previous step until the flow is finished;
because the middle area of the insulator is greatly influenced by illumination, a method for eliminating compensation of a reflection area based on a gray value of a trunk area is adopted to eliminate the influence of illumination on edge extraction, and the method comprises the following specific steps: performing threshold segmentation on the insulator, marking a white area in a binary image, solving a gray average value X of pixel points of the white area at the position in an original image, and performing smooth interpolation on the pixel points of the gray value in an interval (alpha X, 1) in the original image;
and (5): and (3) carrying out feature extraction on the preprocessed regions, wherein the feature extraction comprises extraction of regions on two sides of the insulator and edge detection extraction:
the slope and direction of a surface is defined, mathematically by a gradient,
wherein I (x) represents the original image, J (x) represents the local gradient direction, pointing to the direction of the extremely rapid rise of the luminance function;
considering that the derivation of the image emphasizes the high frequency part and amplifies the noise, the image is smoothed by a low-pass filter before calculating the gradient, a gaussian function is selected as a circularly symmetric filter, and the smoothed gradient graph is written as:
where the general form of the two-dimensional gaussian function is:
in the formula, the parameter σ reflects the bandwidth of the gaussian function;
then connecting the boundary primitives into a chain shape to extract features, wherein the specific processing steps are as follows:
1) 8-connection normalization processing is carried out on the curve obtained by edge detection, namely any pixel point in the line can only have two points on the line at most and is adjacent to the point;
2) eliminating the branch point;
3) if the distance between the end points of any two curves is less than 3 pixel points, and the slope of the end points is close to that of the end point connecting line, connecting the two curves;
in order to extract the two side edges of the complete insulator and avoid the influence of irregular textures of the edges on the edge extraction, a region segmentation algorithm is mainly a region growing algorithm taking background points as starting points, region growing is a process of aggregating pixels or sub-regions into a larger region according to a predefined criterion, the basic idea is to start from a group of growing points, the growing points are single pixels or a certain small region, adjacent pixels or regions with similar properties to the growing points are merged with the growing points to form new growing points, the process is repeated until the growing points cannot grow, and then feature extraction is carried out:
the contour shape features to be extracted are:
a) length of boundary
The boundary length L is the basic attribute of the boundary, the vertical and horizontal stride is the unit length, and the length of the diagonal stride under 8 connectivity is
b) Curvature of boundary
Traversing a plane curve, assuming that A is a point in the neighborhood of point B on the curve, delta is an intersection angle formed by positive tangents of the two points, AB represents the distance between the point A and the point B, and AB defines a continuous boundary curvature k as:
in a binary image, use is made of a point A on the curveiAnd its predecessor point Ai-bpAnd successor point Ai+bsIs approximated as a distance from the point on the curveTwo points at a certain chessboard distance, passing through Ai-bp,Ai,Ai+bsThe boundary curvature is calculated by the precursor digital straight line segmentation determined by the three points, and the algorithm is as follows:
1) computing
lp=||Ai-bp,Ai||,ls=||Ai,Ai+bs||,
δp=|Θp-Θ|,δs=|Θs-Θ|
Wherein lp,lsAre respectively point AiDistance from its predecessor and successor points, Θp,ΘsRespectively a precursor bevel angle and a subsequent bevel angle, deltap,δsRespectively a front driving deflection angle and a subsequent deflection angle;
xi+bs,yi+bsrespectively as successor point Ai+bsThe abscissa and the ordinate of the graph (a),
xi-bp,yi-bpare respectively a front driving point Ai-bpThe abscissa and the ordinate of the graph (a),
xi,yiare respectively point AiThe abscissa and ordinate of (a);
2) in AiBoundary curvature k (A) of pointi) Is composed of
Each pixel point is in a discrete space, and the inclination angle of the tangent line is approximate to the inclination angle of two points on the curve which are at a certain chessboard distance away from the point;
c) bending energy
The bending energy BE of the boundary is the energy required to bend a beam into the desired shape, and the formula for the calculation is the curvature k (A) of the boundaryi) The sum of squares divided by the number of boundary pixels N:
the shape characteristics of the region to be extracted are as follows:
a) area of
The area of the region refers to the number of pixels contained in the closed region;
b) projection (projector)
Defining horizontal and vertical region projections g, respectivelyh(v) And gv(j) Is composed of
I (v, j) is an image function;
c) eccentricity of a rotor
Eccentricity e is characterized by the ratio of the length of the longest chord Q of the region to the longest chord P perpendicular to Q;
d) center moment
The moment of the region is expressed by considering a normalized gray scale image function as the probability density of a two-dimensional random variable and the central moment is expressed as
Where p, q denote the order of the moment, xc,ycIs the coordinates of the center of gravity of the region;
and (6): performing feature detection and matching to complete fault detection of the target;
the fault can be divided into two parts:
1) detecting the abnormality of the edges of the two sides, extracting the shape characteristic area, the projection, the eccentricity and the central moment of the edge areas of the two sides of the insulator according to the step (5), and identifying the abnormality by using a support vector machine;
2) the middle edge anomaly detection, the normal insulator middle edge is smooth, namely the tangent slope of the point on the curve should be monotonously changed, because the curve in the binary image is not a strict curve but is composed of 8 connected pixel points, the change of the inclination angle of the tangent on the pixel point is caused to oscillate, therefore, the inclination angle of each pixel point tangent on the insulator middle edge curve is drawn on a rectangular coordinate system, and the sequence image is subjected to smoothing treatment, namely: the gray value of each pixel point except the end point on the curve is iterated to be the average gray of the point and two adjacent points, the iteration frequency is the length of the curve, the monotonicity of the smooth image is observed to be used as a judgment basis for whether the edge is abnormal or not, and the specific steps of the algorithm are as follows:
traversing a plane curve I, wherein the length of the curve, namely the number of pixel points is n, the point on the curve is p1,p2…pi…pnTo represent;
for any point p on curve Ii2 < i < n-1, the coordinates of the point beingThe slope of the tangent at this point is defined as:
drawing the inclination angle of the tangent line of the point on the curve on a rectangular coordinate system, wherein the abscissa is a point sequence, and the ordinate is the inclination angle of the tangent line of the point, and the inclination angle of the tangent line of the point is obtained byThe value of (a) is subjected to m smoothing iterations:i is more than 1 and less than n, j is the current iteration time, m is INT (beta n), INT is an integer function, and the value range of beta is a constant of 0-1, which indicates that the iteration time and the number of pixel points are in positive correlation;
and drawing the updated inclination angle on a rectangular coordinate system, and if the obtained image is a non-monotone changing curve, indicating the edge fault.
2. The method for detecting the abnormality of the insulator of the high-speed rail contact network based on the image processing as claimed in claim 1, is characterized in that: in the step 4), the value of alpha is 1.5.
3. The method for detecting the abnormality of the insulator of the high-speed rail contact network based on the image processing as claimed in claim 1, is characterized in that: the value of beta is 0.8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910489410.XA CN110288571B (en) | 2019-06-06 | 2019-06-06 | High-speed rail contact net insulator abnormity detection method based on image processing |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910489410.XA CN110288571B (en) | 2019-06-06 | 2019-06-06 | High-speed rail contact net insulator abnormity detection method based on image processing |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110288571A CN110288571A (en) | 2019-09-27 |
CN110288571B true CN110288571B (en) | 2021-03-23 |
Family
ID=68003439
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910489410.XA Active CN110288571B (en) | 2019-06-06 | 2019-06-06 | High-speed rail contact net insulator abnormity detection method based on image processing |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110288571B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113221614A (en) * | 2020-12-23 | 2021-08-06 | 南方电网电力科技股份有限公司 | Power grid insulator damage image identification method based on hybrid neural network |
CN113911885A (en) * | 2021-10-29 | 2022-01-11 | 南京联了么信息技术有限公司 | Elevator anti-pinch method and system based on image processing |
CN114022483B (en) * | 2022-01-08 | 2022-03-25 | 南通欣斯特机械制造有限公司 | Injection molding flash area identification method based on edge characteristics |
CN115272831B (en) * | 2022-09-27 | 2022-12-09 | 成都中轨轨道设备有限公司 | Transmission method and system for monitoring images of suspension state of contact network |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE3616802A1 (en) * | 1986-05-17 | 1987-11-19 | Bergner Richard Gmbh Co | Protection device against the consequences of an insulator fracture in the case of double chains of a high-voltage overhead line |
CN105572541A (en) * | 2015-12-07 | 2016-05-11 | 浙江大学 | High-voltage line patrol fault detection method and system based on visual attention mechanism |
CN106934418A (en) * | 2017-03-09 | 2017-07-07 | 国家电网公司 | A kind of insulator infrared diagnostics method based on convolution Recursive Networks |
WO2018086299A1 (en) * | 2016-11-11 | 2018-05-17 | 广东电网有限责任公司清远供电局 | Image processing-based insulator defect detection method and system |
CN108154072A (en) * | 2016-12-02 | 2018-06-12 | 天津工业大学 | Insulator breakdown of taking photo by plane based on depth convolutional neural networks detects automatically |
CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
CN108828409A (en) * | 2018-08-03 | 2018-11-16 | 南方电网科学研究院有限责任公司 | A kind of fault detection system based on edge calculations |
CN109215020A (en) * | 2018-08-30 | 2019-01-15 | 国网黑龙江省电力有限公司佳木斯供电公司 | Ultra-high-tension power transmission line fault recognition method based on computer vision |
CN109711257A (en) * | 2018-11-27 | 2019-05-03 | 成都宜泊信息科技有限公司 | A kind of banister condition detection method and system based on image recognition |
CN109801284A (en) * | 2019-01-25 | 2019-05-24 | 华中科技大学 | A kind of high iron catenary insulator breakdown detection method based on deep learning |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003163292A (en) * | 2001-08-13 | 2003-06-06 | Halo Lsi Inc | Twin nand device structure, its array operation and its fabricating method |
CN107727658A (en) * | 2017-08-21 | 2018-02-23 | 南京理工大学 | Pantograph crackle on-line measuring device and method based on image procossing |
CN108693453A (en) * | 2018-05-18 | 2018-10-23 | 云南电网有限责任公司电力科学研究院 | A kind of active infrared thermal image detection device and method of composite insulator internal flaw |
-
2019
- 2019-06-06 CN CN201910489410.XA patent/CN110288571B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE3616802A1 (en) * | 1986-05-17 | 1987-11-19 | Bergner Richard Gmbh Co | Protection device against the consequences of an insulator fracture in the case of double chains of a high-voltage overhead line |
CN105572541A (en) * | 2015-12-07 | 2016-05-11 | 浙江大学 | High-voltage line patrol fault detection method and system based on visual attention mechanism |
WO2018086299A1 (en) * | 2016-11-11 | 2018-05-17 | 广东电网有限责任公司清远供电局 | Image processing-based insulator defect detection method and system |
CN108154072A (en) * | 2016-12-02 | 2018-06-12 | 天津工业大学 | Insulator breakdown of taking photo by plane based on depth convolutional neural networks detects automatically |
CN106934418A (en) * | 2017-03-09 | 2017-07-07 | 国家电网公司 | A kind of insulator infrared diagnostics method based on convolution Recursive Networks |
CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
CN108828409A (en) * | 2018-08-03 | 2018-11-16 | 南方电网科学研究院有限责任公司 | A kind of fault detection system based on edge calculations |
CN109215020A (en) * | 2018-08-30 | 2019-01-15 | 国网黑龙江省电力有限公司佳木斯供电公司 | Ultra-high-tension power transmission line fault recognition method based on computer vision |
CN109711257A (en) * | 2018-11-27 | 2019-05-03 | 成都宜泊信息科技有限公司 | A kind of banister condition detection method and system based on image recognition |
CN109801284A (en) * | 2019-01-25 | 2019-05-24 | 华中科技大学 | A kind of high iron catenary insulator breakdown detection method based on deep learning |
Non-Patent Citations (7)
Title |
---|
A Method of Insulator Detection from Video Sequence;Bingfeng Li 等;《2012 Fourth International Symposium on Information Science and Engineering》;20130411;第386-389页 * |
Adaptive Seeded Region Growing for Image Segmentation Based on Edge Detection, Texture Extraction and Cloud Model;Gang Li 等;《International Conference on Information Computing and Applications》;20101231;第285-292页 * |
图像边缘检测在输电线巡检中的应用;刘雅琴;《数字技术与应用》;20131231;第138页 * |
基于Faster R-CNN的航拍图像中绝缘子识别;程海燕 等;《现代电子技术》;20190115;第42卷(第2期);第98-102页 * |
基于数字图像处理的绝缘子串掉片监测方法的研究;杨照光 等;《第三届(2012)全国架空输电线路技术交流研讨会论文集》;20121231;第558-566页 * |
基于自适应形态学的边缘检测及应用;贺萌;《中国优秀硕士学位论文全文数据库 信息科技辑》;20140515;第2014年卷(第5期);第I138-2961页 * |
绝缘子图像的边缘检测;王伟 等;《微计算机信息》;20081231;第24卷(第9-3期);第308-309、154页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110288571A (en) | 2019-09-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110264448B (en) | Insulator fault detection method based on machine vision | |
CN110288571B (en) | High-speed rail contact net insulator abnormity detection method based on image processing | |
CN108921176B (en) | Pointer instrument positioning and identifying method based on machine vision | |
CN110349207B (en) | Visual positioning method in complex environment | |
CN110866924B (en) | Line structured light center line extraction method and storage medium | |
CN109325935B (en) | Power transmission line detection method based on unmanned aerial vehicle image | |
CN111428748A (en) | Infrared image insulator recognition and detection method based on HOG characteristics and SVM | |
CN111814686A (en) | Vision-based power transmission line identification and foreign matter invasion online detection method | |
Song et al. | Micro surface defect detection method for silicon steel strip based on saliency convex active contour model | |
CN110222661B (en) | Feature extraction method for moving target identification and tracking | |
CN107742307A (en) | Based on the transmission line galloping feature extraction and parameters analysis method for improving frame difference method | |
CN110276747B (en) | Insulator fault detection and fault rating method based on image analysis | |
CN110335233B (en) | Highway guardrail plate defect detection system and method based on image processing technology | |
CN109559324A (en) | A kind of objective contour detection method in linear array images | |
CN116630813B (en) | Highway road surface construction quality intelligent detection system | |
CN109359604B (en) | Method for identifying instrument under shadow interference facing inspection robot | |
CN117094916B (en) | Visual inspection method for municipal bridge support | |
CN115631116B (en) | Aircraft power inspection system based on binocular vision | |
CN114581805A (en) | Coating roller surface defect detection method adopting 3D line laser profile technology | |
Qi et al. | Research on the image segmentation of icing line based on NSCT and 2-D OSTU | |
Liang et al. | Automatic defect detection of texture surface with an efficient texture removal network | |
CN115272306A (en) | Solar cell panel grid line enhancement method utilizing gradient operation | |
CN110348307B (en) | Path edge identification method and system for crane metal structure climbing robot | |
CN112288682A (en) | Electric power equipment defect positioning method based on image registration | |
CN111127381B (en) | Non-parallel detection method for pantograph slide plate |
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 |