Pantograph abnormality detection system based on deep learning

Abstract

The application relates to the technical field of pantograph abnormality detection, and discloses a pantograph abnormality detection system based on deep learning. The pantograph abnormality detection system based on deep learning collects train operation signals, train numbers, train numbers and pantograph images captured by a camera through a data acquisition module, a data processing module performs cleaning, denoising and amplification operations on the pantograph image data, a deep learning algorithm model detection module calculates the performance evaluation indexes of the algorithm by setting the performance evaluation indexes, compares a PPYOLO algorithm, a YOLOv3 algorithm with a backbone network of Darknet-53 and a YOLOv3 algorithm with a backbone network of MobileNet, selects an optimal algorithm PPYOLO algorithm for pantograph abnormality detection, and transmits pantograph detection abnormality information to a data recording module for storage, and a data output module outputs the pantograph abnormality information to a visual interface.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of pantograph anomaly detection, in particular to a pantograph anomaly detection system based on deep learning. BACKGROUND

[0002] With the development of science and technology, the railway industry is booming. By the end of 2021, the mileage of railways in China has exceeded 150,000 kilometers, and the mileage of high-speed railways has reached 40,000 kilometers. It is undeniable that railway transportation has been deeply integrated into people's lives. Accelerating the construction of the railway industry will help drive the economic and social development of China.

[0003] At present, most of the railways in China are electrified railways, which are a kind of railway used for electric train operation. In 2021, the number of electric locomotives accounted for 64.1% of the total number of railway locomotives. In the power supply system of electrified railways, the pantograph is an important electrical equipment, which is installed on the roof and connected with the overhead line to obtain electric energy and power the running electric traction locomotive. Therefore, the working state of the pantograph greatly determines the running state of the train. If the pantograph fails, it will seriously affect the traffic safety of the track and the performance of the train electric drive system. Therefore, it is necessary to detect the pantograph.

[0004] At present, China mainly relies on the traditional manual roof inspection method to detect the pantograph. This method requires the train to enter the depot, and after a series of operations such as stopping, power-off, and lowering the pantograph, the staff climbs to the roof for detection. This method has poor detection accuracy and safety hazards, so a new pantograph detection technology is needed to solve the above problems. SUMMARY

[0005] (I) Technical problems solved

[0006] In view of the deficiencies of the prior art, the present application provides a pantograph abnormality detection system based on deep learning, which has a data acquisition module for acquiring train operation signals Lcxh, train numbers Lcch, train numbers Lccc and pantograph images Sdgx captured by a camera, a data processing module for calculating and processing the above data to obtain cleaned pantograph images Qxtx, denoised pantograph images Qztx and enlarged pantograph images Fdtx, a deep learning algorithm model detection module for calculating the performance evaluation index of the algorithm and comparing the PPYOLO algorithm, the YOLOv3 algorithm with a backbone network of Darknet-53 and the YOLOv3 algorithm with a backbone network of MobileNet, selecting the optimal algorithm PPYOLO algorithm for pantograph abnormality detection, and transmitting the pantograph abnormality detection information to a data recording module for storage, and a data output module for outputting the pantograph abnormality information to a visual interface, so that the optimal algorithm model PPYOLO algorithm based on deep learning is used for pantograph abnormality detection, manual climbing for detection is not needed, the accuracy of pantograph abnormality detection is improved, and the advantages of reducing safety hazards and the like are achieved, and the above problems are solved.

[0007] (II) Technical solutions

[0008] To achieve the above object, the present application provides the following technical solutions: a pantograph abnormality detection system based on deep learning, comprising a data acquisition module, a data processing module, a model detection module, a data recording module and an output module.

[0009] The data acquisition module is connected to the data processing module through a network.

[0010] The data processing module is connected to the deep learning algorithm model detection module through a network.

[0011] The deep learning algorithm model detection module is connected to the data recording module through a network.

[0012] The data recording module is connected to the output module through a network.

[0013] The data acquisition module is used for acquiring train operation signals Lcxh, train numbers Lcch, train numbers Lccc and pantograph images Sdgx captured by a camera, and transmitting the acquired data to the data processing module through a network.

[0014] The data processing module receives the train operation signals Lcxh, train numbers Lcch, train numbers Lccc and pantograph images Sdgx captured by a camera transmitted by the data acquisition module, and performs cleaning, denoising and enlarging operations on the pantograph images to obtain cleaned pantograph images Qxtx, denoised pantograph images Qztx and enlarged pantograph images Fdtx.

[0015] The data processing module transmits the train operation signal Lcxh, the train number Lcch, the train number Lccc, the pantograph image Qxtx, the denoising pantograph image Qztx, and the enlarged pantograph image Fdtx to the deep learning algorithm model detection module;

[0016] The deep learning algorithm model detection module sets performance evaluation indicators, calculates the performance evaluation indicators of the algorithm, and compares the PPYOLO algorithm, the YOLOv3 algorithm with a backbone network of Darknet-53, and the YOLOv3 algorithm with a backbone network of MobileNet;

[0017] The deep learning algorithm model detection module detects the enlarged pantograph image Fdtx based on the deep learning PPYOLO algorithm, judges the abnormal condition of the pantograph, and uploads the abnormal information to the data recording module;

[0018] The data recording module stores the pantograph abnormal information transmitted by the deep learning algorithm model detection module in the system storage medium;

[0019] The output module displays the abnormal information of the pantograph detected in the system visual interface by reading the system storage medium of the data recording module.

[0020] Preferably, the data acquisition module acquires real-time images of the pantograph through a camera, and encodes the acquired images into a pantograph image data set, which is {Sdgx1, Sdgx2, Sdgx3,..., Sdgx n}。

[0021] Preferably, the data processing module cleans the pantograph image to obtain the cleaned pantograph image Qxtx, and the cleaning formula is as follows:

[0022] Qxtx out (x,y)={Qxtx in (x+i,y+j:-r<=i,j<=r)}

[0023] In the formula, Qxtx in (x,y) represents the pixel value of the coordinate (x, y) in the input pantograph image, Qxtx out (x,y) represents the pixel value of the coordinate (x, y) in the output pantograph image, and r is the radius of the filter, which is selected as 5.

[0024] Preferably, the data processing module performs a denoising operation on the cleaned pantograph image Qxtx based on the image to obtain the denoising pantograph image Qztx, and the denoising formula is as follows:

[0025]

[0026] In the formula, Qxtx in (x,y) represents the pixel value of the coordinate (x,y) in the input pantograph image, Qztx out (x,y) represents the pixel value of the coordinate (x,y) in the output pantograph image, and r is the radius of the filter, which is selected as 5.

[0027] Preferably, the data processing module performs an amplification operation on the denoised pantograph image Qztx to obtain an amplified pantograph image Fdtx, and the amplification formula is as follows:

[0028]

[0029] In the formula, Fdtx out (x,y) represents the pixel value of the coordinate (x,y) in the output pantograph image, Qxtx in (x,y) represents the pixel value of the coordinate (x,y) in the input pantograph image, and S represents the magnification of the pantograph image, which is selected as 10.

[0030] Preferably, the deep learning algorithm model detection module sets a performance evaluation index, and calculates the performance evaluation index of the algorithm, and the calculation formula is as follows:

[0031]

[0032] In the formula, Recall represents the recall rate, Precision represents the accuracy rate, TP represents that the sample and the prediction result are both positive, FN represents that the sample is positive and the prediction result is negative, and FP represents that the sample and the prediction result are both negative.

[0033] Preferably, the deep learning algorithm model detection module compares the PPYOLO algorithm, the YOLOv3 algorithm with a backbone network of Darknet-53, and the YOLOv3 algorithm with a backbone network of MobileNet, and the comparison results are as follows:

[0034] The average precision value of the PPYOLO algorithm is 99.85%, and the FPS is 39.16 frames per second;

[0035] The average precision value of the YOLOv3 algorithm with a backbone network of Darknet-53 is 98.58%, and the FPS is 33.34 frames per second;

[0036] The average precision value of the YOLOv3 algorithm with a backbone network of MobileNet is 97.93%, and the FPS is 37.28 frames per second.

[0037] Preferably, the deep learning algorithm model detection module detects the enlarged pantograph image Fdtx based on a deep learning-based PPYOLO algorithm, and the detection method is as follows:

[0038] When the deep learning algorithm model detection module detects that the enlarged pantograph image Fdtx has a structural anomaly, the pantograph structural anomaly information is transmitted to the system;

[0039] When the deep learning algorithm model detection module detects that the enlarged pantograph image Fdtx has a suspended foreign matter, the pantograph suspended foreign matter anomaly information is transmitted to the system.

[0040] Preferably, the data recording module is internally provided with a storage medium, and the pantograph anomaly information detected by the deep learning algorithm model detection module is stored.

[0041] Preferably, the output module can read the pantograph anomaly information stored in the data recording module, and transmit the anomaly information to the visual interface for display.

[0042] Compared with the prior art, the present application provides a pantograph anomaly detection system based on deep learning, which has the following beneficial effects:

[0043] The present application collects train operation signals Lcxh, train number Lcch, train number Lccc and pantograph images Sdgx taken by a camera through a data acquisition module, and calculates and processes the above data through a data processing module to obtain a cleaned pantograph image Qxtx, a denoised pantograph image Qztx and an enlarged pantograph image Fdtx. The deep learning algorithm model detection module calculates the performance evaluation index of the algorithm, compares the PPYOLO algorithm, the YOLOv3 algorithm with a backbone network of Darknet-53 and the YOLOv3 algorithm with a backbone network of MobileNet, selects the optimal algorithm PPYOLO algorithm for pantograph anomaly detection, and transmits the pantograph detection anomaly information to the data recording module for storage. The data output module outputs the pantograph anomaly information to the visual interface. Through the optimal algorithm model PPYOLO algorithm based on deep learning for pantograph anomaly detection, manual climbing for detection is not required, the accuracy of pantograph anomaly information detection is improved, and safety hazards are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The present application is a system flowchart. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely below with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0046] At present, the pantograph is mainly detected by the traditional artificial top climbing inspection method. The method requires the train to drive into the depot, and after a series of operations such as stopping, power-off and pantograph lowering, the working personnel climbs to the roof to detect. This method has poor detection accuracy and safety hazards. Therefore, a pantograph abnormality detection system based on deep learning is proposed. Please refer to Figure 1 The system includes a data acquisition module, a data processing module, a model detection module, a data recording module and an output module, wherein:

[0047] The data acquisition module acquires real-time images of the pantograph through a camera, and encodes the acquired images into a pantograph image data set, which is {Sdgx1, Sdgx2, Sdgx3,..., sdgx n};

[0048] Since the number of collected pantograph images is large and the quality cannot be unified, the data processing module is needed to clean the pantograph images to obtain cleaned pantograph images Qxtx. The cleaning formula is as follows:

[0049] Qxtx out (x,y)={Qxtx in (x+i,y+j:-r<=i,j<=r)}

[0050] Data cleaning can effectively identify and correct errors, missing data, and repeated or inconsistent data in the data, thereby improving the accuracy, integrity and consistency of the data, making the data more reliable. In the formula, Qxtx in (x,y) represents the pixel value of the coordinate (x,y) in the input pantograph image, Qxtx out (x,y) represents the pixel value of the coordinate (x,y) in the output pantograph image, and r is the radius of the filter. The radius is selected as 5. The median filter is used to clean the pantograph image, which helps to find hidden patterns and correlations in the data, improves the accuracy and reliability of data mining, and better explores the potential information in the data.

[0051] The removal of noise can reduce errors in image processing and analysis, improve the accuracy of image content recognition and understanding, and facilitate subsequent image-related applications. Therefore, the data processing module removes noise from the pantograph image Qxtx based on the cleaned pantograph image Qxtx to obtain the denoised pantograph image Qztx. The denoising formula is as follows:

[0052]

[0053] The removal of noise can improve the compression effect of the image, and the image can maintain higher quality and clarity under different resolutions and formats. In the formula, Qxtx in (x,y) represents the pixel value of the input pantograph image at coordinates (x,y), and Qztx out (x,y) represents the pixel value of the output pantograph image at coordinates (x,y), and r is the radius of the filter. The radius is selected as 5. The mean filter method is used to denoise the pantograph image. The filter with a radius of 5 can effectively compress the pantograph image and avoid excessive calculation caused by a large radius. The removal of noise can reduce errors in image processing and analysis, improve the accuracy of image content recognition and understanding, and facilitate subsequent image-related applications.

[0054] In order to enhance the clarity of the pantograph image, the data processing module is used to enlarge the denoised pantograph image Qztx to obtain the enlarged pantograph image Fdtx. The enlargement formula is as follows:

[0055]

[0056] Enlarging the image can increase the resolution of the image, thereby highlighting more details and information, making the image look clearer and more vivid. In the formula, Fdtx out (x,y) represents the pixel value of the output pantograph image at coordinates (x,y), and Qxtx in (x,y) represents the pixel value of the input pantograph image at coordinates (x,y), and S represents the magnification of the pantograph image. The magnification is selected as 10. The The formula is used to enlarge the pantograph image by a fixed magnification, making the image clearer and facilitating the next target image detection and recognition work.

[0057] The deep learning algorithm model detection module sets performance evaluation indicators and calculates the performance evaluation indicators of the algorithm. The calculation formula is as follows:

[0058]

[0059]

[0060] The performance evaluation index can accurately reflect the comprehensive performance of the algorithm. In the formula, TP represents that the sample and the prediction result are both positive, FN represents that the sample is positive and the prediction result is negative, and FP represents that the sample and the prediction result are both negative. After determining the recall rate and the accuracy rate, the P-R curve is drawn with the recall rate as the horizontal coordinate and the accuracy rate as the vertical coordinate. The area under the P-R curve of a certain class is the AP (average accuracy) of the class. The average of the APs of all classes is the mAP.

[0061] The deep learning algorithm model detection module compares the PPYOLO algorithm, the YOLOv3 algorithm with a backbone network of Darknet-53, and the YOLOv3 algorithm with a backbone network of MobileNet. The comparison results are as follows:

[0062] The average precision value of the PPYOLO algorithm is 99.85%, and the FPS (frames per second) is 39.16.

[0063] The average precision value of the YOLOv3 algorithm with a backbone network of Darknet-53 is 98.58%, and the FPS is 33.34.

[0064] The average precision value of the YOLOv3 algorithm with a backbone network of MobileNet is 97.93%, and the FPS is 37.28.

[0065] The deep learning algorithm model detection module detects the enlarged pantograph image Fdtx based on the deep learning PPYOLO algorithm. The detection method is as follows:

[0066] When the deep learning algorithm model detection module detects that the enlarged pantograph image Fdtx has a structural anomaly, it transmits pantograph structural anomaly information to the system.

[0067] When the deep learning algorithm model detection module detects that the enlarged pantograph image Fdtx has a suspended foreign matter, it transmits pantograph suspended foreign matter anomaly information to the system.

[0068] The data recording module stores the pantograph anomaly information detected by the deep learning algorithm model detection module through the internal storage medium. The output module can read the pantograph anomaly information stored in the data recording module and transmit the anomaly information to the visualization interface for display. By using a large number of pantograph pictures as training data sets for deep learning, the trained deep learning algorithm model is used to detect the abnormal position of the pantograph, improving the detection accuracy of the abnormal position of the pantograph. At the same time, it is not necessary to climb and search for the abnormal position of the pantograph manually, reducing the safety hazards.

[0069] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A deep learning-based pantograph anomaly detection system, characterized in that: It includes a data acquisition module, a data processing module, a deep learning algorithm model detection module, a data recording module, and an output module; The data acquisition module is connected to the data processing module via a network; The data processing module is connected to the deep learning algorithm model detection module via a network; The deep learning algorithm model detection module is connected to the data recording module via a network; The data recording module is connected to the output module via a network; The data acquisition module is used to acquire train operation signal Lcxh, train number Lcch, train number Lccc, and pantograph image Sdgx captured by a camera, and transmit the acquired data to the data processing module via the network. The data processing module receives the train operation signal Lcxh, train number Lcch, train number Lccc, and pantograph image Sdgx captured by the camera from the data acquisition module, and performs cleaning, denoising, and magnification operations on the pantograph image to obtain the cleaned pantograph image Qxtx, the denoised pantograph image Qztx, and the magnified pantograph image Fdtx. The data processing module transmits the train operation signal Lcxh, train number Lcch, train number Lccc, cleaned pantograph image Qxtx, denoised pantograph image Qztx, and enlarged pantograph image Fdtx to the deep learning algorithm model detection module. The deep learning algorithm model detection module sets performance evaluation metrics, calculates the performance evaluation metrics of the algorithm, and compares the PPYOLO algorithm, the YOLOv3 algorithm with Darknet-53 backbone network, and the YOLOv3 algorithm with MobileNet backbone network. The deep learning algorithm model detection module detects the enlarged pantograph image Fdtx based on the deep learning PPYOLO algorithm, determines abnormalities in the pantograph, and uploads the abnormal information to the data recording module. The data recording module stores the pantograph anomaly information transmitted by the deep learning algorithm model detection module in the system storage medium; The output module reads the system storage medium of the data recording module and displays the detected abnormal information of the pantograph in the system visualization interface.

2. The pantograph anomaly detection system based on deep learning according to claim 1, characterized in that: The data acquisition module captures real-time images of the pantograph using a camera and compiles these images into a pantograph image dataset, which is {Sdgx1, Sdgx2, Sdgx3, ..., Sdgx}. n } 3. The pantograph anomaly detection system based on deep learning according to claim 2, characterized in that: The data processing module cleans the pantograph image to obtain a cleaned pantograph image Qxtx. The cleaning formula is as follows: Qxtx out (x,y)={Qxtx in (x+i,y+j:-r<=i,j<=r)} In the formula, Qxtx in (x, y) represents the pixel value at coordinates (x, y) in the input pantograph image, Qxtx out (x, y) represents the pixel value at coordinates (x, y) in the output pantograph image, and r is the radius of the filter, which is set to a value of 5.

4. The pantograph anomaly detection system based on deep learning according to claim 3, characterized in that: The data processing module performs a denoising operation on the cleaned pantograph image Qxtx to obtain a denoised pantograph image Qztx. The denoising formula is as follows: In the formula, Qxtx in (x, y) represents the pixel value at coordinates (x, y) in the input pantograph image, Qztx out (x, y) represents the pixel value at coordinates (x, y) in the output pantograph image, and r is the radius of the filter, which is set to a value of 5.

5. The pantograph anomaly detection system based on deep learning according to claim 4, characterized in that: The data processing module performs a magnification operation on the denoised pantograph image Qztx to obtain a magnified pantograph image Fdtx. The magnification formula is as follows: In the formula, Fdtx out (x, y) represents the pixel value at coordinates (x, y) in the output pantograph image, Qxtx in (x, y) represents the pixel value at coordinates (x, y) in the input pantograph image, and S represents the magnification factor of the pantograph image, which is set to 10.

6. The pantograph anomaly detection system based on deep learning according to claim 5, characterized in that: The deep learning algorithm model detection module sets performance evaluation metrics and calculates the algorithm's performance evaluation metrics using the following formulas: In the formula, Recall represents recall, Precision represents accuracy, TP represents both the sample and the prediction result being positive, FN represents both the sample and the prediction result being negative, and FP represents both the sample and the prediction result being negative.

7. The pantograph anomaly detection system based on deep learning according to claim 6, characterized in that: The deep learning algorithm model detection module compares the PPYOLO algorithm, the YOLOv3 algorithm with Darknet-53 backbone network, and the YOLOv3 algorithm with MobileNet backbone network. The comparison results are as follows: The PPYOLO algorithm has an average accuracy of 99.85% and a frame rate of 39.16 frames per second. The average accuracy of the YOLOv3 algorithm with Darknet-53 as the backbone network is 98.58%, and the FPS is 33.34 frames per second. The average accuracy of the YOLOv3 algorithm with MobileNet as the backbone network is 97.93%, and the FPS is 37.28 frames per second.

8. The pantograph anomaly detection system based on deep learning according to claim 7, characterized in that: The deep learning algorithm model detection module detects the magnified pantograph image Fdtx based on the deep learning PPYOLO algorithm. The detection method is as follows: When the deep learning algorithm model detection module detects a structural anomaly in the magnified pantograph image Fdtx, it transmits the pantograph structural anomaly information to the system. When the deep learning algorithm model detection module detects a hanging foreign object in the magnified pantograph image Fdtx, it transmits abnormal information about the hanging foreign object to the system.

9. The pantograph anomaly detection system based on deep learning according to claim 8, characterized in that: The data recording module is equipped with a storage medium to store the pantograph anomaly information detected by the deep learning algorithm model detection module.

10. A deep learning-based pantograph anomaly detection system according to claim 9, characterized in that: The output module can read the pantograph abnormality information stored in the data recording module and transmit the abnormality information to the visualization interface for display.

Images