CN103034861B - The recognition methods of a kind of truck brake shoe breakdown and device - Google Patents

Abstract

The invention discloses the recognition methods of a kind of truck brake shoe breakdown, including: from present image, extract the segmentation feature of three angles；Segmentation feature according to described three angles determines the lorry brake shoe characteristic area of present image；The feature of lorry brake shoe is extracted from the lorry brake shoe characteristic area of present image；Use support vector machine (SVM) algorithm that the feature calculation of lorry brake shoe draws the eigenvalue of present image, judge whether lorry brake shoe exists fault according to described eigenvalue and preset Fault Identification value.The invention also discloses the identification device of a kind of truck brake shoe breakdown, use the present invention to be avoided that manually-operated error, it is ensured that the discrimination of fault, Accident prevention occurs timely, thus ensures operation security.

Description

A method and device for identifying faults of freight train brake shoes

technical field

The invention relates to the field of image processing, in particular to a method and device for identifying faults of freight train brake shoes.

Background technique

At present, freight car brake shoe braking is the tread braking method commonly used by domestic railway freight cars. The brake shoe brazing on the freight car brake shoe is mainly used to prevent the freight car brake shoe from falling off. Once the brake shoe brazing is lost, the freight car brake shoe will be damaged during the operation It is very likely to fall off, causing brake failure, and what is more serious is that if the brake shoe of the truck just falls off the rail, it will cause a derailment accident. In recent years, the Ministry of Railways has vigorously promoted the fault image dynamic detection system (TFDS) of freight trains to image the key parts of the running trains, and manually browse the images to complete the fault identification. This method of manual fault identification has problems such as low efficiency and unstable recognition rate, which cannot meet the development requirements of safe train operation.

It can be seen that the use of TFDS in the prior art for brake shoe fault identification relies too much on manual operations, so the recognition rate of faults cannot be guaranteed, and accidents cannot be prevented in time, so that operation safety cannot be guaranteed.

Contents of the invention

In view of this, the object of the present invention is to provide a method and device for identifying faults of freight train brake shoes, which can avoid manual operation errors, ensure the recognition rate of faults, prevent accidents in time, and ensure operational safety.

In order to achieve the above object, technical solution of the present invention is achieved in that way:

The invention provides a method for identifying faults of freight train brake shoes, the method comprising:

Extract segmentation features from three angles from the current image;

Determine the feature area of the brake shoe of the current image according to the segmentation features of the three angles;

Extract the features of the truck brake shoe from the feature area of the truck brake shoe in the current image;

Using the Support Vector Machine (SVM, Support Vector Machine) algorithm to calculate the characteristics of the truck brake shoe to obtain the feature value of the current image, and determine whether the truck brake shoe has a fault according to the feature value and the preset fault identification value.

In the above scheme, the extraction of the segmentation features of three angles from the current image includes: periodically extracting the current image from the TDFS, and performing grayscale projection of the current image at three angles to obtain three projection curves; Filtering, using the maximum value of each filtered projection curve as the segmentation feature of the three angles of the current image.

In the above solution, the determination of the characteristic area of the truck brake shoe of the current image according to the segmentation features of the three angles includes: using the Canny operator to extract the edge information of the current image, and determining the characteristic area of the truck brake shoe according to the segmentation feature with an angle of zero degrees The coordinate values of the left boundary and right boundary of ; according to the segmentation features with angles of -25° and 25°, determine the coordinate values of the upper boundary and lower boundary of the feature area of the brake shoe of the truck.

In the above solution, the extraction of the features of the truck brake shoe from the feature area of the truck brake shoe of the current image includes: using a method based on background area estimation to segment the image in the feature area of the truck brake shoe of the current image to obtain the truck brake shoe. tile area binary image; use pixel notation method to extract the maximum connected area from the binary image of the truck brake shoe area; use Canny operator to extract the truck brake shoe edge profile from the binary image in the maximum connected area; according to the truck Brake shoe edge profile extraction of freight car brake shoe features.

In the above solution, the characteristics of the freight car brake shoe include: the smooth characteristic value of the edge profile of the freight car brake shoe, the concave-convex characteristic value of the edge contour of the freight car brake shoe, the sawtooth degree of the edge contour of the freight car brake shoe, and the firmness of the edge contour of the freight car brake shoe. value, the compactness of the wagon brake shoe edge profile, the circularity of the wagon brake shoe edge profile, the aspect ratio of the wagon brake shoe edge profile, the area of the wagon brake shoe edge profile, and the perimeter of the wagon brake shoe edge profile.

In the above scheme, before extracting the segmentation features of three angles from the current image, the method further includes: using the image of the brake shoe of a truck without a fault and the image of a brake shoe of a truck with a fault to respectively establish positive and negative sample training set; respectively extract the features of all the images in the positive and negative sample training sets to form the nine-dimensional feature vectors corresponding to the positive and negative sample training sets, use the SVM algorithm to calculate the values corresponding to the positive and negative samples, and convert the negative sample training set The corresponding value is used as the fault identification value.

The present invention also provides an identification device for a brake shoe failure of a freight car, which includes: a feature extraction module and an identification module; wherein,

The feature extraction module is used to extract the segmentation features of three angles from the current image, determine the feature area of the truck brake shoe in the current image according to the segmentation features of the three angles, and extract the truck brake shoe feature area from the current image. The feature of the tile, sending the feature of the truck brake shoe to the recognition module;

The recognition module is used to use the SVM algorithm to calculate the feature value of the current image based on the features of the truck brake shoe sent by the feature extraction module, and determine whether there is a fault in the truck brake shoe according to the feature value and the preset fault identification value .

In the above scheme, the feature extraction module is specifically used to periodically extract the current image from the TDFS where it is located, and perform grayscale projection of the current image at three angles to obtain three projection curves; filter all projection curves, and filter The maximum value of each projection curve of is used as the segmentation feature of the three angles of the current image.

In the above scheme, the feature extraction module is specifically used to extract the edge information of the current image using the Canny operator, and determine the coordinate values of the left boundary and the right boundary of the truck brake shoe feature area according to the segmentation feature with an angle of zero degrees; according to the angle For the segmentation features of -25° and 25°, determine the coordinate values of the upper boundary and lower boundary of the feature area of the brake shoe of the freight car.

In the above solution, the feature extraction module is specifically used to use a method based on background area estimation to segment the image in the feature area of the truck brake shoe in the current image to obtain a binary image of the truck brake shoe area; The maximum connected area is extracted from the binary image of the truck brake shoe area; the edge profile of the truck brake shoe is extracted from the binary image in the maximum connected area using the Canny operator; the feature of the truck brake shoe is extracted according to the edge profile of the truck brake shoe.

In the above solution, the feature extraction module is specifically used to extract the smooth feature value of the edge profile of the brake shoe of the truck, the concave-convex feature value of the edge profile of the brake shoe of the truck, the sawtooth degree of the edge profile of the brake shoe of the truck, and the solidity of the edge profile of the brake shoe of the truck. Reliance, the compactness of the edge profile of the truck shoe edge, the circularity of the edge profile of the brake shoe edge of the truck, the aspect ratio of the edge profile of the brake shoe edge of the truck, the area of the edge profile of the brake shoe edge of the truck and the perimeter of the edge profile of the brake shoe of the truck Brake shoe characteristics.

In the above solution, the identification module is also used to use the image of the brake shoe of the truck without the fault and the image of the brake shoe of the truck with the fault to respectively establish the positive and negative sample training sets; extract all the positive and negative sample training sets respectively. The features of the image form the nine-dimensional feature vectors corresponding to the positive and negative sample training sets, and use the SVM algorithm to calculate the corresponding values of the positive and negative samples, and use the values corresponding to the negative sample training sets as the fault identification value.

The identification method and device of the truck brake shoe failure provided by the present invention can automatically extract the segmentation features of three angles from the current image; determine the feature area of the truck brake shoe of the current image according to the segmentation features of the three angles; Extracting the features of the truck brake shoe from the feature area of the truck brake shoe in the image; and then determining whether there is a fault in the truck brake shoe according to the characteristics of the truck brake shoe and the fault identification conditions. In this way, errors in manual operation can be avoided, the recognition rate of faults can be ensured, and accidents can be prevented in time to ensure operational safety.

Description of drawings

Fig. 1 is the schematic flow chart of the identification method of the truck brake shoe fault of the present invention;

Fig. 2 is a schematic diagram of the composition and structure of the identification device for the failure of the truck brake shoe of the present invention;

Figure 3 is a table of test results.

detailed description

The basic idea of the present invention is: extract the segmentation features of three angles from the current image; determine the feature area of the brake shoe of the current image according to the segmentation features of the three angles; extract the feature area of the brake shoe of the truck from the current image The characteristics of the brake shoes: use the SVM algorithm to calculate the characteristics of the brake shoes of the truck to obtain the feature value of the current image, and determine whether there is a fault in the brake shoe of the truck according to the feature value and the preset fault identification value.

The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

The identification method of freight car brake shoe fault of the present invention, as shown in Figure 1, comprises the following steps:

Step 101: Extract segmentation features from three angles from the current image.

Specifically, the current image is periodically extracted from TDFS, and three projection curves are obtained by gray-scale projection of the current image at three angles; all projection curves are filtered, and the maximum value of each filtered projection curve is used as the current image Segmentation features from three angles;

Here, the three angles are the set three angles of -25°0° and 25°;

The grayscale projection includes: expanding the current image to obtain an expanded image; performing projection calculation on the expanded image, and discretizing the image obtained by the projection calculation to obtain a projection curve.

The current image is expanded to fill pixels with a grayscale of zero at the edge of the image to avoid errors caused by the edge of the image during projection. The following formula can be used for calculation:

$\∀(i,j)\∈R^2,$ $I^{\′}(i,j)=\left\{\begin{array}{cc}I(i,j)& (i,j)\∈\mathrm{\Ω}\\ 0& (i,j)\∉\mathrm{\Ω}\end{array}\right.;$ Among them, Ω represents infinity.

Wherein, the projection calculation can use the formula:

in, $i=\sqrt{r^2+l^2}\cos [\mathrm{\θ}+a\mathrm{the\; tan}\left(\frac{l}{r}\right)],$ $j=\sqrt{r^2+l^2}\sin [\mathrm{\θ}+a\mathrm{the\; tan}\left(\frac{l}{r}\right)],$ Suppose a coordinate system is established, the origin of the coordinates is o, the height of the current image I is m, and the width is n, is the projection direction, the 2D image I will be projected to the projection line The starting point is o, and perpendicular to each other. r is any point on the projection line, p is a point passing through r, and Parallel straight line, l is any point (i, j) above the projection line The vertical distance of , I(i, j) is the pixel gray value of the image at point (i, j), and the angle θ between the projection line and the x-axis is the projection direction angle, θ is -25°0° or 25° .

The discretization of the image obtained by the projection calculation can be calculated using the following formula:

in, θ is -25° 0° and 25°, The height of the current image I is m, the width is n, δ _θ (r) represents the grayscale projection of the image on the projection direction angle θ, and it can reflect the grayscale change of the image in the direction of the projection direction angle θ.

The formula for filtering all projection curves can be used:

$h\left(x\right)=\left\{\begin{array}{ccc}1& x_2\&le;x\&le;x_1& \\ -1& x_3\&le;x<x_1\&cup;& x_2<x\&le;x_4\end{array}\right.;$ Among them, x ₁ , x ₂ , x ₃ , x ₄ are constants, satisfying $\left\{\begin{array}{c}|x_1-x_2|=h\\ |x_3-x_4|=2h\end{array}\right.,$ H is the estimated width of the segmentation feature.

Step 102: Determine the feature area of the brake shoe of the current image according to the segmentation features of the three angles.

Specifically: use the Canny operator to extract the edge information of the current image, and determine the coordinate values of the left boundary and right boundary of the feature area of the truck brake shoe according to the segmentation feature with an angle of zero degree; according to the segmentation feature with an angle of -25° and 25° , to determine the coordinate values of the upper boundary and lower boundary of the characteristic area of the freight train brake shoe.

Here, according to the segmentation feature with an angle of zero degrees, determining the coordinate values of the left boundary and the right boundary of the truck brake shoe feature area includes: using the segmentation feature with an angle of zero degrees to divide the edge information of the current image into two sub-images, respectively Calculate the sum of the pixel gray values of the two sub-images, compare the sum of the pixel gray values of the two sub-images, if the sum of the pixel gray values of the left sub-image is larger, the right The coordinate value of the boundary is equal to the abscissa value of the segmentation feature of zero degrees, minus the preset right boundary distance value, and the coordinate value of the left boundary of the freight car brake shoe feature area is equal to the coordinate value of its right boundary minus the preset width value; Otherwise, the coordinate value of the left boundary of the characteristic area of the truck brake shoe is equal to the abscissa value of the segmentation feature of zero degrees, plus the preset left boundary distance value, and the coordinate value of the right boundary of the characteristic area of the truck brake shoe is equal to the coordinate value of its left boundary Add the preset width value;

For example, assuming that the edge information of the current image is I, and the segmentation feature with an angle of zero degree is l ₂ in the figure, the edge information I of the current image is divided into two left and right sub-images, which are I ₁ and I ₂ respectively; the two sub-images I are calculated respectively The sum of pixel gray values of ₁ and I ₂ edge1 and edge2; if edge1>edge2, then B _rigth =l ₂ -h ₁ , B _left =B _rigth -N; when edge1<edge2, B _left =l ₂ + h ₂ , B _right = B _left +N; among them, l ₂ is the abscissa value of the zero-degree segmentation feature, h ₁ is the preset right border distance value, h ₂ is the preset left border distance value, and N is the preset set width value;

in, $\mathrm{edge}1=\underset{j=0}{\overset{N_1}{\mathrm{\&Sigma;}}}\underset{i=0}{\overset{m_1}{\mathrm{\&Sigma;}}}{I}_1(i,j),$ $\mathrm{edge}2=\underset{j=0}{\overset{N_2}{\mathrm{\&Sigma;}}}\underset{i=0}{\overset{m_2}{\mathrm{\&Sigma;}}}{I}_2(i,j);$ Where I ₁ (i, j) is the pixel gray value of the sub-image I ₁ at point (i, j), M ₁ , N ₁ are the height and width of I ₁ ; I ₂ (i, j) is the sub-image I ₂ is the pixel gray value at point (i, j), M ₂ , N ₂ is the height and width of the sub-image I ₂ .

According to the segmentation features with angles of -25° and 25°, determining the coordinate values of the upper boundary and the lower boundary of the feature area of the truck brake shoe includes: setting the straight line _l1 corresponding to the segmentation feature with an angle of -25° and the image The abscissa value of the intersection point of the upper boundary is top1, the abscissa value of the intersection point of the lower boundary is bot1; the abscissa value of the intersection point of the line l ₃ corresponding to the segmentation feature with an angle of 25° and the upper boundary of the image is top3, and the abscissa value of the intersection point of the lower boundary is bot3 , calculate the ratio ratio=|(top1-top3)/(bot1-bot3)|; when ratio<1, the brake shoe area of the truck is at the top of the current image; when ratio>1, the brake shoe area of the truck is at the current The bottom position of the image; according to the mechanical structure characteristics of the brake shoe brazing area of the truck, the coordinate values of the upper boundary and the lower boundary are extracted according to the relaxation principle. Wherein, the relaxation principle is an existing technology, which is used to ensure that the selected characteristic area of the brake shoe of the truck can fully represent the brake shoe of the truck, and its implementation method will not be repeated here.

Step 103: Extracting the features of the brake shoes of the freight train from the feature area of the brake shoes of the freight train in the current image.

Specifically, use the method based on background area estimation to segment the image in the feature area of the truck brake shoe in the current image to obtain a binary image of the truck brake shoe area; use the pixel labeling method to extract the maximum A connected area; using a Canny operator to extract the edge profile of the truck brake shoe from the binary image in the maximum connected area; extracting the features of the truck brake shoe according to the edge profile of the truck brake shoe.

Here, the method for estimating the background area is a prior art, which is not repeated here; the pixel marking method is a prior art, which is not repeated here; the Canny operator is a prior art, which is not repeated here.

The characteristics of the truck brake shoe include: the smooth eigenvalue of the edge profile of the truck brake shoe, the concave-convex eigenvalue of the edge profile of the truck brake shoe, the sawtooth degree of the edge profile of the truck brake shoe, the fixed value of the edge profile of the truck brake shoe, the Compactness of shoe edge profile, circularity of freight car brake shoe edge profile, aspect ratio of freight car brake shoe edge profile, area of freight car brake shoe edge profile and perimeter of freight car brake shoe edge profile.

Wherein, the calculation of the smooth eigenvalue of the edge profile of the truck brake shoe includes: calculating the center point of the edge profile of the truck brake shoe, and expressing the edge profile of the truck brake shoe as a polar coordinate with the center point as the center; Discretize the edge profile of the truck brake shoe expressed in polar coordinates to obtain the discretized truck brake shoe edge profile; after performing normalized calculation on the discretized truck brake shoe edge profile, perform Sorting, and then performing wavelet decomposition on the sorted results to obtain wavelet coefficients, and using wavelet coefficients to calculate smooth eigenvalues;

The calculation of the central point of the edge profile of the truck brake shoe may include: assuming that the edge profile of the truck brake shoe is x, then using the relative arc length s as a parameter, the edge profile of the truck brake shoe can be expressed as: where 0≤s≤1, is the edge profile of the truck brake shoe; then the formula for calculating the center point of the edge profile of the truck brake shoe is $\stackrel{\&Right\; Arrow;}{x_0}={\&Integral;}_0^1\stackrel{\&Right\; Arrow;}{x\left(\mathrm{the\; s}\right)}\mathrm{ds}=({\&Integral;}_0^{1}x\left(\mathrm{the\; s}\right)\mathrm{ds},{\&Integral;}_0^1\mathrm{the\; y}\left(\mathrm{the\; s}\right)\mathrm{ds}),$ where 0≤s≤1.

Said expressing the edge profile of the freight car brake shoe as polar coordinates with the central point as the center can be expressed as:

$\left\{\begin{array}{c}\mathrm{\&theta;}\left(\mathrm{the\; s}\right)=\&angle;(\stackrel{\&Right\; Arrow;}{x}\left(\mathrm{the\; s}\right)-\stackrel{\&Right\; Arrow;}{x_0})\\ r\left(\mathrm{the\; s}\right)={\left|\right|\stackrel{\&Right\; Arrow;}{x}\left(\mathrm{the\; s}\right)-\stackrel{\&Right\; Arrow;}{x_0}\left|\right|}_2\\ \stackrel{\&Right\; Arrow;}{z}\left(t\right)=(\mathrm{\&theta;}\left(\mathrm{the\; s}\right),\left(r\left(\mathrm{the\; s}\right)\right).\end{array}\right.,$ Among them, ∠ represents the polar angle of the vector in the polar coordinate system.

The discretization calculation of the edge profile of the truck brake shoe expressed in polar coordinate form may be to obtain the polar coordinates (r(s _i ), θ(s _i )) of each discrete point on the profile, where i=0, 1,...N-1, N is the number of discrete points.

The normalized calculation can use the formula:

The sorting according to preset rules may include: record θ'(s _i ) as θ'(i), r'(s _i ) as r'(i), and reorder r'(i), to obtain : R(i)=r′(η(i)), so that for There is θ(η(i))<θ(η(j));

Performing wavelet decomposition on the sorted results to obtain wavelet coefficients may include: $d\left(i\right)=r^{\&prime;}\left(\mathrm{\&eta;}\left(i\right)\right)\&CircleTimes;\mathrm{\&psi;}\left(i\right);$

The smooth feature value calculated by using wavelet coefficients may be: E=||d(i)|| ₂ .

The formula used for calculating the concave-convex eigenvalue of the edge profile of the freight car brake shoe is: Among them, c(n), n=1, 2····Sequence points of the edge profile of N truck brake shoes, c(n)=||c(n)-c ₀ ||, n=1, 2,... , N, $c_0=\frac{1}{N}\underset{\mathrm{no}=1}{\overset{N}{\mathrm{\&Sigma;}}}c\left(\mathrm{no}\right);$

The formula for calculating the rectangularity of the edge profile of the truck brake shoe can be: Wherein A ₀ is the area of the contour, and A _MER is the area of the smallest circumscribed rectangle of the contour, and the acquisition methods are all prior art, and will not be repeated here.

The formula for calculating the reliability of the edge profile of the freight car brake shoe can be: Among them, A represents the area of the contour, and CA represents the area of the smallest convex polygon. The acquisition methods are all existing technologies, and will not be repeated here.

The compactness calculation formula of the edge profile of the freight car brake shoe can be: Wherein, P represents the perimeter of the contour, and A represents the area, and the acquisition methods are all in the prior art, and will not be repeated here.

The formula for calculating the circularity of the edge profile of the freight car brake shoe can be:

in, _μR is the average distance from the center of gravity of the region to the boundary points. _δR is the mean square error of the distance from the center of gravity of the region to the boundary point.

The aspect ratio of the edge profile of the brake shoe of the truck is the ratio of the width to the length of the smallest circumscribed rectangle of the edge profile of the brake shoe of the truck;

The area of the edge contour of the brake shoe of the freight train can be obtained by counting the number of pixels within the contour range of the brake shoe edge of the freight train.

The perimeter of the edge profile of the brake shoe for a freight car may be: the boundary length of the edge profile of the brake shoe for a freight car.

Step 104: Using the SVM algorithm to calculate the feature value of the brake shoe of the truck to obtain the feature value of the current image, and determine whether there is a fault in the brake shoe of the truck according to the feature value and the preset fault identification value.

Here, the method for determining the fault identification value is: use the image of the brake shoe of the truck without fault and the image of the brake shoe of the truck to establish the positive and negative sample training sets respectively; Features, which form the nine-dimensional feature vector corresponding to the positive and negative sample training set, use the SVM algorithm to calculate the corresponding value of the positive and negative sample, and use the value corresponding to the negative sample training set as the fault identification value. Wherein, the implementation of the SVM algorithm is an existing technology, and details are not described here.

The determining whether there is a fault in the truck brake shoe according to the eigenvalue and the preset fault identification value includes: if the eigenvalue of the current image is equal to the fault identification value, then determining that the truck brake shoe is faulty; otherwise, determining that the truck brake shoe is not faulty.

As shown in Figure 2, the present invention provides an identification device for a truck brake shoe fault, which includes: a feature extraction module 21 and an identification module 22; wherein,

The feature extraction module 21 is used to extract the segmentation features of three angles from the current image, determine the feature area of the brake shoe of the current image according to the segmentation features of the three angles, and extract the feature area of the brake shoe of the truck from the current image. The characteristics of the brake shoe, sending the characteristics of the brake shoe of the freight car to the identification module 22;

The recognition module 22 is used to use the SVM algorithm to calculate the feature value of the current image on the features of the truck brake shoe sent by the feature extraction module 21, and determine whether the truck brake shoe is based on the feature value and the preset fault identification value. There is a glitch.

The feature extraction module 21 is specifically used to periodically extract the current image from the TDFS where it is located, perform three-angle grayscale projections on the current image to obtain three projection curves, filter all projection curves, and filter each projection curve The maximum value in the curve is used as the segmentation feature of the three angles of the current image. Wherein, the three angles are the set three angles of -25°, 0° and 25°.

The feature extraction module 21 is specifically used to expand the current image to obtain an expanded image; perform projection calculation on the expanded image, and discretize the image obtained by projection calculation to obtain a projection curve.

The feature extraction module 21 is specifically used to fill pixels with a grayscale of zero at the edge of the image to avoid errors caused by the edge of the image during projection, and to expand the current image, which can be calculated using the following formula:

$\&ForAll;(i,j)\&Element;R^2,$ $I^{\&prime;}(i,j)=\left\{\begin{array}{cc}I(i,j)& (i,j)\&Element;\mathrm{\&Omega;}\\ 0& (i,j)\&NotElement;\mathrm{\&Omega;}\end{array}\right.;$ Among them, Ω represents infinity.

The feature extraction module 21 is specifically used to perform projection calculation using the following formula:

in, $i=\sqrt{r^2+l^2}\cos [\mathrm{\&theta;}+a\mathrm{the\; tan}\left(\frac{l}{r}\right)],$ $j=\sqrt{r^2+l^2}\sin [\mathrm{\&theta;}+a\mathrm{the\; tan}\left(\frac{l}{r}\right)],$ Suppose a coordinate system is established, the origin of the coordinates is o, the height of the current image I is m, and the width is n, is the projection direction, the 2D image I will be projected to the projection line The starting point is o, and perpendicular to each other. r is any point on the projection line, p is a point passing through r, and Parallel straight line, l is any point (i, j) above the projection line The vertical distance of , I(i, j) is the pixel gray value of the image at point (i, j), and the angle θ between the projection line and the x-axis is the projection direction angle, θ is -25°0° or 25° .

The feature extraction module 21 is specifically used to discretize the image obtained by projection calculation using the following formula:

in, θ is -25° 0° and 25°, The height of the current image I is m, the width is n, δ _θ (r) represents the grayscale projection of the image on the projection direction angle θ, and it can reflect the grayscale change of the image in the direction of the projection direction angle θ.

The feature extraction module 21 is specifically configured to use the following formula to filter all projection curves:

$h\left(x\right)=\left\{\begin{array}{ccc}1& x_2\&le;x\&le;x_1& \\ -1& x_3\&le;x<x_1\&cup;& x_2<x\&le;x_4\end{array}\right.;$ Among them, x ₁ , x ₂ , x ₃ , x ₄ are constants, satisfying $\left\{\begin{array}{c}|x_1-x_2|=h\\ |x_3-x_4|=2h\end{array}\right.,$ H is the estimated width of the segmentation feature.

The feature extraction module 21 is specifically used to extract the edge information of the current image using the Canny operator, and determine the coordinate values of the left boundary and the right boundary of the truck brake shoe feature area according to the segmentation feature with an angle of zero degrees; ° and 25° segmentation features, determine the coordinate values of the upper boundary and lower boundary of the feature area of the brake shoe of the freight car.

The feature extraction module 21 is specifically used to divide the edge information of the current image into two left and right sub-images using a segmentation feature with an angle of zero, calculate the sum of the pixel gray values of the two sub-images respectively, and compare the two sub-images. The size of the sum of pixel gray values, if the sum of the pixel gray values of the left sub-image is larger, the coordinate value of the right boundary of the feature area of the truck brake shoe is equal to the abscissa value of the zero-degree segmentation feature, minus the preset The right boundary distance value, the coordinate value of the left boundary of the characteristic area of the truck brake shoe is equal to the coordinate value of the right boundary minus the preset width value; otherwise, the coordinate value of the left boundary of the characteristic area of the truck brake shoe is equal to that of the segmentation feature The abscissa value, plus the preset left boundary distance value, the coordinate value of the right boundary of the freight car brake shoe feature area is equal to the coordinate value of its left boundary plus the preset width value;

For example, assuming that the edge information of the current image is I, and the segmentation feature with an angle of zero degree is l ₂ in the figure, the edge information I of the current image is divided into two left and right sub-images, which are I ₁ and I ₂ respectively; the two sub-images I are calculated respectively The sum of pixel gray values of ₁ and I ₂ edge1 and edge2; if edge1>edge2, then B _rigth =l ₂ -h ₁ , B _left =B _rigth -N; when edge1<edge2, B _left =l ₂ + h ₂ , B _right = B _left +N; among them, l ₂ is the abscissa value of the zero-degree segmentation feature, h ₁ is the preset right border distance value, h ₂ is the preset left border distance value, and N is the preset set width value;

in, $\mathrm{edge}1=\underset{j=0}{\overset{N_1}{\mathrm{\&Sigma;}}}\underset{i=0}{\overset{m_1}{\mathrm{\&Sigma;}}}{I}_1(i,j),$ $\mathrm{edge}2=\underset{j=0}{\overset{N_2}{\mathrm{\&Sigma;}}}\underset{i=0}{\overset{m_2}{\mathrm{\&Sigma;}}}{I}_2(i,j);$ Where I ₁ (i, j) is the pixel gray value of the sub-image I ₁ at point (i, j), M ₁ , N ₁ are the height and width of I ₁ ; I ₂ (i, j) is the sub-image I ₂ is the pixel gray value at point (i, j), M ₂ , N ₂ is the height and width of the sub-image I ₂ .

The feature extraction module 21 is specifically used to set the angle as the line ₁₁ corresponding to the segmentation feature of -25 ° and the image upper boundary intersection abscissa value is top1, and the lower boundary intersection abscissa value is bot1; the angle is 25 ° The abscissa value of the intersection point of the straight line l ₃ corresponding to the segmentation feature and the upper boundary of the image is top3, and the abscissa value of the intersection point of the lower boundary is bot3, and the calculated ratio ratio=|(top1-top3)/(bot1-bot3)|; when ratio<1 When , the truck brake shoe area is located at the top of the current image; when ratio>1, the truck brake shoe area is located at the bottom of the current image; according to the mechanical structure characteristics of the truck brake shoe area, the upper boundary is extracted according to the relaxation principle and the coordinates of the lower boundary.

The feature extraction module 21 is specifically used to use a method based on background area estimation to segment the image in the feature area of the truck brake shoe in the current image to obtain a binary image of the truck brake shoe area; Extract the maximum connected area from the binary image of the tile area; use the Canny operator to extract the edge profile of the truck brake shoe from the binary image in the maximum connected area; extract the features of the truck brake shoe according to the edge profile of the truck brake shoe.

Here, the method for estimating the background area is a prior art, which is not repeated here; the pixel marking method is a prior art, which is not repeated here; the Canny operator is a prior art, which is not repeated here.

The feature extraction module 21 is specifically used to calculate the center point of the edge profile of the brake shoe of the truck, and express the edge profile of the brake shoe of the truck as a polar coordinate with the center point as the center; Perform discretization calculation on the edge profile of the brake shoe to obtain the discretized edge profile of the brake shoe of the truck; after performing normalized calculation on the discretized edge profile of the brake shoe of the truck, sort it according to the preset rules, and then sort it to get The results of wavelet decomposition to obtain wavelet coefficients, using wavelet coefficients to calculate the smooth eigenvalues;

Wherein, the calculation of the central point of the edge profile of the truck brake shoe may include: assuming that the edge profile of the truck brake shoe is x, then using the relative arc length s as a parameter, the edge profile of the truck brake shoe can be expressed as: where 0≤s≤1, is the edge profile of the truck brake shoe; then the formula for calculating the center point of the edge profile of the truck brake shoe is $\stackrel{\&Right\; Arrow;}{x_0}={\&Integral;}_0^1\stackrel{\&Right\; Arrow;}{x\left(\mathrm{the\; s}\right)}\mathrm{ds}=({\&Integral;}_0^{1}x\left(\mathrm{the\; s}\right)\mathrm{ds},{\&Integral;}_0^1\mathrm{the\; y}\left(\mathrm{the\; s}\right)\mathrm{ds}),$ where 0≤s≤1.

Said expressing the edge profile of the freight car brake shoe as polar coordinates with the central point as the center can be expressed as:

$\left\{\begin{array}{c}\mathrm{\&theta;}\left(\mathrm{the\; s}\right)=\&angle;(\stackrel{\&Right\; Arrow;}{x}\left(\mathrm{the\; s}\right)-\stackrel{\&Right\; Arrow;}{x_0})\\ r\left(\mathrm{the\; s}\right)={\left|\right|\stackrel{\&Right\; Arrow;}{x}\left(\mathrm{the\; s}\right)-\stackrel{\&Right\; Arrow;}{x_0}\left|\right|}_2\\ \stackrel{\&Right\; Arrow;}{z}\left(t\right)=(\mathrm{\&theta;}\left(\mathrm{the\; s}\right),\left(r\left(\mathrm{the\; s}\right)\right).\end{array}\right.,$ Among them, ∠ represents the polar angle of the vector in the polar coordinate system.

The discretization calculation of the edge profile of the truck brake shoe expressed in polar coordinate form may be to obtain the polar coordinates (r(s _i ), θ(s _i )) of each discrete point on the profile, where i=0, 1,...N-1, N is the number of discrete points.

The normalized calculation can use the formula:

The sorting according to preset rules may include: record θ'(s _i ) as θ'(i), r'(s _i ) as r'(i), and reorder r'(i), to obtain : R(i)=r′(η(i)), so that for There is θ(η(i))<θ(η(j));

Performing wavelet decomposition on the sorted results to obtain wavelet coefficients may include: $d\left(i\right)=r^{\&prime;}\left(\mathrm{\&eta;}\left(i\right)\right)\&CircleTimes;\mathrm{\&psi;}\left(i\right);$

The smooth feature value calculated by using wavelet coefficients may be: E=||d(i)|| ₂ .

The formula used for calculating the concave-convex eigenvalue of the edge profile of the freight car brake shoe is: Among them, c(n), n=1, 2····Sequence points of the edge profile of the N truck brake shoe, c′(n)=||c(n)-c ₀ ||, n=1, 2, ···, N, $c_0=\frac{1}{N}\underset{\mathrm{no}=1}{\overset{N}{\mathrm{\&Sigma;}}}c\left(\mathrm{no}\right);$

The formula for calculating the rectangularity of the edge profile of the truck brake shoe can be: where A ₀ is the area of the contour and A _MER is the area of the smallest circumscribing rectangle of the contour.

The formula for calculating the reliability of the edge profile of the freight car brake shoe can be: Among them, A represents the area of the contour, and CA is the area of the smallest convex polygon.

The compactness calculation formula of the edge profile of the freight car brake shoe can be: Among them, P represents the perimeter of the contour, and A represents the area.

The formula for calculating the circularity of the edge profile of the freight car brake shoe can be:

in, _μR is the average distance from the center of gravity of the region to the boundary points.

_δR is the mean square error of the distance from the center of gravity of the region to the boundary point.

The aspect ratio of the edge profile of the brake shoe of the truck is the ratio of the width to the length of the smallest circumscribed rectangle of the edge profile of the brake shoe of the truck;

The area of the edge contour of the brake shoe of the freight train can be obtained by counting the number of pixels within the contour range of the brake shoe edge of the freight train.

The perimeter of the edge profile of the brake shoe for a freight car may be: the boundary length of the edge profile of the brake shoe for a freight car.

The identification module 22 is specifically used to save the fault identification value.

The identification module 22 is specifically configured to determine that the brake shoe of the truck is faulty if the feature value of the current image is equal to the fault identification value; otherwise, determine that the brake shoe of the truck is not faulty.

The identification module 22 is also used to establish positive and negative sample training sets by using images of truck brake shoes without faults and images of faulty truck brake shoes; respectively extracting the features of all images in the positive and negative sample training sets to form The nine-dimensional feature vectors corresponding to the positive and negative sample training sets, the values corresponding to the positive and negative samples calculated using the SVM algorithm, and the values corresponding to the negative sample training sets are used as fault identification values.

The above device can be installed in the management device of TDFS as a logic module.

By using the method and device provided by the present invention, the test selects 75 truck brake shoe fault sample images, and 75 non-fault sample images set up a sample set for classifier training. For the adaptability of the test recognition algorithm, the idea of cross-validation (cross-validation) is adopted. Verification is also called cross-comparison. First, the training sample v is divided into equal parts, one of which is used as a test set, and the other v-1 is used as a training set. Rotate in turn until each sample has a test set, that is, v The process of training and prediction times. Therefore, the accuracy of cross-validation is the average value of v times) to carry out the fault recognition test, a total of 10 experiments are done, and the average value of the experimental results is the final recognition result, as shown in Figure 3. It can be seen that after using the method and device for identifying a truck brake shoe failure provided by the present invention, the identification time is less than two seconds, which is obviously less than manual identification through images, and the missed detection rate and false detection rate are very low.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention.

Claims (6)

1. A method for identifying a truck brake shoe failure, characterized in that the method comprises: Extract segmentation features from three angles from the current image; Determine the feature area of the brake shoe of the current image according to the segmentation features of the three angles; Extract the features of the truck brake shoe from the feature area of the truck brake shoe in the current image; Using the support vector machine SVM algorithm to calculate the characteristics of the truck brake shoe to obtain the feature value of the current image, and determine whether the truck brake shoe has a fault according to the feature value and the preset fault identification value; Wherein, before extracting the segmentation features of three angles from the current image, the method also includes: using the images of the brake shoe of the truck without fault and the image of the brake shoe of the truck to respectively establish positive and negative sample training sets; The features of all the images in the positive and negative sample training sets form the nine-dimensional feature vectors corresponding to the positive and negative sample training sets. The values corresponding to the positive and negative samples calculated using the SVM algorithm are used as the values corresponding to the negative sample training sets. fault identification value; Wherein, extracting the segmentation features of three angles from the current image includes: periodically extracting the current image from the fault image dynamic detection system TDFS, and performing three-angle grayscale projection on the current image to obtain three projection curves; for all The projection curve is filtered, and the maximum value of each filtered projection curve is used as the segmentation feature of the three angles of the current image; The grayscale projection includes: expanding the current image to obtain an expanded image; performing projection calculation on the expanded image, and discretizing the image obtained by the projection calculation to obtain a projection curve; The said determination of the freight car brake shoe characteristic region of the current image according to the segmentation features of the three angles includes: using the Canny operator to extract the edge information of the current image, and determining the left boundary and the The coordinate value of the right boundary; according to the segmentation features with angles of -25° and 25°, determine the coordinate values of the upper boundary and lower boundary of the characteristic area of the brake shoe of the truck; The determining the coordinate values of the left boundary and the right boundary of the feature area of the truck brake shoe according to the segmentation feature with an angle of zero degrees includes: using the segmentation feature with an angle of zero degrees to divide the edge information of the current image into two left and right sub-images, and calculating the two sub-images respectively. The sum of the pixel gray values of two sub-images, compare the sum of the pixel gray values of the two sub-images, if the sum of the pixel gray values of the left sub-image is larger, the right boundary of the truck brake shoe feature area The coordinate value is equal to the abscissa value of the segmentation feature of zero degrees, minus the preset right boundary distance value, and the coordinate value of the left boundary of the freight car brake shoe feature area is equal to the coordinate value of its right boundary minus the preset width value; otherwise, The coordinate value of the left boundary of the characteristic area of the truck brake shoe is equal to the abscissa value of the segmentation feature of zero degrees plus the preset left boundary distance value, and the coordinate value of the right boundary of the characteristic area of the truck brake shoe is equal to the coordinate value of the left boundary plus preset width value; According to the segmentation features with angles of -25° and 25°, determining the coordinate values of the upper boundary and the lower boundary of the feature area of the truck brake shoe includes: setting the straight line _l1 corresponding to the segmentation feature with an angle of -25° and the image The abscissa value of the intersection point of the upper boundary is top1, the abscissa value of the intersection point of the lower boundary is bot1; the abscissa value of the intersection point of the line l ₃ corresponding to the segmentation feature with an angle of 25° and the upper boundary of the image is top3, and the abscissa value of the intersection point of the lower boundary is bot3 , calculate the ratio ratio=|(top1-top3)/(bot1-bot3)|; when ratio<1, the brake shoe area of the truck is at the top of the current image; when ratio>1, the brake shoe area of the truck is at the current The bottom position of the image; according to the mechanical structure characteristics of the brake shoe brazing area of the truck, the coordinate values of the upper boundary and the lower boundary are extracted according to the relaxation principle. 2. The method according to claim 1, wherein said extracting the features of the truck brake shoe from the feature area of the truck brake shoe of the current image comprises: using a method based on background area estimation to convert the truck brake shoe of the current image The image segmentation in the brake shoe feature area obtains the binary image of the truck brake shoe area; uses the pixel marking method to extract the maximum connected area from the binary image of the truck brake shoe area; uses the Canny operator to extract the maximum connected area from the binary image in the maximum connected area Extracting the edge contour of the freight car brake shoe; extracting the features of the freight car brake shoe according to the edge contour of the freight car brake shoe. 3. The method according to claim 2, wherein the characteristics of the freight car brake shoe include: the smooth eigenvalue of the edge profile of the freight car brake shoe, the concave-convex eigenvalue of the edge profile of the freight car brake shoe, the The degree of sawtooth, the fixed value of the edge profile of the truck brake shoe, the compactness of the edge profile of the truck brake shoe, the circularity of the edge profile of the truck brake shoe, the aspect ratio of the edge profile of the truck brake shoe, the area of the edge profile of the truck brake shoe and The perimeter of the profile of the edge of the truck brake shoe. 4. An identification device for a truck brake shoe fault, characterized in that the device comprises: a feature extraction module and an identification module; wherein, The feature extraction module is used to extract the segmentation features of three angles from the current image, determine the feature area of the truck brake shoe in the current image according to the segmentation features of the three angles, and extract the truck brake shoe feature area from the current image. The feature of the tile, sending the feature of the truck brake shoe to the recognition module; The recognition module is used to use the SVM algorithm to calculate the feature value of the current image based on the features of the truck brake shoe sent by the feature extraction module, and determine whether there is a fault in the truck brake shoe according to the feature value and the preset fault identification value ; The recognition module is also used to establish positive and negative sample training sets respectively by using the images of the brake shoe of the truck without fault and the image of the brake shoe of the truck with fault; respectively extract the features of all the images in the training set of positive and negative samples to form a positive , the nine-dimensional feature vector corresponding to the negative sample training set, the values corresponding to the positive and negative samples calculated using the SVM algorithm, and the value corresponding to the negative sample training set as the fault identification value; The feature extraction module is specifically used to periodically extract the current image from the TDFS where it is located, and perform grayscale projection of three angles on the current image to obtain three projection curves; filter all projection curves, and filter each projection curve The maximum values in all are used as the segmentation features of the three angles of the current image; the grayscale projection includes: expanding the current image to obtain an expanded image; performing projection calculation on the expanded image, and performing projection calculation on the image obtained by the projection calculation. Discrete to get the projection curve; The feature extraction module is also specifically used to extract the edge information of the current image using the Canny operator, and determine the coordinate values of the left boundary and the right boundary of the truck brake shoe feature area according to the segmentation feature with an angle of zero degrees; ° and 25° segmentation features, determine the coordinate values of the upper boundary and lower boundary of the feature area of the brake shoe of the freight car; The determining the coordinate values of the left boundary and the right boundary of the feature area of the truck brake shoe according to the segmentation feature with an angle of zero degrees includes: using the segmentation feature with an angle of zero degrees to divide the edge information of the current image into two left and right sub-images, and calculating the two sub-images respectively. The sum of the pixel gray values of two sub-images, compare the sum of the pixel gray values of the two sub-images, if the sum of the pixel gray values of the left sub-image is larger, the right boundary of the truck brake shoe feature area The coordinate value is equal to the abscissa value of the segmentation feature of zero degrees, minus the preset right boundary distance value, and the coordinate value of the left boundary of the freight car brake shoe feature area is equal to the coordinate value of its right boundary minus the preset width value; otherwise, The coordinate value of the left boundary of the characteristic area of the truck brake shoe is equal to the abscissa value of the segmentation feature of zero degrees, plus the preset left boundary distance value, and the coordinate value of the right boundary of the characteristic area of the truck brake shoe is equal to the coordinate value of the left boundary plus preset width value; According to the segmentation features with angles of -25° and 25°, determining the coordinate values of the upper boundary and the lower boundary of the feature area of the truck brake shoe includes: setting the straight line _l1 corresponding to the segmentation feature with an angle of -25° and the image The abscissa value of the intersection point of the upper boundary is top1, the abscissa value of the intersection point of the lower boundary is bot1; the abscissa value of the intersection point of the line l ₃ corresponding to the segmentation feature with an angle of 25° and the upper boundary of the image is top3, and the abscissa value of the intersection point of the lower boundary is bot3 , calculate the ratio ratio=|(top1-top3)/(bot1-bot3)|; when ratio<1, the brake shoe area of the truck is at the top of the current image; when ratio>1, the brake shoe area of the truck is at the current The bottom position of the image; according to the mechanical structure characteristics of the brake shoe brazing area of the truck, the coordinate values of the upper boundary and the lower boundary are extracted according to the relaxation principle. 5. The device according to claim 4, wherein the feature extraction module is specifically configured to use a method based on background area estimation to segment the image in the characteristic region of the truck brake shoe of the current image to obtain the truck brake shoe Region binary image; using pixel notation method, extracting the maximum connected area from the binary image of the truck brake shoe area; using Canny operator, extracting the edge profile of the truck brake shoe from the binary image in the maximum connected area; according to the truck brake Extraction of brake shoe features from the tile edge profile. 6. The device according to claim 4, wherein the feature extraction module is specifically used to extract the smooth feature value of the edge profile of the brake shoe of the truck, the concave-convex feature value of the edge profile of the brake shoe of the truck, and the edge profile of the brake shoe of the truck. The sawtooth degree, the fixed value of the edge profile of the truck brake shoe, the compactness of the edge profile of the truck brake shoe, the circularity of the edge profile of the truck brake shoe, the aspect ratio of the edge profile of the truck brake shoe, the area of the edge profile of the truck brake shoe and the perimeter of the edge profile of the truck brake shoe as a feature of the truck brake shoe.