CN114648520A - Method, system, electronic device and storage medium for detecting track defects - Google Patents

Abstract

The invention discloses a track defect detection method, system, electronic equipment and storage medium. The track defect detection method comprises: separately performing singular value decomposition, saliency detection and gradient amplitude calculation on the track image to be processed to obtain several corresponding feature images; performing weighted fusion on the several feature images to obtain a weighted fusion feature image; determining a defect area in the track image to be processed based on the weighted fusion feature image. The track defect detection method provided by the present invention performs singular value decomposition, saliency detection and gradient amplitude calculation processing on the track image to be processed, and weights and fuses several obtained feature images, thereby realizing the combination of visual multi-feature fusion. The unsupervised machine vision processing method can efficiently detect the surface defects of railway tracks, and at the same time, it is suitable for the detection of complex defects of various scales and shapes.

Description

Track defect detection method, system, electronic device and storage medium

technical field

The invention relates to the technical field of rail transit operation and maintenance inspection, in particular to a rail defect detection method, system, electronic device and storage medium.

Background technique

The rapid development of railway construction has made the total mileage of transportation increase year by year. However, while enhancing the passenger and freight capacity, it also brings high cost of patrol operation and maintenance. Among them, due to the continuous high-strength extrusion and friction with the wheels, as well as the long-term erosion of natural factors such as rain, snow, wind and frost, the surface of the rail track will have many pit defects of different degrees, which will affect the stability of the train and cause the running of the train. Security risks. Therefore, regular inspection of rail surface defects is necessary.

The traditional rail surface defect detection mainly relies on artificial visual inspection to check the more serious rail sections. However, this method is not only inefficient and time-consuming, but also prone to human omissions. In contrast, using machine vision technology to perform automatic defect detection on the scanned rail surface images can greatly improve work efficiency, shorten time periods, and significantly reduce labor costs.

The track defect detection method used in the prior art is to perform image reconstruction on the track image to obtain the difference image thereof, and obtain the defect area based on the difference information between the original image and the reconstructed image. However, this method has the following defects: (1) The image reconstruction is very sensitive to the selection of the reconstruction algorithm model and its parameters, and the quality of the reconstructed image is very easily disturbed by the background clutter. The adaptability is poor; and (2) the image difference method is greatly interfered by background noise and texture clutter, and it is difficult to judge and distinguish small defects and background clutter, and it is easy to introduce false alarms and cause false detection.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a method, system and electronic device for detecting track defects in order to overcome the defects of low quality of reconstructed images, resistance to false detection and easy occurrence of false detections when using image reconstruction for track defect detection in the prior art. equipment and storage media.

The present invention solves the above-mentioned technical problems through the following technical solutions:

In a first aspect, the present invention provides a method for detecting a track defect, and the method for detecting a track defect includes:

Perform singular value decomposition, saliency detection and gradient magnitude calculation on the track image to be processed to obtain several corresponding feature images;

performing weighted fusion on the several feature images to obtain weighted fusion feature images;

Defect areas in the to-be-processed track image are determined based on the weighted fusion feature image.

Preferably, the several feature images include singular pixel contrast-enhanced feature images;

The steps of the singular value decomposition specifically include:

Calculate the difference between the grayscale values of several target pixels and reference pixels in the track image to be processed to obtain a singular pixel contrast enhancement feature image, wherein the difference between the reference pixel and the target pixel is The distance is less than the first threshold;

And/or, the several feature images include singular line contrast-enhanced feature images;

The steps of the singular value decomposition specifically include:

Calculate the difference between the grayscale values of several target pixel rows and reference pixel rows in the track image to be processed to obtain a singular row contrast-enhanced feature image, wherein the difference between the reference pixel row and the target pixel row is The distance is less than the second threshold.

Preferably, the several feature images include visual saliency feature images;

The steps of the significance detection specifically include:

Fourier transform and phase spectrum calculation are performed on the track image to be processed to obtain a visual saliency map;

Smooth filtering is performed on the visual saliency map to obtain a visual saliency feature image.

Preferably, the several feature images include probability distribution images used to characterize non-edge regions;

The steps of calculating the gradient amplitude specifically include:

Calculate the gradient magnitude of each pixel in the track image to be processed to obtain a gradient magnitude image;

Perform smooth filtering and inversion operations on the gradient magnitude image to obtain the probability distribution image.

Preferably, the step of determining the defect area in the track image to be processed based on the weighted fusion feature image includes:

Perform adaptive threshold segmentation on the weighted fusion feature image to obtain a binary image;

A mathematical morphological opening operation is performed on the binary image to determine the defect area.

Preferably, the method also includes:

Get the initial orbit image;

Filter processing is performed on the initial track image to obtain the track image to be processed.

Preferably, the filtering process includes L ₀ gradient minimization filtering and median filtering.

In a second aspect, the present invention provides a system for detecting track defects, the system for detecting track defects includes:

The image processing module is used to separately perform singular value decomposition, saliency detection and gradient magnitude calculation on the track image to be processed to obtain several corresponding feature images;

a feature fusion module, for performing weighted fusion on the several feature images to obtain a weighted fusion feature image;

A defect detection module, configured to determine a defect area in the track image to be processed based on the weighted fusion feature image.

In a third aspect, the present invention provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implements the above-mentioned track defect when executing the computer program detection method.

In a fourth aspect, the present invention provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the above-mentioned method for detecting a track defect.

The positive improvement effect of the present invention is that: the track defect detection method provided by the present invention performs singular value decomposition, saliency detection and gradient amplitude calculation processing on the track image to be processed, and performs weighted fusion of several obtained feature images, In this way, an unsupervised machine vision processing method combined with visual multi-feature fusion is realized, which can efficiently detect the surface defects of railway tracks, and at the same time effectively overcome the background clutter interference in the track image under the load environment, and has strong adaptability to the environment and detection effect. It is stable and suitable for complex track defect detection of various scales and forms; and it does not need to collect samples for training in practical applications, and has strong applicability and low application costs.

Description of drawings

FIG. 1 is a first schematic flowchart of a method for detecting a track defect according to Embodiment 1 of the present invention.

FIG. 2 is an initial track image scanned in according to Embodiment 1 of the present invention.

FIG. 3 is a partial schematic flowchart of the method for detecting track defects according to Embodiment 1 of the present invention.

FIG. 4 is a filtered track image to be processed according to Embodiment 1 of the present invention.

FIG. 5 is a first schematic flowchart of singular value decomposition in step S1 of the track defect detection method according to Embodiment 1 of the present invention.

FIG. 6 is a contrast-enhanced feature image of singular pixels obtained through singular value decomposition according to Embodiment 1 of the present invention.

FIG. 7 is a second schematic flowchart of singular value decomposition in step S1 of the method for detecting a track defect according to Embodiment 1 of the present invention.

FIG. 8 is a singular line contrast-enhanced feature image obtained by singular value decomposition according to Embodiment 1 of the present invention.

FIG. 9 is a schematic flowchart of the significance detection in step S1 of the track defect detection method according to Embodiment 1 of the present invention.

FIG. 10 is a visual saliency feature image obtained through saliency detection according to Embodiment 1 of the present invention.

FIG. 11 is a schematic flowchart of gradient amplitude calculation in step S1 of the method for detecting track defects according to Embodiment 1 of the present invention.

FIG. 12 is a probability distribution image of a non-edge region obtained through gradient magnitude calculation according to Embodiment 1 of the present invention.

FIG. 13 is a multi-feature weighted fusion feature image obtained by weighted feature fusion according to Embodiment 1 of the present invention.

FIG. 14 is a schematic flowchart of sub-steps of step S3 of the track defect detection method according to Embodiment 1 of the present invention.

FIG. 15 is a schematic diagram of a defect area obtained after threshold segmentation and morphological processing according to Embodiment 1 of the present invention.

FIG. 16 is a schematic block diagram of a track defect detection system according to Embodiment 2 of the present invention.

FIG. 17 is a schematic structural diagram of an electronic device for implementing a method for detecting track defects according to Embodiment 3 of the present invention.

Detailed ways

The present invention is further described below by way of examples, but the present invention is not limited to the scope of the described examples.

Example 1

The present embodiment discloses a method for detecting rail defects, which is based on the fusion processing of various features of computer vision, and can effectively detect the scanned image of the rail surface to determine the defect area on the rail surface.

Specifically, as shown in Figure 1, the detection method of the track defect includes:

S1. Perform singular value decomposition, saliency detection and gradient magnitude calculation on the track image to be processed, respectively, to obtain several corresponding feature images;

S2, weighted fusion of several feature images to obtain weighted fusion feature images;

S3. Determine the defect area in the track image to be processed based on the weighted fusion feature image.

Usually, the input scanning initial track image I is shown in Figure 2, which includes many interference factors. If it is directly processed, it will seriously reduce the accuracy of the final detection result. Therefore, as a preferred embodiment , as shown in Figure 3, the above method also includes:

S101, obtain initial orbit image I;

S102. Perform filtering processing on the initial track image I to obtain the track image I ⁰ to be processed.

Preferably, L ₀ gradient minimization filtering and median filtering are used in sequence to perform image preprocessing to eliminate background clutter and noise interference, thereby obtaining the track image I ⁰ to be processed as shown in FIG. 4 .

Specifically, the directly obtained input scan image I is firstly subjected to L ₀ gradient minimization filtering to obtain a filtered image S, and then the image S is filtered with a window scale of w _m × h _m to obtain the track image I to be processed. ⁰ , where w _m represents the window width of the median filter and h _m represents the window height of the median filter.

计算该点与其相邻像素在x轴和y轴方向的灰度差异，定义梯度测度表达式为Among them, the calculation method of L ₀ gradient minimization filter is: for a certain pixel p in the image, I _p and S _p represent the pixel values of image I and image S at point p, respectively, and the gradient

Calculate the grayscale difference between the point and its adjacent pixels in the x-axis and y-axis directions, and define the gradient measure expression as

Then the objective function expression of L ₀ gradient minimization filter is:

表示对X方向计算偏导，

表示对Y方向计算偏导，#{·}表示获取满足条件的像素点的数量，λ表示权重系数。in,

Indicates that the partial derivative is calculated in the X direction,

Indicates that the partial derivative is calculated in the Y direction, #{·} indicates the number of pixels that satisfy the condition, and λ indicates the weight coefficient.

In this embodiment, the value of the weight coefficient λ is 0.05, and the value of the median filter window scale is 3×8. It should be noted that the above values are only used for illustration, and are not limited to this. In the specific implementation process, the weight coefficient λ and the median filter window scale can be arbitrarily selected according to actual needs.

In this embodiment, the track image I ⁰ to be processed includes three different track defects a, b, and c, respectively. The direction parallel to the track is taken as the X direction of the track image I ⁰ to be processed, and the direction perpendicular to the track is taken as the Y direction of the track image I ⁰ to be processed.

For step S1, after performing singular value decomposition, saliency detection and gradient magnitude calculation on the track image I ⁰ to be processed respectively, several corresponding feature images I ¹ , I ² , I ³ . . . In can ^be obtained.

In a preferred embodiment, the above-mentioned several feature images I ¹ , I ² , I ³ . . . In include singular pixel contrast ^- enhanced feature images I ¹ ;

Step S1, as shown in Figure 5, specifically includes:

S11. Calculate the difference between the grayscale values of several target pixels in the track image I ⁰ to be processed and the reference pixels to obtain the singular pixel contrast enhancement feature image I ¹ , wherein the difference between the reference pixel and the target pixel is The distance is less than the first threshold.

Step S11 includes statistical calculation of the difference between each target pixel point and the gray value of the reference pixel point in the X direction in the local area where it is located, so as to obtain the singular pixel contrast enhancement feature image ^I1 of the track image ^I0 to be processed (as shown in Figure 6 ). shown). Wherein, the local area is determined by the position of the reference pixel point whose distance from the target pixel point does not exceed the first threshold, and its specific range can be arbitrarily set as required.

则奇异像素对比增强特征图像I¹中对应像素点p的灰度值为Specifically, the calculation method of the singular pixel contrast-enhanced feature image I ¹ is: set the coordinates and grayscale values of the pixel p in the track image I ⁰ to be processed as (x _p , y _p ) and

Then the gray value of the corresponding pixel p in the singular pixel contrast enhancement feature image I ¹ is

Wherein, Δ is a positive integer representing the first threshold, and in this embodiment, Δ takes an integer value of 0.1 times the length of the track image to be processed in the I ⁰ X direction. It should be noted that the value of Δ here is only used for illustration, and is not limited to this. In the specific implementation process, Δ can be any positive integer value according to actual needs.

In a preferred embodiment, the above-mentioned several feature images I ¹ , I ² , I ³ . . . In include a singular line contrast ^- enhanced feature image I ² ;

Step S1, as shown in Figure 7, specifically includes:

S12. Calculate the difference between the grayscale values of several target pixel rows and the reference pixel row in the track image I ⁰ to be processed, to obtain the singular row contrast-enhanced feature image I ² , wherein the difference between the reference pixel row and the target pixel row is The distance is less than the second threshold.

Step S12 includes statistically calculating the difference between the grayscale values of each target pixel row in the X direction and the reference pixel row in the X direction in the local area where it is located, so as to obtain the singular pixel row contrast enhancement feature image ^I2 of the track image ^I0 to be processed. (as shown in Figure 8). The local area is determined by the position of the reference pixel row whose distance from the target pixel row does not exceed the second threshold, and its specific range can be arbitrarily set as required.

Specifically, the calculation method of the singular pixel row contrast enhancement feature image I ² is: assuming that all gray values of the target pixel row x in the track image I ⁰ to be processed constitute a feature vector f _x ∈ R ^Y , then the singular pixel row contrast enhancement feature The grayscale value of each pixel point corresponding to the target pixel row x in the image I ² is

Wherein, Δ is a positive integer representing the second threshold. In this embodiment, Δ takes an integer value of 0.1 times the length of the track image to be processed in the I ⁰ X direction. It should be noted that the value of Δ here is only used for illustration, and is not limited to this. In the specific implementation process, Δ can be any positive integer value according to actual needs.

In a preferred embodiment, the above ^- mentioned several feature images I ¹ , I ² , I ³ . . . In include a visually significant feature image I ³ ;

Step S1, as shown in Figure 9, specifically includes:

S131, performing Fourier transform and phase spectrum calculation on the track image I ⁰ to be processed to obtain a visual saliency map;

S132. Perform smooth filtering on the visual saliency map to obtain a visual saliency feature image I ³ .

FIG. 10 shows the above-mentioned visual saliency feature image I ³ , wherein the specific calculation method of the visual saliency feature image I ³ is: first, extract the Fourier phase spectrum of the track image I ⁰ to be processed

P=P{F(I ⁰ )}

Among them, F(·) represents the Fourier transform, and P(·) represents the phase spectrum calculation.

Then, the inverse Fourier transform is performed according to the phase spectrum P to obtain the visual saliency map

SF=||F ^-1 (e ^i·P )|| ²

Here, F ^-1 (·) represents the Fourier transform.

Finally, a Gaussian smoothing filter is applied to the visual saliency map SF to obtain a visual saliency feature image I ³ .

In this embodiment, the window size of the Gaussian smoothing filter is an integer value of 0.2 times the length of the track image to be processed in the I ⁰ Y direction, and the Gaussian kernel σ takes a value of 0.25 times the size of the filter window. It should be noted that the above values are only used for illustration, and are not limited to this. In the specific implementation process, the window size of the Gaussian smoothing filter and the Gaussian kernel σ can be arbitrarily selected according to actual needs.

In a preferred embodiment, the above-mentioned several feature images I ¹ , I ² , I ³ . . . In include a probability distribution image I ⁴ for characterizing non ^- edge regions;

Step S1, as shown in Figure 11, specifically includes:

S141, calculate the gradient magnitude of each pixel in the track image I ⁰ to be processed to obtain a gradient magnitude image;

S142. Perform smooth filtering and inversion operations on the gradient magnitude image to obtain a probability distribution image I ⁴ .

FIG. 12 shows the above probability distribution image I ⁴ , wherein the specific calculation method of the probability distribution image I ⁴ is as follows: first, the gradient magnitude of each pixel in the track image I ⁰ to be processed is calculated to obtain the track image I ⁰ to be processed. The corresponding gradient magnitude image is then subjected to Gaussian smoothing filtering, and finally the inversion operation is used to obtain the probability distribution image I ⁴ representing the non-edge region.

In this embodiment, the window size of the Gaussian smoothing filter is 5×10, and the Gaussian kernel σ=3.0. It should be noted that the above values are only used for illustration, and are not limited to this. In the specific implementation process, the window size of the Gaussian smoothing filter and the Gaussian kernel σ can be arbitrarily selected according to actual needs.

For step S2, the geometric weighting calculation method is used for the singular pixel contrast enhancement feature image I ¹ , the singular line contrast enhancement feature image I ² , the visual saliency feature image I ³ and the non-edge probability distribution image I ⁴ obtained above to obtain multiple Feature weighted ^fusion feature image If, as shown in Figure 13.

Specifically, the calculation expression of the multi-feature weighted ^fusion feature image If is:

Among them, N(·) represents numerical normalization, and α ₁ , α ₂ , α ₃ , and α ₄ respectively represent the geometric weights corresponding to each feature image.

In this embodiment, the values of the weights α ₁ , α ₂ , α ₃ , and α ₄ are 1.1, 1.0, 0.9, and 1.0, respectively. It should be noted that the above values are only used for illustration, and are not limited thereto. In the specific implementation process, the geometric weight corresponding to each feature image can be arbitrarily selected according to actual needs.

For step S3, in a preferred embodiment, as shown in Figure 14, it includes:

S31, perform adaptive threshold segmentation on the weighted fusion feature image I ^f to obtain a binary image;

S32, perform mathematical morphological opening operation on the binary image to determine the defect area.

Specifically, the adaptive threshold expression is

T(I ^f )=mean(I ^f )+k·std(I ^f )

Among them, mean( ) represents the mean value of the pixel grayscale, std( ) represents the standard deviation of the pixel grayscale, and k is the standard deviation weight coefficient, which usually ranges from 3.0 to 5.0.

The image effect after threshold segmentation and morphological processing is shown in Figure 15, where the white area is the final determined defect area.

In this embodiment, the value of k is 4.0, and in the mathematical morphology opening operation after binary segmentation, a 4×4 scale circular template is used for the erosion operation, and then a 7×7 scale circular template is used for expansion operate.

It should be noted that the detection method for rail defects provided by the present invention has been verified by experiments, and the detection accuracy of the surface defects of the rail is relatively high. In the performance experiment, by testing 195 pieces of rail surface scanning image data with a total of 326 defects in various shapes, sizes and appearances, a total of 338 defects were detected, including 309 real defects and 29 false detections. Therefore, the detection accuracy rate is 91.4%, and the detection recall rate is less than 94.8%.

Therefore, the track defect detection method provided by the present invention performs singular value decomposition, saliency detection and gradient amplitude calculation processing on the track image to be processed, and weights and fuses several obtained feature images, thereby realizing the combination of visual multiple The unsupervised machine vision processing method of feature fusion can efficiently detect the surface defects of railway tracks, and at the same time effectively overcome the background clutter interference in the track image under the load environment, has strong adaptability to the environment, stable detection effect, and is suitable for a variety of It can detect complex track defects of scale and shape; and it does not need to collect samples for training in practical application, and has strong applicability and low application cost.

Example 2

This embodiment discloses a track defect detection system, which is based on computer vision-based multi-feature fusion processing, and can effectively detect the scanned image of the track surface to determine the defect area on the track surface.

Specifically, as shown in Figure 16, the detection system for the track defect includes:

The image processing module 1 is used to separately perform singular value decomposition, saliency detection and gradient magnitude calculation on the track image to be processed, so as to obtain several corresponding feature images;

Feature fusion module 2, used for weighted fusion of several feature images to obtain weighted fusion feature images;

The defect detection module 3 is used to determine the defect area in the track image to be processed based on the weighted fusion feature image.

Usually, the input scanning initial track image I is shown in Figure 2, which includes many interference factors. If it is directly processed, it will seriously reduce the accuracy of the final detection result. Therefore, as a preferred embodiment , as shown in Figure 3, the above system also includes a preprocessing module 4 for:

Get the initial orbit image I;

The initial track image I is filtered to obtain the track image I ⁰ to be processed.

Preferably, the preprocessing module 4 sequentially uses L ₀ gradient minimization filtering and median filtering to perform image preprocessing to eliminate background clutter and noise interference, thereby obtaining the track image I ⁰ to be processed as shown in FIG. 4 .

Specifically, the preprocessing module 4 first performs L ₀ gradient minimization filtering on the directly obtained input scan image I to obtain a filtered image S, and then applies w _m × h _m window scale median filtering to the image S to obtain the Process the track image I ⁰ , where w _m represents the window width of the median filter and h _m represents the window height of the median filter.

表示该点与其相邻像素在x轴和y轴方向的灰度差异，定义梯度测度表达式为Among them, the calculation method of L ₀ gradient minimization filter is: for a certain pixel p in the image, I _p and S _p represent the pixel values of image I and image S at point p, respectively, and the gradient

Represents the grayscale difference between the point and its adjacent pixels in the x-axis and y-axis directions, and the gradient measure expression is defined as

Then the objective function expression of L ₀ gradient minimization filter is:

表示对X方向计算偏导，

表示对Y方向计算偏导，#{·}表示获取满足条件的像素点的数量，λ表示权重系数。in,

Indicates that the partial derivative is calculated in the X direction,

Indicates that the partial derivative is calculated in the Y direction, #{·} indicates the number of pixels that satisfy the condition, and λ indicates the weight coefficient.

In this embodiment, the value of the weight coefficient λ is 0.05, and the value of the median filter window scale is 3×8. It should be noted that the above values are only used for illustration, and are not limited to this. In the specific implementation process, the weight coefficient λ and the median filter window scale can be arbitrarily selected according to actual needs.

In this embodiment, the track image I ⁰ to be processed includes three different track defects a, b, and c, respectively. The direction parallel to the track is taken as the X direction of the track image I ⁰ to be processed, and the direction perpendicular to the track is taken as the Y direction of the track image I ⁰ to be processed.

The image processing module ¹ can obtain several corresponding feature images I ¹ , I ² , ^{I 3} .

In a preferred embodiment, the above-mentioned several feature images I ¹ , I ² , I ³ . . . In include singular pixel contrast ^- enhanced feature images I ¹ ;

The image processing module 1 is specifically used for:

Calculate the difference between the gray values of several target pixels and reference pixels in the track image I ⁰ to be processed to obtain the singular pixel contrast enhancement feature image I ¹ , wherein the distance between the reference pixel and the target pixel is less than first threshold.

Specifically, the difference between each target pixel point and the gray value of the reference pixel point in the X direction in the local area where it is located is statistically calculated to obtain the singular pixel contrast enhancement feature image I ¹ of the track image I ⁰ to be processed (as shown in Figure 6 ). shown). Wherein, the local area is determined by the position of the reference pixel point whose distance from the target pixel point does not exceed the first threshold, and its specific range can be arbitrarily set as required.

则奇异像素对比增强特征图像I¹中对应像素点p的灰度值为Specifically, the calculation method of the singular pixel contrast-enhanced feature image I ¹ is: set the coordinates and grayscale values of the pixel p in the track image I ⁰ to be processed as (x _p , y _p ) and

Then the gray value of the corresponding pixel p in the singular pixel contrast enhancement feature image I ¹ is

Wherein, Δ is a positive integer representing the first threshold, and in this embodiment, Δ takes an integer value of 0.1 times the length of the track image to be processed in the I ⁰ X direction. It should be noted that the value of Δ here is only used for illustration, and is not limited to this. In the specific implementation process, Δ can be any positive integer value according to actual needs.

In a preferred embodiment, the above-mentioned several feature images I ¹ , I ² , I ³ . . . In include a singular line contrast ^- enhanced feature image I ² ;

The image processing module 1 is specifically used for:

Calculate the difference between the grayscale values of several target pixel rows and the reference pixel row in the track image I ⁰ to be processed to obtain the singular row contrast-enhanced feature image I ² , wherein the distance between the reference pixel row and the target pixel row is less than second threshold.

Specifically, the difference between the grayscale values of each target pixel row in the X direction and the reference pixel row in the X direction in the local area where it is located is calculated statistically to obtain the singular pixel row contrast enhancement feature image ^I2 of the track image ^I0 to be processed. (as shown in Figure 8). The local area is determined by the position of the reference pixel row whose distance from the target pixel row does not exceed the second threshold, and its specific range can be arbitrarily set as required.

Specifically, the calculation method of the singular pixel row contrast enhancement feature image I ² is: assuming that all gray values of the target pixel row x in the track image I ⁰ to be processed constitute a feature vector f _x ∈ R ^Y , then the singular pixel row contrast enhancement feature The grayscale value of each pixel point corresponding to the target pixel row x in the image I ² is

Wherein, Δ is a positive integer representing the second threshold. In this embodiment, Δ takes an integer value of 0.1 times the length of the track image to be processed in the I ⁰ X direction. It should be noted that the value of Δ here is only used for illustration, and is not limited to this. In the specific implementation process, Δ can be any positive integer value according to actual needs.

In a preferred embodiment, the above ^- mentioned several feature images I ¹ , I ² , I ³ . . . In include a visually significant feature image I ³ ;

The image processing module 1 is specifically used for:

Perform Fourier transform and phase spectrum calculation on the track image I ⁰ to be processed to obtain a visual saliency map;

The visual saliency map is smoothed and filtered to obtain the visual saliency feature image I ³ .

FIG. 10 shows the above-mentioned visual saliency feature image I ³ , wherein the specific calculation method of the visual saliency feature image I ³ is: first, extract the Fourier phase spectrum of the track image I ⁰ to be processed

P=P{F(I ⁰ )}

Among them, F(·) represents the Fourier transform, and P(·) represents the phase spectrum calculation.

Then, the inverse Fourier transform is performed according to the phase spectrum P to obtain the visual saliency map

SF=||F ^-1 (e ^i·P )|| ²

Here, F ^-1 (·) represents the Fourier transform.

Finally, a Gaussian smoothing filter is applied to the visual saliency map SF to obtain a visual saliency feature image I ³ .

In this embodiment, the window size of the Gaussian smoothing filter is an integer value of 0.2 times the length of the track image to be processed in the I ⁰ Y direction, and the Gaussian kernel σ takes a value of 0.25 times the size of the filter window. It should be noted that the above values are only used for illustration, and are not limited to this. In the specific implementation process, the window size of the Gaussian smoothing filter and the Gaussian kernel σ can be arbitrarily selected according to actual needs.

In a preferred embodiment, the above-mentioned several feature images I ¹ , I ² , I ³ . . . In include a probability distribution image I ⁴ for characterizing non ^- edge regions;

The image processing module 1 is specifically used for:

Calculate the gradient magnitude of each pixel in the track image I ⁰ to be processed to obtain a gradient magnitude image;

Smooth filtering and inversion operations are performed on the gradient magnitude image to obtain a probability distribution image I ⁴ .

FIG. 12 shows the above probability distribution image I ⁴ , wherein the specific calculation method of the probability distribution image I ⁴ is as follows: first, the gradient magnitude of each pixel in the track image I ⁰ to be processed is calculated to obtain the track image I ⁰ to be processed. The corresponding gradient magnitude image is then subjected to Gaussian smoothing filtering, and finally the inversion operation is used to obtain the probability distribution image I ⁴ representing the non-edge region.

In this embodiment, the window size of the Gaussian smoothing filter is 5×10, and the Gaussian kernel σ=3.0. It should be noted that the above values are only used for illustration, and are not limited to this. In the specific implementation process, the window size of the Gaussian smoothing filter and the Gaussian kernel σ can be arbitrarily selected according to actual needs.

The feature fusion module 2 adopts the geometric weighting calculation method for the singular pixel contrast enhanced feature image I ¹ , the singular line contrast enhanced feature image I ² , the visual saliency feature image I ³ and the non-edge probability distribution image I ⁴ obtained from the image processing module 1 Perform weighted fusion to obtain a multi-feature weighted ^fusion feature image If, as shown in Figure 13.

Specifically, the calculation expression of the multi-feature weighted ^fusion feature image If is:

Among them, N(·) represents numerical normalization, and α ₁ , α ₂ , α ₃ , and α ₄ respectively represent the geometric weights corresponding to each feature image.

In this embodiment, the values of the weights α ₁ , α ₂ , α ₃ , and α ₄ are 1.1, 1.0, 0.9, and 1.0, respectively. It should be noted that the above values are only used for illustration, and are not limited thereto. In the specific implementation process, the geometric weight corresponding to each feature image can be arbitrarily selected according to actual needs.

For the defect detection module 3, in a preferred embodiment, it is used for:

Perform adaptive threshold segmentation on the weighted fusion feature image I ^f to obtain a binary image;

Perform mathematical morphological opening operations on binary images to determine defect areas.

Specifically, the adaptive threshold expression is

T(I ^f )=mean(I ^f )+k·std(I ^f )

Among them, mean( ) represents the mean value of the pixel grayscale, std( ) represents the standard deviation of the pixel grayscale, and k is the standard deviation weight coefficient, which usually ranges from 3.0 to 5.0.

The image effect after threshold segmentation and morphological processing is shown in Figure 15, where the white area is the final determined defect area.

In this embodiment, the value of k is 4.0, and in the mathematical morphology opening operation after binary segmentation, a 4×4 scale circular template is used for the erosion operation, and then a 7×7 scale circular template is used for expansion operate.

It should be noted that the detection system for rail defects provided by the present invention has been verified by experiments, and the detection accuracy rate of surface defects of the rail is relatively high. In the performance experiment, by testing 195 pieces of rail surface scanning image data with a total of 326 defects in various shapes, sizes and appearances, a total of 338 defects were detected, including 309 real defects and 29 false detections. Therefore, the detection accuracy rate is 91.4%, and the detection recall rate is less than 94.8%.

Therefore, the track defect detection system provided by the present invention separately performs singular value decomposition, saliency detection and gradient amplitude calculation processing on the track image to be processed through the image processing module, and weights the obtained several feature images through the feature fusion module. , so as to realize the unsupervised machine vision processing method combined with visual multi-feature fusion, which can efficiently detect the surface defects of the railway track, and at the same time effectively overcome the background clutter interference in the track image under the load environment, and has strong adaptability to the environment. The effect is stable, and it is suitable for complex track defect detection of various scales and shapes; and it does not need to collect samples for training in practical application, and has strong applicability and low application cost.

Example 3

FIG. 17 is a schematic structural diagram of an electronic device according to Embodiment 3 of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method for detecting a track defect provided in Embodiment 1 when the processor executes the program. The electronic device 40 shown in FIG. 17 is only an example, and should not impose any limitation on the function and scope of use of the embodiment of the present invention.

As shown in FIG. 17 , the electronic device 40 may take the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 40 may include, but are not limited to: the above-mentioned at least one processor 41 , the above-mentioned at least one memory 42 , and a bus 43 connecting different system components (including the memory 42 and the processor 41 ).

The bus 43 includes a data bus, an address bus, and a control bus.

Memory 42 may include volatile memory, such as random access memory (RAM) 421 and/or cache memory 422 , and may further include read only memory (ROM) 423 .

The memory 42 may also include a program/utility 425 having a set (at least one) of program modules 424 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, which An implementation of a network environment may be included in each or some combination of the examples.

The processor 41 executes various functional applications and data processing by running the computer program stored in the memory 42, such as the method for detecting track defects provided in Embodiment 1 of the present invention.

The electronic device 40 may also communicate with one or more external devices 44 (eg, keyboards, pointing devices, etc.). Such communication may take place through input/output (I/O) interface 45 . Also, the model-generating device 40 may also communicate with one or more networks (eg, a local area network (LAN), a wide area network (WAN), and/or a public network such as the Internet) through a network adapter 46 . As shown, the network adapter 46 communicates with other modules of the model-generated device 40 via the bus 43 . It should be understood that, although not shown, other hardware and/or software modules may be used in conjunction with the model-generated device 40, including but not limited to: microcode, device drivers, redundant processors, arrays of external disk drives, RAID (disk) array) systems, tape drives, and data backup storage systems.

It should be noted that although several units/modules or sub-units/modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. Indeed, the features and functions of two or more units/modules described above may be embodied in one unit/module according to embodiments of the present invention. Conversely, the features and functions of one unit/module described above may be further subdivided to be embodied by multiple units/modules.

Example 4

This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for detecting a track defect provided in Embodiment 1 is implemented.

Wherein, the readable storage medium may include, but is not limited to, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, or any of the above suitable combination.

In a possible implementation manner, the present invention can also be implemented in the form of a program product, which includes program codes, when the program product runs on a terminal device, the program code is used to cause the terminal device to execute the implementation The method for detecting track defects provided in Embodiment 1.

Wherein, the program code for executing the present invention can be written in any combination of one or more programming languages, and the program code can be completely executed on the user equipment, partially executed on the user equipment, as an independent The software package executes on the user's device, partly on the user's device, partly on the remote device, or entirely on the remote device.

Although the specific embodiments of the present invention are described above, those skilled in the art should understand that this is only an illustration, and the protection scope of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principle and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims (10)

1. A method for detecting a track defect, the method comprising:

respectively carrying out singular value decomposition, significance detection and gradient amplitude calculation on the orbit image to be processed to obtain a plurality of corresponding characteristic images;

performing weighted fusion on the feature images to obtain weighted fusion feature images;

and determining a defect area in the rail image to be processed based on the weighted fusion characteristic image.

2. The method of detecting rail defects of claim 1 wherein said plurality of feature images comprises singular pixel contrast enhanced feature images;

the singular value decomposition step specifically includes:

calculating the difference between the gray values of a plurality of target pixel points and reference pixel points in the orbit image to be processed so as to obtain a singular pixel contrast enhancement feature image, wherein the distance between the reference pixel points and the target pixel points is smaller than a first threshold value;

and/or the plurality of feature images comprise singular row contrast enhanced feature images;

the singular value decomposition step specifically includes:

and calculating the difference between the gray values of a plurality of target pixel rows and reference pixel rows in the orbit image to be processed so as to obtain a singular row contrast enhancement feature image, wherein the distance between the reference pixel row and the target pixel row is smaller than a second threshold value.

3. The method of detecting rail defects according to claim 1, wherein the plurality of feature images includes a visually significant feature image;

the step of significance detection specifically comprises:

performing Fourier transform and phase spectrum calculation on the orbit image to be processed to obtain a visual saliency map;

and performing smooth filtering on the visual saliency map to obtain a visual saliency characteristic image.

4. The method of detecting rail defects according to claim 1, wherein the plurality of feature images includes a probability distribution image for characterizing non-edge regions;

the step of calculating the gradient amplitude specifically comprises:

calculating the gradient amplitude of each pixel point in the orbit image to be processed to obtain a gradient amplitude image;

and carrying out smooth filtering and negation operation on the gradient amplitude image to obtain the probability distribution image.

5. The method for detecting track defects according to claim 1, wherein the step of determining the defect region in the to-be-processed track image based on the weighted fusion feature image comprises:

performing self-adaptive threshold segmentation on the weighted fusion characteristic image to obtain a binary image;

and performing mathematical morphology on the binary image to determine the defect area.

6. The method of detecting a track defect of claim 1, further comprising:

acquiring an initial track image;

and carrying out filtering processing on the initial orbit image to obtain the orbit image to be processed.

7. Method for detecting track defects according to claim 6, characterized in that said filtering process comprises L₀Gradient minimization filtering and median filtering.

8. A rail defect detection system, comprising:

the image processing module is used for respectively carrying out singular value decomposition, significance detection and gradient amplitude calculation on the orbit image to be processed so as to obtain a plurality of corresponding characteristic images;

the characteristic fusion module is used for carrying out weighted fusion on the characteristic images so as to obtain weighted fusion characteristic images;

and the defect detection module is used for determining a defect area in the rail image to be processed based on the weighted fusion characteristic image.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of detecting a track defect according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method of detection of a track defect according to any one of claims 1 to 7.

Images