CN111462045B - A method for detecting defects of catenary support components - Google Patents

Abstract

The invention discloses a method for detecting defects of a catenary support assembly, and a method for detecting defects of a catenary support assembly, comprising the following steps: step 1: constructing a data set of catenary load-bearing cable bases and oblique stay wire hooks; step 2 : Use the Faster RCNN convolutional neural network to locate the target, and obtain the positioning results of the catenary bearing cable base and the cable-stayed hook; Step 3: According to the positioning results and the structural information of the catenary cable-bearing cable base and the cable-stayed hook, obtain the oblique cable hook. The image of the candidate area where the cable is located; Step 4: Use Hough transform to locate the image of the candidate area of the cable to obtain the positioning result of the cable, and perform the loose defect detection of the cable according to the straight line detection result; Step 5: According to the image processing method and According to the detection result, the installation defect detection is performed on the base of the bearing cable; the invention improves the detection efficiency and precision of the components, can effectively detect whether the cable-stayed wire has a loose fault, has high detection efficiency, and simplifies the difficulty of fault detection.

Description

A method for detecting defects of catenary support components

technical field

The invention relates to the technical field of high-speed rail image intelligent detection, in particular to a method for detecting defects of catenary support components.

Background technique

As the advantages of high-speed railway become more and more prominent, countries around the world are building a large number of railways. The railway foundation is mainly composed of two parts: the catenary system and the railway system. The catenary system is mainly responsible for the power supply of high-speed locomotives, and it contains a large number of supporting components, such as insulators, stable booms, supporting wire hooks, etc. These support assemblies are used to withstand mechanical loads, electrical insulation, etc. However, due to the high-speed movement of the locomotive and the influence of the external environment, the catenary support assembly may lose parts, cause loose parts, cracks, etc., which will pose a major threat to the safe operation of high-speed trains.

At present, the detection methods used for the state defects of catenary parts at home and abroad mainly include: naked eye observation method, laser testing method, eddy current method, ultrasonic method and so on. These detection methods have achieved certain results, but many methods have problems such as inaccurate measurement, high risk, complicated operation, expensive and bulky equipment, heavy detection tasks, and poor anti-interference ability. The research on non-contact pantograph-catenary detection technology based on image processing technology can realize the development of pantograph-catenary detection device that does not interfere with driving safety. Zhang Guinan proposed a high-speed railway pole insulator detection and identification method based on Contourlet Transform (CT, Contourlet Transform) and Chan-Vese model. Han Ye uses a deformable part model to locate the rod insulator. Yang Hongmei uses template matching surf to detect insulator faults. With the improvement of computer computing power and information collection ability, many algorithms based on deep learning have been proposed. Liu Kai et al. used Faster RCNN to detect the damage of the base of the bearing cable. Wang Liyou et al designed a network with high detection accuracy to detect whether the equipotential line is loose. Liu, et al. employ a technique combining depthwise separable convolution with an object detection network to detect hanging string faults. These deep learning-based detection methods show fast detection speed and high detection accuracy. However, few people have proposed corresponding methods for cable-stayed faults. In this case, fast and accurate detection of catenary support components Target localization and defect detection methods are particularly important.

SUMMARY OF THE INVENTION

Aiming at the problems existing in the prior art, the present invention provides a method for detecting the defect of a catenary support assembly, which can realize the fast detection of the installation state of the bearing cable base and the loose fault of the cable-stayed cable.

The technical scheme adopted in the present invention is:

A method for detecting defects of a catenary support assembly, comprising the following steps:

Step 1: Build a dataset of catenary cable bases and cable-stayed hooks;

Step 2: Use the Faster RCNN convolutional neural network to locate the target, and obtain the location results of the catenary load-bearing cable base and the cable-stayed hook;

Step 3: According to the positioning result in Step 2 and the structure information of the catenary bearing cable base and the cable hook, obtain the image of the candidate area where the cable is located;

Step 4: Use Hough transform to locate the candidate area image of the cable-stayed wire obtained in step 3, obtain the positioning result of the cable-stayed wire, and perform the looseness defect detection of the cable-stayed wire according to the straight line detection result;

Step 5: According to the image processing method and the detection result in Step 4, perform installation defect detection on the base of the bearing cable obtained in Step 2.

Further, the specific process of the step 2 is as follows:

S11: Perform a convolution operation on the input image to obtain a feature map;

S12: Extract the RoI of the region of interest through the region proposal network;

S13: classify and locate the RoI.

Further, the specific process of the step 3 is as follows:

S21: According to the Faster RCNN positioning result, obtain the relative position of the catenary bearing cable base and the cable-stayed hook;

S22: According to the coordinates and relative position of the prediction frame, intercept the image of the candidate area where the diagonal bracing line is located.

Further, the specific process of step 4 is as follows:

S31: Binarize the image of the candidate region where the diagonal bracing line obtained in step 3 is located;

S32: Extract the edge of the binarized image by the Candy algorithm, and use the Hough transform to detect the horizontal wrist arm; according to the straight line detection result and the angle of the horizontal wrist arm, rotate the bearing cable base to the horizontal direction;

S33: The Hough transform θ angle is limited, and according to the Hough transform result, select the peak distribution of the θ angle with the top n statistic; Normal, otherwise loose.

Further, the specific process of step 5 is as follows:

S41: Convert the catenary cable base image into the original grayscale histogram;

S42: Perform binarization, dilation, erosion and filling processing on the original grayscale histogram in step S41 in sequence to eliminate the background;

S43: Determine the opening direction of the base of the bearing cable according to the pulse signal generated by scanning in the horizontal direction;

S44: According to the opening direction of the bearing cable base obtained in step S43 and the detection direction of the oblique stay wire determined in step 4, if the two directions are the same, the bearing cable base is installed correctly, otherwise the installation of the component is defective.

The beneficial effects of the present invention are:

(1) The present invention can quickly and accurately detect the catenary support components (the load-bearing cable base and the cable-stayed hook) by detecting the bad state of the high-speed rail catenary load-bearing cable base and the cable-stayed wire through the deep convolutional neural network and the image processing method. ), at the same time, using the structural relationship of the components to quickly extract the candidate area of the cable-stayed line, which improves the efficiency and accuracy of component detection;

(3) The present invention combines the Hough variation characteristics of the straight line with the structure of the cable-stayed wire according to the structural characteristics of the cable-stayed wire, so as to effectively detect whether the cable-stayed wire has a loose fault;

(3) The method of the present invention can effectively detect the installation failure of the load-bearing cable base and the looseness of the cable-stayed cable, the correct detection efficiency is high, and the difficulty of failure detection is simplified.

Description of drawings

Fig. 1 is the schematic flow chart of the method of the present invention.

FIG. 2 is a schematic diagram of on-site acquisition of an image of a high-speed railway catenary support and suspension device according to an embodiment of the present invention.

FIG. 3 shows the positions of the base of the load-bearing cable and the hook of the cable-stayed cable located by the Faster RCNN convolutional neural network according to the embodiment of the present invention.

FIG. 4 is an area diagram of the oblique stay line and a result diagram of the location of the oblique stay line according to the embodiment of the present invention.

FIG. 5 is a rotation diagram of a horizontal wrist arm according to an embodiment of the present invention.

FIG. 6 is a grayscale histogram of background elimination according to an embodiment of the present invention.

FIG. 7 is a diagram of preprocessing such as binarization, corrosion, and expansion in an embodiment of the present invention.

FIG. 8 is a schematic diagram and a result diagram of an angle-based fast detection method in an embodiment of the present invention.

FIG. 9 is a diagram showing a result of Hough transform of the oblique-braced wire in an embodiment of the present invention.

FIG. 10 is a peak distribution diagram of Theta in the embodiment of the present invention.

FIG. 11 is a schematic diagram of the installation defect detection of the base of the bearing cable according to the embodiment of the present invention.

FIG. 12 is an example diagram of the detection result of the installation defect of the bearing cable base in the embodiment of the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

As shown in Figure 1, a method for detecting defects of catenary support components includes the following steps:

Step 1: Build a dataset of catenary cable bases and cable-stayed hooks;

The support and suspension devices of the catenary of the high-speed railway were imaged by the comprehensive train inspection vehicle of the special train, and the data set of the catenary load-bearing cable base and the cable-stayed hook was established.

Step 2: Use the Faster RCNN convolutional neural network to locate the target, and obtain the location results of the catenary load-bearing cable base and the cable-stayed hook;

The specific process is as follows:

S11: Perform a convolution operation on the input image to obtain a feature map;

S12: Extract the RoI (RoI, Region of Interests) through the Region Proposal Network (RPN, Region Proposal Network);

S13: classify and locate the RoI.

Step 3: According to the positioning result in Step 2 and the structure information of the catenary bearing cable base and the cable hook, obtain the image of the candidate area where the cable is located;

The specific process is as follows:

S21: According to the Faster RCNN positioning result, obtain the relative position of the catenary bearing cable base and the cable-stayed hook;

S22: According to the coordinates and relative position of the prediction frame, intercept the image of the candidate area where the diagonal bracing line is located.

Step 4: Use Hough transform to locate the candidate area image of the cable-stayed wire obtained in step 3, obtain the positioning result of the cable-stayed wire, and perform the looseness defect detection of the cable-stayed wire according to the straight line detection result;

The specific process is as follows:

In order to analyze the opening direction of the cable base, the image should be rotated to the horizontal according to the angle of the horizontal wrist arm.

S31: Binarize the image of the candidate region where the diagonal bracing line obtained in step 3 is located;

S32: Extract the edge of the binarized image through the Candy algorithm, and use the Hough transform to detect the horizontal wrist arm; according to the straight line detection result and the angle of the horizontal wrist arm, rotate the bearing cable base to the horizontal direction.

In order to improve the detection speed and detection accuracy of the cable-stayed wire, a fast Hough transform detection scheme based on the candidate area is proposed. According to the structural information of the support wire component, the angle of the support wire must be within the range of the connection angle between the base of the cable and the hook of the cable-stayed wire. Inside. Using the candidate area obtained in step 3, the value range of the Hough transform θ angle is constrained and limited, so as to eliminate the interference of straight lines from other angles and ensure the accuracy and efficiency of detection.

S33: The Hough transform θ angle is limited, and according to the Hough transform result, the peak distribution of the θ angle in the top n ranks of the statistics is selected (the top five ranks are selected in this embodiment); if the standard deviation of the θ angle is less than the set threshold, the peak value If the number is greater than the set threshold, the cable is normal, otherwise it is loose. Theoretically, the straighter the line segment is, the smaller the angle fluctuation is, and the number of peaks is within a certain range, it can be judged whether the detection line is bent or deformed. Therefore, the standard deviation of the Theta angle and the number of peaks are used as indicators to detect loosening defects. If the standard deviation is less than 1, and the number of peaks is greater than 6000, the diagonal cable is determined to be normal, otherwise, it is determined to be a loose defect.

Step 5: According to the image processing method and the detection result in Step 4, perform installation defect detection on the base of the bearing cable obtained in Step 2.

The specific process is as follows:

S41: Convert the catenary cable base image into the original grayscale histogram; for the obtained catenary cable base image, the original grayscale histogram can be obtained by the following formula:

The median pixel of the image is calculated by:

Among them, I _{(i, j)} is the gray value of the image pixel, L represents the gray level, (M, N) represents the size of the image, and Rank() represents the ranking function of the image gray value. Pixels below the median pixel in the image are set to 0 to obtain an image with background noise removed.

S42: Perform binarization, dilation, erosion and filling processing on the original grayscale histogram in step S41 in sequence to eliminate the background;

S43: Determine the opening direction of the base of the bearing cable according to the pulse signal generated by scanning in the horizontal direction;

By scanning in the horizontal direction to generate a pulse signal, the defect of the reverse installation of the base of the bearing cable can be detected. First, the cable base is mounted on the horizontal arm, so the image starts to scan down from the central axis A of the horizontal arm. When the opening direction of the fixed hook of the base of the load-bearing cable is the left side, a pulse signal first appears at B, and then continues to scan, and two pulse signals appear at A and B. Otherwise, the detected opening is directed to the right. Therefore, according to this characteristic, the direction of the positioning hook of the bearing cable base can be determined. According to the determined opening direction of the base of the load-bearing cable and the detection direction of the cable-stayed wire determined in step 4, determine the installation defect of the base of the load-bearing cable. If the directions of the two are the same, the base of the bearing cable is installed correctly; otherwise, the component is installed incorrectly and is installed in reverse.

S44: According to the opening direction of the bearing cable base obtained in step S43 and the detection direction of the oblique stay wire determined in step 4, if the two directions are the same, the bearing cable base is installed correctly, otherwise the installation of the component is defective.

FIG. 2 is a schematic diagram of the present invention for collecting images of a high-speed rail catenary suspension device on-site. Figure 3 shows the base of the load-bearing cable and the cable-stayed hook located using the FasterRCNN convolutional neural network. According to the Hough transform, etc., the positioning result of the oblique stay line is obtained, as shown in Figure 4.

To rotate the image of the base of the load-bearing cable, it is necessary to first binarize the image. Next, extract the edge of the binarized image through the Candy algorithm, and use the Hough transform to detect the horizontal arm. Then, according to the detection result, the base of the load-bearing cable is rotated to horizontally, as shown in Figure 5.

The background of the base of the cable is eliminated, and the pixels below the median value of the image are set to 0 to obtain the image after eliminating the background. The result is shown in Figure 6; where a is the grayscale histogram before the background is eliminated, and b is the elimination The resulting grayscale histogram.

After image preprocessing, binarization, erosion, and dilation, the image results are shown in Figure 7.

For the looseness detection of the pull-down wire, a fast detection method based on Hough transform is proposed according to the distribution of the inclined-pull wire with a certain angle range. As shown in Figure 8. Figure 9 shows the Hough transform result of the oblique cable. According to the peak distribution result of theta, as shown in Figure 10, it can be determined whether the oblique cable is in a loose state.

The installation defect detection of the base of the load-bearing cable will produce different pulse results according to the opening direction of the load-bearing cable. The detection method of its installation defect is proposed. The schematic diagram is shown in Figure 11. Figure 12 shows an example graph of the results of installation defect detection.

The invention detects the bad state of the bearing cable base and the cable-stayed wire of the high-speed rail catenary by means of the deep convolutional neural network and the image processing method. Can quickly and accurately detect catenary support components (bearing cable base and inclined cable hook). At the same time, the component structure relationship is used to quickly extract the candidate area of the cable-stayed wire, which improves the efficiency and accuracy of component detection. According to the structural characteristics of the cable-stayed wire, combining the Hough change characteristics of the straight line with the structure of the cable-stayed cable can effectively detect whether the cable-stayed cable is loose or not. It can effectively detect the installation fault of the load-bearing cable base and the loose fault of the cable-stayed cable, and the correct detection rate is high, which simplifies the difficulty of fault detection.

Claims (1)

1. a detection method for catenary support assembly defect, is characterized in that, comprises the following steps: Step 1: Build a dataset of catenary cable bases and cable-stayed hooks; Step 2: Use the Faster RCNN convolutional neural network to locate the target, and obtain the location results of the catenary load-bearing cable base and the cable-stayed hook; S21: Perform a convolution operation on the input image to obtain a feature map; S22: Extract the RoI of the region of interest through the region proposal network; S23: classify and locate the RoI; Step 3: According to the positioning result in Step 2 and the structure information of the catenary bearing cable base and the cable hook, obtain the image of the candidate area where the cable is located; S31: According to the Faster RCNN positioning result, obtain the relative position of the catenary bearing cable base and the cable-stayed hook; S32: according to the coordinates and relative position of the prediction frame, intercept the image of the candidate area where the diagonal bracing line is located; Step 4: Use the Hough transform to locate the candidate area image of the cable-stayed wire obtained in step 3, obtain the positioning result of the cable-stayed wire, and perform the looseness defect detection of the cable-stayed wire according to the straight line detection result; S41: Binarize the image of the candidate region where the diagonal bracing line obtained in step 3 is located; S42: Extract the edge of the binarized image by the Candy algorithm, and use the Hough transform to detect the horizontal wrist arm; according to the line detection result and the angle of the horizontal wrist arm, rotate the bearing cable base to the horizontal direction; S43: The Hough transform θ angle is limited, and according to the Hough transform result, the peak distribution of the θ angle with the top n statistic is selected; if the standard deviation of the θ angle is less than the set threshold, and the number of peaks is greater than the set threshold, then the diagonal line Normal, otherwise loose; Step 5: According to the image processing method and the detection result in Step 4, perform installation defect detection on the base of the bearing cable obtained in Step 2; S51: Convert the catenary cable base image into an original grayscale histogram; for the obtained catenary cable base image, the original grayscale histogram is obtained by the following formula:

The median pixel of the image is calculated by:

Among them, I _{(i, j)} is the gray value of the image pixel, L is the gray level, (M, N) is the size of the image, and Rank() is the ranking function of the gray value of the image, which will be lower than the median value of the image. The pixel of the pixel is set to 0 to obtain an image with background interference removed; S52: Perform binarization, expansion, erosion, and filling processes on the original grayscale histogram in step S51 to eliminate the background; S53: Determine the opening direction of the base of the bearing cable according to the pulse signal generated by scanning in the horizontal direction; S54: According to the opening direction of the bearing cable base obtained in step S53 and the detection direction of the cable-stayed wire determined in step 4, if the two directions are the same, the bearing cable base is installed correctly; otherwise, the bearing cable base is installed incorrectly.

Images